The Formulation Scientist’s Toolkit

Prof Steven Abbott

Steven Abbott TCNF Ltd & University of Leeds

With thanks for inspiration and encouragement from:

• Prof Massimo Bonino University of Florence;

• Prof Gerardo Palazzo University of Bari;
• Dr Seishi Shimizu University of York.

Timeline:

• First demo: February 2023

• Alpha version: July 2023
• Beta release: September 2023
• v1.0: 16 October 2023
• v1.01: 30 October 2023 – minor tweaks
• v1.02: 29 January 2024 – a few updates/corrections

Introduction
When you have to fix something around the house, you often find that a “simple” job turns out to be hard
because it requires a specific tool you don’t have in your toolkit. You go out and buy the right tool, finish
the job … and maybe the tool then sits at the bottom of the toolkit, unused … till the day you have a similar
job. It takes a few minutes to find it, but then the job is easily finished.

Formulation science isn’t so different. Many problems can be solved quickly if you have the right scientific
tool. But formulation science is very broad and none of us can know every relevant bit of science or have
the time to search in the literature for a tool that might help.

That’s why I’ve created this Toolkit. There are over a hundred scientific tools ready for you to use. They
are described in the Science chapters. And there’s a set of representative Products that each use relevant
Science chapters. But there’s a problem – how do you find what you need; what links everything up?
Behind the scenes there’s the FST Matrix made from Product columns and Science rows, with an X
showing intersections of interest – the cover gives you an idea. Using the concept of a “self-assembling
book” the matrix ensures that the Science is linked to the Products and the Products are linked to the
Science. Within the text, these inter-Science and inter-Product links also take you parts of the book you
want to explore. All the (1000+) links are auto-checked so you can click with confidence. The text also has
links to relevant external apps that open in your browser, so you can explore a wide formulation space with
a few clicks.

Navigation
In a PDF there aren’t the familiar browser-style navigation tools. But Alt-Left-Arrow takes you
back to where you were, and Alt-Right-Arrow takes you forward. However, unlike a browser, the
PDF lets you open the bookmark or page views to help you navigate. And there is always Ctrl-F to

2 FST
 The Formulation Scientist’s Toolkit

search for what you are looking for.

Each time you go to a new Science chapter you find a set of hyperlinks to those Product chapters that use
the same science. These might suggest alternative ways of thinking about your own issues. And each
Product chapter has its own set of hyperlinks to Science chapters. This double hyperlinking is an important
feature of the FST. Science is universal, so seeing your own types of problems in different contexts can
often bring you fresh insights. Something which might be common knowledge in industry A might be
almost unknown in industry B, even though the core science is the same.

By widening your knowledge of how your core science applies to other products, you know where to look
for fresh knowledge. You also become more flexible and employable.

The FST Community

Welcome! You are part of the community. You will quickly find things you don’t like, things missing,
things you’d like changed. No problem. Let me know. Behind the scenes the self-assembling book goes
from matrix to final PDF. So missing bits of science, missing links, new Science or Product topics, new
matrix links, or just errors in the text - at the press of a few buttons a new version of the book can be
created and uploaded to the FST site.

A Toolkit this big is bound to have repetition, omissions, biases and errors, so don’t hesitate to point out
any faults you find.

Send comments, suggestions, corrections to me at steven@[Link].

My thanks to Prof J-M Aubry who suggested numerous improvements from the beta release and helpful
corrections/additions for v1.02.

AI and the future of the FST

ChatGPT, Bard and similar tools emerged while I was writing the FST. The question immediately arose: is
the FST unnecessary, are the AIs our Toolkits? So I’ve put plenty of time into learning how to ask suitable
questions via GPT4 + plugins (the most powerful combination of tools at the time of writing). The FST has
been improved through this process because GPT4 is more methodological (and sometimes tedious) in
answering interesting questions, so has spotted some details that I’d not thought of and was happy to add.
I’ve also asked it some tricky detailed questions about issues that cross conventional boundaries, and got
impressive replies. But I could only ask those questions because I knew the answers, and knew why they
were important to formulators. I’ve not found a way for it to generate the sort of content that I think
formulators need beyond their routine training.

Will the FST become irrelevant as the AIs get better? Of course! Till then, I’ll carry on doing my best to
improve the depth and breadth of the content. After all, the AIs will be reading the FST so in a small way
I’m helping them get smarter.

Abbreviations
The list is not exhaustive.

• AI Artificial Intelligence

FST 3
 • APG Alkyl Polyglycoside
• API Active Pharmaceutical Ingredient
• BP Boiling Point
• CAPB Cocamidopropyl betaine
• Cc The “characteristic” value of a surfactant in HLD theory
• DoE Design of Experiments
• EACN The Equivalent Alkane Carbon number in HLD theory
• HLD Hydrophilic Lipophilic Difference
• HSP Hansen Solubility Parameters
• Mc Critical Entanglement Molecular Weight
• ML Machine Learning
• MVol Molar Volume
• MW Molecular Weight
• PCA Principal Component Analysis
• PE/PP Polyethylene/Polypropylene
• PEG/PPG Same as PEO/PPO
• PEO/PPO Polyethyleneoxide/Polypropyleneoxide
• PSA Pressure Sensitive Adhesive
• PVOH Polyvinylalcohol
• SLS Sodium Lauryl Sulfate; SLES Sodium Laureth Sulfate (extra EOs)
• Tg Glass transition temperature

4 FST
 Contents

Contents
- The Products -................................................................................................................................................ 8
Adhesive .......................................................................................................................................................... 9
Deodorant Sticks............................................................................................................................................ 18
Dishwashing Liquids ..................................................................................................................................... 21
Emulsion Paint ............................................................................................................................................... 25
Emulsion Polymers ........................................................................................................................................ 32
Fragrances ...................................................................................................................................................... 40
Hair Conditioners........................................................................................................................................... 45
Hair Shampoos............................................................................................................................................... 48
Inkjet Inks ...................................................................................................................................................... 54
Laundry Liquids............................................................................................................................................. 61
Lipstick .......................................................................................................................................................... 67
Mascara .......................................................................................................................................................... 73
Microencapsulation........................................................................................................................................ 80
PSA ................................................................................................................................................................ 99
Pharma Formulation....................................................................................................................................... 91
SkinCare....................................................................................................................................................... 110
Soaps and Washing ...................................................................................................................................... 121
Solvent-Based Paint ..................................................................................................................................... 127
Sun Screens.................................................................................................................................................. 135
Surface Cleaning.......................................................................................................................................... 138
Toothpastes .................................................................................................................................................. 141
Water-based Adhesive.................................................................................................................................. 144
- The Science - ............................................................................................................................................. 149
Absorbency_Capillarity ............................................................................................................................... 150
Absorbency_Porosity and Darcy ................................................................................................................. 151
Absorbency_Swelling .................................................................................................................................. 153
Adhesion_Adhesion promoters.................................................................................................................... 155
Adhesion_Butt ............................................................................................................................................. 158
Adhesion_Crack resistance.......................................................................................................................... 160
Adhesion_Crosslinking................................................................................................................................ 162
Adhesion_Entanglement .............................................................................................................................. 166
Adhesion_Intermingling .............................................................................................................................. 168
Adhesion_JKR ............................................................................................................................................. 170
Adhesion_Peel ............................................................................................................................................. 173
Adhesion_Shear ........................................................................................................................................... 176
Adhesion_Surface energy Interactions ........................................................................................................ 179
Adhesion_Testing ........................................................................................................................................ 183
Cleaning_Bleaching and Oxidation ............................................................................................................. 187
Cleaning_Boundary removal ....................................................................................................................... 190
Cleaning_Contact Angles ............................................................................................................................ 192
Cleaning_Enzymes ...................................................................................................................................... 195
Cleaning_Solvents ....................................................................................................................................... 196
Cleaning_Surfactants ................................................................................................................................... 199
Cleaning_Temperature effects ..................................................................................................................... 202
Coating_Dewetting Theory.......................................................................................................................... 203

FST 5
Coating_Levelling Theory ........................................................................................................................... 206
Coating_Pinholing Theory........................................................................................................................... 209
Diffusion_Barrier Properties........................................................................................................................ 211
Diffusion_Basic Diffusion ........................................................................................................................... 216
Diffusion_Concentration-Dependent Diffusion........................................................................................... 220
Diffusion_Diffusion Coefficients................................................................................................................. 223
Diffusion_Diffusion into Skin and Hair....................................................................................................... 226
Diffusion_Permeation OTR and WVTR ..................................................................................................... 228
Dispersions_DLVO...................................................................................................................................... 231
Dispersions_ODC ........................................................................................................................................ 235
Dispersions_PVC and CPVC....................................................................................................................... 237
Dispersions_Rheology (High shear) ............................................................................................................ 239
Dispersions_Rheology (Low shear)............................................................................................................. 243
Dispersions_Scheutjens-Fleer...................................................................................................................... 246
Dispersions_Settling .................................................................................................................................... 250
Dispersions_Solubility................................................................................................................................. 253
Dispersions_Zeta potential........................................................................................................................... 256
Dissolution_Crystallization.......................................................................................................................... 258
Dissolution_Dissolution Kinetics ................................................................................................................ 264
Dissolution_Dissolution Rheology .............................................................................................................. 268
Dissolution_Flory-Huggins ......................................................................................................................... 272
Dissolution_Gases........................................................................................................................................ 276
Dissolution_Hansen Solubility Parameters.................................................................................................. 278
Dissolution_Ideal solubility ......................................................................................................................... 285
Dissolution_Kirkwood-Buff ........................................................................................................................ 288
Dissolution_Solubilizers and Hydrotropes .................................................................................................. 290
Evaporation_Basics...................................................................................................................................... 295
Evaporation_Diffusion limited .................................................................................................................... 297
Evaporation_Enthalpy of Vaporization........................................................................................................ 300
Evaporation_Heat and Mass Transfer.......................................................................................................... 302
Evaporation_Humectants............................................................................................................................. 305
Evaporation_Marangoni .............................................................................................................................. 307
Evaporation_Spray Drying .......................................................................................................................... 310
Evaporation_Temperature and Antoine Coefficients................................................................................... 315
Flow_Atomization ....................................................................................................................................... 317
Flow_Basic viscosity ................................................................................................................................... 319
Flow_Couette Flow...................................................................................................................................... 321
Flow_Creep and Relaxation......................................................................................................................... 322
Flow_G' and G''............................................................................................................................................ 325
Flow_Ohnesorge .......................................................................................................................................... 329
Flow_Oscillatory rheology .......................................................................................................................... 331
Flow_Particle Viscosity ............................................................................................................................... 335
Flow_Poiseuille Flow .................................................................................................................................. 337
Flow_Polymer Viscosity.............................................................................................................................. 339
Flow_Rotational rheology ........................................................................................................................... 343
Flow_Shear dependent................................................................................................................................. 345
Flow_Stefans Squeeze ................................................................................................................................. 348
Flow_TTS-WLF .......................................................................................................................................... 353
Flow_Thixotropy ......................................................................................................................................... 350

6 FST
 Contents

Flow_Yield Stress ........................................................................................................................................ 356

Fragrance_Activity Coefficients .................................................................................................................. 358
Fragrance_Barrier Properties ....................................................................................................................... 360
Fragrance_Vapour Pressure ......................................................................................................................... 363
Gelling_Networks and Percolation .............................................................................................................. 366
Gelling_Syneresis ........................................................................................................................................ 371
Heat_Conductive.......................................................................................................................................... 375
Heat_Convective.......................................................................................................................................... 378
Heat_Radiative............................................................................................................................................. 380
Humidity_Water mechanical isotherm......................................................................................................... 382
Humidity_Water vapor isotherm.................................................................................................................. 384
Mechanical_Friction .................................................................................................................................... 389
Mechanical_Hardness .................................................................................................................................. 392
Mechanical_Modulus................................................................................................................................... 396
Mechanical_Surface Roughness .................................................................................................................. 398
Optics_Colour .............................................................................................................................................. 400
Optics_Gloss ................................................................................................................................................ 405
Optics_Scattering and Opacity .................................................................................................................... 407
Particles_Basic behaviour............................................................................................................................ 409
Particles_Porosity ........................................................................................................................................ 413
Particles_Sintering ....................................................................................................................................... 415
Particles_Size distribution ........................................................................................................................... 417
Separation_Filtration.................................................................................................................................... 419
Separation_Settling and Centrifugation....................................................................................................... 421
Surfactancy_Anti-foaming........................................................................................................................... 423
Surfactancy_CMC and Langmuir ................................................................................................................ 428
Surfactancy_CPP and phases....................................................................................................................... 438
Surfactancy_Coacervation ........................................................................................................................... 431
Surfactancy_Critical Capillary Number ...................................................................................................... 440
Surfactancy_Dynamic Surface Tension ....................................................................................................... 442
Surfactancy_Emulsification......................................................................................................................... 445
Surfactancy_Emulsion Coalescence ............................................................................................................ 449
Surfactancy_Emulsion Creaming and Flocculation..................................................................................... 452
Surfactancy_Emulsion Inversion ................................................................................................................. 454
Surfactancy_Emulsion Ostwald................................................................................................................... 457
Surfactancy_Foam Drainage........................................................................................................................ 460
Surfactancy_Foam Ostwald ......................................................................................................................... 464
Surfactancy_Foam Rheology....................................................................................................................... 468
Surfactancy_Foaming .................................................................................................................................. 470
Surfactancy_HLD ........................................................................................................................................ 472
Surfactancy_Interfacial Tension and Rigidity ............................................................................................. 478
Surfactancy_Phase Diagrams....................................................................................................................... 481
Surfactancy_Roll-up and Eötvös Number ................................................................................................... 485
Thickeners_Associative Thickeners ............................................................................................................ 487
Thickeners_Polymeric Thickeners............................................................................................................... 491
Thickeners_Wormlike Micelles ................................................................................................................... 495
- Extras -....................................................................................................................................................... 498
Analytical Techniques.................................................................................................................................. 499
Open Analyzer ............................................................................................................................................. 506

FST 7
- The Products -
You can explore the FST via the product categories. Your specific products might not feature but the idea is
to take inspiration from those product classes that are close to yours.

Experience shows that exploring outside your usual product domain is a great way to find fresh inspiration.
In too many formulation domains, everyone else is working with the same, tired ideas. If that applies to
you, splash out and learn the ideas that are common in domains that might interest you.

It also helps with career advancement and employability to have knowledge of a broader set of products
and the core science elements that connect them.
 Adhesive

Adhesive
Links
Adhesion_Adhesion promoters, Adhesion_Butt, Adhesion_Crack resistance, Adhesion_Crosslinking,
Adhesion_Entanglement, Adhesion_Intermingling, Adhesion_JKR, Adhesion_Peel, Adhesion_Shear,
Adhesion_Surface energy Interactions, Adhesion_Testing, Diffusion_Basic Diffusion, Dissolution_Flory-
Huggins, Dissolution_Hansen Solubility Parameters, Evaporation_Basics, Evaporation_Diffusion limited,
Flow_G' and G'', Flow_Oscillatory rheology, Flow_Polymer Viscosity, Flow_Rotational rheology,
Flow_Shear dependent, Flow_Stefans Squeeze, Flow_TTS-WLF, Flow_Thixotropy, Flow_Yield Stress,
Mechanical_Surface Roughness

Making a good adhesive is hard. Look at the list of things you need to know to be able to formulate
well. But it is even harder if the formulation team aren’t familiar with the core science behind a
good adhesive. Those working on Water-based Adhesive and PSA have specific challenges
discussed in those separate chapters.

Adhesion is a Property of the System

This phrase is key to understanding adhesion. The best adhesive in the wrong system is useless. A mediocre
adhesive in the right system will be fine for the application.

Here are two examples where “the system” has a big influence. The first compares the adhesion (via pure
surface energy) of two pieces of super-smooth rubber, tested in peel, shear and butt modes:

The app does the calculations for you:

[Link]

In the case of peel, the force needed to break the bond is 1 mN. Pulling the same system apart under shear
requires 1 N – 1000x times more even though the same surface energy is involved. Pulling apart as a butt
joint requires 32 N, 32,000 times more.

More interesting is that although (of course) the peel force depends only on the surface energy, the shear
and butt joints depend on the modulus and thickness of the polymer, neither of which is an “adhesive”
property.

Another example is discussed more fully in

Adhesion_Shear. The “lap shear joint” test has two
adherends stuck together via an adhesive with a length
L of overlap.

As you can see, a force F is applied to each end to induce shear in the joint. It is obvious that the joint will
(eventually) fail in shear (it’s called a shear test!) and that the more adhesive, the longer L, the stronger the
bond (it’s a lap shear test!).

Both intuitions are wrong. Adhesion is a property of the system, and in this system the joints generally fail
in peel and after a minimum L, the overlap makes no difference to the strength of the joint.

Which brings us to Adhesion_Testing in general. Many standard adhesion tests give numbers that are
irrelevant for the specific application. So when you test for adhesion you need to think of your system and
find tests that are relevant to that system. If you put a carbon composite into a standard lap shear test it fails
catastrophically. Yet carbon composites are wonderfully strong in shear. They fail because the “shear” test
is, in fact, a peel test, which carbon composites are known to fail easily. For the true shear behaviour of a
carbon composite you need to use the “double overlap test” which is less convenient and far more
informative.

A final example of system thinking is connected to the notes on adhesive packages

telling you to apply only a thin coating. Why would those selling adhesives want us
to use less of it?

When you pull up on this butt joint (this time with adhesive, the one above was pure
surface energy), the force, F, needed to break the joint of area A, for an adhesive
3E
with modulus E and Poisson ratio ν depends on the bulk modulus K = 1 − 2ν , the (dissipative) work of
adhesion W and adhesive thickness d, given by:

F=A √ KWd
10 FST
 Adhesive

The thinner the adhesive, the stronger the bond.

Why did I say “(dissipative) work of adhesion”? Think about it. Most adhesives have similar surface
energies, so if W was a classic surface energy work of adhesion, we could use just about any adhesive. But
a good adhesive, as we shall see, is able to dissipate crack energy, so it really matters that you have an
adhesive suited to the types of crack forces that joint will receive during use.

Handling the adhesive

Whether the adhesive is a solution or a self-soluble mix of two components, the adhesive has to be handled
from production to application and the drying/curing process itself. At the very least you need three sets of
information:

• Flow_Rotational rheology,
• Flow_Shear dependent.
• Flow_Polymer Viscosity,

The basic rheology gives core information about the flow properties. The shear dependent data give the
behaviour during different parts of the process. Pouring from a pot might have a shear rate of 1/s but
squeezing a thin film between adherends might be 1000/s. And the sorts of polymers that are likely to give
strong adhesion because of entanglement (see below) are automatically going to have a rapid increase in
viscosity at concentrations above a critical value. Something as simple as a desirable reduction of solvent
by 10% might lead to a catastrophic increase in viscosity.

Each property changes with temperature and with drying/curing in use. Without a core set of such data, it is
easy for an adhesive to be unusable at some critical stage in its use.

Things that aren’t important for adhesion

Surface energy is 10000x too small to be of relevance. A
reasonable surface energy is 40 mJ/m² while a reasonable
pressure sensitive adhesive is 400 J/m² - and that’s not even
a “strong” adhesive. It follows that measuring surface
energies is a waste of time. The idea that “if it doesn’t wet it can’t stick” is also wrong. It is trivially shown
that all liquids wet all surfaces (so water wets Teflon) when sandwiched between two surfaces. In the
diagram the liquid has a 90° contact angle so “doesn’t wet”, yet it nicely wets between two surfaces of the
same material.

People ask why I get so angry about surface energies. It’s because 50 years of adhesion science
have been wasted by people obsessing about measuring them, rather than thinking about what’s
really going on. The day I realised, after many hours of surface energy measurements, that not
once in my professional career had they ever been of use to me, was the day I started my search for
ideas that really do help solve adhesion problems

The one good reason for (usually) quick measures of surface energy is a test for obvious surface
contamination. Everything discussed in this chapter assumes that you have A and B when you are sticking
A to B. Surface contamination would give not A-B but A-X-B. So if some drops of water+ethanol, or some

FST 11
dynes pens can tell you day after day that your surface is the same as before, that’s fine. And if today it’s
lower then that’s a signal that something has changed, which is usually a bad thing. But if the surface
energy is higher you don’t celebrate – you realise that something has changed, which is usually a bad thing.

If corona/plasma is your preferred way to remove contamination from the surface, and if surface energy is
good at showing the removal, that’s great. Adhesion is a property of the system and any system with junk
that hasn’t been removed is unlikely to give good adhesion. Focus on junk removal, not surface energy.

Surface roughness does not, and cannot, help adhesion via “extra surface area” and “mechanical
interlocking”. Although a surface profile might look like this:

[Link]

when you explore this diagram in Mechanical_Surface Roughness you find that the extra surface area
(compared to a smooth surface) is 0 and that the chances of mechanical interlocking are similarly 0. Both
ideas are an illusion based on mixed axis scales for x and y.

Old ideas of electrostatics have proven to be unhelpful.

Chemical bonds as a sole strategy are a failure. If you have complete chemical bonding you gain 1 J/m²,
tiny compared to our 400 J/m² earlier, see [Link]
A low level of chemical bonds is a potent source of strong adhesion, for reasons explained shortly. Yes, too
few chemical bonds might give poor adhesion, but too many (> ~ 1%) make things worse.

Dissipation is key
Adhesion fails when a crack propagates along the interface. Although you can easily start a crack, if it
doesn’t propagate, the adhesion is fine. If the energy in the crack is dissipated into the system then the crack
won’t propagate:

12 FST
 Adhesive

A pure surface energy bond has no way to dissipate crack energy. As found from hysteresis in
Adhesion_JKR JKR experiments, intermingling of polymer across a boundary, Adhesion_Intermingling,
can help dissipate energy via friction. A lot of strong adhesion comes from Adhesion_Entanglement where
polymer chains are so entangled that if you try to pull them apart they cause motion over many nm which
in turn dissipates energy. In these cases, the generic idea of viscous dissipation is helpful. This is
particularly relevant to the crack resistance of PSA systems.

To get a full picture of your cured adhesive, you need to do some basic Flow_Oscillatory rheology, to
measure Flow_G' and G''. Dissipation is via G" (the loss modulus), but for a strong adhesive you need a
large G', the elastic modulus. Getting the balance right is hard if you use rheology and impossible if you
don’t.

Thinking seriously about how your adhesive system might or might not dissipate crack energy is a
productive process. Because of Flow_TTS-WLF a system that works well at, say, room temperature and
“normal” test speeds might fail easily at low temperatures or high test speeds because dissipation can no
longer take place.

Removing your own cracks

Another way to increase the strength of your adhesive system is to
remove anything that provides an excuse for a crack to form and
propagate. As explained in Adhesion_Crack resistance, the stress needed
1
to open an existing crack of radius r is proportional to √r . If your joint
contains super-small defects then maybe you will be OK, but a larger defect from, say, an air bubble can be
devastating. Putting thought into removing air bubbles (or lumps of general dirt) can often give you more
adhesion than trying to improve the formulation of the adhesive itself.

Reaching across the interface

In order to dissipate energy across an interface there must be ways for molecules of adhesive and adherend
to cross the interface.

The simplest way to achieve this is a PolymerA-PolymerA interface created by heat sealing. Heat the
interface above the MPt of the polymer and keep the surfaces in contact for 0.5s and on rapid cooling
without accidentally pulling the two surface apart, you have effectively a solid A polymer. It is self-
entangled, with the bond approximately as strong as the polymer itself. You can explore the thermal effects
via the app:

FST 13
[Link]

“It is self-entangled, with the bond approximately as strong as the polymer itself.” That’s true for
heat sealing. When it comes to 3D printing with hot, extruded, polymers there is an unfortunate
trade-off. At lower temperatures, there isn’t enough heat & time to allow the polymer chains to
entangle, so adhesion is low. At higher temperatures, there’s plenty of entanglement, but the whole
structure sags and deforms under its own weight. It turns out that the temperatures used in practice
give only 1 or 2 entanglement lengths at the interface, too few to give strong adhesion. These 3D
printed structures easily fracture along the interlayer-planes – they can’t dissipate enough crack
energy.

If you can’t use heat, you can use solvents. Typically the adhesive will be in a solvent chosen for good
compatibility, so you can get a high concentration of adhesive. If the solvent cannot attack the adherend
surface then no entanglement can take place. That is why many adhesives have a solvent optimized for a
range of compatible polymers (typically excluding polyethylene and polypropylene), where there is
sufficient “bite” into the adherend polymer surface without destroying it.

Choosing the right solvent for mutual compatibility can be done rationally via Dissolution_Hansen
Solubility Parameters. Getting the right evaporation rate is described in Evaporation_Basics and
minimizing the amount of trapped solvent is discussed in Evaporation_Diffusion limited.

If you can’t get polymers across an interface (for example, when sticking to a metal) then you can resort to
chemical bonds. As with everything in life, too much of a good thing is a bad thing. It turns out that if you
have a lot of chemical bonds, adhesion is poor. Why? Because the interface is so brittle (think of glass) that

14 FST
 Adhesive

a crack easily travels along it. The chemical bonding done correctly at a low level is there to deliver
entanglement into the polymeric adhesive. No matter whether the bonding is between groups naturally on
the surface or those added with Adhesion_Adhesion promoters, you just need enough bonding to create
plenty of tangles. It’s a common observation that when you try especially hard to increase the level of
adhesion promoter, or heat the system to encourage more bonding, the adhesion gets worse.

Adhesive mechanical properties

Assuming the interface is good, what else is needed from the adhesive? That depends on the system.

For many structural adhesives, a key requirement is strength, i.e. a high Mechanical_Modulus. This will be
especially useful in guarding against Flow_Creep and Relaxation, the slow distortion of the bond with
prolonged loads. Often these adhesives will be required to keep those good mechanical properties over a
large temperature range. If the temperature range is large then stresses from mismatch of thermal expansion
coefficients of adhesive and adherend can be a problem. The more rigid the adhesive, the more critical it is
to have a good match of expansion coefficients as there is no ability of the adhesive to relax under local
thermal stresses.

So structural adhesives are the highly crosslinked systems such as epoxies or UV. Achieving sufficient
Adhesion_Crosslinking is difficult because as the crosslink density increases it becomes harder for
subsequent links to be formed. Shrinkage during crosslinking risks damaging the bond – the low shrinkage
of epoxies makes them especially desirable.

Conversely, many adhesives are better when they are relatively low modulus and flexible. For example,
wood expands and contracts drastically with temperature and humidity. So an adhesive that is able to
stretch and shrink is often preferable. Similarly, you do not want a rigid adhesive for shoes, handbags and
many autoparts where light weight and flexibility are part of the design.

Some years ago I was researching optimal wood adhesives. I kept finding papers using
sophisticated structural adhesives, while observing that no one was using them. The PVA used in
common wood glues is notable for being a relatively poor polymer, especially compared to the
wonder polymers used in all those academic papers that everyone was ignoring. That’s when I
found the true explanation – as wood expands/shrinks with humidity, low-strength PVA can flow to
accommodate the changes. The joints made from wizzo adhesives break under the stresses.

Once again, because of Flow_TTS-WLF, any single measure of mechanical properties is meaningless –
you have to know the temperature and time-scale of any likely threat to the adhesive.

Adhesive chemical properties

For non-crosslinked adhesives, resistance to solvents can be straightforwardly estimated via
Dissolution_Hansen Solubility Parameters. For crosslinked adhesives the problem isn’t one of dissolution
but rather of swelling. The Absorbency_Swelling guide introduces you to Flory-Rehner theory which
depends on the Flory-Huggins χ parameter which links directly to Hansen Solubility Parameters.

Resistance to aggressive chemicals is a more specialist topic and you probably know a lot already for your
specific system.

FST 15
The real challenge for everyone is water and humidity. Water is a small molecule available in large
quantities from a humid atmosphere so it can diffuse into the adhesive, where it might cause some
weakening of structure (e.g. by plasticisation). See Diffusion_Permeation OTR and WVTR. But more
serious is diffusion to the interface. Relatively small changes to the polymer(s) at the interface can have a
significant effect on local stresses, relaxation times and viscous dissipation, with potentially significant
reduction in adhesion. If you are relying on relatively few chemical bonds (too many are bad in other ways)
and if they are susceptible to hydrolysis then you have big problems.

So make sure you have the minimum of spare -OH and other polar groups at the interface, and if you are
using chemical bonds across the interface, try to make them multi-functional. Maleic acid and, especially,
trialkoxysilanes, are favourite adhesion promoters because their multifunctionality gives them extra
resilience against hydrolysis.

Stefan’s Squeeze
An under-appreciated difficulty with adhesives is getting the right, even, thin layer of adhesive across the
whole joint. It seems easy – add a central blob with a slight excess of adhesive, squeeze, and the job is
done.

Unfortunately, Flow_Stefans Squeeze tells us that it is hard to get things right. If you feel the urge to jump
off a bridge with a bungee cord attached to a lump of steel stuck to another lump of steel with 1 large drop
of superglue – don’t do it. You will die. But if you have the same volume of superglue supplied as 10 drops,
Stefan is on your side and you will have a satisfying dive and rebound.

Finding the right pattern of drops of adhesive for the best uniformity with zero chance of trapping air
bubbles is a scientifically unsolved puzzle. The closest I can find is a way to think about dissolving the air
bubble during the pressure of the squeeze: [Link]
[Link].

Short summary
Now you know that Adhesion is a Property of the System and have seen examples of how our intuitions
about the system can often be wrong, you have some core ideas for tackling your own adhesive issues.
We’ve given you no direct advice about your specific adhesive polymers, crosslinkers, promoters, surface
preparation. Each of these is super important, but this isn’t a guide to formulation specifics, it’s a guide to
formulation principles. This is especially important for adhesives where historically there has been too
much focus on the specifics, without stepping back and seeing the whole problem. If it turns out that you
can fix an adhesion problem by changing, say, the modulus of the adherend or removing crack-inducing air
bubbles, that saves a lot of time which might have been wasted trying to improve an already good
formulation.

I once helped fix an adhesion problem without changing the adhesive or the adherends. Instead, a
polymer layer further away from the site of adhesion failure was to blame. When that remote layer
was LDPE, the standard “tear test” deformed the whole structure and stresses got concentrated at
the adhesive interface. When it was HDPE, the same test passed with no problem – stresses went
where they were meant to, and the adhesive was fine. I have to say it again – Adhesion is a
property of the system.

16 FST
 Adhesive

FST 17
Deodorant Sticks
Links
Dissolution_Hansen Solubility Parameters, Dissolution_Solubilizers and Hydrotropes, Flow_Oscillatory
rheology, Flow_Shear dependent, Flow_Yield Stress, Fragrance_Activity Coefficients, Fragrance_Vapour
Pressure, Gelling_Networks and Percolation, Gelling_Syneresis, Optics_Scattering and Opacity

We have two formulation decisions:

• The type of deodorant package to use

• The matrix for the deodorant package

Because antiperspirants (stopping sweat from reaching the surface of the skin) raise too many
regulatory complexities, they are not included in the discussions.

The Deodorant package

The simplest is a general-purpose pleasant fragrance, along with any marketing claims about the fragrance
being extracted from or plausibly related to some suitably exotic and natural source.

More sophisticated is the recognition that many fine perfumes include malodorous molecules such as
scatole. Somehow a hint of malodour can be converted into a desirable perfume with the right combination.
The distinctive underarm odours are specific hexenoic, octanoic and hydroxyhexanoic acids plus
sulfanylalkanols. Amazingly a specific combination of terpene-derived esters works with the malodours to
create a pleasant fragrance.

An attractive semi-natural ingredient is zinc ricinolate, based on natural ricinoleic acid. Although other zinc
ingredients work adequately, the ricinolate seems especially good at intercepting the carboxylic acids and
the sulfurous molecules and removing the odours. The reasons are unclear, but the efficacy is impressive.

The sweat itself is odourless. The malodorous molecules are produced by enzymes breaking down some
natural (odourless) glutamides into glutamine and the offending carboxylic acids. If your fragrance package
includes glutamides that are broken down to odourless molecules, you can overwhelm the enzymes so they
don’t touch the odour-producing one. Or you can add gluatamide mimics that bind to and block the
enzymes.

So far there is no controversy. If your matrix is water based then you must add some preservatives.
Although these are designed to inhibit microbes the rules of the game say that these are preservatives and
not antimicrobial. But if you go one logical step to add antimicrobials to the deodorant to zap the skin
microbes that produce the enzymes that cause the odours then you are in difficult territory. You are
potentially offending two very different groups:

1. The regulatory authorities that require proof of efficacy, safety and absence of build-up of
resistance to the antimicrobials
2. The users who are increasingly aware that their skin biome is there not just to create malodours –
 Deodorant Sticks

it should be a healthy part of their overall defence system against threats to the outside of the skin.

Using potassium alum, “natural extracts” and maybe chelators such as a salt of EDTA or GLDA may
merely upset some microbes (e.g. the chelators starve them of iron) so they aren’t really antimicrobials.
And maybe they don’t upset the skin biome too much. And maybe the good microbes have already been
eliminated by excess washing with powerful soaps, so zapping the bad microbes is reasonable. Sorting out
these issues is an interesting balance of science and marketing claims.

It is scientifically crazy that we damage our skin biomes with too much hot, soapy washing then
try to fix the odours which might well be caused by excess “bad microbes”, using an array of
chemicals which might themselves further disrupt the biome. This leads to a formulation
bifurcation. One way is to add pre-/pro-biotics to the soaps and deodorants, the other way is to
formulate minimalist soap/deodorant packages.

The matrix
We have to choose between water-based and oil-based

Water-based

With water as a low-cost evaporative component along with propylene glycol or dipropylene glycol as the
(effectively) non-volatile carriers of the relatively small % of chemicals in the deodorant package, our key
need is to gel the system. With “sodium stearate” (actually a mix of different alkane chain lengths, giving a
range of properties) we have, essentially, cheap soap holding everything together. Minor ingredients such as
chelators, preservatives and antioxidants can easily be incorporated. The problem is that the fragrances are
generally hydrophobic. So how do we compatibilise the fragrance with the hydrophilic portion?

First, why do we want to make them compatible?

• Because incompatibility leads to Mie scattering (Optics_Scattering and Opacity) and consumers
like clear gels which suggest no possibility of staining of clothes
• Because incompatible fragrances have a high activity coefficient (Fragrance_Activity
Coefficients) and will evaporate too quickly, losing claims of long-lasting action
• Because incompatibilities can lead to syneresis of the gel – with fragrances or water being
eliminated from the gel during storage, producing an undesirable surface liquid and a poor
consumer experience. Although the Gelling_Syneresis chapter concentrates on particle gels, the
general background to syneresis is still valid.

It would be nice to say (given how they feature in many parts of the FST) that the problem of compatibility
could be fixed by suitably matching the Dissolution_Hansen Solubility Parameters of the fragrances and
the gel. Indeed, some of the relevant product literature invokes HSP. But that can’t be true. HSP and other
solubility approaches are “mean field” theories, meaning an absence of long-range order. We can be certain
that gelling agents such as sodium stearate are acting with plenty of long-range order. As mentioned above,
they are merely soap and although there isn’t a clear water/oil distinction relevant to normal soap use, there
has to be some aspect of head/tail ordering to produce the gelation. Although it is possible to claim that the
popular long chain dimethylamide additives have an HSP that matches some fragrances, it seems more
plausible that the long chains have interesting interactions with the various long chains in the “sodium

FST 19
stearate” creating, for example, local pockets of lower crystallinity that allow the fragrance molecules to
escape the hostile aqueous environment.

Instead we have to invoke the ideas of “solubilizers” Dissolution_Solubilizers and Hydrotropes.

Unfortunately, these have been subject to decades of fruitless hand waving rather than solid science.
Although the newer Dissolution_Kirkwood-Buff statistical thermodynamic approaches can work well to
describe solubilization, we have no predictive framework to invoke for solving problems like these.

How do we proceed under these circumstances? By being intelligent and inquisitive. We don’t know how
these solubilizers might work, but we can be certain that they will impact the gelation behaviour – this is
certain because they are interfering with long-range order. So we need to be ready to look for patterns of
desirable increases of solubility of the fragrances (greater optical clarity) while attending to changes in
gelation characteristics. Which, as so often in the FST takes us to rheology, discussed below.

Oil based

Such formulations eliminate the need for preservatives and avoid fragrance incompatibility issues. We use
any of a myriad of pleasant, light plant oils, some waxes for strength and cetearyl alcohol as a thickening
and “glide” agent. Zinc ricinolate fits in easily along with antioxidants such as tocopherol that consumers
love to see. If you don’t like to add fragrances (which imply some artificiality) you can add essential oils
instead. They are fragrances, but you don’t have to call them that.

The first downside of these formulations is that the stick is opaque, which points towards the second
downside that all this oily gunk can create a white stain your clothes, and that the stain is relatively hard to
remove compared to the water-soluble aqueous formulation ingredients.

Feeling good
The different chain lengths in the different “sodium stearates” give different levels of crystallinity and
different responses to temperature during storage and during application to the warm skin surface. The
same applies to the different waxes (beeswax, carnauba, shea butter etc.) and oils for the non-aqueous
products. The formulations must have a Flow_Yield Stress behaviour nicely tuned to feeling good during
application – not too stiff, not too runny, not applying too much or too little.

Clearly this needs the help of considerable amounts of smart rheology. Equally clearly, there is no way to
go directly from rheology to a successful formulation. You need to build up a smart map of how the
different formulation ingredients and rheological measurements map on to the product performance. This
needs measurements of good and bad products, internal and competitor materials.

How to do this is described in the Lipstick chapter. Although lipstick and deodorant sticks are different
products, science is universal and the problems of cost, marketing claims, ingredient fads, rheology
measurements, skin feel, mapping across complexity are so similar that it’s a good idea, not mere authorial
laziness, to send the reader to lipstick.

20 FST
 Dishwashing Liquids

Dishwashing Liquids
Links
Cleaning_Boundary removal, Cleaning_Contact Angles, Cleaning_Surfactants, Cleaning_Temperature
effects, Flow_Basic viscosity, Surfactancy_Anti-foaming, Surfactancy_CMC and Langmuir,
Surfactancy_CPP and phases, Surfactancy_Dynamic Surface Tension, Surfactancy_Foam Drainage,
Surfactancy_Foam Ostwald, Surfactancy_Foam Rheology, Surfactancy_Foaming, Surfactancy_HLD,
Surfactancy_Roll-up and Eötvös Number

The official test for a dishwashing liquid is not how well it washes dishes, but how many greasy
plates can be washed before the foam created at the start of the test finally disappears. Plenty of
formulations superior at washing dishes are therefore rejected in favour of foamy formulations.

So although this chapter will explain how to get the best formulation for washing, to be relevant to
the real world it has to focus on foaming.

Foaming
The science of Surfactancy_Foaming is remarkably simple. Take any reasonable surfactant solution, mix
air with it, then put in enough energy to break down large bubbles to small ones … and you get an
impressive, long-lasting foam. More than good enough to pass a standard dish washing test. The physics of
Surfactancy_Foam Rheology, Surfactancy_Foam Drainage and Surfactancy_Foam Ostwald converge on
the fact that small bubbles have a high yield strength (a “strong” foam), slow drainage and a greater
resistance to Ostwald ripening – bigger bubbles getting bigger at the expense of smaller one.

So why are so many dishwashing liquids based around SLES and CAPB? Because they happen to give
relatively large volumes of relatively stable foam when tested under standard dish washing conditions.
Also, because they are cheap. But they are cheap because they are used in vast quantities – the argument is
a bit circular.

There are no satisfactory explanations for why a few surfactant combinations give foaming characteristics
that are ideal for the standard dish washing market. There are plenty of ad hoc explanations, but a good
explanation would let formulators predict newer, more interesting formulations … and we don’t have this.

Are the standard formulations good because they have good Surfactancy_Anti-foaming resistance? No. The
science of anti-foaming remains relatively crude and there is no reason to assume that SLES/CAPB is
particularly good, especially when it’s not clear how much the fats in the washing water present a classic
anti-foam challenge. Anyone who has tried to add anti-foamers to reduce foaming will find (as explained in
the anti-foam chapter) that a good anti-foam is hard to find.

The grease eventually wins because surfactant molecules end up wrapped around the fat globules. It is an
interesting fact that SLES is not especially good at wrapping up fat globules (as we shall see, it is far too
hydrophilic), so maybe the long-lasting nature of the foam is due to the fact that the surfactant isn’t doing a
great job at cleaning.
 It is very easy to be negative about marketing-driven formulations for consumer products. So it
was great to have the chance to talk to an experienced European formulator making products for
those large cut-price supermarkets who demand top performance at low cost. So everything
unnecessary is eliminated – what matters is cost and performance. The feedback was clear. They
could make a superior dishwashing product no problem, but because “performance” is defined by
the foam test, they have no choice but to make technically inferior “foamy” products.

The Sinner circle

According to Dr Herbert Sinner of Henkel, you can clean with any combination of
Temperature, Time, Mechanical Action and Chemicals. The mechanical action is
necessary for Cleaning_Boundary removal which, thanks to the no-slip boundary
condition is impossible without such action. Many dishes will clean themselves
adequately if left to soak for a time. And temperature is good for dissolving
soluble residue on the dishes as well as helping to melt off fats and grease. So what
are the distinctive chemical aspects to the cleaning (rather than foaming) roles of
the surfactants?

Reducing surface tension?

There are those who like measuring Cleaning_Contact Angles and are concerned about Surfactancy_CMC
and Langmuir in order to get the lowest contact angle for the least amount of surfactant. They tell us that
wetting and cleaning are strongly related. But they aren’t. A dish in the washing bowl is fully wetted. When
scrubbing with the brush, the power of the washing action overwhelms relatively weak surface tension
forces.

The time we need a small-ish contact angle is during the final rinse when we would like the water to drain
off evenly. It’s not often stated, but this happens best if the hydrophobic tail of the surfactant is
contaminating the dish (OK, it’s only a chemical monolayer), leaving the hydrophilic head to let the rinsing
water (generally fresh water, low in surfactants) drain as a nice sheet.

As has often been pointed out, if CMC was super important then we would all be using non-ionic
surfactants which have CMCs far below those of the anionics that dominate the dish washing world.

What about Surfactancy_Dynamic Surface Tension. For dynamic processes, the static surface tension
measured during CMC tests is irrelevant – fresh interface is created on the millisecond timescale and it can
often take seconds before an equilibrium surface tension is reached. Interestingly, the secret to a good
dynamic surface tension is a high CMC; a “bad” surfactant might actually be better. But as surface tension
effects are of minimal importance for dish washing, we don’t have to worry too much about dynamic
surface tension effects.

Real cleaning
In a dish washer, or a washing machine, where foam is actively not wanted, the formulations can focus on
real cleaning. Dish washers use plenty of caustic to attack fat and grease. Washing machines use an array of
enzymes to chop up the lipids, proteins and starches on the clothes. We can’t use caustic or enzymes in dish
washing liquid. What could we use?

22 FST
 Dishwashing Liquids

To get fats to “roll up” from a surface is straightforward if you formulate for a Eötvös number:
Surfactancy_Roll-up and Eötvös Number. Translated, this means a low Surfactancy_Interfacial Tension and
Rigidity which is trivially obtained by formulating so that the Hydrophilic Lipophilic Difference,
Surfactancy_HLD is zero. When you explore HLD (which balances oiliness, surfactancy and temperature)
you quickly find that a surfactant such as SLES is hopeless – it is far too hydrophilic. At best it can wrap
around whatever fat blobs you can create, so does help to clean. But compared to the sorts of surfactants
chosen for washing machines (where roll-up and therefore HLD balance is critical) dish washing
formulations are sub-optimal.

You can buy real dish washing formulations that are effective rather than foamy, but they are not popular
and cannot claim to be good, because the standard test is the foamability one.

Making the product

At first it sounds trivial. Add the different surfactants to water, mix a bit to create a concentrate and the
product is ready to be delivered into the final containers either via dilution during the final step or by
selling a concentrate to reduce the environmental footprint.

However, concentrated surfactants can form difficult phases, on their own or in mixtures. If these phases
are hexagonal or cubic then they are semi-solid and difficult to disperse/dissolve. The problem can arise at
any stage in dilution from concentrate to the final formula.

[Link]

In this diagram, the concentrate (green dot at the right) is being diluted
to the final form (green dot to the left) and passes through a cubic phase
(cyan circle).

It’s important for this, and many other reasons, to be familiar with
Surfactancy_Phase Diagrams.

I’d known that hexagonal or cubic phases were “viscous” but had imagined that claims for how
difficult they were to handle, or the dangers of them appearing during dilution we being
exaggerated … till on a lab visit I got a chance to compare an impressively solid hexagonal phase
to the two liquid samples sitting next to it, one higher and one lower concentration. Mixing these
two liquids produced the impossible-to-handle gel. There is a difference between “knowing”
something and knowing something.

Doing something new

The pressures on you to produce low-cost, foamy products force you back to SLES-CAPB or whatever
“sulfate-free” equivalent you can find to replace SLES. The pressures to go green require either genuinely
greener surfactant types or clever schemes to offset the carbon footprint and environmental damage of your
current type. But genuinely greener products don’t give the same foamy signals desired by the consumers
and required to pass the irrational cleaning tests.

FST 23
This forces you to adopt one of three novel strategies:

• Use honesty in marketing to explain to consumers that your formulation cleans better and greener,
despite having few bubbles. Convert a bug (no foam) into a feature (Look! No Foam!!!)
• Find some novel dispensing device to get lots of bubbles from a formulation that doesn’t foam
well in the standard test.
• Find a magic additive that transforms your formulation from low foaming to high foaming

I once had to find an additive to transform a surfactant from no-foam to high-foam. A simple,
high-throughput, small-volume test allowed us to try a wide variety of “co-foamers” – something
impossible with slow, high-volume conventional tests. I tried 3 that seemed to me to be reasonable
– with no effect. A colleague suggested a 4th. The test produced so much foam I thought I’d
accidentally used SLES so carefully repeated the test, with the same amazing amount of foam. To
our dismay we haven’t found a good explanation for the effect. But it shows that such effects are
possible – and that imaginative HT tests are a good idea.

24 FST
 Emulsion Paint

Emulsion Paint
Links
Adhesion_Adhesion promoters, Adhesion_Crosslinking, Adhesion_Entanglement,
Adhesion_Intermingling, Coating_Dewetting Theory, Coating_Levelling Theory, Coating_Pinholing
Theory, Dispersions_DLVO, Dispersions_Zeta potential, Dissolution_Hansen Solubility Parameters,
Evaporation_Basics, Flow_Polymer Viscosity, Humidity_Water mechanical isotherm, Humidity_Water
vapor isotherm, Optics_Gloss, Optics_Scattering and Opacity, Surfactancy_Emulsion Inversion,
Thickeners_Associative Thickeners, Thickeners_Polymeric Thickeners

The basics of paint have been discussed in Solvent-Based Paint. Here we focus on those special
features that arise because the paint is water-based.

From dispersed to insoluble

For simplicity we will call the dispersed blobs of paint polymer emulsion drops, even those that aren’t
strictly emulsions. By doing this we can use much of the world of emulsion science without bothering too
much about the fine details.

The problem at the heart of these emulsion systems is that the individual drops must be infinitely stable
when dilute, yet should form a continuous film of polymer once the water has (largely) disappeared. Each
property on its own is not a problem – getting both right is the challenge.

A stable dispersion
For these aqueous systems, Dispersions_DLVO theory is adequate.

[Link]

The particles can be stabilised via:

 • Charges, typically via an anionic surfactant. If the zeta potential ([Link]
practical-solubility/[Link]) is somewhere over 30 mV then the particle has a good chance of
being stable.
• Steric effects, a long hydrophilic (typically ethoxylate) chain sticking out.

Arguably, charge-charge repulsion is the stronger stabilization method because steric repulsion can be
changed into attraction (bridging or depletion flocculation) via extraneous polymers. The downside is that
those water-loving ionic groups are not desirable in the final paint because they attract water from the
environment.

The dispersion also has to be stable against settling or creaming:

[Link]

Stokes law tells us that the time for a particle (the pigment or emulsion drop) of radius r and density ρp to
fall a distance h in water of viscosity η and density ρw (and therefore density different Δρ = ρp- ρw) is given
by:

hη
t=
2.18gΔρr2

In the screen shot, a modest particle density is assumed. The volume fraction input, φ, allows the use of the
Richardson-Zaki formula where the velocity is reduced by (1-φ)5.65. A higher volume fraction also
increases the viscosity, as discussed next.

To reduce settling, in addition to the necessity, for other reasons, of a high viscosity, smaller particles are
helpful. Unfortunately with densities like 4.2 and 2.7 for TiO2 and CaCO3 respectively, pigments are
trickier to keep in suspension.

26 FST
 Emulsion Paint

Viscosity and rheology

[Link]

A complication of understanding the viscous behaviour of these systems is that we have two types of
dispersed particles – polymer emulsion and pigments. If, for simplicity, we use a single value for the
volume fraction, φ, of “particles”, then over 60%, the viscosity increases, depending on the model and on
the internal viscosity of the droplet, by about 40x. In this region the close packing limit for rigid spheres is
being reached and the precise increase to higher φ values is too complex to model. In any case, these
models are for smooth spheres with no attractive forces between them. As we try to go to a higher volume
fraction required for a practical paint, we lose any good theory and have to rely on rheology to compare/
contrast formulations.

Although in production and QC it is necessary to use simple measurements of “viscosity”, no one in

formulation should ever measure “the” viscosity of a paint. Viscosity depends on multiple factors so
formulations should be, at a minimum, be tested via flow curves.

The standard flow curve Flow_Shear dependent is a single sweep from low to high shear rate. For paints
the reverse sweep is required to get information about thixotropy Flow_Thixotropy.

Because low shear performance (e.g. avoidance of sagging if applied to a vertical surface) is important, the
Levelling & Sagging app at [Link] is
useful. This emphasises the need for careful measurements in the 0.01/s to 5/s range.

For the most information with the least effort, checking out Flow_G' and G'' is effective. The balance of
elastic (G') and lossy (G") shear properties offers insights not so much into any specific formulation, but
into trends that occur with changes of concentration or ingredients. This means that you set up a few
experiments with different concentrations or ingredients and see the G' & G" trends. From these you get a
deeper understanding of how quickly/slowly paint properties change during, say, drying.

Thickeners
Controlling behaviour in the can, on the brush/roller, in the minutes after application and, for vertical
surfaces, controlling sag, usually requires smart thickeners as part of the formulation package. There is a
choice between Thickeners_Associative Thickeners and Thickeners_Polymeric Thickeners, with the

FST 27
associative type being generally preferred because their kinetics of breaking and re-forming are normally
faster than those of classic polymers.

Film formation
It is helpful to introduce a neutral, and slightly unusual, word into the discussions: sintering. There are
many terms used to describe the joining of blobs that are packed close together, but they lack a coherent
theory to capture the key parameters that influence the phenomenon. The Rumpf equation for sintering
describes the key elements behind the joining of particles, and allows us to identify how changes in our
system can affect film formation for better or worse.

[Link]

The Rumpf equation tells you how overlap x (in the diagram you see 2x) changes over time t for particles
with radius r and “viscosity” η and surface energy γ. The equation includes any applied external force, F,
which we can assume to be 0.

2
( xr ) =
t
η (0.8 γ
r + 0.4
F
πr2 )
The use of x/r in the formula is deliberate because we’re generally interested in the fractional sintering – a
value of x on its own doesn’t convey much.

There are a few interesting aspects to this equation:

• Contrary to instincts, a low surface energy, γ, reduces the rate of sintering. This isn’t surprising
because the process is driven by surface energy. The surfactants necessary to disperse the blobs
now become unhelpful. However, the surfactant effect on surface energy is likely to be no more
than a factor of 2, so it’s not worth worrying about.
• Smaller particles fuse faster. Unfortunately they are also harder to keep separate in the original
formulation, so there are trade-offs
• We need to attend to the viscosity of our emulsion particles (here we ignore the complexities of
the pigment particles). In addition to this being a difficult thing to measure (I have found no
helpful literature on how to do this), the variation of viscosity with water or concentration of
“coalescing solvent” is an added complexity.

The apps associated with the app above are all related to food science. But physics is universal and the idea
that the Tg changes dramatically with moisture content which in turn is governed by the water-vapour
isotherm means that there is a logical chain for investigation. Your paint is designed to dry out, but what is
the water-vapour isotherm – how dry is dry? And what is the effect of small amounts of water on the Tg? If

28 FST
 Emulsion Paint

film formation is complete long before the water has evaporated then these questions are irrelevant. The
important thing is to know if/when film formation happens.

If the particle Tg is not affected by the water (an advantage for long term performance) then either the Tg
has to be low enough for the film forming temperature to be close to room temperature. Or you have to use
a coalescing solvent. The solvent must have a lower volatility than water so its concentration increases
sufficiently to lower the Tg and decrease η. It must also have a good Dissolution_Hansen Solubility
Parameters compatibility with the polymer.

Good, but not too good. The simple story is that the coalescing agent reduces the viscosity of the whole
particle. But that would require a large amount of agent, which then has to escape the paint over days,
weeks, months. What is really wanted is for the agent to soften the outer portion of the particle, without
going too deep inside. If, as seems likely, the outer portion has more hydrophilic components then its HSP
will feature a higher δH than the bulk. So the HSP of the coalescing agent should be optimised for a
Distance to that higher δH system. At the same time, the δH should not be so high that it partitions too
readily into the remaining water which itself might have an HSP closer to that of the polymer thanks to the
various soluble components that increase % concentration as the water evaporates.

Everyone wants “low VOC” paints. But they also want paints that form wonderfully tough, water-
resistant films. Any film former that remains in the paint is a potential form of weakness, so having
them volatile is optimal for paint performance. How can you have a volatile film former while
legally claiming “low VOC”? Any molecule with a BP > 250°C or VP less than 0.01 kPa at 20°C
[those values are confused in the literature so don’t quote them] is officially classed as zero VOC
so you can add as much of it as you like to your paint.

Mud cracking
If the drying process happens too quickly, before the sintering takes place,
then the paint can start to show mud crack patterns, such as the one shown
here, taken from the app, [Link]
[Link], which describes the phenomenology while giving little advice
about how to solve the cracking issue. It’s interesting that the cracks join at
right angles, though sometimes you get patterns that crack at 120°.
Interesting, but not so helpful.

For some formulation insights we need to turn to the mud-cracking app.

[Link]

Here we calculate the CCT, the Critical Crack Thickness. Above the CCT, cracking is likely – telling us

FST 29
that one way to reduce cracking is to coat thinner.

The cracking is driven by a capillary pressure, P, calculated from the particle radius R, the surface tension γ,
the contact angle θ and the close packed fraction φ.

2γ 3φcosθ
P= r 2(1 − φ)

Already we can see that cracking is reduced if we reduce γ and increase r.

The critical crack thickness depends on a on a "coordination number" M, which is related to φ and the "how
well the particles are stuck together” parameter G. One well-known formula is from Tirumkudulu, and tells
us that:

( )(
½
GMφR3 2γ ½
CCT = 0.64 2γ PR)
To increase the CCT (so we can get thicker coatings without cracking), we can either give up on otherwise
desirable small radius particles or increase G. For entirely the uninteresting systems studied by academics,
G is the “shear modulus” arising from pure particle-particle interactions. For the rest of us, G is more
related to dissipative shear resistance (the “energy release rate”) from polymers that are (the emulsions) or
surround (additives) the particles. Unless we use polymers that are especially weak, G is likely to be high if
we give the system time to sinter. If our coalescing solvents and/or surfactants also reduce γ (which ends up
1
as a √ γ dependency) that can also help.

The advice, from the web page is not super helpful, but decades of mud-cracking research hasn’t come up
with anything more profound:

• Although the correlation with R is good in simple systems, real-world systems correlate better
with SSA, Specific Surface Area. For spherical particles the conversion from SSA in m²/cm³ is
1/R when R is in μm so for your real particles you can try using 1/SSA instead of a nominal R.
However, the SSA data has a dependency of CCT on ~SSA-2.2 rather than the expected 1 from
½
(R3)
R½
• Shape, over a range of typical particles, makes relatively small difference.
• Soft particles have no CCT - they simply fuse.
• A mix of soft and hard particles can start to crack when hard particles start to exceed 50% and can
percolate across the film.
• Lowering the (dynamic) surface tension isn't a straightforward win. That will also decrease the
contact angle, which increases P.
• Experiments often show small effects of drying rates - but these are often super-slow idealised
experiments or for slow-drying paints. For coatings that dry in seconds it's highly likely that
"happy" formulations that keep moving particles for as long as possilbe, will crack less than
"unhappy" ones. Putting it another way ...
• ... the longer the available "relaxation time", the more viscous flow that can reduce stress.
• A bi- or multi-modal particle distribution where small particles fill gaps between larger particles
obviously help increase CCT.
• You can (at least in principle) measure stress build-up in a coating using the cantilever approach
described in the Stress from Bending app, [Link]

30 FST
 Emulsion Paint

[Link].
• Extra polymers can sometimes hinder (if they make G too small) but mostly help as they can link
up otherwise weakly-connected particles. There's a subtle issue here described in the Ceramic
Green Strength app, [Link]
[Link]. If the polymer congregates at the intersections of the particle it has a bigger effect on
strength than if "wasted" by coating most of the particle.

FST 31
Emulsion Polymers
Links
Dispersions_DLVO, Dispersions_Zeta potential, Flow_Polymer Viscosity, Surfactancy_Dynamic Surface
Tension, Surfactancy_Emulsification, Surfactancy_Emulsion Inversion, Surfactancy_HLD

Typical uses of emulsion polymers are described in Water-based Adhesive and Emulsion Paint.
The variety of polymerisation techniques and their complexities are beyond the scope of this
chapter. Instead we focus on the “emulsion” part of making them and provide a toolkit within a
toolkit.

What’s in this toolkit?

In addition to basics such as surfactant MW, Krafft Point and Cloud Point, you need:

• A knowledge of surfactant o/w partition coefficients

• Knowledge of dynamic surface tension effects
• A method & data to tell you how to respond to:
◦ Changes in the oil (e.g. acrylates, methacrylates)
◦ Changes in the temperature
◦ Changes in salinity (ionic surfactants change their own salinity…)
◦ Changes in surfactant type and surfactant mixtures
◦ Changes in surfactant efficiency
◦ Changes in polar oils (defined later) and other additives
◦ A knowledge of interfacial tensions in delicate balanced emulsion systems
◦ A knowledge of curvature effects
◦ Thermodynamic AND Concentration inversions

Why a toolkit in the FST? There wasn’t enough usable science for a training session to a large
corporation on emulsion polymerisation (which relies on “this is how we’ve always done it”). So
the idea of a toolkit emerged, with a reverse guarantee: Success cannot be guaranteed if you adopt
the toolkit’s science, but you can guarantee that life will be unnecessarily difficult without it.

Partition coefficients
Ideally, your surfactant should be at the o/w interface. But if it is too soluble in the water or oil, it will be
mostly wasted. Measuring the partition coefficient should be routine, but it sounds difficult. In fact it is
delightfully simple so should be routine:
 Emulsion Polymers

[Link]

Put a small amount of surfactant into a tube containing a known volume of water. Measure its
surface tension, γorig, in the example above that’s 30 mN/m. Now add the same volume of oil,
shake and allow to equilibrate. Take out (most of) the water phase and measure its surface
tension, γequil. Separately you will have measured the surface tension versus concentration for the
surfactant (Langmuir isotherm), a typical experiment used to measure the CMC and the saturated
surface tension γc. From a calculation based on the Langmuir isotherm you can work out the
starting concentration (in this example 6.6 µM) and the equilibrium concentration 2.7 μM. From
this the partition coefficient, Kp = Coil/Cwater, turns out to be 1.5.

Dynamic surface tension, DST

Your polymerisation is a dynamic process so a knowledge of how fast or slow the surfactant is to get to the
interface will be helpful. Modern bubble pressure tensiometers can measure DST over a good range of
time-scales. How do you interpret the data? Although it is common to analyse DST to get ad hoc
parameters such as those of Rosen, [Link] there is a
more insightful technique:

FST 33
[Link]

The app text explains how you extract the core surfactant value Γm which is the surface excess
concentration and via the diffusion coefficient D (which doesn’t change much over a wide range of
surfactants) you can derive the characteristic diffusion time τD. You then find that to get a low value of τD
you need a surfactant with a relatively high CMC, a fact that is little appreciated.

Hydrophilic-Lipophilic Difference, HLD

To access most of the effects that are important during an emulsion polymerisation, the Surfactancy_HLD
approach is the only one powerful enough to be of help. The HLB approach has long-since been discredited
and should not be used.

Briefly, HLD says that the balance of a surfactant, oil and water system depends on salinity S, temperature
T, a measure of oiliness EACN and a characteristic value for the surfactant Cc. That balance is the HLD
given by:

HLD = F(S) − [Link] − α(T − 25) + Cc

There are 3 sets of parameters:

• For ethoxylates, F(S) = 0.13S, α = -0.06 and k~0.16

• For ionics, F(S) = ln(S+Surf), α = 0.01 and k~0.16. The extra term in F(S) is because the ionic
surfactant creates its own contribution to salinity.
• For Spans, APGs, Polyglycerols, F(S) ~ 0, α = 0 (maybe -0.01 for polyglycerols), k~0.16

There is some debate about k values but given the many other uncertainties we don’t have to worry about it
here.

The EACN, Equivalent Alkane Carbon Number, gives us a measure of the oiliness of the oil. If your
acrylate behaves (in surfactancy terms) like octane, then its EACN = 8. Going in the more hydrophilic
direction we get EACNs of 3 or -1. That’s because it is a scale of virtual hydrocarbons, we’re not proposing
that you can make an emulsion with a molecule containing -1 carbon atoms.

34 FST
 Emulsion Polymers

The Cc should be a value provided by your surfactant supplier. It has been a battle to get them to measure
and provide these numbers but slowly users are getting their way. As described in the HLD chapter, users
can set up simple scans (next paragraph) to check whether the “same” surfactant (different batch or
different supplier) has the same Cc and, when it doesn’t (commercial surfactants are mixtures) they can
compensate using the HLD equation.

Scans are routinely used to measure EACNs and Ccs. Across a set of tubes you systematically change one
component of the HLD equation (e.g. S) with two knowns kept constant (e.g. T and EACN) and from the
tube where HLD ~ 0 (i.e. the Type III transition between Types I and II) you can calculate the unknown (in
this case the Cc).

[Link]

These scans can be used to answer many different questions, from the Cc values of new surfactants
(typically broad scans) to changes in Cc between batches of the “same” surfactant (typically a high-
precision scan) and, by looking at what happens when you add the same small amount of the ingredient to
each tube, how that ingredient changes the HLD (the position of the transition tube) and the efficiency of
that system (the size of the intermediate phase). A detailed examples covering polar oils is described below.

For your emulsification process, you first decide what HLD values you require during the different stages
of your process. If, for example, you want a classic o/w emulsion then an HLD of -0.5 to -1 is optimal. If
you are using microemulsions then you need HLD ~ 0.

Now we can go through each aspect of the toolkit, listed above.

Changing oil

If for the past year you have used an oil with EACN = 3 and you now want to polymerise an oil with
EACN = -1, you can instantly work out that the Cc has to reduce by 4x0.16 = 0.66. The beauty of HLD is
that Cc values are a molar-weighted average of individual Cc values. So if you previously had an X:Y ratio
of two surfactants, you can shift the Cc by 0.66 by going to X+δ:Y-δ.

Changing temperature

If you wanted to have an o/w emulsion at room temperature then a microemulsion at the reaction

FST 35
temperature of 55°C then you can choose an ethoxylate system. The ethoxylate coefficient of -0.06 means
that a 30°C change gives an HLD change of +1.8. This means that you need to start with an inefficient
hydrophilic system of HLD = -1.8 to achieve your microemulsion at 55°C. If you happen to go to 65°C
then you are encouraging a phase inversion to w/o which may not be what you want.

The point of the previous paragraph is that many ethoxylate-based emulsion polymerisations are
undergoing very large shifts in HLD … and because people have fixated on HLB, which contains no
explicit T-dependence, they have been unaware of this issue.

If, on the other hand, you wanted to move to a greener APG, if your emulsion system required a T-induced
change in HLD during the process, your APG will fail because APGs have no T-effect on HLD.

For those who like to stay in control of their systems, the large T-dependency of ethoxylates is a good
argument for not using them.

Changes in salinity

For non-ionic systems, addition of salts, for whatever reason, has little or no effect on the HLD. The Ln()
dependence for ionic systems means that if you start with a low salt concentration (maybe just the
surfactant’s own salt contribution) of S ~ 0.001 g/100cc, an extra 0.001 g/100cc changes the HLD by
Ln(0.002)-Ln(0.001) a change of 0.7! Hence it is a good idea to have a modest amount, say 0.01 g/100cc of
“background salt” to stabilise the system.

When you have to choose, for other reasons, between ionics and non-ionics, those who want to keep in
good control of their systems might want to take into account whether a background salt would be
acceptable in the overall process and final product.

Changes in surfactants

Those (and that seems to be most formulators) who are forced to make changes to their surfactant package
for cost, greenness, supply chain or other reasons can seriously reduce the pain of the transition by using
HLD.

If you are swapping within the same class of surfactants then by knowing the Cc you can instantly know
how much change there is from the previous one and what to do about it. If you are swapping class then
you need to think deeper if you are relying on T effects or if you have the need for significant amounts of
salts.

Changes in surfactant efficiency

The full theory, HLD-NAC, shows not only where you have to be in surfactant space, that’s the HLD part,
but how efficient the system will be when you’re there, that’s the NAC part. You can have two surfactants
which produce identical HLD values but where one (there’s a special class called “extended surfactants”) is
far more efficient than another. There is also a phenomenon that the efficiency for a given surfactant
decreases as the EACN of the oil increases. You can add “linkers”, or “cosurfactants” that increase the
efficiency, but read this caution, then the next section on polar oils, before trying these.

There are few known rules for success in finding additives to improve an HLD system but there is

36 FST
 Emulsion Polymers

a key rule for avoiding failure: Additives may change HLD, efficiency or both, so test additives
within a scan.

If you don’t use a scan, then if you additive produces a wonderful improvement but changes the
HLD, the result in your tube will be worse and you will have missed that a small change in your
surfactant blend to restore the HLD would give you better results.

This brings us to the next issue.

Changes in polar oils and additives

Suppose you have to add some octanol for some reason. This is an oil, so might have an EACN, it’s a
“linker”, it’s even a surfactant. So how do you incorporate it into the HLD equation? The general answer is:
you can’t. There are no known rules for these sorts of molecules which are conveniently called “polar oils”.

However, because HLD is based on scans, it is easy to set up a scan around your current system, with the
central tube being where HLD = 0, then you add the same small amount of the polar oil to each tube and
see how things change. You can repeat the additions to get a quantitative measure of how both the HLD and
the efficiency change. Here we start with the reference scan at the top right and see 4 possible outcomes:
No change, change to HLD only, change to phase volumes (efficiency) only, change to both.

Testing additives would be easy if they changed only the phase volumes. Given that it is likely the that they
will change both HLD and phase volumes, the experiments are a bit slower, while the chances of success
are much higher.

Curvature

As HLD transitions from negative through zero to positive, the curvature changes from negative, o/w to
neutral, microemulsion, to positive, w/o. For control of your emulsion polymerisation, knowing where you
are in that curvature space is clearly important.

The change also affects the interfacial tension. If a typical IFT between an oil and water is 5 mN/m at a
large HLD , then as the HLD decreases the IFT might decrease to 1 but around HLD = 0 it might decrease
to 0.01.

If you have a specific emulsion polymerisation method you are trying to control, it would be a good idea to
know that you were safely in the > 1 mN/m zone for one kind and in the ~ 0.01 mN/m zone for another.

The other reason for avoiding the low IFT region is that if you stray a little further, you are in a zone of
opposite curvature and you risk a nasty phase inversion.

FST 37
Inversion

There are times when you want to carry out controlled emulsion phase inversions. HLD tells you how to do
it via T, S, EACN or CC. This is a thermodynamic inversion. The other way to do it is via having too little
of the bulk phase, so the emulsion is forced to switch to being the “wrong” way round. This is a
concentration inversion. Each on its own is a useful process when done deliberately and a disaster if done
accidentally.

The reason there is a risk of this happening with emulsion polymerisation is that we would generally like
the least water and the most emulsion in the final product. Take this too far and the oil phase will take over
as the bulk phase and you have an undesirable w/o batch in your reactor.

The famous diagram worked out by Salager helps us see where our formulation is within inversion space:

[Link]
practical-surfactants/
[Link]

The uncertainties around the

edges of the non-HLD
transitions are due to the
complex kinetics and hysteresis
common with these transitions.
The point of the diagram is that
if, like many formulators, you
are unaware of the HLD effects,
you can get transitions when
you don’t expect them. It is a
good idea to use the diagram to
ensure you are safely in your desired zone or that you get the thermodynamic or concentration inversion
you require.

Smith-Ewart
For those doing classic emulsion polymerisation, there is a popular idea that you can describe it using the
Smith-Ewart formula. Although an app exists, it is not clear what value it adds. It is provided for those who
might find it useful:

38 FST
 Emulsion Polymers

[Link]

Water whitening
Because emulsion polymers are created and coated from water, they will have at least a modest attraction
for water in their final form. If this is a general absorption that might create issues around long-term
stability. It is also observed that atmospheric water builds up as little droplets. These then scatter light,
leading to water whitening.

Because obvious explanations such as attraction to the surfactants used do not seem to be useful, the app
can merely model the effect and suggest tricks to reduce it:

[Link]

The relevant equations and references are provided in the app.

FST 39
Fragrances
Links
Diffusion_Barrier Properties, Evaporation_Basics, Evaporation_Enthalpy of Vaporization,
Evaporation_Temperature and Antoine Coefficients, Fragrance_Activity Coefficients, Fragrance_Barrier
Properties, Fragrance_Vapour Pressure

A fragrance has been selected by a specialist team. Now it’s your job to formulate it into the
product. There are plenty of challenges ahead.

Controlled disappearance
The point of a fragrance is to evaporate. As the different fragrance molecules have different volatilities, the
mixture that reaches the nose changes over time. Rather than fight against this inevitability, the fragrance is
envisaged to have a certain top note (the instant hit of the most volatile), middle note (“the” fragrance) and
bass note (the lingering memory of the fragrance).

As a formulator you might be asked to provide some control over this aromatic symphony. The starting
point is to know what would happen to a simple mix of the aroma molecules. So you need to know their
Antoine Coefficients (Evaporation_Temperature and Antoine Coefficients ) in order to calculate their
vapour pressures at the relevant temperature (e.g. 20°C for a typical cleaning product, 35°C for something
on the skin). The three coefficients, AA, AB and AC are merely fitting constants that describe the vapour
pressure of the pure liquid over a relevant temperature range, usually up to the boiling point. Databases of
Antoine values tend to be in units of °C and mm/Hg so that’s what we will use here. The vapour pressure,
p, at temperature T is given by:

AB
log10(p) = AA − AC + T

In the space above the fragrance, component 1 will have a pressure p1 which is a fraction p1/Σp of the total
pressure. If the fragrance is measured in weight fractions, wn, then the individual pressure values need to be
corrected by the molecular weight, Mn, of each component:

∑p=
wnpn
Mn

At a given pressure, the mass, mn, of the nth molecule in a volume V of the vapour is the number of moles
per litre of vapour multiplied by the molecular weight, M, so, using 760 mm/Hg as atmospheric pressure
and 24.7 as the number of moles per litre of an ideal gas at 25°C:

pnMnV
mn = 760 x 24.7

The rate at which the solvents are removed depends on the air velocity. The dependence is complex and is
described in Evaporation_Basics.
 Fragrances

Putting these things together we can calculate how a complex fragrance mixture will change over time. The
app, which, for simplicity, only has 8 ingredients, chosen from a list of 40+, shows the principle:

[Link]

The ethanol used to create the fragrance has, in this example, disappeared in less than 1 minute. After 1hr,
only 15% of the original formulation remains, and it’s made up of the three low volatility components,
carvone, linalool and citronellal.

It’s not ideal

The calculations have assumed ideal solutions, i.e. where the activity coefficients (Fragrance_Activity
Coefficients) of everything are 1. For a fragrance delivered from ethanol, once the ethanol has gone, the
assumption is not unreasonable – because molecules with very different solubility properties and chemical
functionalities tend not to be useful as fragrance molecules.

This changes once the fragrance is formulated into other products. A candle is a very different solubility
environment from a shower gel. More hydrophilic molecules will have a higher vapour pressures than
hydrophobic ones within a candle, and vice-versa in the shower gel. A difference in activity coefficients of
2 is not unreasonable, meaning that an ingredient might be twice as volatile in one formulation than
another. This would mean a very different aroma.

This is such a significant effect that when a product range uses the “same” aroma for, say, candles and
shower gels, the formulations have to be significantly different.

Indeed, a reader kindly sent me the following which shows that subtle effects occur even within a single
type of formulation, such as shower gels:

“An interesting observation to note is the phenomenon of partitioning of fragrance molecules within
micelles in scented shower gels and shampoos. Hydrophobic molecules (such as limonene) position
themselves in the core of the micelle, while more polar molecules (aldehydes, alcohols) prefer the micelle's

FST 41
outer layer. Consequently, the olfactory sensation experienced by consumers using these hygiene products
varies depending on the nature of the surfactants employed (e.g., SDS vs SLES) and the polarity of the
fragrant molecules.”

It is an ongoing debate whether these different variants can be designed via art & craft or whether a series
of GC head space analyses allow the activity coefficients to be worked out and the formulae adjusted
accordingly.

Fragrance migration
As discussed in the Diffusion_Basic Diffusion chapter, diffusion is simple to understand. It is driven by the
concentration gradient, the difference in concentration, Δc, divided by the distance, x, over which the
gradient exists. For a given gradient, the flux (amount per unit time) is simply a diffusion coefficient, D,
times the gradient.

If we are concerned with fragrance/flavour molecules escaping, so-called flavour

scalping, then we can assume that the external concentration is zero and that we
have a packaging film from a single ingredient (we will modify this later) of
thickness x.

Different flavour molecules have different MWs and we can make a basic
assumption that the diffusion coefficient is proportional to MW, so a molecule with
MW = 200 will diffuse at half the speed of one with MW = 100. The precise
dependence on MW is debatable, but because the concentration factor is much
more important, we don’t need to worry too much.

The concentration is really a partition coefficient between the internal environment and the packaging film.
If the internal environment is dry then we are talking about Henry constant partitioning. If it is liquid then
we have differential partition between solution and packaging.

To create a full model of any specific system is complex, so to illustrate the principles we use an app that
makes the following (drastic) approximations:

• The fragrance has a maximum of 8 ingredients, chosen from a list of ~ 40

• The total amount of fragrance is assumed to be sitting on the packaging film as a liquid film with a
g/m² equivalent to the total amount in your package.
• The partition coefficient depends only on the Dissolution_Hansen Solubility Parameters Distance
between each fragrance molecule and the packaging material (you specify its 3 HSP)
• The maximum concentration for a Distance = 0 molecule is 1 vol%, with a fall off given by exp(-
(Distance/Radius)²) where the Radius is the distance over which solubility in the polymer goes
from maximum to (effectively) 0. An “open” polymer has a large Radius, a “closed” one has a
small Radius.
• Concentrations are too low to create concentration-dependent diffusion coefficients.
• The concentration on the outside remains as zero.
• The diffusion coefficient for a molecule of MW = 100 varies from 10-10 to 10-8 cm²/s (a unit
more common than m/s) via a slider from 1 to 100.
• Diffusion coefficients go as 100/MW.

Although you may object to some or all of those restrictions, remember that the app is meant as an

42 FST
 Fragrances

illustration of broad effects – and is much better than the usual alternative which is no model at all.

[Link]

Fragrances in water
Most fragrance molecules are, at best, only slightly soluble in water. As we increasingly move away from a
solvent-based world, how do we formulate fragrances in water?

One obvious way is via a classic oil-in-water emulsion as described in Surfactancy_Emulsification. This is
generally not appreciated for stand-alone fragrances because of the scattering of an emulsion, and the
problems of long-term stability when you can’t use high viscosities to stabilise the emulsion drops. For
personal care products with their relatively large amounts of surfactant, there isn’t much of a problem as the
surfactant molecules are more than enough to keep the fragrance happy.

A way to solve both the visible scattering and longevity problem is to use a microemulsion. Using
Surfactancy_HLD theory it’s not hard to create a reasonable microemulsion. The difficult part is to find
(via the NAC part of HLD-NAC theory) an efficient formulation using the minimum of surfactant. The
tricks (rational cosurfactants, linkers or extended surfactants) described in the HLD chapter can make this
possible. As fads of what are or are not acceptable surfactants come and go, reformulating with acceptable
alternatives is relatively easy if you know the Cc values of the available alternatives, something that will
become easier as suppliers are persuaded to provide these values routinely.

The idea of using the science of “solubilizers”, “hydrotropes”, “surfactantless emulsions” and so forth is
generally hopeless because there is so much confusion about what these terms mean, and so little usable
theory. That is changing, as described in Dissolution_Solubilizers and Hydrotropes where
Dissolution_Kirkwood-Buff Kirkwood-Buff theory allows us to find out what is going on inside our

FST 43
formulation.

And many of these ideas are hopeless in practice. Who wouldn’t want a “surfactantless emulsions”? It
sounds such a great idea. You can do this with “solvents” but solvents are bad. So you do it with
hydrotropes or co-surfactants or solubilizers because they are nice words. But they are still solvent-like
molecules and, worse, you often have to use a lot of them to get the emulsion you like. Terms like
“surfactantless emulsions” are great for academic papers and grant proposals, but usually not so brilliant for
real-life formulation.

Why formulators dislike fragrances

Many formulations are smart balances of competing properties. Thickening by Thickeners_Wormlike
Micelles, clever emulsification via Surfactancy_HLD and efficient Surfactancy_Anti-foaming are examples
of such balances.

Things are fine till marketing come along with a new fragrance, carefully tuned to capture some essence of
the zeitgeist. Taking out the old 0.1% and adding the new 0.1% often destroys the previously excellent
formulation, and formulators have few tools for understanding why. It’s worse than that. Many fragrance
ingredients are common (low price is vital) so it might be that the new, destructive, fragrance isn’t so
different from the old one. But even the “same” fragrance molecule can be a mixture. So it may be that the
problem comes from one sub-component of one of 20+ ingredients.

My view, based on the “polar oil” problem discuss in the HLD chapter, is that the fragrance effects are due
to subtle partitioning effects not in the tail (these are boring and predictable for such low levels of
molecules) but in the critical area between the head and tail. The scientifically relevant app, but one too
hard to use in practice is based on Tchakalova’s CIT technique: [Link]
surfactants/[Link].

It seems to me that the future is to take inspiration from the CIT approach and find ways to identify such
effects swiftly via curvature-critical measurements.

44 FST
 Hair Conditioners

Hair Conditioners
Links
Diffusion_Diffusion into Skin and Hair, Evaporation_Humectants, Flow_Basic viscosity,
Surfactancy_Phase Diagrams, Thickeners_Wormlike Micelles

If your hair has been damaged or its natural protective oils have been removed by aggressive Hair
Shampoos then a fix exists via a conditioner. If the hair is suffering from excess conditioner then
that problem can be fixed by a conditioner conditioner.

The problem to be solved

The hair follicle and the scalp’s biome do a good job in keeping your hair in good condition. A combination
of free and bound 18-methyl eicosonoic acid (18-MEA) makes the hair highly hydrophobic, with easy slip
between individual hairs. Regular shampooing strips off the free 18-MEA and aggressive hair drying and
combing helps to remove the bound 18-MEA. The cysteines in the hair keratin become oxidised into
cysteic acids. The hair is, therefore, not in good shape.

A root-cause fix for this is to shampoo less frequently (a water wash will remove most “dirt” and a light
shampooing will remove excess oil), and dry/comb more gently. The follicles and scalp biome can look
after things rather well as they have had millions of years to work out how to do this.

The alternative fix is to:

• Add a convenient oil to coat the hair;

• Add a long-chain cationic molecule (quaternary ammonium, “quat”) to lock onto the cysteic acids.
As a bonus, the resulting salt is hygroscopic, adding some anti-static functionality for free;
• Ensuring that the formulation is low pH to encourage the cysteic acid form and to ensure that non-
quat amines such as stearamidopropyl dimethyl amine are in their cationic form;
• Optionally (but they are becoming unacceptable), add some silicones for that wonderful silicone
feel;
• Optionally (but they aren’t as good as silicones) add some “natural oils” to smooth out rough hair
surfaces;
• Optionally include some “strengthening” polymers.

As has been known for decades, this can be done with ~2.5% of cetyl alcohol and ~ 1% of cetrimonium
chloride, plus <1 % of a thickening gum such as HEC, some fragrance and whatever preservative package
fulfils the triple role of (a) stopping bad microbes from growing in the formulation, (b) without harming all
the good microbes in, say, the scalp biome and (c) conforming to whatever is the latest trend in preservative
claims.

If you want a “deep” conditioner, just add a lot more cetyl alcohol (etc.) and a lot more quat to emulsify it.
As a bonus you later get to sell conditioner conditioner to restore the hair to some form of normality.

The silicones undeniably give a beautiful soft feel to the hair but too much of a good thing is a bad thing,
especially because silicones (being insoluble) build up on the hair over time. They are also surprisingly
sticky as the build up, so hair attracts more dust and dirt. In any case, regulations are taking us to silicone-
free formulations as silicones are seen as allergenic and toxic for aquatic life.

With the need to replace silicones, formulators have to master the art of getting drops of “natural oils”
from, say, rice bran or coconut to be delivered from emulsions, maybe trapped in coacervates, and then
once on the damaged hair fibre, to spread spontaneously – something the silicones do and the oils tend not
to do. Although relatively simple Cleaning_Contact Angles thinking can be used, the fact that the hair
surface changes from relatively good quality near the root to highly damaged near the tip, and that
spreading can be interrupted by surface roughness, the formulator will find this difficult.

Polymers (beyond the thickeners added because consumers have been conditioned to believe that viscous =
luxurious) will, as claimed, “strengthen” the hair or “add body” but in general are just more gunk that
sooner or later has to be removed via a conditioner conditioner.

Via clever coacervate science (Surfactancy_Coacervation) these ingredients can be added to the shampoo
itself – allowing the shampoo to remove natural oils (18-MEA) and replace them with, say, cetyl alcohol
which, while arguably “natural” is not natural to the hair system. For these systems the coacervate trick
requires polymeric versions of the cationics to be used, polyquats (synthetic or natural guar gum) instead of
simple quats. As above, these have the upside of providing “body” and the downside of being “gunk”.

The formulator’s problem

The simplicity and efficacy of a standard conditioner formulation is a problem for marketing. Consumers
should be smart enough to buy the least amount of the lowest-price me-too product and that’s not what
marketing want.

Therefore the formulator has to add ingredients that allow marketing to claim, via plausible-sounding
“logic”, that this conditioner will condition especially well:

• Protein from whatever source sounds both sciencey and green, vegan etc;
• Anything with keratin in the title (“keratin polypeptides”) because hair is made from keratin so
more keratin polypeptides must be good for it;
• Stearyl alcohol and cetearyl alcohol (a mix of stearyl and cetyl alcohol) because more ingredients
must be better;
• Aloe vera because everyone likes aloe vera because the polysaccharide acemannan is from aloe
vera;
• Vitamins and antioxidants because who doesn’t want extra vitamins in their keratin and maybe the
antioxidants will help reduce the damage to keratin cysteines;
• Extract of some exotic plant because exotic plants must be good.

Although those ingredients can all be thrown in with little obvious problem, the more stuff that’s added, the
more unexpected problems can arise. Maybe the exotic plant extract forms a complex with the acemannan
in the aloe vera, but only after 1 month of storage. That’s a tough problem to identify. And all those extra
nutrients may make your previously excellent preservative package incapable of passing microbial tests
(the nutrient boost might overwhelm the preservative effect) so a difficult round of balancing antimicrobial
efficacy against marketing trends has to take place.

As explained in the Shampoos chapter, well-controlled experiments on shampoos show that the average
consumer forms their judgement based on the packaging rather than what’s in the formulation – the same

46 FST
 Hair Conditioners

standard shampoo is “better” if provided in an expensive package and consumers score randomly when
judging a standard or expensive shampoo delivered in the same standard package. We can be confident that
many expensive conditioners, packaged accordingly, exert a similar placebo effect.

Conditioner-Conditioner
If there is too much build-up of conditioner residues then the consumer has to buy yet another conditioner
to remove those residues. Removing a mix of silicones, quats and oils requires some powerful chemistry
with risk of damage to the user’s hair and scalp … so you probably have to provide a post conditioner-
conditioner conditioner.

Maybe consumers will work out that if they treated their hair more carefully, using fewer products, less
frequently, they won’t need multiple levels of hair warfare.

Although management and marketing would object to a strategy of urging customers to purchase less,
science is on the side of those companies who find a way to do more with less.

FST 47
Hair Shampoos
Links
Cleaning_Boundary removal, Cleaning_Contact Angles, Cleaning_Surfactants, Cleaning_Temperature
effects, Flow_Basic viscosity, Surfactancy_Anti-foaming, Surfactancy_CMC and Langmuir,
Surfactancy_CPP and phases, Surfactancy_Coacervation, Surfactancy_Dynamic Surface Tension,
Surfactancy_Foam Drainage, Surfactancy_Foam Ostwald, Surfactancy_Foam Rheology,
Surfactancy_Foaming, Surfactancy_HLD, Surfactancy_Roll-up and Eötvös Number, Thickeners_Wormlike
Micelles

A basic shampoo is very simple – just about any soap or surfactant will get rid of the small
amounts of dirt and oils on our hair. For a mixture of marketing and functional reasons, a typical
formulation is much more complex.

No-poos and dry shampoos

Many people use no shampoo other than when their hair gets super dirty after some tough job. Water alone
does a good-enough job at keeping their hair in good condition. Others use dry shampoos – powders that
absorb excess oil and, via combing, help remove other bits of dirt.

So for those who choose to use a shampoo, the absolute minimum of ingredients should do a good job of
cleaning whatever needs to be cleaned. Using a minimum would be good for the pocket and good for the
planet. It is worth noting that a good Laundry Liquids cleans at ~ 200ppm while a typical shampoo is used
at 20,000ppm (i.e. 2%), an astounding over-use of a complex product for a simple task.

For those who want a compromise position using half the amount of shampoo, half as often cuts the carbon
footprint 4x and saves money as well.

The essentials
Because of the high concentrations, applied in high doses, we can use baby shampoos, soap bars, simple
SLS formulations and super-complex formulations, each with a different class of surfactants (sorbitan
ethoxylates, fatty acid salts, SLS, complex mix) and get adequate cleaning of the generally small amounts
of dirt and oils that need to be removed for a clean hair.

Users with hard water and those whose hair picks up various metal salts from the urban environment will
benefit from small amounts of chelation from EDTA or a greener alternative.

All aqueous formulations need a preservative package, discussed below.

Used sparingly, the simplest formulation, surfactant, chelator, preservative, will achieve the nominal
requirements for a shampoo.
 Hair Shampoos

Packed with features

For whatever reason, consumers have decided that their shampoo should be packed with extra features,
none of which is relevant to the core task of cleaning excess dirt without destroying the hair’s own
protective biome and oils:

• Viscosity to give a luxury feel;

• Foaming partly to help disperse the viscous surfactant, partly because users associate “foaming”
with “cleaning”;
• Oils to replace natural hair oils removed by the shampoo;
• Silicones (or, as these are becoming unacceptable, “natural oils”) to smooth out hair that is
partially damaged by over-use of shampoos;
• Quats to avoid static from hair that is over-cleaned;
• Polymers to add “body” to hair that might have been damaged by over-cleaning;
• Fragrances for a feel-good factor;
• Probiotics and their equivalents;
• Ingredients with no functionality other than to provide marketing claims.

Let’s discuss each in turn.

Viscosity
There is little consumer utility in making a shampoo viscous. Some say it stays better on the hand during
application in a shower, but this isn’t much of a justification. It is anti-functional because it makes it harder
to spread evenly through the hair. But consumers have been conditioned to associate viscosity with luxury,
so most shampoos are thickened.

Because most Thickeners_Polymeric Thickeners leave unpleasant sticky residues (though they might be
used to create “body”), the most usual thickener is NaCl, via the magic of Thickeners_Wormlike Micelles.
The salt can also provide some mild water softening effects so this cheap thickener is very attractive.

The downside of this trick is that many alternative surfactant formulations that might be considered
greener, don’t have the ability to form wormlike micelles. So formulators are trapped. They want to use
newer, greener surfactants but have to preserve a functionless feature that consumers expect.

As is known to everyone who has formulated via salt thickeners, the other ingredients can have subtle
effects on the thickening. If a new fragrance happens to provide optimal thickening at a lower salt
concentration and Production add the standard amount of salt, the formulation might end up in the “past
thickening” state and the entire batch is an expensive write-off. If the additive interferes with the thickening
then maybe an extra bit of salt will save the batch – but maybe it simply won’t thicken and the batch is a
write-off.

Because the rheology of salt-thickened formulations contains time/speed/temperature complexities, it is

especially tricky to check for subtle impacts of extra ingredients. A simple QC check with a single viscosity
might say that a new additive is OK. Only a proper rheological scan, unlikely to be used in QC would
reveal the subtle problems.

Some basic flow curve rheology, Flow_Rotational rheology, will allow judgement about flows through
production (for example, calculating Flow_Poiseuille Flow through pipes) and for the end-user experience

FST 49
(pourability through the shampoo dispenser).

Foaming
Any surfactant molecule that is sitting in a foam wall is contributing nothing to cleaning the hair. Foam is,
therefore, not just useless but has a negative impact on efficacy … except where the foam helps distribute
the viscous surfactant more evenly through the hair.

But because decades of marketing has convinced consumers (see also Dishwashing Liquids where the
standard test is foam lifetime rather than cleaning efficiency) that “foam = clean”, formulators are forced
to abandon otherwise excellent surfactants for those that create lots of stable foams in standard tests that
may not apply to the real world.

As described in Surfactancy_Foaming, most surfactants will foam adequately if sheared efficiently. The
shear of a viscous surfactant rubbed between the user’s hands and the shear from fingers rubbing surfactant
through the hair is very different from the standard tests such as a blender, a cylinder of solution being
regularly inverted, or a stream of bubbles rising inside a cylinder.

This means that formulations are optimised for a non-core function via tests that don’t replicate the real
phenomenon. This problem is an opportunity. Finding imaginative ways to deliver foaming to consumers
using less conventional surfactants might be an interesting way to use the laws of physics to your
advantage. It’s probably better than trying yet another combination to create an equivalent of SLES/CAPB
to mimic what everyone else is doing.

Oils
Whether the added oils are “natural” or not, they aren’t the oils that the consumer’s scalp naturally produce.
The chosen mix will be based partly on hard-to-make-scientific sensorial effects and partly on marketing
claims for whatever oil happens to be fashionable.

In a formulation containing so much surfactant, the challenge of incorporating the oil into the formulation
is not very hard. The real problem is if/when the oils interfere with the all-important salt thickening
mechanism. Again, a marketing-driven non-necessity presents formidable formulation challenges for which
we have few rational tools other than plenty of rheology.

Silicones
The highly desirable silky feel from silicones has to be balanced with the increasing need to claim
“silicone-free” formulations. The non-volatile silicones suffer the problem of build-up over time. They
even become sticky – attracting dust and dirt, making the hair worse. Like so many such problems, this is
an opportunity to formulate “deep cleaning” shampoos to remove the problems created by the regular
shampoos. Scientifically, the best surfactants for silicones are silicone-based surfactants. That’s because the
“hydrophobic tail” of a conventional surfactant is an alkane … and silicones prefer silicones to alkanes. If
silicone surfactants are unacceptable then relatively high Cc (hydrophobic) surfactants are needed to be
able to remove the silicones.

Some silicones can also be quats.

With the need to replace silicones, formulators have to master the art of getting drops of “natural oils”, such

50 FST
 Hair Shampoos

as those from rice bran or coconut, to be delivered from emulsions, maybe trapped in coacervates (see
below), and then once on the damaged hair fibre, to spread spontaneously – something the silicones do and
the oils tend not to do. Although relatively simple Cleaning_Contact Angles thinking can be used, the fact
that the hair surface changes from relatively good quality near the root to highly damaged near the tip, and
that spreading can be interrupted by surface roughness, the formulator will find this difficult.

Quats
Quaternary amines polymers (“polyquats”, synthetic and natural like guar gum) are salts, so attract water
and provide a conductive pathway along the surface of the hair. So they provide both anti-frizz and anti-
static properties. Although hair nearer the root is neutral as it hasn’t been damaged, nearer the tip the hair
contains more anions (e.g. sulfates from oxidation of cysteine groups in the keratin) so the quats are nicely
attracted to these damaged areas, giving the chance to provide “body” to the weakened fibres as well as a
degree of smoothing of the roughened surface.

They can be delivered via Surfactancy_Coacervation, the concentration-dependent solubility/insolubility of

complexes of the polyquats with anionic surfactants such as SLES. Unfortunately, the science of
coacervates remains rudimentary and there seem to be no useful guides as to how to formulate these effects
rationally. They can clearly have a big effect on wormlike micelles – they can destroy them by removing
surfactant, or create complex equivalents. Again, disentangling these effects scientifically seems to be
beyond our knowledge.

Polymers for body

It seems irrational to want to clean junk from the hair and replace it with other stuff intended to stiffen the
hair. But especially for those who use a lot of hair care products, the damage caused to the hair by the
aggressive cycles of chemicals, hot water and hair dryers is enough to require repair work using a coating
of polymer.

The polyquats can do some of this. Other uncharged polymers might also perform this function as long as
they don’t act as glue between strands. For those who can’t use salt thickening for their surfactant
formulation, a “body” polymer might also provide the required thickening.

Although not as strong as the interactions of polyquats with anionics, polymers can show significant
interactions with surfactants so a well-intentioned addition of a polymer might have a bad effect on other
subtle properties of the formulation such as thickening. The extended discussion in the Coacervates chapter
covers some of these issues.

Fragrances
The general challenges of formulating a fragrance, including avoidance of flavour scalping through the
packaging, are discussed in the Fragrances chapter. With so much surfactant in a shampoo, getting the
relatively low quantities of fragrances into the formulation is not much of a problem.

As with the other ingredients, the effects on the liquid crystal (wormlike) phases created by salt thickeners
can be subtle and difficult to diagnose and formulate around.

FST 51
Probiotics
As with Soaps and Washing, after years of selling products which aggressively interfere with the skin and
scalp biomes (to use a convenient shorthand term), industry wants to sell products to restore our systems to
their optimal natural balance.

The “no-poos” who don’t used shampoos claim that it takes a few weeks for their systems to recover to a
natural balance, with their biome then doing a good job of caring for their hair. There’s an obvious selection
bias here as this doesn’t work for those who try, but fail, to go shampoo-free. Who knows whether their
natural biome isn’t great for their hair, or whether the biome that evolves is out of balance because of the
history of shampoo use.

Whether the addition of one or more “pre-biotics” (food for your microbes) or actual biotics (live microbes)
can create a biome as natural and healthy as yours (presumably) would have been without lots of shampoos
is a tricky question. Even harder for the manufacturer is making a solid claim. Use vague terms, with no
proof of efficacy, and people might buy the product. Make precise claims, based on extensive safety and
efficacy trials, and it’s probably a pharmaceutical product.

Preservatives
For any aqueous product that contains ingredients that are nutritious to microbes (and that very much
includes the surfactants used in shampoos as well as the oils, vitamins etc. added for whatever purposes)
there needs to be something in the formulation to stop the microbes from growing. By definition, these
preservatives have a negative effect on living cells, which means that it’s highly likely that some of them
could have some negative effect on some human cells. It makes no difference if the preservative is entirely
synthetic or is found in nature – it has to have some bad effects on living cells, that’s its job.

Formulating a preservative into a shampoo is not hard. Proving its efficacy involves tedious standard tests,
but they are routine. The hard thing is finding a way to say that your shampoo contains ingredients that are
harmful to living cells without scaring the customers who often believe that all “chemicals” are harmful
and only “natural” things are good.

The only way out of this dilemma is to have a formulation with insignificant amounts of water, as discussed
soon.

Ingredients for marketing claims

If your product proudly claims that it contains extract of mountain blossom and has a picture of a mountain
to prove it, and if, indeed, some mountain blossom has been extracted and added to the shampoo, then,
assuming you know that mountain blossom contains no harmful chemicals, your product can be sold. If
consumers think that extract of mountain blossom must be good (otherwise, the logic goes, you would not
be proudly adding it), then you are probably OK. Only “probably”. Authorities are so irritated by marketing
claims designed to fool the customers that they have become more strict even on things that merely imply,
but don’t explicitly state, some benefit to the consumer that cannot, in fact, be substantiated.

As a formulator, your job is to add what marketing want you to add. As long as extract of mountain
blossom is present only at the minimum that allows the marketing claim, it should pose no real formulation
problem.

52 FST
 Hair Shampoos

The right packaging

Sadly, well-controlled experiments using either bland, industry-standard shampoo formulations in different
packaging, or bland versus expensive formulations in standard packaging, have shown that consumers
overwhelmingly rate the effect of the shampoo according to their expectations from the packaging. It might
be distressing to the formulator that all those extra ingredients added to their top-of-the-range product are
not detectable to the average consumer, but it’s what is expected from the science.

If you think that I’m making up these assertions, get hold of a copy of the Robbins Chemical and
Physical Behaviour of Human Hair. The author has a lively sense of what is good science and
what is mere marketing.

Dry shampoos
Dry shampoos mean you don’t have to ship large amounts of useless water, with the packaging and
transport costs the water requires. They can also be free of preservatives as microbes are unlikely to grow
in an environment largely free of water.

There are two broad types.

• Absorbent powders such as starch that attract excess oils and can then be combed out of the hair,
leaving it feeling OK. The tricky formulation problem is getting the right Particles_Size
distribution. If the particles are, on balance, too small then the product is a dust hazard and hard to
comb out of the hair. If, on balance, they are too large, then they feel “gritty” and the surface area
to volume ratio is too small to be an effective remover of oil.
• Shampoo bars, which are generally old-fashioned soap with whatever ingredients (e.g. excess
glycerol from the manufacture or excess oil) make it less alkaline and aggressive, plus the usual
fragrances and colourants for a feel-good factor.

Neither presents a tough formulation challenge, though they are each a marketing challenge as is the
challenge of a soap dish that is convenient and keeps the bar relatively dry without falling into the sink or
shower.

FST 53
Inkjet Inks
Links
Absorbency_Porosity and Darcy, Dispersions_DLVO, Dispersions_ODC, Dispersions_Rheology (Low
shear), Dispersions_Zeta potential, Evaporation_Basics, Flow_Ohnesorge, Flow_Rotational rheology

Although it is easy to specify and create a workable inkjet ink, the frustration is with the difficulty
of getting perfect drops, infinite lifetime as a stored ink and zero problems within the head and on
the nozzle. Following everything in this chapter will get you a workable ink, the rest is hard work.

The limits of Ohnesorge

As discussed in Flow_Ohnesorge, you can’t get a good inkjet drop if the combination of drop size, velocity,
viscosity and surface tension takes it outside the core area defined by Ohnesorge, Reynolds and Weber
numbers:

[Link]

For the velocities and (small) drop sizes from typical commercial inkjet heads, the limits of surface tension
are modest, maybe 30-60 mN/m is fine, with lower tensions more problematical for interactions with the
nozzle plate than for Ohnesorge. The limits on viscosity are tougher 5-25 mPa.s are often quoted, with
something like 15 mPa.s being a common target. Outside these ranges the drops can’t form or satellite
drops are formed excessively.

For typical UV inkjet formulations, at room temperature the viscosity limit is tough to achieve from the
limited set of usable UV monomers. Heads routinely run at 40°C to take advantage of the significant drop
in viscosity with temperature that is typical of acrylates.

For water-based formulations, the equilibrium surface tension is a worthless value, what matters is the
 Inkjet Inks

dynamic surface tension (Surfactancy_Dynamic Surface Tension) measured, typically, via the bubble
pressure tensiometer.

In principle, viscosity should be judged using a proper rotational viscometer Flow_Rotational rheology to
check for properties over a range of shear rates. In practice most reasonable inkjet inks have happily dull
rheologic properties at the rates and strains that can be measured. What’s happening at the super-high
extensional flows coming out of the nozzle is outside the capabilities of this book.

Basic knowledge of particle viscosities at low shear, Dispersions_Rheology (Low shear) can help
understand the modest effect of particle volume fraction up to, say, 30%, with suitable caution that a change
in viscosity of the base formulation, without particles, from, say, 5 to 7 cP, which is a trivial 2 cP change
will take a 30% formulation from 13 to 19 cP (13 * 7/5) which is a significant jump for an inkjet ink.

The basic particle viscosity curve assumes no attractive particle-particle interactions. Getting the right
dispersant is vital, using either the Dispersions_ODC approach for non-aqueous systems or standard
Dispersions_DLVO for aqueous ones.

Another constraint on formulation comes from subtle internal issues of specific heads. Having read
the specs of a specific head, we created a reactive ink that delivered exactly the desired effect …
until the head suddenly stopped working. An enquiry to the head manufacturer led to them asking
about the pH of the ink. “Ah, we failed to put into the spec that the pH must not be below 5.5”. The
chemistry of this specific ink required a pH of 5. A project that had so much promise had to be
abandoned, thanks to a pH difference of 0.5.

Early drop dynamics

Assuming you are in the right Reynolds/Weber/Ohnesorge range, success in drop behaviour is not
guaranteed. Complex interactions while leaving the nozzle, and complex fluid dynamics as the stretched-
out drop pulls in on itself can cause many wild things to happen. There seem to be many observations via
high-speed drop watchers and plenty of computer simulations, but nothing appable that can help the
formulator. So let’s assume you have a nice drop with, at worst, a single satellite following harmlessly
directly behind (some satellites shoot off to the side, which is most undesirable).

Flying through the air

There are three issues of interest about the drop’s flight from the nozzle to the substrate:

1. By how much it gets slowed down by air resistance

2. By how much it is influenced by gravity (answer, the effect is insignificant)
3. By how much it is deflected from its straight path to the substrate by the fact that the head is
moving in the cross-web and along-web directions.

FST 55
[Link]

The app answers all 3 questions, but the gravity and deflection issues are not relevant to this book. What is
important is that a typical 10pl drop (you define drop diameter and its volume is calculated for you) coming
out of the head at 8 m/s lands on the substrate after a 10mm gap at 1.3m/s.

Contrary to instinct, to a typical drop, hitting a surface at 1.3 m/s is a gentle landing, so there’s no dramatic
3
splashing. Instead it quickly forms a hemispherical cap of √2 the diameter of the full drop, in this case a 27
µm drop becomes 34 μm on the surface. What happens after that needs another app.

56 FST
 Inkjet Inks

Drop spread

[Link]

What is scary about drop spread is how fast it is. We assume a 90° contact angle as the drop hits the surface
and we’ve set the starting radius to 17 µm. Within 50 ms (you can use your mouse to confirm this) the drop
has spread to 92 µm diameter. After 1 s it is 134 µm. As explained in the app, the dramatic fast early speed
is due to the fact that for a contact angle of θ, surface tension σ and viscosity η, the spreading velocity is:

θ3σ
v= η

That θ³ dependency is very harsh if you want to avoid drop spread.

Given the limits of surface tension and viscosity for an ink, drop spread cannot be seriously controlled via
the ink. It’s the substrate that matters most.

If you make sure that the equilibrium contact angle of the ink is, say, 50° then you find:

FST 57
that the drop goes near instantly to 48 μm, and stays there pinned at the 50° limit. Inkjet printing is one of
the rare portions of formulation science where contact angle is of crucial importance.

You can explore the consequences of this via another app:

[Link]

Here you can explore the impact of final drop size on the optical density (OD) of the print. Too little dot
gain (contact angle too high) and the dots don’t join up. Too much and your print becomes a mess.

The other way to stop drop spread is to absorb the ink into a porous substrate:

58 FST
 Inkjet Inks

[Link]

The standard Absorbency_Porosity and Darcy model seems to over-predict the sorption (70% in 1 ms). The
Davis & Hocking version predicts only 23%, but that’s still going to have an impact on drop spread as there
may well be some pinning implications.

Both Darcy and D&H depend on the distance, d, travelled in time t into the porous medium with pore
radius r.

d= √ rσcos2η(θ)t
D&H tells us that the volume, V, absorbed depends on the starting radius of the drop, R via:

V=
πR2
2 ( 2d −
d2
D )
It is worth noting that for a nanoparticle ink, the chances are that the particles will block the pores, so
absorption stops very quickly.

There are so many subtle ways for drops to fail to arrive at the desired location that naïve dreams
of “printed electronics” via inkjet have regularly failed. I have been astonished that so much
funding went to so many companies that stood zero chance of success. The one success story is
OLEDs, but the printers cost in the $10’s of millions. Those with printers at home note the large
amount of ink consumed during “necessary head cleaning cycles”. It’s not that the printer
manufacturers are especially greedy, it’s just that they are paranoid about losing a single nozzle
which may lead to losing a customer.

Dot join-up
Given the interesting calculations of drop spread, the next logical question is about dot join-up. It turns out
that the rule is trivial: if the drops touch then they join and it’s essentially instant. That’s frustratingly
imprecise, but seems to be the case. One interesting calculation involves the task of writing a nice straight
line:

FST 59
[Link]

The screenshot shows that these dots have joined to create an uneven pattern. With some variations in
spacing, drop size and contact angle you can get separate dots or the desired uniform straight line.

Is that it? I’ve tried hard to find more examples of reliable inkjet science we can all use, but it doesn’t seem
to exist.

60 FST
 Laundry Liquids

Laundry Liquids
Links
Cleaning_Bleaching and Oxidation, Cleaning_Boundary removal, Cleaning_Contact Angles,
Cleaning_Enzymes, Cleaning_Surfactants, Cleaning_Temperature effects, Dissolution_Solubilizers and
Hydrotropes, Flow_Basic viscosity, Fragrance_Activity Coefficients, Fragrance_Barrier Properties,
Fragrance_Vapour Pressure, Surfactancy_Anti-foaming, Surfactancy_CMC and Langmuir,
Surfactancy_Dynamic Surface Tension, Surfactancy_Foaming, Surfactancy_HLD, Surfactancy_Interfacial
Tension and Rigidity, Surfactancy_Phase Diagrams, Surfactancy_Roll-up and Eötvös Number

Modern laundry detergents are based on complex trade-offs of science, cost, environmental
footprint and marketing claims. They are over-engineered for their typical requirements which are
to freshen up hardly-soiled clothes, but must give satisfactory results with any tough soil that the
user happens to provide. Much of what is discussed applies to solid laundry powders, but their
formulation challenges are relatively easier, so the focus here is on the more challenging liquids.

A basic wash
Given some soft domestic water and a typical load of hardly-soiled clothes, just about any low-foaming
surfactant in the fairly hydrophobic range (discussed with more precision shortly) will render the clothes
clean by any rational standard. Any hydrophilic dirt will be removed by the water and typical hydrophobic
dirt particles will attract the tail of most surfactants, leaving a sufficiently hydrophilic head sticking out to
keep the particle suspended. The difficult challenge of removing oil on the surface is discussed below. The
surface tension reached by the surfactant in an ideal lab test is irrelevant to what is found in a washing
machine. It’s nice to have a lower surface tension, but whether it’s 36 or 41 mN/m is of no great
significance.

The low-foaming need is paramount. Consumers have been conditioned to associate “foaming” with
“cleaning”, which is irrational. In a washing machine, there is no need to impress the consumer with foam
and the formulation can deal with the obvious fact that foaming interferes with cleaning by tying up
surfactant, impeding free flow of cleaning water, and overflowing the machine.

Of more importance than the surfactant is the fact that overcoming the no-slip boundary condition
(Cleaning_Boundary removal) is necessary to get any cleaning. This all comes down to the amount of
agitation in the machine which depends on drum design, orientation (vertical drums are far less efficient),
on the relative volumes of fabrics and water, and on the relative absence of foam.

To cope with hard water, a typical chelator such as EDTA or an environmentally friendlier one will keep
the calcium/magnesium salts from interfering and “builders” such as phosphates, silicates or polyacrylics
are included to help with chelation and tie up ions such as iron or copper that can discolour the wash or the
washing machine.

Such a product would have a minimum environmental footprint and would be low cost. It would also be
unsellable.
What consumers want
The first requirement is to mask the usual “chemical” smell from traces of surfactants and any smell of
damp clothes. So it needs a fragrance (Fragrances). A typical fragrance molecule will be nicely soluble in
the concentrated detergent and will happily dissolve in the washing water … to disappear down the drain
unless is strongly partitions onto or into the fibres of the clothes, something that is hard to guarantee. Via
Microencapsulation most of the formulation problems disappeared as the capsules are readily trapped by
just about any fabric and their slow release of fragrance via leakage or via rupture through handling of the
fabric gives the desirable long-lasting properties. The transition from microcapsules that produce
microplastics to ones that are fully biodegradable is happening as this book is being written.

The next requirement is that whatever soil is on any fabric will just disappear in the wash. To make this
happen we need the following:

• A surfactant that can produce efficient roll-up of oils and grease, Surfactancy_Roll-up and Eötvös
Number.
• Optionally a lipase enzyme to break up fatty acids and therefore reduce the need for the surfactant
to do all the removal of fats. It’s optional because a formulation that doesn’t then remove the
resulting fatty acids can have a slight “sweaty” odour.
• A protease enzyme to break up protein stains including blood and grass.
• An amylase to break up starch residues, usually from food stains.
• Optionally a cellulase to remove cellulose microfibres that make the fabric look “dull”. The
removal also removes and dirt or stain that’s on those microfibres.
• A bleaching system (Cleaning_Bleaching and Oxidation) that magically destroys all coloured
stains (including wine, blood, tea/coffee, grass) while never harming any of the fabrics and never
causing fade to any of the colours of the fabrics themselves.
• A high alkalinity is great for destroying many of the soils but risks damaging the fabrics. A
modest alkalinity (often provided via the package of builders) helps a little and is especially good
for solubilizing the fatty acid odours from stale sweat and those released by lipases.

Consumers also want their detergents to be chemical free, natural and low price. And although the most
natural parts of most formulations are the enzymes, many consumers want “bio-free” formulations,
recognising that just because something is natural doesn’t mean that it’s harmless. A mix of lipase, protease
and amylase sitting on your skin could, indeed, result in plenty of damage.

The obvious fact that all residues of surfactants, enzymes and bleaches can be removed via thorough
rinsing meets the requirement to reduce water usage during a wash. As is hopefully well known, 3 rinses
with 1/3 the amount of water is better than 2 rinses with 1/2 the amount, which is more efficient than one
rinse with the full amount – hence the trend to more rinse steps with a good spin after each rinse. Smart
washing machines can adjust the volume of water and number of rinses based on various sensors (such as
turbidity), but formulators need to formulate for dumb machines too.

The right surfactant system for oil removal

As we know from general cleaning, the choice of surfactant isn’t of great importance – you can clean most
things adequately with SLS. But for removing oils from fabrics you need to be in the right part of surfactant
space, as defined by HLD (Surfactancy_HLD) to achieve the necessary super-low interfacial tension γ
(Surfactancy_Interfacial Tension and Rigidity ) between water and oil which gives you a large Eötvös
number:

62 FST
 Laundry Liquids

ΔρgR2
E= 2γ

The interfacial tension becomes super-low (100x lower so

Eötvös is 100x larger) only when the Hydrophilic-Lipophilic
Difference is close to zero, and because HLD depends on the
oiliness of the oil (its EACN), the salinity of the formulation
(all those chelators/builders that are added), the temperature of
the wash and the characteristic value, Cc, of the surfactant
blend, the focus for the formulator is to get the Cc right.

Under any typical set of values for oil, salinity and temperature,
you need relatively high Cc (hydrophobic) surfactants which,
happily, tend to be poor at stabilizing the water-air interface so
give the low foaming necessary for a washing machine.

The temperature dependence of the much-used non-ionic ethoxylates was a key aspect of making them
efficient at typical 40-60°C washes that were common in the past. In the example shown here, an
ethoxylate with a Cc of 1 gives excellent washing at 50°C because HLD ~0 while at 30°C it is useless
because HLD is ~-1. To work at 30°C needs a surfactant with Cc ~ 2.

[Link] & [Link]

surfactants/[Link]

The problem now is that consumers might use a low temperature wash at a high temperature and the
interfacial tension at 50°C will now be ~1 mN/m instead of 0.01.

There are two broad approaches to solving this issue.

1. Create a temperature-insensitive combination of ethoxylate and ionic surfactants. Because the T

effect of an anionic is 6 times smaller, in the opposite direction, a suitable anionic-ethoxylate ratio
might give adequate temperature independence. Unfortunately, simple HLD cannot be used for
this because anionic/ethoxylate mixes show non-ideal behaviour. Instead, a careful experimental
program of HLD scans is required.
2. Use something like an APG or polyglycerol formulation because these have very low and low T

FST 63
 dependence respectively. Finding effective packages of these that are in the right HLD space and
have the required efficiency is a challenge.

Anti-foams for detergents

Even though efficient, lower foaming surfactants can be used, there is still a general requirement for some
Surfactancy_Anti-foaming. Interestingly, the challenge isn’t so much to find a really good anti-foam, but
rather one that does only a modest job – and that is a difficult challenge. Scientifically a good anti-foam is
perfect for washing. Unfortunately, consumers might see an absence of foam through the window of their
machine and think that the detergent must be useless. Had the marketing departments of megacorps not
wasted decades persuading consumers that foam = clean, we could have better formulations for most of our
consumer products. Given the expectation of some foam being visible, finding a reliable not-quite-good-
enough anti-foam is a frustration for the detergent formulators and suppliers of anti-foams. In the end, it
comes down to trial-and-error and “experience”, both of which can be upended when, say, a “bad” sulfate-
based surfactant is replaced by a “good” sulfate-free equivalent. Worse than this is when marketing want a
new fragrance for the product. As happens so often in other parts of personal care, 0.1% of a new fragrance
can wreck an anti-foam formulation that worked wonderfully for 0.1% of a different fragrance.

Why can’t we design (or purchase) anti-foam packages that are right first time? Read the anti-foam chapter.

The right surfactant for a concentrated product

Concentrates have become the norm, meaning that less water and packaging are required per wash. There
are two problems with concentrates – of which the first is obvious and the second is hidden, requiring more
work:

1. Some surfactants get too viscous at the desired concentration so they cannot be used without
whatever ad hoc tricks destroy whatever strong self-association is causing the high viscosity.
2. Some surfactants are fine at high viscosity, fine (of course) at low viscosity, but catastrophic at the
intermediate concentrations that arise when the concentrate is diluted in the washing machine. The
problem is formation of cubic or hexagonal phases which are semi-solid. As there is no way to
predict these (ideas around Critical Packing Parameter (Surfactancy_CPP and phases) are far too
naïve for real-world detergent formulations), you have to do the work to explore the phase
diagram (Surfactancy_Phase Diagrams). As the formulations tend to be complex, a ternary phase
diagram is often needed.

You might find something like this:

64 FST
 Laundry Liquids

[Link]

The concentrate on the X:S axis to the right is OK, the diluted formulation at the end of the arrow is OK,
but the cyan blob represents an hexagonal phase which will result in blobs of surfactant floating in the wash
and ending up on the clothes at the end.

Obtaining a phase diagram is often seen as difficult, but the [Link]

surfactants/[Link] page, courtesy of Seth Lindberg at P&G shows that you can obtain the relevant
information relatively easily.

The trade-offs
Hot water makes enzymes and bleaches work faster and the kinetics, if not the thermodynamics, of
surfactancy are improved. Soluble dirts are more soluble, tough greases start to melt so the only downside,
which is overwhelming, is the combination of direct energy costs for the user and the impact on the planet.

With enough effort, enzymes can be tuned to be efficient at cold water temperatures and catalysts can
encourage the peroxide bleaches to work faster … or you just add more of them, with an obvious cost
penalty. If the temperature isn’t high enough to melt some grease, maybe the lipase will be able to break it
up.

A longer wash cycle will be especially useful for the enzymes and bleaches, but unappreciated by the busy
end users.

If the surfactants, bleaches, enzymes, chelators/builders and fragrance package interact in some
unfavourable ways then one solution is to package the individual components inside a “pod”. The pods are
made from a water-soluble polymer such as PVOH which will hopefully dissolve quickly. Unfortunately,
pod residues can sometimes get caught in the fabrics and, because they are denied the flow of water
required to dissolve them completely, end up as gunk on the clothes. This can happen if the PVOH dries out

FST 65
during storage, making it less responsive to the first seconds of contact with the water. One potential fix is
to add Evaporation_Humectants that keep a fairly stable amount of water even if the local atmosphere is of
low RH, though the app throws some doubt on the standard explanation for the efficacy of humectants.

66 FST
 Lipstick

Lipstick
Links
Adhesion_Adhesion promoters, Adhesion_Surface energy Interactions, Cleaning_Boundary removal,
Cleaning_Solvents, Cleaning_Surfactants, Cleaning_Temperature effects, Diffusion_Barrier Properties,
Diffusion_Basic Diffusion, Diffusion_Diffusion into Skin and Hair, Diffusion_Permeation OTR and
WVTR, Dispersions_ODC, Dispersions_PVC and CPVC, Dispersions_Rheology (Low shear),
Dispersions_Settling, Flow_Basic viscosity, Flow_G' and G'', Flow_Oscillatory rheology, Flow_Rotational
rheology, Flow_TTS-WLF, Flow_Yield Stress, Optics_Colour

Lipstick contains its own contradiction: the colour should be indestructible on the lips, never
transfer to an object such as a cup or glass, yet be fully and easily removable when required,
leaving no stain/residue on the lips.

The basic formulation

For this chapter we will assume that the pigment is a given and that it is the optimal size for hiding power
(too small and there’s insufficient scattering, Optics_Scattering and Opacity) and colour intensity (too large
and the interior of the particles adds no extra absorption) and has an excellent dispersant
(Dispersions_ODC ) optimized for the oil/wax environment. The pigment might be ground using one or
more of the oils as a convenient medium.

A possible exception to this is that the common mica pigments may be providing global properties such as
slip, barrier or strength properties beyond their colourant, gloss and sparkle effects. If these extra effects are
important, you can include them in the formulation space mapping discussed at the end of this chapter.

The wax component provides the necessary semi-solid form at room and skin temperature (conventionally
said to be 32°C) while being meltable at a reasonable temperature (conventionally 55-75°C) to be liquid for
combining the ingredients and then pouring into a mould to produce the desired shape.

The oils provide the desired softness, emollience and general feel-good effects.

Melting together some beeswax, coconut oil, shea butter and just about any hydrophobic colourant gives an
adequate lipstick when poured into a tube and allowed to cool.

A suitable fragrance and an antioxidant such as BHT or vitamin E (tocopherol) to stop the oils being
oxidised completes the basic formulation.

Even without extra functional ingredients, discussed below, the formulator is faced with a daunting
formulation challenge. There are many waxes and oils, and they interact with each other in multiple ways,
especially in terms of the crystallinity of the waxes as the lipstick cools in the mould. Inevitably this means
we need to use Differential Scanning Calorimetry (DSC) to get core information and have at least some
idea of crystallization theory.

The subjective feel of the lipstick during and after application is too subtle to be properly analysed in the
lab, but we have to find measurements that at least correlate, so that means rheology.

We address each of those in turn, after first thinking about what should be tested.

This was a difficult chapter to write. It is very easy to go blah blah with truisms, but that doesn’t
help. There are no easy, relevant formulae to provide an app-route through formulation space.
Instead we identify those measurements that must be relevant to the formulation, then work out
how to map those data against real-world behaviour of products that are known to be good or bad,
your own and competitors. This is an unfamiliar approach, but a powerful one.

Core data
It is tempting to throw together some typical formulations and measure the key data. Although, as
discussed below, it’s important to measure some real formulations, especially the low, medium and high-
end competitors’ formulations, to get out of the habit of adding too many ingredients too soon, it is
necessary to get some core data on simpler sub-components.

So choose a few waxes (ones favoured by marketing for their claims) and a few oils (similarly chosen by
marketing) and gather pair-wise data to see what is affecting what in what ways. For the future
digitalisation of your product development, you can build on core data from key ingredients. It is hard to
build on the data from complex mixtures selected for historical reasons such as “we’ve always done it this
way”.

DSC, MPts etc.

The DSC plots the mW/mg of sample needed to
raise the temperature by 1° at the given heating
rate, °/s. Without melting, the value (when
multiplied by the timestep) is simply the heat
capacity of the material. The latent heat of fusion is
responsible for the large increase in power needed
to maintain the heating rate. Even a “pure” wax
melts over a significant temperature range and a
typical mix of waxes for a lipstick melts, as here,
over wide range. This example shows two peaks,
nominally for two types of waxes. In reality, the
number of peaks found depends strongly on
heating rate – the slower the rate the more
individual peaks can be found. The integration gives us the total latent heat of fusion, in J/g. A fully
amorphous wax would show a small peak, the fully crystalline equivalent would show a large peak, so the
area is used as some indication of the degree of crystallinity.

Any single graph should be treated with caution. This “heating” curve will be different from the “cooling”
curve because not only is there supercooling (so the MPt peaks will be at a lower T), there is also a
different degree of crystallization. Both of these will be cooling-rate dependent. Scientifically we might
measure “isokinetic” cooling where the DSC cools the sample at a fixed rate. In production the cooling is
not only non-isokinetic (cooling rate is faster at higher temperature) but also faster at the outside of the

68 FST
 Lipstick

sample than in the core that remains hot for longer.

What do we do with the data? In principle we can relate the experience of the lipstick in use to:

• The onset temperature (first sign of something beyond standard heat capacity);
• What’s happening at 32°C
• The number of MPts and the maximum MPt
• The “degree of crystallization”, with a higher degree making it harder to apply the lipstick while
making it more resistant to accidental transfer.

We return to these ideas later.

Rheology
We know that lipstick must be a combination of elastic and plastic properties – it must be relatively rigid
(elastic) yet flow when applied to the lips (plastic). The standard way to examine the relative values is via
oscillatory rheology, measuring G' (elastic) and G" (plastic), Flow_G' and G''. Most uses of G'/G" data rely
on “small strain” data, where the oscillation is in the linear region of, say, 0.02% strain, SAOS (Small
Angle Oscillatory Shear). Small strain data are scientifically simple and we know how to analyse them. For
lipsticks we must look at data up to 100% strain. This takes us into LAOS (Large Angle Oscillatory Shear)
techniques with their Lissajous figures and Chebyshev coefficients. With modern software, we can extract
“simple” G'/G" data (as shown in the plots) or any other feature such as strain stiffening and shear thinning
and find ways to use the parameters to help characterise the complex trade-offs of lipstick.

Typically a G'/G" curve would be measured at a “reasonable” oscillation rate of, say, 1/s that matches the
application process, though in an ideal world it would be measured at a range of frequencies representative
of typical challenges to the lipstick. Measurement at 32°C makes sense as a reference point.

To complement the DSC data, a standard low-strain, 1/s measurement should be made across a relevant
temperature range to get some idea of the intrinsic physical properties (elastic/plastic) during processing.
Here are some typical scans showing what to expect. On the left is a less-routine LAOS sweep up to 100%
strain, on the right is a routine SAOS sweep over a convenient range of temperatures:

In both cases, at low strains and low temperatures, the lipstick is mostly an elastic solid, with G' 3x larger
than G". A key property of lipstick is that it should flow under strain. The strain sweep shows that above
1% strain it starts to become much weaker and above 10% strain G" dominates. The T sweep shows that

FST 69
not a lot happens to G' and G" in the DSC region below ~55°C, above which where there is a lot of melting
behaviour and then, at these low strains, the lipstick finally starts to soften significantly.

What can you do with these data? Again, on their own, not much. We discuss below how best to use them.

Adhesion and abhesion

Although we could readily add adhesion promoters to the formulation that would provide strong bonding to
the lips, these would result in a stained lip as the bulk is removed. That is unacceptable.

The only source of adhesion, therefore, is dissipation, just as is the case for pressure sensitive adhesives,
PSA. We measure the potential for dissipation via the balance between G' and G", so our rheology will
provide insights into the adhesive properties. As is the case for PSAs there is no simple formula relating G'
& G" to strong adhesion, just the notion that too much or too little of either is a bad thing. That unhelpful
remark points, again, to the discussions below.

Abhesion, the tendency to not stick is easily attained by having a high G' and low G", a brittle, high-wax
formulation. This means that we either have to compromise or use the opposite effect by providing
something that migrates to the surface to provide a thin (sub-µm is good enough) layer with low G' and G"
and which, ideally, doesn’t include much pigment. Although this can be achieved with silicones, they are
becoming less acceptable, so some oil partition is desirable – without being excessive and showing up as
syneresis or “weeping” from the lipstick (Gelling_Syneresis ). All upsides have downsides – a thin layer on
the surface of the bulk lipstick might make it too slippery to apply to the lips or might give a disconcerting
“slip” when first brought into contact.

It is common to measure variants of adhesion and abhesion: tack and friction coefficient.

• A probe tack test pushes a flat-ended cylinder (don’t use a spherical end!) onto a block of lipstick
with a known force then measures the force needed to separate it. The peak force and the
integrated force from any “stringing” of the lipstick give two measures of “tack”. Too little tack
and adhesion to the lip is likely to be low, too much and the lipstick is obviously “sticky” and
unacceptable.
• The friction coefficient (Mechanical_Friction) can be measured with various degrees of
sophistication, but placing a block of lipstick onto a surface of interest, tilting the surface and
finding the angle at which the block starts to slide is simple and sufficiently accurate. Tests on a
simulated skin surface will give an idea about applying the lipstick, and tests on a glass surface
will give an idea of whether the lipstick will slip away when taking a drink from a glass.

Finally, adhesion of the lipstick to itself (cohesion) can be measured via a standard “breaking load” test –
the weight needed to snap a sample.

Ingredient compatibility
Although we could use a tool such as Dissolution_Hansen Solubility Parameters to make sure that like is
compatible with like in the formulation, we don’t need to bother. All the ingredients to be used in
significant quantities have to be waxy or oily, which means an absence of any functionalities except the
esters holding fatty acids and alcohols together. As is widely acknowledged, trying to incorporate castor oil
with the extra -OH groups from the ricinoleate chains comes with its dangers.

Those adding polymers as film formers or adhesion promoters might add polyacrylates, but they do so

70 FST
 Lipstick

ensuring that a good fraction of the acrylate groups are based on long-chain molecules.

User data
Via a panel of trained or (giving a different perspective) untrained testers it is possible to score a lipstick for
properties such as:

• Ease of application
• Covering of lip imperfections
• Degree of “bleed”
• Feel on the lips
• Resistance to transfer
• Longevity on the lips
• Removability from the lips

These are the properties that matter, though different consumers have different preferences for the balances
of these features. It is not obvious how to connect the individual properties to the core science, nor how to
tune the ensemble of properties for the right part of consumer space. That is the challenge in the final
section.

Two other properties:

• Uniformity of colour
• Shine

are relatively stand-alone, though if shine is reduced for a deliberate matt look, this might help with
reducing transfer.

The uniformity of colour requires a good dispersant package for the pigments which, as mentioned at the
start, is out of the scope of this chapter, though is relatively straightforward to arrange via the right
dispersion science.

For simplicity these two factors are not discussed further.

Navigating through lipstick formulation space

Ignoring the pigment package, antioxidants and fragrances, the challenge is to link our core science values
to individual properties and to enable the trade-offs of consumer preferences.

The need for this arises because formulation space is vast. If a typical “lean” formulation has “only” 3
waxes and 3 oils they have to be chosen from a palette of, say, 10 of each with multiple ratios of each
component to be explored. Some reduction in that palette can be made if marketing insist on “chemical
free”, “all natural” or “low cost for high margin”, but that still leaves an impossible space to explore,
especially as the user tests are slow and expensive and, maybe, too late in the development process.

Let’s assume we have set ourselves up to do the minimax set of scientific tests on a wide range of samples.
Minimax is the minimum number of different experiments giving us the maximum number of parameters
relevant to optimising in formulation space. The experiments themselves need to be routine and reliable,
with data analyses (e.g. from LAOS sweeps) automated to produce the parameters we want to use. A
proposed minimax is:

FST 71
 • DSC to give you a map of the melting points and crystallinity
• SAOS and LAOS experiments. If you can’t manage these then a rotational viscometer
(Flow_Rotational rheology) able to cope with these semi-solids, with their yield stresses
(Flow_Yield Stress) and shear dependence (Flow_Shear dependent) might work.
• Friction coefficient, either simple tilt or a more sophisticated speed/load-dependent device must
surely feature.
• The tack test, with due thought about initial loads and probe speeds, provides nice data for
relatively little effort.

In the early days, don’t bother about your own formulations. Buy in, and test, a big range of lipsticks of
well-known brands with plenty of user information about them: Soft, Long Lasting, Budget, Luxury,
Natural, … which you can approximately map onto the user data classifications defined for your own
formulations.

Now you have the data analytics challenge of mapping scientific values against user perceptions. At the
start this will be “dumb mapping” as you don’t know what you are doing. You can do this via some
mindless algorithm that produces a PCA or ML or AI fit to every measured parameter. You gain precision
but lose insight. Instead, the first task is to reduce internal duplication. If analytical parameter A correlates
strongly with analytical parameter B, and if some reflection tells you that they are different ways of looking
at the same physics, then choose whichever is easiest to get and remove the other from the fitting. And try
to do fits to sets of parameters your scientific intuition says should work well. If the fits are OK then it’s
usually better to go with scientific meaning rather than overfitting to higher accuracy. With these scientific
fits, outlier data points are a valuable challenge to the theory. If they force you to modify your theory, that’s
called good science.

Now you have a validated “smart mapping” process, you can use it in reverse with your new formulations.
As you alter ingredients and their ratios, you see how they are affecting key scientific parameters and can
quickly sort out a response surface that should get you to a desired point in formulation space. Because you
are using minimax analytics, you can justify the resources needed to do this mapping – you already know
that you are not doing worthless or duplicate tests (they have been eliminated by the original analysis) and
that what you are measuring provides a direct route to whatever optimal balance marketing have requested.

The point of smart mapping is that it can make itself smarter. At first you might have no idea why waxes
ABC work together better than DEF in the context of oils UVW yet it’s the other way round with oils XYZ.
As the data accumulate you can start sub-mapping of the wax and oil domains. Each individual component
has its own melting point, viscosity etc. so now these parameters feed into the overall map, improving both
the map and the scientific understanding of which features interact and why.

Why not let the AI do all this? Because unless you have set up to do this sort of mapping and assembled a
minimax set of reliable data points, you don’t have datasets worth plugging into the AI. In the end, the
smart thing to do might be to ask the AI not only to provide the formulation predictions but to explain why
they work. But you first have to do the hard work

72 FST
 Mascara

Mascara
Links
Adhesion_Adhesion promoters, Adhesion_Surface energy Interactions, Cleaning_Boundary removal,
Cleaning_Solvents, Cleaning_Surfactants, Cleaning_Temperature effects, Diffusion_Barrier Properties,
Diffusion_Basic Diffusion, Diffusion_Diffusion into Skin and Hair, Diffusion_Permeation OTR and
WVTR, Dispersions_ODC, Dispersions_PVC and CPVC, Dispersions_Rheology (Low shear),
Dispersions_Settling, Flow_Basic viscosity, Flow_G' and G'', Flow_Oscillatory rheology, Flow_Rotational
rheology, Flow_TTS-WLF, Flow_Yield Stress, Optics_Colour

How do you rationally formulate for a suite of properties such as curling, volume, shine, smudge/
crack resistance, water resistance, along with smooth/easy application, long pot-life and “natural”?
By taking a step back from formulations with too many ingredients and thinking it through.

If 6 ingredients are good, 12 must be better

You can make a workable mascara at home with some carbon black, coconut oil, aloe vera gel and
beeswax, just 4 ingredients. If you want to go wild you can get extra volume from bentonite clay and
maybe some extra smoothness from shea butter which takes us up to 6.

But that shea butter is mostly oleic and stearic acid, so it’s really 7 ingredients, and fatty acids can go
rancid, so you’d better add some antioxidant, maybe Vitamin E as that sounds good in the ingredients list,
taking us to 8. And that aloe vera gel is more than 90% water so you’d better add a preservative package
which might be two ingredients such as phenoxyethanol and caprylyl glycol. So we are up to 10
ingredients.

Finally, to be really natural, add some essential oil which takes us to 11 ingredients if we are allowed to
count the oil as 1 rather than 10+ ingredients.

Looking at any commercial mascara we easily reach 12+ ingredients and when film forming polymers are
added for anti-smudge and water resistance we can head to 16+.

For example, instead of just beeswax we might find a blend of C18-36 acid triglyceride, White beeswax,
Stearic acid, Copernicia cerifera and Euphorbia cerifera. Why that specific combination of 5 waxy
materials was chosen is often unknown, but the assumption has to be that someone with magical
formulation fingers found that it gave just the right balance of final properties.

This was a difficult chapter to write. It is very easy to go blah blah with truisms, but that doesn’t
help. There are no easy, relevant formulae to provide an app-route through formulation space.
Instead we identify those measurements that must be relevant to the formulation, then work out
how to map those data against real-world behaviour of products that are known to be good or bad,
your own and competitors. This is an unfamiliar approach, but a powerful one.
Maybe as a formulation scientist it is a better idea to take a step back from the assumptions of complex
mixtures and find ways to map out the formulation domain in a way that builds a foundation for all future
formulations.

If we look at the desired properties we can plan for a minimax formulation – minimum (affordable)
ingredients for the maximum performance. At each step we can identify potential contradictions and
resolve them rationally rather than exploring an ever-wider range of ingredient tweaks that might get us
there eventually.

Taking a step back – the Water phase

We can start with the most dangerous ingredient – water. We have to make it harmless by choosing a
preservative package via whatever objective or marketing-led process makes sense. We know that
preservatives will interact with other parts of the formulation, so let’s find out those interactions sooner
rather than later.

Assuming the water needs to be thickened to make a stable emulsion and to stop the mascara dripping
during application and drying, the starting point is a marketing-led decision on the type of thickener.
Xanthan gum, gum Arabic and similar might be good for “naturals”, hydroxyethylcellulose is artificial but
some consider it natural, and something like polyvinylalcohol is purely synthetic.

Given a choice of one or two thickeners, some standard rheology, Flow_Rotational rheology, Flow_Shear
dependent, Flow_Thixotropy, Flow_Yield Stress will quickly capture the performance landscape of the
key properties we need to understand:

• Low-shear viscosity and yield stress

• Shear-dependent viscosity & thixotropy
• Concentration-dependence of low shear viscosity

You can find out the sort of values required for the first two by measuring whatever commercially
successful products are seen as market leaders. You will need a viscosity somewhat lower than the real
product because the emulsion from your other ingredients increases the viscosity by a factor easily
estimated from the desired volume fraction using the app described in Flow_Particle Viscosity.

[Link]

74 FST
 Mascara

If we assume that the Yaron, Gal-Or formula applies, that our wax+oils has a 1000 cP viscosity and that we
have 20% of it, then an initial viscosity of ~50 cP rises to ~125 cP. If you started with 500 cP then it would
rise to 1250 cP – the effect of emulsion particles is a proportionality factor, in this case just over 2.

The shear-thinning behaviour is controlled by the degree of entanglement of the polymer which, in turn,
depends how far above its entanglement molecular weight (see Adhesion_Entanglement ) you are. You
want some shear-thinning to allow the thick formulation to flow easily when being applied to the lashes.
You don’t want thixotropy (slow recovery back to full viscosity) because that would allow the formulation
to sag on the lashes.

Sorting out concentrations and MW to give you the desired properties is tedious, but this is giving you
permanent data on relatively pure systems that can be applied to other formulations. This is very different
from trying to optimise these things within a complex formulation.

The final pieces of data are the concentration-dependences of the low-shear viscosity when you
systematically vary concentrations of key ingredients, below and above what you expect to be reasonable.
You need this for two conflicting purposes:

1. Desirable: A high concentration-dependence gives a rapid increase in viscosity when drying on the
lashes so the final mascara properties are attained before drips or sags can happen;
2. Undesirable: That same high concentration-dependence gives a rapid increase in viscosity in the
mascara container over days or weeks of use.

Building up this concentration-dependent and MW-dependent information is tedious, but it is timeless data
for current and future formulations.

Taking a step back – the Oil phase

Again the starting point is marketing requirements for oils, fatty acids and waxes that are acceptable for the
desired products. Paraffin or mineral oil might be preferred for inertness and lack of impact on food crops;
or coconut oil might be preferred because it’s natural.

The fatty acids might be dual use – surfactants/emulsifiers as triethanolamine salts in the formulation, or
simply present as individual acids or blends such as shea butter. Because the triethanolamine salts revert to
the acids on drying, for the moment we consider only their fatty acid properties in the final oil phase.

For the moment we can focus on the issues of what individual waxes do in the context of the chosen oil
phases. Whether we need to create a blend is something to be decided if there is a problem that only a blend
can solve. Again, the choice of natural versus synthetic waxes is mostly a marketing issue.

Across a reasonable range of oil/fat/wax ratios there are 3 key questions to be answered:

1. Over a user-relevant temperature range (say, 10-35°C) are the key mechanical properties stable
enough to provide protection against drooping (high T) and flaking (low T)?
2. How do the wax crystals compare to your desired size (not too small to be irrelevant, not too large
to be obviously visible), number (not too few, not too many) and stability against disappearance or
growth during storage?
3. Given that cooling rates for a batch might vary, how sensitive is the crystal distribution to fast or
slow cooling

FST 75
Crystallization in the real formulation might be affected by the pigments (discussed later), but it’s still
important to get to know the core oil/wax interactions.

The “key mechanical properties” can be most conveniently measured across a relevant temperature range
via oscillatory rheometry Flow_Oscillatory rheology to get the G' & G" values Flow_G' and G''. An over-
brittle, poor adhering mascara (maybe showing up only at lower T) will have high G' (elastic) and small G"
(plastic) values and an over-soft one (maybe showing up only at higher T) will be low G' with a lot of G".

The trick is to find a “strong” formulation that can retain “body” and “curl” and be anti-smudge, while
being soft (“rich”) and flexible. By gathering G' & G" (probably at a convenient 1/s frequency) over the,
say, 10-35°C range, any alarming changes in properties can be picked up quickly.

What are the ideal G' and G" values? Doing the same cycles of measurements on dried commercial
mascaras of known low, medium and high quality will give some indications.

With a few examples you can have a look at whether low-level additives in the oil phase, such as
tocopherol antioxidants, have unexpected effects. It is better to understand unexpected interactions at this
relatively simple phase of building knowledge than to discover them late in the formulation cycle.

Taking a step back – Pigment effects

If you have put in the effort to ensure that your pigment will be nicely dispersed in your hot oil phase then,
at the 5-10% addition range, they should make little difference to the oil. This can be seen from the
Dispersion Viscosity app where increases of, say, 25% in viscosity might be expected. If, at least for
generally spherical particles, you see a doubling of viscosity this is a signal that the dispersion is poor,
producing strong pigment-pigment interactions that you don’t want. A well-dispersed pigment (see Optimal
Dispersant Concentration, Dispersions_ODC) in the few 100nm size range gives optimal optical
performance (colour, hiding power, see Optics_Scattering and Opacity) and it is likely that it will offer
minimal interference with crystallization when typical crystals are in the multi-µm range.

Your formulation is complicated enough without having to take into account particle-particle, particle-wax,
particle-oil and particle-water interactions. Obtaining pigments with good oil dispersants, or grinding them
in the presence of some of your oil along with a good dispersant, should not pose great difficulties.

An exception is the micas. As described in Flow_Particle Viscosity, particles, such a mica, with a high
aspect ratio (length/thickness) rapidly form a percolated network of high viscosity. Even with good
dispersion, there will be significant viscosity increase, with a poor dispersant the formulation will be
unusable. Because a typical bentonite clay is hydrophilic, micas and colour-coated micas need specific
dispersants with a mica-friendly head and a long alkane tail. Tallow quats were a good example, though
tallow is out of favour in many formulations.

After the investment in all those G' & G" measurements of basic oil/wax mixes, some follow-up
experiments with pigmented versions will either provide confirmation that the pigments are mostly
harmless to your formulation, or alert you to specific issues, such as particle seeding producing different
crystallinities, or micas needing (probably) more oil and less wax to achieve the same balance of properties.

Taking a step back – the Emulsifiers

Although the emulsifier ends up in the dried-on mascara as (probably) an extra bit of oil/wax, choosing it
for its anticipated contribution to the final product is not yet a priority. We want an emulsifier that gets us to

76 FST
 Mascara

a great o/w emulsion with the least effort. Given that emulsification takes place at something like 80-90°C
in order to melt the waxes (unless our previous work encourages the use of a lower MPt wax) we need to
use a theory that understands the “oiliness” of the oil/wax solution as well as the temperature effect on the
characteristics of the surfactant (which is the word we will now use instead of the vague word
“emulsifier”). This automatically means HLD: Surfactancy_HLD.

Experience with HLD tells us that the Equivalent Alkane Carbon Number (EACN) that defines the oiliness
of these types of formulations is likely to be in the range of 15 – depending on the ratio of simple alkanes
(maybe we’re adding isododecane), esters, triglycerides and waxes. However, we know that relatively small
quantities of “polar oils”, which include fragrances and some preservatives, can change (usually reduce) the
EACN, but we can’t say by how much.

We also know that we have to think carefully about the classes of surfactants we intend to use. The
ethoxylates become much more hydrophobic at these high temperatures, with the danger of creating w/o
emulsions which are not good when you want a rapidly drying o/w emulsion. The anionics (including the
fatty-acid salts) become somewhat more hydrophilic, which is not so good for these oily oils. And APGs
and polyglycerols are temperature insensitive. Whatever the class, we need its characteristic (Cc) defined at
25°C.

To get efficient o/w emulsification we need an HLD ~ -0.25 at the high temperature. For normal o/w
emulsions we want to ensure HLD ~ -1 to give stability against coalescence and flocculation. An HLD near
-0.25 gives rapid emulsification but also relatively rapid de-emulsification, so we need to get from -0.25 to
-1 for a stable room-temperature emulsion. But given that our emulsion will be semi-solid particles, as long
as we cool quickly, we can benefit from the low Surfactancy_Interfacial Tension and Rigidity with a small,
negative HLD to give us a fine emulsion with little need for high energy dispersion.

Once again, marketing have to rule on which classes of surfactants are acceptable. Once we know the
classes we do a quick HLD calculation to tell us what starting Cc we need to give us a good emulsification
then we find the 2-surfactant blend (usually no single surfactant is a perfect match) that gets us to that
point. This should be done via manufacturers’ lists of Cc values, but industry is only slowly giving us the
data we need.

Looking at the list of Cc values on [Link] we find that

the fatty acid salts are probably of little use (though the table only lists the Na salts, not the amine salts) as
they are too hydrophilic, even more so at high T. The Spans, some APGs and polyglycerols look in the right
sort of range and their temperature insensitivity is a bonus. There are plenty of ethoxylates that can do a
good job at 80°C but they will be much too hydrophilic at low temperatures, which might be a problem.

Given the double uncertainty of EACN and Cc values, how can we formulate rationally? By doing
intelligent scans:

FST 77
[Link]

Here we have 10 tubes containing a 50:50 mix of our oil and water+package phases. We know we have
some nice hydrophobic sugar- or glycerol-based surfactants, maybe in the Cc range of 1 to 3. So in each
tube we place 4% of consecutive mixes of the surfactants to take us from a nominal Cc = 1 to Cc = 3. We
shake up the tubes in their 80°C water bath and allow them to phase separate. When we see that (in this
case) tube 6 contains the 3 phases characteristic of HLD = 0, we have a good idea of where we need to be
and, specifically, the mix in tube 5 (HLD ~ -0.25) should be a good starting point for emulsifying our
mascara.

This is highly idealised and reality doesn’t give you a great scan right first time. The payoff of working to
get good scans is that it is then easy to, say, change preservative packages, or make gradual changes to your
oils/waxes. If the optimum tube changes slightly then you have expanded your range of easy-to-swap
ingredients if some sudden new requirement emerges. If the optimum tube changes significantly then you
now know that ingredient swapping will require more care, with adjustments to the surfactant package to
accommodate the change.

Bringing it all together

It has been a lot of work. But now you have a palette of scientific information that you can start to combine
in real formulations. And you have the basis for a rational digitalisation process. When the individual
components come together with only modest new interactions, you know in advance how to tune some
ratios to quickly get you to the desired end point. When they come together and give very different
properties than those you expect, that can be welcomed by optimists as a sign of an interesting effect –
interestingly bad or good, it’s all valuable information.

Had you tested that formulation without the preliminary data, there would be no way of knowing that some
synergy or antagonistic effect had taken place – it would be just another confusing datapoint in mascara
formulation space.

78 FST
 Mascara

Bells and whistles

You might still be required to add some exotic plant extract for marketing reasons. With the efficient
science-based methodology, you can quickly test its effect on whichever phase it is going to be added in
(water or oil) and with luck you will see, at worse, the need only for a minor tweak.

The film-forming polymers for greater water resistance are more of a problem. Some are water-based
(acrylics), with a strong pH dependence, e.g. requiring the triethanolamine salt to be water soluble,
reverting to insolubility when the triethanolamine evaporates, leaving the insoluble acid form. Others are
oil based and some formulations even use mixes of water-based and oil-based polymers to make life even
more complicated.

Getting these polymers to be right, not too rigid, not too glossy/plastic-looking, not too hard to remove is a
whole topic beyond the scope of this chapter. But maybe the rational approach discussed here can get you
to the desired properties without those extra ingredients. A glossy, water-resistant, well-adhering oil/wax
combination is a real possibility and marketing can put “no nasty polymers” onto the packaging for the
many customers who like pure simplicity.

FST 79
Microencapsulation
Links
Absorbency_Porosity and Darcy, Diffusion_Basic Diffusion, Diffusion_Concentration-Dependent
Diffusion, Diffusion_Diffusion Coefficients, Dispersions_Settling, Dissolution_Flory-Huggins,
Dissolution_Hansen Solubility Parameters, Dissolution_Ideal solubility, Dissolution_Solubilizers and
Hydrotropes, Flow_Basic viscosity, Flow_Particle Viscosity, Flow_Polymer Viscosity, Fragrance_Activity
Coefficients, Fragrance_Barrier Properties, Fragrance_Vapour Pressure

We have 3 broad ways to make microcapsules:

1. Covering solids with a spray

2. Making a solution or emulsion and spray drying it
3. Making an emulsion and adding a shell

Here we are interested in not so much in microencapsulation itself but the core science that a
formulator can use to influence the process during making and then using the capsules.

Release from the capsule

There is no point in making a microcapsule if its
release properties are inappropriate for the desired
application. So let’s make sure we understand
release mechanisms before working out the best
way to make the capsules. Although what follows
sounds as if it is specific to monocore, there is
little difference in principle if we have a polycore
or matrix form.

Mechanical fracture

A typical microcapsule will need to be robust enough to withstand mechanical stresses during manufacture
and general use, yet weak enough to rupture under the desired stress conditions, such as when scratching a
film or when a fabric is being worn. So it should be standard practice to measure the relevant parameters
and compare them to the expected stresses.

However, the science is murky and indirect. For example, the force, F, needed to deform a capsule of radius
R by amount δ depends on the Poisson ratio ν which is either 0.3 or 0.5 (in reality it makes little difference
which you choose) and the modulus E of the capsule:

F=
√
4E R δ1.5

3(1 − ν2)

This Hertz model seems to say that larger capsules are stronger – bigger R means bigger F – but because
 Microencapsulation

2 1
we are interested in the stress (force per unit area), we have to divide by πR , meaning that there is a
R1.5
dependence on effective strength – bigger capsules are easier to break.

What about the modulus, E? This is some combination of the modulus of the contents and the modulus of
the shell, with the combination depending on the ratio of shell thickness to capsule radius and the ratio of
the shell modulus to that of the contents. Details are again sketchy but we can probably assume that if the
contents are purely liquid then doubling the shell thickness will double the strength. And maybe doubling
the modulus of the shell material will double the strength … for a constant δ.

This leaves us with δ itself. If you add an extra shell of silica to a typical shell of melamine-formaldehyde
then you increase both the shell thickness and its modulus. But as the shell is now more brittle, you will
reduce δ. It’s the classic dilemma – “stronger” can mean “more brittle” which can mean, on balance,
weaker.

This leaves the formulator with the general idea that thicker, stronger shells and smaller capsules will be
either more robust (good) or hard to break for the user (bad). If you aim for thicker shells you automatically
decrease the loading of your ingredient, which is not good. If you aim for stronger shells, these might be
brittle, leading to a smaller δ and a weaker capsule. All of which, sadly, says that you have to resort to trial
and error. That is not a sentence we want to include in the FST, but we can only work with whatever the
literature provides.

Porous escape

Approximately speaking, if you have a porous capsule, escape will be “fast”. Although we could think of
applying Absorbency_Porosity and Darcy, with the typical thin walls and small capsule sizes, release will
be effectively instantaneous.

Diffusion

The laws of diffusion (Diffusion_Basic Diffusion) tell us that the steady-state rate of diffusion or flux, F,
out of a capsule with shell thickness h, diffusion coefficient D and concentration gradient C is:

DC
F= h

Because, for practical/economic reasons, you can’t change h by more than a factor of ~2 in the real world,
you can’t make much of a difference with shell thickness.

The next factor is the diffusion coefficient, D. Although it would be nice to aim for a superb, highly
crystalline polymer with D < 10-10 cm²/s, most shells are created very rapidly and are too disordered for
such a low D. Although crosslinking can lead to low D values, again a typical crosslinked shell does not
have an impressively low D. Larger molecules have a lower D value. If D for a molecule with MW = 100 is
4.109 then, assuming a MW² dependency, a molecule of 200 Daltons will have D = 1.109. Clearly D is an
important factor, but if the shell’s other desirable properties don’t come with a low value of D, what else
can you do?

The final factor is the solubility of the molecule in the shell. The rate of diffusion is driven by the
concentration gradient, C, from inside to outside, and we can define this as the saturated concentration of
the molecule within the inner few nm of the shell if we assume that the concentration outside is effectively

FST 81
zero. Because of the small effects (factor of ~2) of shell thickness and MW and with only a modest ability
to lower D for a given type of capsule, your flux is largely driven by C, which can be a factor of 1000
because the range of solubilities can be anything from, say, 10% down to 0.01%.

How do you control the solubility? As so often in this book, the simplest way is via Hansen Solubility
Parameters, HSP: Dissolution_Hansen Solubility Parameters. By knowing the HSP of your shell polymer
and that of the molecule(s) inside the shell, it is easy to calculate the HSP Distance which tells you whether
the solubility is low, medium or high. An important consideration here is that if you have a mixture of
molecules you can calculate the average Distance and, therefore, get a general idea of solubility. But you
can also look at the individual Distance values and estimate (along with their individual MWs) which
molecules will diffuse quickly or slowly. If you have a fragrance formulation, big differences in HSP and
MW may show up as a fragrance that changes over time – with the smaller, closer molecules escaping
sooner than the larger, more distant ones.

Mixed modes

Typically, the amount released at time t, mt is related to the total mass, m0, by:

mt
m0 = ktn

For Fickian release, n = 0.5 (classic square root of time) and for the more general Ritger-Peppas (R-P)
model, n can be anything from 0.35 (somehow constrained) to 0.6 (maybe swelling helping diffusion) to 1
(the much-desired zeroth order release).

If we release into an infinite sink, so that the molecular concentration outside the capsule is always
(effectively) zero, we can obtain graphs that go from mt = 0 to mt = m0 as in these graphs:

Our porous release might give just burst kinetics. Our

mechanical restraints might give us triggered release.
We might get pure Fickian (or R-P if k ≠ 0.5) with a
steady curve, or we might get a mix of burst (maybe
molecule caught in or on the shell) and Fickian or R-P.

DC
The release coefficient, k, is basically F = h so the
different graphs (ignoring the bursts) represent either
different diffusion coefficients, D, saturated
concentrations C or wall thicknesses h.

If k is purely a function of the shell, then the size of the

capsule makes no difference. If k depends on the core
then h will be some complex mix of core and shell
sizes, so larger capsules will show slower release. Although we could produce some fancy equations for
this, in reality we are unlikely to know all the parameters so we can treat k as just another constant.

If our encapsulated molecule is a solid then k is a complex mixture of solvent (usually water) coming
through the shell, dissolving the solid and the concentration gradient then driving out the molecule. Once
again we can subsume that into some generic k. This is arguably more Higuchi kinetics than Fickian
kinetics, but as they each have a t0.5 dependency, we won’t really notice the difference.

82 FST
 Microencapsulation

If our aim is to keep the molecules in the capsule even though the capsule is in some environment where
release is possible (e.g. a fragrance capsule in a liquid laundry detergent) then we must reduce C. Although
we say that this is a concentration gradient, in reality it is a gradient against activity. So if the saturated
concentration in the external environment is very low (e.g. a fragrance oil in water), after a small mt the
concentration has reached saturation and the activity gradient is now 0, so F is 0 and k is 0.

Strong barriers

If diffusion through our shell is too fast for the intended application then we can try a few more tricks:

• Evaporate a shell of Al
• Coat with some sort of sol-gel to give something like a silica coating
• Incorporate nanoclays that provide a tortuosity barrier, as explained in Diffusion_Barrier
Properties

Each of these has obvious downsides in terms of cost and complexity. The tortuosity barrier is especially a
problem for spray-dried capsules, as discussed below.

Polycore and matrix capsules

Whether we have monocore, polycore or matrix capsules makes no obvious difference to the science. The
bursting behaviour is just as inscrutable, and the release mechanisms can follow any of the curves shown
above. This is not to say that these different capsule types are of no consequence. For a given shell, matrix
and molecule, the performance might depend strongly on the capsule type. The point here is that each
capsule type is able to produce each of the different release modes. It is likely that a monocore capsule will
tend to be Fickian and that a matrix capsule might show interesting R-P behaviour, but when you add in the
effects of capsule size distribution, damage during processing, variabilities in shell thickness and integrity,
difference between capsule types become blurred.

Making the microcapsules

We will look at the 3 most typical ways to make microcapsules:

1. Coating onto a particle

2. Spray drying from solution or dispersion/emulsion
3. Adding a core/shell to an emulsion

Coating onto a particle

There is surprisingly little that can usefully be said. Assuming, for the moment, that you are using a
fluidized bed, then your particles can’t be too large (can’t be fluidized at reasonable air velocities) or too
small (uncontrollable) and, therefore, you need control over the size distribution so all your particles are in
the viable range, see Particles_Size distribution. Then you need to spray your coating using some sort of
atomizer where, again, the bad news is that there is little useful formulation science (Flow_Atomization)
other than the obvious advice to have low viscosity and an absence of viscoelasticity.

Whether you use top spray or bottom spray or a Wurster device is usually a given so you have to formulate
with what you have.

Spray drying

FST 83
 We have the same problem as with coatings. Spray drying,
mostly from water, is mostly about getting a good atomizer
and good airflows, neither of which is under the formulator’s
control. In this section we don’t worry if we have a solution
or an emulsion. The emulsion aspects are described in the
next section.

There are difficult tensions over which we can have some control.

• Production will always demand higher concentrations (less water to evaporate) and lower
viscosity (easier atomization) which are usually contradictory.
• They also want a rapid transformation on the outside (crust formation) so that particles that touch
each other or the wall don’t stick, without that crust inhibiting rapid evaporation of the water. Such
inhibition can slow down the process or, at worst, lead the water to explode as the internal
temperature exceeds 100°C.
• We want maximum hot air temperature for rapid drying, but we don’t want to destroy the contents
of the particle, such as enzymes or friendly biome microbes.

Let’s try to do something about those contradictions.

Maximum concentration, minimum viscosity

If our water contains particles or (it’s the same thing) emulsion drops, then the same app shows the
viscosity dependence on volume fraction, φ and on the viscosity of the dispersed particle (a few cP for a
typical oil and for a solid particle 10,000 cP is good enough:

[Link]

This tells us that without particle-particle interactions, just them getting in each other’s way, the viscosity
doesn’t change much from water’s 1cP till about 50% volume fraction when it reaches 10cP and is starting
to grow rapidly.

But be careful. If your “water” starts at 10cP because of some soluble additive, then at 50% volume
fraction the viscosity is now 100cP – the crowding effect is multiplicative, not additive. If you really need a
few % more particles, do whatever it takes (and it’s often relatively easy) to reduce your “water” viscosity
by a few cP. This is an important and little-known trick.

84 FST
 Microencapsulation

If, however, your particles show significant interactions because, for example, you have a poor dispersant
on solid particles, or because (and here is one warning about nanoclays for extra barrier properties) they
have a high aspect ratio (length/thickness) then viscosities can shoot up.

We can explore these ideas in Flow_Particle Viscosity which includes this app:

[Link]

As you play with the sliders (the N and Fractal Dimension sliders describe particle association, there is also
the Aspect Ratio slider) you can see two things:

• It is scarily easy to shoot up to super-high viscosities at low shear.

• But if your pumps are good enough to create high shear in the atomizer nozzle, then maybe you
will be OK.

If you are going to rely on the very strong shear thinning of particle formulations, make sure you have
tested many likely variations in production batch formulations. It is very easy to go from a 1000x low shear
increase in viscosity to 10000x with a small change in dispersant or aspect ratio.

If you are dealing with polymer solutions then you need to be familiar with Flow_Polymer Viscosity as
shown in the app:

FST 85
[Link]

The core idea is not so much that higher polymer concentrations lead to rapid increase in viscosity (we all
know that), nor that high MW polymers are worse than low MW (again we know that). The key is MW
compared to Mc, the “critical entanglement MW”. You can have two similar polymers, with similar
solubilities and similar MWs, but one of them will be much more viscous. This is because the number of
tangles depends on MW/Mc. If the two polymers have Mc values differing by a factor of 2, then the one
with the lower Mc is twice as tangled so will be more viscous.

In many applications (Adhesion_Entanglement) you want lots of entanglement, so you have a real fight
when increasing concentration and MW. For spray drying you may not need so much entanglement so can
choose either a lower MW or a polymer with a large Mc, so a high MW polymer (desirable for other
reasons) becomes acceptable.

However, if you want a low diffusion coefficient for your encapsulated molecule in the polymer, all other
things being equal, an entangled system leads to lower diffusion. Everything in formulation is a trade-off.

Another way to decrease viscosity of a polymer solution is to decrease its maximum solubility. Compared
to a “happy” polymer with chains reaching out and tangling, the “unhappy” polymer is more coiled up on
itself so isn’t so tangled. The obvious flaw in this approach is that you need a high solubility to get enough
polymer into solution.

Fast drying, all the way to dry

Everyone who dries coatings wants maximum drying speed with minimum formation of a crust that inhibits
further drying. See Evaporation_Basics and Evaporation_Diffusion limited for a more detailed discussion.

The key idea is that in “constant mode”, which is evaporation-limited drying, you are relying on a
combination of heat and mass transfer, readily arranged in a modern drying system. As soon as you become
“diffusion limited”, you can’t dry faster than the solvent gets through the crust, which becomes slower as
the crust thickens, then there is nothing you can do other than increase temperature because diffusion goes
faster at higher temperatures. High temperatures of a coating or a particle are generally not desirable.

86 FST
 Microencapsulation

So how do we keep our crust “open” enough to keep the drying evaporation limited? There are a few
standard tricks:

• Add a “high boiler” (e.g. propylene glycol) in which your formulation remains at least partially
soluble. This keeps the crust “open” for much longer.
• Use the formulation tricks for highest concentration with minimum viscosity to keep your
formulation “happy” for as long as possible. You don’t care that the shell is now very high
viscosity, you just want to make sure that there is still enough water for diffusion to be rapid and
not rate limiting.
• Add some junk (e.g. rough particles) that stops the crust getting too compact. There’s an obvious
downside if this makes the capsules too leaky, but with some thought you can get a reasonable
compromise. If those particles are sticking out they will also reduce capsule-capsule and capsule-
wall adhesion while the drying is incomplete.

This problem of the crust becoming too good a barrier is especially severe for those trying to use nanoclays
as tortuosity barriers. There are no obviously good workarounds for this.

Keeping cool

How do we spray dry heat-sensitive enzymes or microbes in 140°C (413°K) air, Tgas, without destroying
them? A handy approximate formula (found in the app below) gives us the answer for water with a BP, Tb
= 373°K. It tells us that the temperature of the microcapsule, Tmc is given by:

Tb 0.68
Tmc = 137 ( )
373 log10 (Tair) − 45
For Tb=373 then we have 137 log10(413)-45 = 313°K = 40°C. Yes, our rapidly evaporating water is only at
40°C, so our enzyme or microbe inside the capsule is OK. If, however, we create a barrier crust then the
capsule rapidly approaches 140°C and we have failed. Avoiding diffusion-limited drying really is important
for spray drying.

A spray drying app

For those keen to explore some of the complexities of a pharma spray-drying system, this app gives more
than enough to think about:

[Link]

FST 87
The meaning of the multiple inputs and outputs are described on the app page.

Why only this one app? Because the spray drying community has not provided any other usable models.

From emulsions to capsules: Getting the emulsions

One route to an emulsion-based microcapsule is to make an emulsion and spray dry it. The discussions here
are tangled up with the more general principle of using the emulsion as the basis for the whole capsule. If
you are oriented towards simple spray drying, you can ignore the other aspects.

In other parts of the FST we treat emulsification as a relatively simple process of putting an oil into water
or water into oil. Here we focus on “oil” in water, but now our oil is much more complex; it might be the
content of the capsules, or it might be an oil containing the real content dissolved within it. In these cases, a
shell has to be provided around the emulsion. Or it might be a hydrophobic pre-polymer emulsion
containing the content, with the aim being to create a “matrix” capsule consisting of content dissolved in
the crosslinked polymer, with no need for a shell. Or it might be a hybrid – a matrix plus a shell.

The FST approach to emulsification is to use HLD theory, Surfactancy_HLD, as the backdrop to
understanding Surfactancy_Emulsification itself.

The reason for using HLD is that we can characterise the oiliness of the oil (its EACN, Equivalent Alkane
Carbon Number), the characteristic properties of the surfactant (blend), Cc, and any salinity, S and
temperature effects, T. We can tune the system to get HLD ~ 0 to allow easy emulsification with low
energy, then move to HDL ~ -1 to give a stable o/w emulsion. This rational approach avoids the trial and
error necessitated by classic surfactant science which has no coherent way to balance EACN, Cc, S and T
effects.

The obvious problem applying this to microencapsulation is that other than for the first type of capsule
delivering a simple oil, we don’t have classic oils that are likely to follow nice EACN rules. What is the
EACN of a simple oil containing a 20-component fragrance and some crosslinkable pre-polymer? No one
knows.

Here is where some patience in the early formulation steps saves a lot of time later on.

The assumption in what follows is that you already know the types of surfactant you prefer to use: anionics,
cationics, ethoxylates, sugar-based, polyglycerols, and their level of greenness. This means that you have
some idea of how temperature and/or salinity will affect your chosen surfactant class. It is also assumed
that you have some choice, within your range, of hydrophilic and hydrophobic variants. And although you
might not know the Cc values of all your chosen class of surfactants, you have at least some idea of the Cc
values of at least two of them. If this paragraph is mostly meaningless, then invest some time in the HLD
chapter before proceeding.

88 FST
 Microencapsulation

[Link]

The idea is that we start with our base, carrier, oil. Using two surfactants, (in this example, one hydrophilic
(Cc = -1) and one hydrophobic (Cc = 3)) we do a scan of 50:50 o:w of stepwise mixtures from left to right
and find in which tube we change from HLD < 0 to HLD > 0. We are not interested in 50:50 o:w nor are we
especially interested in the base oil (though it plays an important role so if we change it, everything will
change along with it). Instead we are pleased to have one nice point in the formulation map relevant to our
spray drying. Although we say “water” you might choose to start with whatever aqueous phase you will be
using (containing a range of ingredients) or you might prefer to use pure water then (not discussed further)
do a similar process to find the effects of the extra ingredients.

Now we repeat the scan with a small % of, say, the oil-soluble molecule we wish to deliver. Maybe the scan
shifts one tube to the left. We then double the % and the scan shifts 2 tubes. We now know that our
molecule is making things effectively more hydrophilic, and we have some approximate rule for the shift in
HLD versus concentration.

We now repeat this but adding our pre-polymer. Maybe this shifts things to the right.

Assuming you are doing this via some sort of robotic system (the scans are simple but tedious), you quickly
build up a map of what’s going on. You can now see if the effects of the molecule and the pre-polymer are a
linear mix (which is nice) or have some non-ideal mixture interactions.

Why is it worthwhile to have done all this? There are three reasons:

1. If by some rational trick you can set HLD ~ 0 during the emulsification stage you can get a fine
emulsion easily, with little expenditure of time and energy. You then need to complete the trick by
shifting -1 < HLD < -0.5 to get a stable emulsion for further processing (discussed shortly). This is
easy (but costs energy) to do via raising the temperature in ethoxylate systems (so-called PIT
method) and requires other tricks for ionics and the sugars/glycerols.
2. If you choose to emulsify with -1 < HLD < -0.5 (without the HLD ~ 0 trick) then you get the
most emulsion for least cost/effort by using the correct surfactant blend
3. Any time you change your formulation, you can readily adjust your surfactant blend to stay in the

FST 89
 optimum zone for emulsification.

By bringing control to the emulsification step and by understanding why well-intentioned changes in other
parts of the system (a different oil, prepolymer etc.) can disrupt the emulsification, the whole process of
product development and improvement is easier. Suppose, for example, that marketing want you to
abandon, say, ethoxylates because of some internet scare. Without the understanding from HLD it would be
a nightmare to find a replacement using, say, APGs. With the understanding, you can readily find an APG
blend that gives an approximation to the equivalent ethoxylate blend and if you have relied on PIT you
don’t have to waste time with fruitless temperature cycles – you have to get to HLD ~ 0 by Cc blending.

Why do we want -1 < HLD < -0.5? What we really want is high concentrations of stable emulsion particles
that don’t give Surfactancy_Emulsion Coalescence or Surfactancy_Emulsion Creaming and Flocculation.
We know that when HLD ~ 0 the interfacial tension is wonderfully low for creating emulsions but that also
means that the emulsions coalesce easily. When HLD > 0 we are in the w/o region that we must avoid, And
when HLD < -1 the surfactant is likely to be very happy in the water rather than at the interface, so is less
efficient for stabilizing the natural o/w curvature we have obtained.

From emulsions to capsules: Getting the capsules

There seems to be little incisive information for the formulator. If you are creating the shell via an
interfacial polymerization, say an isocyanate in the emulsion core and a polyamine in the continuous phase
then you have complicated kinetics and thermodynamics with few reliable tools to inform your choices. If
you are creating a water-insoluble polymer via water-soluble pre-polymers then it’s the same thing – you
have complicated kinetics and thermodynamics leading, one hopes, to the polymer forming around the
emulsion drops rather than as gunk within the reactor.

And if you are relying on complex coacervation, Surfactancy_Coacervation, you are entering a world with
plenty of general ideas, lots of handwaving and little usable theory.

Let me know of any FST-style science that can provide some usable ideas so this chapter doesn’t just end
like this.

90 FST
 Pharma Formulation

Pharma Formulation
Links
Diffusion_Concentration-Dependent Diffusion, Diffusion_Diffusion into Skin and Hair,
Dissolution_Crystallization, Dissolution_Dissolution Kinetics, Dissolution_Dissolution Rheology,
Dissolution_Flory-Huggins, Dissolution_Hansen Solubility Parameters, Dissolution_Ideal solubility,
Dissolution_Kirkwood-Buff, Dissolution_Solubilizers and Hydrotropes, Flow_Basic viscosity,
Flow_Couette Flow, Flow_Oscillatory rheology, Flow_Poiseuille Flow, Flow_Rotational rheology,
Flow_Shear dependent, Flow_Yield Stress, Surfactancy_Coacervation, Surfactancy_Emulsification,
Surfactancy_Emulsion Coalescence, Surfactancy_Emulsion Creaming and Flocculation,
Surfactancy_Emulsion Ostwald, Surfactancy_HLD, Surfactancy_Phase Diagrams

To deliver a dose of an Active Pharmaceutical Ingredient, API, other than via direct injection
(ignored in this chapter) requires it to be dissolved or dispersed in some delivery medium, either as
a solid or liquid formulation which dissolves at the desired rate once ingested, or as an external
cream or patch for delivery through the skin. Here we look at a range of challenges to getting a
viable delivery vehicle for your chosen API

Getting a pure API

Assuming you want to purify by crystallization, the first task is to go to the Dissolution_Crystallization
chapter to discover why something that seems so simple is so tricky. It is made all the harder because the
standard theory is provably wrong, and there hasn’t been an adjustment to the newer ways of thinking about
how to obtain the right crystals at the right speed from the right solvents in the right polymorphic form.

Given that controlling solubility either via temperature or evaporation is key, having a good theory of
solubility and temperature dependence is vital. The starting point is the idea of ideal solubility,
Dissolution_Ideal solubility, where, approximately, the maximum possible solubility is given by a
combination of MPt and Heat of Fusion. If you have molecule with strong crystal packing then both the
MPt and heat of fusion will be large so solubility, whatever you do, will be low. Some (not very good
tricks) for dealing with this are discussed below.

Assuming you have a sufficiently high ideal solubility, finding the right solvent requires a good solubility
tool, which means COSMO-RS. The starting point for COSMO-RS calculations is either the ideal
solubility or the measured solubility in one solvent. What you need to find is a solvent with a rather low
solubility at the lowest temperature of your crystallization process (it might be near 0°C) and a high
solubility at the highest reasonable temperature. The constraint is the relatively small number of solvents
generally acceptable to pharma. It would be useful if there were understandable rules for why different
solutes have different temperature dependencies of solubility, but it mostly comes down to subtle entropic
effects that are not captured in any intuitive way.

Getting the right polymorph

As discussed in the Crystallization chapter, your API crystals can grow in different crystalline forms,
polymorphs. The different crystal packing of the different forms produces different shapes, MPts and
enthalpies of fusion and, automatically, different solubilities. Given that pharma is all about certainty of
dose, it is a requirement to produce the same polymorph every time. It is a catastrophe when something
changes in a plant and the polymorph that has been grown for the past 10 years is replaced by a different
one caused by, it is assumed, minute traces of seeds of the new one brought in to the plant as a contaminant
(yes, such stories are true).

The problem isn’t so much that the different solubility makes the new polymorph unacceptable – changes
in solubility between polymorphs are usually minor. The problem is the kinetics of dissolution. If your drug
has been designed around a release profile of, say, 1hr, then a factor of 2 in the kinetics is unacceptable.

The kinetics are, in turn, affected by two effects:

1. The way the crystals interact with the excipient (e.g. lactose) used in creating a tablet can have a
profound effect on the tablet’s rate of dissolution, affecting porous diffusion, mechanical strength
and available surface areas of contact.
2. Crystal dissolution is limited by the rate at the slowest face, and if the crystal packing along this
face is especially strong, dissolution is slowed down.

These effects end up affecting the “diffusion layer” in the standard Noyes-Whitney dissolution model:

[Link]

That diffusion layer is an imaginary distance between saturated concentration at the surface of the crystal
and the current bulk concentration – the greater it is, the slower the dissolution.

Trying to increase aqueous solubility

There are two types of problem of getting your API delivered in an aqueous environment.

1. The API is relatively happy in water, but the ideal solubility is less than the required dose;

92 FST
 Pharma Formulation

2. The ideal solubility is fine in some solvents, but the API isn’t happy in water.

We discuss each in turn.

Ideal solubility too low

If the API is a salt, then changing the counterion to give a worse crystal packing and a higher solubility is a
root-cause cure, but we now have a different molecule so we go back to an earlier stage in development.

Assuming we can’t change the molecule, then we

have two tricks, neither of which works well in
practice.

The first uses the Ostwald-Freundlich relationship

which shows that for super-small particles with a
large surface area to volume ratio, the surface
energy effects overwhelm the crystalline packing
effects and solubility increases. Where S0 is the
original (low) solubility, the solubility S for a radius
r, surface energy γ, molar volume Vm and the usual
gas constant/temperature term RT then

( )
S 2γVm
ln S0 = rRT

This sounds a great idea till you see, from the above graph, how fine you need to grind your particles.

Even if you succeed at getting your 5nm particles,

Ostwald intervenes, this time in terms of Ostwald
Ripening where larger particles grow at the
expense of smaller ones – driven by the same
surface energy forces that enabled the higher
solubility. The theory tells us that the average
radius at time t, ⟨rt⟩ depends on the starting radius
⟨r0⟩, the surface energy γ, the diffusion coefficient
of a typical molecule through the solvent (~10-6
cm²/s), the mole fraction solubility of the solute
(obviously a low value, but still finite), and, as
above, Vm and RT:

3
⟨ rt ⟩ 8γDct
3 = 9RT
⟨ r0 ⟩

With c = 0.0001 mole fraction, a low solubility, then our 5nm particle grows to 250nm in 7 minutes.

The obvious fact that Ostwald ripening essentially destroys the prospects of using Ostwald-Freundlich to
enhance solubility has not deterred countless researchers attempting to fight the laws of thermodynamics …
and losing.

FST 93
The second trick is to crash-precipitate the API so that it solidifies as amorphous particles. In the lab you
can inject a solution in a water-miscible solvent, maybe with some ultrasonication, into water and remove
the solvent via, say, dialysis.

With no crystal packing energy, their solubility is greatly increased (easily 10x). By making them
nanosized, but in the 100’s of nm range, their kinetic solubility is also increased. But we hit the same
problem of Ostwald ripening, the particles, over minutes or hours steadily grow not only to a slower-
dissolving size but also risk the chance of crystallizing over time without the constraints of the nano-sized
particles.

Fortunately, as described in more detail in Surfactancy_Emulsion Ostwald there is a neat trick. If you can
co-dissolve a very water-insoluble molecule into the API then as the API starts to migrate from smaller to
larger particles, the concentration of the co-solute increases and the thermodynamics (call it osmotic
pressure) brings the ripening to a halt.

Assuming that the amorphous particles are indefinitely stable against crystallization, it might seem that
everyone would be using amorphous formulations. For lab and development experiments where having
high concentrations of an API is useful, these formulations are of proven utility. It is worth the effort to find
a pharma-acceptable Ostwald inhibitor. For end-user formulations there are a few problems:

• Regulators are happier with crystalline purity compared to amorphous materials

• Everyone is nervous about shelf-life. What might trigger the amorphous material to crystallize,
maybe sitting on a shelf in a warm pharmacy?
• If there is a trace of crystalline material then a sort of Ostwald effect takes the API from the
amorphous particles and the crystals grow rapidly, with crystal packing forces overriding the small
osmotic pressure effect of the inhibitor.

Solubilizers

If the ideal solubility is OK but the API doesn’t like water, then one obvious approach is to add some co-
solvent, such as DMSO. This might have been acceptable for liquid formulations, but there are obvious
issues of acceptability. And this doesn’t help for a solid tablet where we need to increase solubility in the
stomach.

So we add a “solubilizer” – a vague term that has caused much confusion. The root cause of the confusion
is the well-known fact that some surfactants can increase the solubility of an API when it associates with
the surfactant micelles. Tweens are the classical type of surfactant used for this effect.

94 FST
 Pharma Formulation

[Link]

In this example we have Tween 80. The API has a starting solubility of 0.6mM and with 1% Tween we
have a 4x increase. The hardly-visible kink in the curve near the start is where concentration rises above
the CMC – below the CMC there is no solubilization effect.

The usual explanation for the solubilization is that the API is hydrophobic (which is why it’s insoluble in
water) and the Tween’s C18 tail is hydrophobic so “therefore” the API is dissolved into the tail. This is
obvious nonsense. The tail is something like hexadecane and the API is probably less soluble in
hexadecane than in water. So where is the API? That’s a good question. With the PEO head of Tween 80
there’s a good chance that at least some of the API is happier in that environment than in the water. In the
app, if you know the solubility enhancement at a given % surfactant you can slide the MSR slider (Molar
Solubility Ratio) till you get the correct increase. Then you can look down at the number of solutes per
micelle and, from the solute’s size, what the chances are that they are filling the hydrophobic core or
occupying the surfactant’s head surface area. In reality it is likely to be a bit of both.

When we move away from real surfactants we enter a confusing world. Small molecule solubilizers are
often called “hydrotropes” though the word has multiple fuzzy meanings. A typical hydrotrope experiment
shows an S-shaped curve, as in the app, with an MHC, minimum hydrotrope concentration that reminds
people of the CMC.

FST 95
[Link]

Instantly we then get theories saying that the hydrotropes produce pseudo-micelles and that’s a pseudo-
explanation of the hydrotrope effect. As explained in Dissolution_Solubilizers and Hydrotropes this
explanation is not only wrong (urea is a hydrotrope but shows no self-clustering) but backwards – those
hydrotropes that self-cluster are less efficient than they otherwise would be. The real explanation, via
Kirkwood-Buff theory is “solute-induced hydrotrope clustering”. Although it is great to have a correct
theory, sadly we have no predictive model based on it. At least by removing the bad theories there is hope
that someone will put the effort into finding ways to predict the optimum hydrotrope.

Another class of solubilizer is represented by polymers ranging from the PPO-PEO-PPO triblocks to the tri-
polymer, Soluplus, with its combination of PEO, polyvinylcaprolactam and polyvinylacetate. In its solid
form it is friendly to many APIs as the HSP average of those three polymers places it nicely in the
relatively high δD, δP and δH range of many APIs – so they have good solubility for controlled release. But
Soluplus isn’t fully soluble in water so provides a complex environment for APIs to partition into in some,
as yet, unspecified manner. Hopefully someone will carry out a Kirkwood-Buff analysis of the effects of
Soluplus, which would help disentangle the solubilization mechanism.

Despite decades of research, finding a solubilizer is still mostly trial-and-error. To sort things out via big
data would require an approach that looks at different systems in a unified manner, and that can only mean
a Kirkwood-Buff analysis of the key interactions. With high-throughput and some thought, it shouldn’t be
so hard.

Tablets
Grind the API with an industry-standard excipient such as lactose, along with any sub-ingredients, add the
right amount of the mix to a tablet press, squeeze, eject the tablet and repeat. It sounds very easy. However,
squeeze a bit too hard and the tablet won’t dissolve at the required speed. Squeeze a bit too lightly and the
tablet lacks the strength to traverse the manufacturing process without falling apart.

Sadly, the state of the art in understanding how particles adhere remains rudimentary, Particles_Basic
behaviour. Different batches of the “same” lactose, without the API, can show very different compaction
and dissolution behaviours for reasons that can’t be pinned down via our standard ways of looking at the
particles. It is interesting that Inverse Gas Chromatography (IGC) can reliably distinguish between two
batches of the “same” lactose which show different tablet compaction properties. Frustratingly, no causal

96 FST
 Pharma Formulation

chain of logic has been identified to explain this, or to provide a root-cause fix.

When you then add the API and small batch-to-batch variations, the chances of unexpected tablet
behaviour increase.

The only reason for this section is to point out to the reader that their frustrations with tablet formation are
not unique to them, and the reason they can’t find the science to allow them to predict, optimize and
troubleshoot tablet properties is that the science doesn’t exist.

Controlled release
Dissolving an API in a polymer is a proven way to provide a controlled release from whatever polymeric
form is chosen. The ideal is usually a continuous, steady release. The worst case is a burst, and a typical
case is a minor burst followed by a relatively high rate falling off with time. Mostly we see a Fickian square
root of time dependency on the cumulative amount, i.e. a steady fall-off of rate. We can see these in a pair
of graphs – cumulative and daily release.

Ideal release can only be obtained by some sort of core-shell system where the release rate is controlled by
the shell. Ignoring that, to overcome burst kinetics we need the API and the polymer to be reasonably
compatible so that whatever method we use to get the API into the polymer results in a fairly homogeneous
distribution. For this, knowing the HSP Dissolution_Hansen Solubility Parameters of the API and the
polymer and calculating the HSP Distance gives us a good idea of (in)compatibility.

The square root of time behaviour can be obtained by pure Fickian diffusion from the API fully dissolved
in the polymer, Diffusion_Basic Diffusion. But it can also be found from Higuchi kinetics where the
solubility is much lower than the loading, so solid API particles are evenly distributed within the polymer.
The key assumption behind Higuchi is that the rate of diffusion of the low level of API dissolved in the
polymer is slower than the speed at which the polymer particles can re-stock the API in the polymer.

Skin delivery
The science discussed in SkinCare is generic, applying as much to getting an API through the skin and into
the bloodstream as it does to delivering the desired ingredients in a skin cream. The key is knowing relative
solubilities and diffusion coefficients of key ingredients. If you can’t get enough API into your delivery
system, then you can’t deliver it effectively. If the API is compatible with elements in the delivery system,
such as a Tween, that happen to be too large to diffuse through the skin, then the API will be trapped on the
surface. If the different components of the formulation diffuse through the skin at different rates, then the
composition on the surface will change; in the worst case this can cause a slow-to-diffuse (large MW) API

FST 97
to be stranded on the surface.

I learned this the hard way. A test formulation applied to my face used a “fast” solvent, DMI,
which disappeared relatively quickly. The large API reached its solubility limit on the skin surface
and crystallized … in the form of very painful sharp needles. I later learned of a real product recall
in countries with high humidity. The cream worked fine everywhere else, but the high humidity
caused the API to crystallize out … again into painful needles.

This is all very simple when stated, but for historical reasons the pharma skin world remained obsessed
with LogP (Octanol/Water partition coefficient), ignoring straightforward science.

Another obsession was with surfactant-assisted delivery and the supposed wonderful attributes of
liposomes and the many other “ome” variants. The surfactants aren’t surfactants after 20min when the
water has gone, and there never was a reason why a large liposome would transport an active through the
skin.

A special problem was that testing focussed on delivering from infinite doses of water-based formulations,
measured with Franz cells, when something applied to the skin is finite dose and the water has all gone
after ~20min. The more obviously sensible approach of examining what is on the skin surface over time
was never adopted, so they missed the fact that water disappears, that APIs can get trapped, that fast-
moving cosolvents might strand the API on the surface.

It is no coincidence that despite decades of research following the old methodology, the number of APIs
delivered through the skin remains very small.

Science bandwagons are an enormous waste of precious resources. It takes only a few minutes to
realise that surfactants on the skin are no longer surfactants when the water has gone in a timescale
far shorter than skin delivery. But this didn’t stop decades of naïve papers comparing skin delivery
between o/w with w/o emulsions. It equally took decades for the “ome” bubble to burst despite it
being obvious that they couldn’t do what paper after paper claimed that they must do.

Franz cells continued to be used long after it was clear that they provided the minimum relevant
information for the maximum work. Confocal Raman is a welcome but expensive way to get rather
more information. But imaginative ways to directly sample what’s on the skin surface over time,
which should be the gold-standard approach, are still rudimentary.

98 FST
 PSA

PSA
Links
Adhesion_Peel, Adhesion_Shear, Adhesion_Testing, Flow_G' and G'', Flow_Oscillatory rheology,
Flow_TTS-WLF, Mechanical_Hardness, Mechanical_Modulus

Pressure Sensitive Adhesives rely on a tricky set of compromises. They must flow easily to create
good surface contact, yet must put up a resistance to being stretched and broken. They can only be
optimized via good rheology, but you can’t (yet) create a great PSA via pure rheology because it’s
a big multi-physics challenge. They offer a reverse guarantee: the right science cannot guarantee
success, but the wrong science guarantees failure. So it’s good to get the science right.

PSAs aren’t pressure sensitive

The adhesion of a PSA doesn’t depend (much) on the pressure you apply to them. They get most of their
adhesion via spontaneous flow into perfect contact with the substrate. At a minimum, therefore, a PSA
must meet the Dahlquist criterion of a 1Hz G' (see below) no greater than 0.1 MPa. This automatically
means that PSAs are weak polymers.

Dahlquist tells us that if we have a surface like this with an

average radius of curvature R and height h then the PSA will
flow into it if its modulus is less than the critical value Gc given
by:

√
R
Gc = W
h3

You can do the calculation here (an alternative formula is provided, linked to other roughness measures as
described in Mechanical_Surface Roughness):

[Link]

There are four key take-aways from this:

1. if you have a very rough surface, most PSAs will fail because their G' will be in the 0.1 MPa
region, too high to flow into a rougher surface.
2. If you want a “stronger” PSA and increase its modulus, it will fail because it won’t flow into
contact on a typical surface.
3. To get both “strength” and Dahlquist you need tricks like using a heat-applied PSA so that G'
meets Dahlquist during application and reaches a higher value during use.
 4. You need to understand G' and G'' behaviour via rheology

How does a PSA stick?

Given that the PSA must be weak, how can it provide strong adhesion? As explained in the Adhesive
chapter, the trick is to use viscous dissipation to stop the crack energy from running along the interface. In
the language of standard PSA rheology (to be discussed shortly), we get viscous dissipation via G'', the loss
component. But although you need viscous behaviour, you can’t make a good PSA from a purely viscous
material – you need plenty of elastic strength, G'. You now see an instant contradiction – you can’t have
lots of G' because you will fail Dahlquist.

All this can only be disentangled via rheology. Here we have the reverse guarantee in the introduction. If
you don’t understand and control your rheology, failure is guaranteed. So although, because of the
complexities of the system, you can’t guarantee success via rheology, you are at least reducing the chances
of failure.

Reverse guarantees? Smart mapping? Why this strange idea of reverse guarantees and the later
one of smart mapping? Because there are times when we know that a direct formula/app-based
approach can’t work. These phrases deliberately alert you that “business as usual” cannot work and
you have to adopt a different mindset to efficiently navigate the complexities of these formulations.

Rheology for PSAs

Although for other aspects of handling (e.g. during coating) you need Flow_Basic viscosity and
Flow_Rotational rheology, here we focus on Flow_Oscillatory rheology to measure the elastic (G') and loss
(G'') moduli which you need to juggle, Flow_G' and G''.

Basically you apply an oscillatory stimulus (e.g. a stress) and measure the response (in this case a strain).
You might find something like this:

[Link]

The response is lagging behind the stimulus in this case. If the material was purely elastic (all G') it would

100 FST
 PSA

respond in phase. If it were purely viscous, lossy, (all G'') it would be 90° out of phase. Here it is 45° out of
phase which means it is an equal mixture of G' and G'', which is the sort of zone you need to be in for a
typical PSA. The ratio of G''/G' is called tanδ. When tanδ = 0, you have pure elastic, when it is very large
you have mostly lossy behaviour.

Temperature, Time, WLF

Although the measurement is valid, in terms of a PSA it is meaningless without further information. A PSA
at low temperature isn’t (unless specifically designed for this) a PSA. The polymer is too rigid, G' is too
high. A PSA at high temperature isn’t a PSA because the polymer is too soft, both G' and G'' are too low.

Just as important is that if, at room temperature, you test the PSA at very low speed, it behaves as if it was
at high temperature – it’s soft and useless. If you test at very high speed, it behaves as if it was at low
temperatures – it’s brittle and useless.

Although your PSA may be designed for a modest

temperature range, it will experience various timescales in
use. If it is perfect when tested slowly yet fails at high
speed, then that might make it useless for a specific
application. Here are some typical timescales and an
indication of how G' changes over those timescales. We
shall shortly see what those G' numbers are, but you
already know from Dahlquist that the Tack G' at 100/s (i.e.
1/s) must be ~0.1 MPa.

We can take this graph and look at it through the Chang

Window, which says that in use you care about G' and G" in the region between 100/s (Peel test) and 0.01/s
(Creep).

FST 101
[Link]

A standard PSA has G'/G" criteria broadly in the middle – in this example the central brown box represents
the rheology of the product, with a range slightly smaller than Chang, but maybe that’s a positive choice to
optimize performance near the centre. There are 3 views of the same information, this is the “Full” version
with a lot of other information added via Bartholomew.

Experts in PSA formulation are clear that Chang is another reverse guarantee. If you are in your desired
window you still might not achieve success, but if you are outside, failure is likely.

A deliberate feature of Chang is that the values are defined at room temperature. So a high temperature
PSA must have G'/G" values way too high when measured at room temperature (top right corner) which
will shift to the centre when in use at those higher temperatures.

The right level of crosslinks & tangles

If you have a highly crosslinked or entangled formulation (in many ways the two things are the same) it
will have a high G', spoiling the chance of meeting Dahlquist. A low level of crosslinks or tangles will be
far too weak to be an effective PSA. So the formulator has to find ways to tune the level of crosslinks and
tangles. With rubber formulations it’s convenient to use styrene triblocks because the crystalline styrene
regions act as crosslinks – and you can tune properties by the styrene-to-rubber ratio and (because the
rubber itself can tangle) the MW of the blocks between the styrenes. For acrylics you can add low levels of
di-acrylates – or rely on entanglement of high MW components. For UV PSAs you can create the required
low level by tuning the levels of mono, di-, tri- functional acrylates.

102 FST
 PSA

An important aspect of these crosslinked systems is strain hardening discussed below.

As a beautiful proof that too much of a good thing is a bad thing, there is one type of UV PSA used in large
quantities with a life between use and throwing away measured in minutes. It solves a problem in silicon
chip manufacture where you need to flip 100s or 1000s of chips in order to carry out a process on the
reverse side. You roll the PSA over the chips and can easily pick them up. After the required process, you
probably need to transfer them to another bit of PSA for a different process. You apply the new PSA but
how do you unstick the previous one? You give it a blast of UV that fully cures it. Now it is a brittle film
with near-zero adhesion. It sounds unlikely but I once spent a couple of days on such a production line
marvelling at the process while also helping to solve some technical issues with the process.

The Ideal PSA

When you get off a train near Amsterdam and the stranger who is there to meet you says “I’m a
rheologist, would you like a beer?” the obvious thing is to forget the rheologist part and say yes to
the beer. However, the beer was accompanied by some heavy rheology talk which was only about
5% comprehensible. But that 5% was enough to realise that something important was being taught
… and that led to the app below. The wonderful rheologist is acknowledged in the app.

There is an amazing law of physics that says that “temperature is equivalent to time”. This is captured as
TTE (Temperature Time Equivalence), TTS (Temperature Time Superposition) or WLF (Williams, Landel
and Ferry who produced a convenient formula we all use), Flow_TTS-WLF.

If you do a set of rheology measurements at one temperature but different speeds, or different temperatures
and one speed (or any other combination) you can re-plot them as if they were your chosen speed over a
large temperature range, or a chosen temperature over a large speed range. So a single rheology plot can
capture what happens to your PSA over a set of conditions way beyond anything you can test in the lab. If
you are not doing WLF analysis of your PSA rheology, you are making formulation life very difficult for
yourself.

For the classic rubber-based PSAs, the rheology can be captured in one “Ideal PSA” plot:

FST 103
[Link]

Here you are going from -80°C to 140°C and from 105 /s to 10-5 /s. If you explore the app you can learn a
lot about plasticizers and tackifiers. In short, a plasticizer is some junk that takes your polymer’s G' down
into Dahlquist, but without adding anything else desirable to the formulation. A tackifier brings your G'
down, but also raises the Tg, giving you a desirable higher temperature performance.

This is an idealised dataset and in the real world you will not obtain high quality data over such a large
range, but the point is that a standard rheometer with a reasonable temperature controller can give a lot of
data for little work. Although any single plot is hard to interpret, as you compare the data from
formulations with different sets of product trade-offs you will learn how to take your current formulation to
a different desired location in PSA space.

Although your rheometer software can do the WLF transformations for you, the WLF apps give you a good
feel for what is going on:

[Link]

On the left we see the raw data from the rheometer at a range of frequencies and temperatures, on the right

104 FST
 PSA

we see them all combined into a single curve described by WLF. While doing so we go from data gathered
over 3 orders of magnitude frequency to the single curve over 11 orders of magnitude.

During some work on an unfamiliar rheometer we couldn’t find the WLF package in the
rheometer’s software. In my hotel room that evening I had to work out how to do the transform in
Excel. Although it’s trivial, it seemed very hard. In the morning I presented the fit – and they also
had an overnight email explaining where to find WLF in their software. Fortunately the results
were the same.

But that spreadsheet has been downloaded many times from my WLF page because everyone else
finds their first WLF to be as hard as I did.

Rheology is not enough

You can’t formulate a PSA without rheology, but there is a fundamental issue with the rheological data
which limits its applicability: it works over small strains (% stretching of the PSA), say 0.1%, while a PSA
in action can easily see 100% strains.

To understand the behaviour at high strains, the ideal test

is on a tensile tester. This is relatively unpopular, but
gives the most direct information. More popular is the
probe tack test which has a tricky relationship to PSA
performance (described below) but is quick and
convenient. In either case you get curves something like
this:

The portion of the curve in the lower left is the elastic domain where you can get G'. After that you get
various curves that show the yield stress σ then various elongations and failures. We certainly don’t want
curve #1 because that’s just an elastic solid. We probably don’t want #4 because that is a weak solid.
Something like #2 or #3 is more suitable for a PSA. The slight strain hardening in #3, discussed shortly,
may or may not be desirable, it depends on the specific application.

As with rheology, you can’t tell a lot from a single curve from on PSA, but as you compare and contrast
good and bad PSAs you start to see the features you require.

Curve #3 shows a slight increase in stress with strain. This is strain hardening. Because the system is
crosslinked/entangled, as you stretch it, it puts up some resistance. There is little agreement about how
much strain hardening is optimal, but if (because you don’t do tensile tests) you have no idea if your system
strain hardens or not, it’s hard to know how to optimise the level of crosslinks and entanglements.

Unfortunately, the WLF work in the hotel room showed that WLF G'/G" curves for their product
and a competitor’s were identical, even thought the competitor’s product was superior. Some
impromptu tests on an Instron-style machine with a light load cell showed obvious differences
between the tensile curves: one showed strain hardening; the other didn’t. These differences could
be translated into formulation insights which resulted in an improved, now market-leading,

FST 105
 product.

Testing PSAs
As the Adhesion_Testing topic shows, most adhesion tests provide little of the information you need to
know on how to improve a formulation. This is true for PSA testing.

For example, the classic PSA shear test shown in the diagram is only a test
for passing the shear test. It has very little relationship with the shear
properties of the PSA for a good reason – it often fails in peel, a fact
common in shear testing. So the formula shown in the image is not
applicable, much to the puzzlement of people who think that the shear test
provides information about shear behaviour.

If you really want shear behaviour, there are at least two tests (see
[Link] that can
do a better job at revealing the true shear behaviour. Shear properties are
important, that is why, when you are optimizing the science, you should do the real shear tests rather than
the industry-standard one. Of course your customers demand the industry-standard one, but first you have
to get the real tests right.

There are two tests that are scientifically valueless

yet often used for QC testing: Rolling Ball and
Loop Tack Tests. Although simple and convenient,
there is a risk that their lack of scientific meaning
allows you to miss crucial changes in the PSA
properties.

Then there is the probe tack test:

106 FST
 PSA

The probe tack test has lots of things happening at once, and because it is essentially a butt joint test, it is
not directly relevant to the most important aspect of a PSA which is its peel. Using the probe test as an
indirect tensile test has its merits, and the Simple Model described in [Link]
practical-adhesion/[Link] while anything but simple, is interesting. A recently developed direct
link between probe and peel tests is described in this app, [Link]
adhesion/[Link], though it is still really for experts.

Which brings us to the peel test. This is at last a direct

measure of what most of our customers (think they) want.
Because adhesion is a property of the system, the peel test
is not as simple as it might seem:

When we are peeling we can see the dramatic fingering

part. But the compression zone is generally unknown
despite, as we shall see, its importance. Although you can explore more in Adhesion_Peel, the core ideas
can be found in the peel app:

FST 107
[Link]

The right-hand edge of the app is just before the first finger appears in the previous image. For simplicity
we’re looking at a 90° peel – you can change the angle at any time.

The tensile stress is shown on the left-hand Y axis. You see that 2 mm ahead of the peel the stress is 0, as
you would expect. But 0.6 mm ahead, the stress is -0.27 MPa, i.e. the adhesive is in compression. This is
non-intuitive but very real. At the right-hand edge you reach the maximum stress, pulling the adhesive
away.

The important fact to note here is that if you change the modulus or thickness of the carrier film, or the
thickness of the adhesive, you change everything. In other words the peel adhesion properties of your
adhesive depend on 3 parameters that are not part of the rheology of the adhesive itself. Yes, you must get
the rheology right, but you should also attend to other aspects of your tape. For example, reduce the
modulus of the tape and the curve “tightens up” – the compression gets deeper and closer to the peel edge,
and the maximum tensile stress increases, reducing overall adhesion.

So a well-intentioned cost reduction via a cheaper, weaker, thinner backing tape can result in an expensive
upgrade to the PSA formulation to restore peel strength to its original value.

I was asked about a customer complaint that the peel strength of a batch was lower than specified.
In-house tests on QC samples at 25°C showed that the batch was in spec. It turned out that the
customer was in a hot country and their lab was at 30°C. Using the additional calculator near the
end of the WLF app, [Link] it was easy to
show that if the customer performed the test 2.5x faster, that would be equivalent to them
performing the test at standard speed at the specified 25°C lab conditions. The customer was
astonished when the test at that speed proved the product to be in spec.

Release liners
Because silicone release liners have a low surface energy, it is usually stated that they are release liners

108 FST
 PSA

because of the low surface energy. This is wrong. Fluoro-surfaces with the same surface energy show poor
release. The real explanation has been known since the 1990s but the myth of surface energy still
dominates.

The way that silicone release works is related to the compression zone ahead of the peel front!

If you put some fluorescent beads into the PSA and watch
what happens as the compression zone arrives, you see that
beads next to the surface and in the middle of the adhesive
move no problem. This doesn’t sound surprising, but it
contravenes the “no slip boundary condition” which says
that there is no fluid flow at a surface. With the
fluoropolymer you see the normal behaviour, the beads at
the surface are stationary.

Because in the fluoropolymer case there is differential motion as the compression zone arrives and leaves,
there is viscous loss of crack energy, so the adhesive remains in contact. But on the silicone, there is no
differential motion, no viscous loss, so the crack is able to travel easily along the interface.

The way the silicone overcomes the no-slip condition is by being relatively mobile itself. It’s not the
surface energy but the surface mobility that matters.

If you are trying to find a replacement for silicone release liners, finding ways to get surface mobility via
some clever science will be more productive than trying to find alternative chemistries with low surface
energies.

At the end of a talk on adhesion science in front of a large audience, someone asked a question to
which I could only answer that I didn’t even know what the question meant. Catching up over
coffee with the (it turned out) distinguished professor who’d asked the question, led to a 15 min
lecture on the science of PSAs. There is a direct link from that admission of total ignorance,
through that 15 min lecture, to this chapter. The distinguished professor is Mikhail Feldstein and
those trying to formulate PSAs without the magnificent 3-volume Handbook of PSAs and Products
by Benedek and Feldstein are making life unnecessarily hard for themselves.

There are two types of ignorance. The first is that a topic has never come up in one’s journey
through life; so admitting ignorance is a chance to learn exciting new stuff. The second type is
refusal to engage with correct ideas because they go against what you think you know. Although
most of us are tempted by this second type, it’s generally not a good way to make progress.

FST 109
SkinCare
Links
Diffusion_Basic Diffusion, Diffusion_Diffusion Coefficients, Diffusion_Diffusion into Skin and Hair,
Dissolution_Flory-Huggins, Dissolution_Hansen Solubility Parameters, Dissolution_Ideal solubility,
Dissolution_Solubilizers and Hydrotropes, Evaporation_Basics, Evaporation_Humectants,
Evaporation_Temperature and Antoine Coefficients, Flow_Basic viscosity, Flow_G' and G'',
Flow_Oscillatory rheology, Flow_Poiseuille Flow, Flow_Rotational rheology, Flow_Shear dependent,
Flow_TTS-WLF, Flow_Thixotropy, Flow_Yield Stress, Fragrance_Barrier Properties, Fragrance_Vapour
Pressure, Optics_Colour, Optics_Scattering and Opacity, Surfactancy_Dynamic Surface Tension,
Surfactancy_Emulsification, Surfactancy_HLD, Surfactancy_Interfacial Tension and Rigidity

Although skin care involves cleaning the skin, that is covered in Soaps and Washing. Here we look
at formulations designed to make the skin be, or look, better.

What problem are we trying to solve?

Skin has evolved to be an adequate solution to the twin problems of keeping good things in and bad things
out of the body. Although it is routinely said that skin is an amazing barrier, it is relatively poor. It easily
lets water in and out (by design) and many liquid chemicals such as solvents can easily get through. It is
also readily damaged by a range of common household chemicals such as caustic, bleach, enzymes in
laundry detergents.

But because the outer part of the outer stratum corneum is designed to be shed continuously it’s a smart
system that can repair itself with little help from ourselves. The skin provides the surface with a range of
nutrients via the hair follicles, which add to the nutrients provided to the skin biome. This array of
interacting microbes, each looking after their own interests, provides mutual assistance against threats to
the biome and the skin.

The skin will remain in a good state if we don’t carelessly expose the skin to excessive threats from
sunlight, chemicals and abrasion, and if we remove dirt without devastating the skin and its biome via
excessive (even if well-intentioned) cleansing.

So the key to a healthy skin is to do as little as possible. Washing with water cleans away most dirt and
most sweat (which is mostly salts dissolved in water), and the lightest dose of soap does the rest, unless
we’ve undertaken a really dirty job without good skin protection.

As is well known, if the skin needs repair from temporary exposure to an aggressive environment (e.g. too
much sun), there are 3 ingredients proven to be of use:

1. Water
2. Glycerol
3. Mineral oil (or some modern equivalent for those who don’t like mineral oil)

The water does rather little because it mostly evaporates after 20-30 minutes from any reasonable
 SkinCare

formulation – though the glycerol is a good humectant (Evaporation_Humectants) and keeps some of the
water as well as interact itself inside the stratum corneum to make it more fluid, less dry. Glycerol is not
quite perfect – it ends up feeling sticky on the skin. The mineral oil provides some occlusion to stop water
loss from within the skin … and lets you have plenty of glycerol without the sticky effect.

Excellent water/glycerol/oil emulsions with, perhaps, a little added fragrance have been around for decades
and pose no new formulation challenges. The choice of surfactant for the emulsion will change with
fashions (and is discussed below) and maybe some Thickeners_Associative Thickeners or
Thickeners_Polymeric Thickeners need to be added to keep the emulsion from separating and to give the
required “luxury” Flow_Yield Stress behaviour. But, again, such formulations have been known for
decades.

Because all aqueous formulations containing nutrients (oils, surfactants, fragrances …) are attractive to
microbes, the product must contain a proven preservative package. Whether the preservative ingredients are
entirely synthetic, entirely natural or a mix cannot alter the fact that they are designed to stop living
organisms from proliferating. They therefore, by definition, pose a potential threat to the living cells in and
on our skin including, especially, our microbiome. The formulation problem is trivial – proven packages of
preservatives are well known. The problem is a marketing one – finding a package that is provably great at
inhibiting bad microbes whilst being entirely friendly towards the good cells and microbes and sounding
good to the end users.

So this chapter could/should end at this point if we define the problem to be solved in the rational manner
above. Instead, the problem to be solved is “How to offer hope to consumers that their skin will look like
the images seen in advertisements?” Now we enter a much more complex world.

Cosmetic or Pharmaceutical?
Although the ~15µm top layer of the skin, the stratum corneum, SC, is far from dead (only the top few µm
of skin cells are dead, but the skin biome is happy in that environment) it is a convenient fiction that if our
formulation only interacts with the SC then we’re probably not doing much damage (though biome
research increasingly questions that assertion). If we want to add an oil, and if that oil is a large molecule
that mostly stays on the surface, then that’s fine. We can also add water, glycerol and the ingredients of
NMF, natural moisturizing factor, already in the SC, such as amino acids, urea and pyrrolidones.

A large surfactant molecule such as a Tween is too bulky to migrate through the SC so is safely on top.

Vitamins might seem OK, but normally our bodies receive vitamins at low concentrations absorbed from
the gut, not in relatively high concentrations delivered to a part of the body not used to them (apart from
Vitamin D which is manufactured in the skin).

The problem arises when you want to do something significant to boost the skin’s function by sending
chemicals into the viable epidermis or deeper into the dermis. These chemicals are now in your body, with
few barriers to getting into the numerous blood vessels near the surface of the skin. If they have a specific
biological function then they are pharmaceuticals. There is a strong tradition of trying to deliver APIs
(Active Pharmaceutical Ingredients) through the skin so aspiring to improve the skin in this manner is
laudable … provided you then follow the regulations surrounding pharmaceuticals. As these regulations are
(rightly) difficult, the skin care industry somehow has to promise that the wonder chemicals they add are
doing great things to the skin biology without being pharmaceuticals.

This used to be easy because regulations were lax. You could claim to be doing wonderful things to, say,

FST 111
“boost DNA”. Who wouldn’t want a skin cream that boosted their DNA? Well, anyone with common
sense. Cancer cells “boost DNA”. So you’d want to be sure that the boost from the cosmetic formulation
was 100% upsides and 0% downsides. Which means pharma-grade proofs of efficacy and safety.

So claims are now carefully crafted by lawyers and marketing experts familiar with just how far they can
make something sound sciencey and biologically effective without making a specific pharmaceutical claim
that would need a decade of hard work to validate – in an industry that changes fashion in months.

Cosmetic claim validation

I once was asked to step in to deliver a 1hr training course on cosmetic claim validation. Being
entirely ignorant of the topic I had to quickly educate myself. The task was made easier by the
existence of a UK database of challenges to cosmetics claims, along with the decisions, and
reasons, delivered by the authorities. The large companies frequently pushed claims to the limits
and beyond. But “beyond” only meant a slap on the wrist after weeks or months of a campaign that
was probably ending anyway.

One type of claim is “Contains X proven to do Y”. Although the claim is factual, it is now generally
regarded as unacceptable if, as is usually the case, X does Y only at concentrations far higher than allowed
in the formulation. Retinol, for example, is proven to do wonderful things to the skin at high levels that,
inevitably, come with the risk of pharma-scale side effects. So retinol can only be added at levels that have
an insignificant effect.

“Contains extract of [some exotic plant]” is a claim that leads the consumer to conclude that it has been
added because this exotic extract must do something wonderful. This is really an ingredient rather than
functional claim so is probably OK. “Tap into the healing power of [some exotic plant]” isn’t claiming that
the product provides any healing power, but because it’s obviously designed to give the impression of an
(unproven and, at the given concentrations, unlikely) healing power it’s probably unacceptable.

What about serious claims?

By using only ingredients approved by the cosmetics standards, the issue of cosmetic safety can be largely
bypassed. So now the claims of, say, “proven to reduce wrinkles” need to be validated.

In some cases, you don’t even need statistics to validate the claim. Any reasonable “filler” formulation (by
definition) fills the spaces between wrinkles, and everyone can agree that wrinkles have vanished for as
long as the filler stays in place.

For most other claims, the industry demands, and gets, reasonable statistics, taking into account sample size
and statistical differences. Those lucky enough to have a wonder formulation will need only a small sample
size. The more common formulations need a larger sample size (expensive) from which to extract a
statistically significant effect.

One reason for this rigour is industry standards. But a key force driving those standards has been
competitors who have been taken aback by a claim from a competitor that sounds too good to be true and
who then come together to show (there are famous historical precedents) that the claim is indeed invalid.

Claims can be subjective measures such “X% of users found that Y improved the Z aspect of their skin”.

112 FST
 SkinCare

The challenge here is to find a relevant control. For typical skin creams, a water/glycerol/mineral oil control
performs distressingly well. Tricks like packaging the control in a way that makes it look cheap (or using
no fragrance), and the real product in a luxury package (with a luxury fragrance) benefit strongly from the
placebo effect so their objectivity has to be questioned.

Or they can be objective measures such as:

• Changes in TEWL (Trans Epidermal Water Loss)

• Objective reductions in wrinkles
• Changes in elasticity
• Changes in colour (e.g. for age spot reduction)

Here the key issue is that the claim “90% users showed a reduction in wrinkles” can be factually correct but
irrelevant – the scientific measure is correct but the consumer cannot see the subtle differences detected by
the equipment.

The history of cosmetics claims shows that either they are great results based on levels of chemicals that
take them into the pharmaceutical zone (as with retinol) or are marginal results of no significance to the
user. This claim is rather obvious – if any company had a truly superior product and the patents to protect
it, they would dominate the field.

Your job as a formulator is to make sure that at least one molecule of the wonder additive is present
(“Contains X” must be factually correct, and presumably one molecule makes it correct) … and more
importantly that the additive doesn’t mess up your carefully balanced formulation, discussed below.

Getting your molecule through the skin

For decades, the standard technique for measuring the penetration of a desired molecule, let’s call it an
“active”, through the skin used an “infinite dose Franz cell”, i.e. on one side of the skin sample was a
container with a large amount of the (aqueous) formulation and on the other side was some buffer that
could be sampled over, say, 24 hours to measure how much of the active made it through the skin.

This has been maximum work for minimum relevant information. For any skin formulation that is water-
based, the water evaporates after 20-30min so the interactions with the skin are nothing like those in the
Franz cell. And much of the formulation that was on the skin disappears after a few hours through the
contacts of daily life. So decades of research has produced data of essentially zero value to the problem.
One confusion from this work was that LogP (water-octanol partition coefficient) was a key determinant of
skin permeation. It simply isn’t. Another confusion (helped by LogP) was that molecules permeated via the
“lipid only” route through a “brick and mortar” skin. Again, this is simply untrue, not least because the
“lipid” portion is full of molecules such as cholesterol containing plenty of polar functionalities.

Another confusion was the role of surfactants. When the water doesn’t evaporate in the Franz cell, the
surfactant is a surfactant. But in a real formulation, when the water goes, the surfactant is no longer a
surface active agent as there is no longer a meaningful water:oil interface. The surfactant therefore becomes
another molecule in the ingredient mix. If that surfactant is a Tween, a large molecule that does not
penetrate the SC, then on the surface you have plenty of alkyl chain, plus a lot of ethoxylate oligomers that
can act as an attractive environment for some of the active ingredients. They are then stuck in this
environment and cannot deliver any of the desired benefits to the skin. This is an obvious fact, but one that
was not obvious to decades of researchers who kept insisting that surfactants acted as surfactants on the
skin.

FST 113
As we know from Diffusion_Basic Diffusion, diffusion depends on the concentration gradient across the
barrier plus the diffusion coefficient of the molecule in the barrier. The diffusion coefficient can be
approximated by something like a MW² effect, leading to the rule of thumb that anything above 500
Daltons won’t permeate much. Obvious exceptions to this rule are discussed below.

To get a high concentration gradient needs three things:

1. A high concentration of molecule in the original formulation (once the water has evaporated);
2. A good partition coefficient into the barrier;
3. A semi-liquid environment at the surface to deliver a continual supply of the ingredient to the skin
surface.

To address the first part, if we consider the formulation after the water has evaporated to be a mix of oils,
emollients, hydrotropes, surfactants (such as the Tween example above), humectants (and more) then we
need to estimate the solubility environment and how friendly it is to the active ingredient. The only viable
tool to do this (and hence, much used in the skin world) is Dissolution_Hansen Solubility Parameters. We
can create a volume % weighted average of the formulation to get its HSP then compare it to the HSP of
the desired ingredient. If the HSP Distance is small, they will be compatible, if far, incompatible.

To address the second part, we need to estimate the solubility environment of the skin. Far from being
“lipid” (which would imply HSP parameters in the range of [16, 2, 2] typical of ingredients such as IPM
(isopropyl myristate), it turns out that the skin is something like [17, 8, 8]. For those who believe in the
lipid-only route, this non-lipid value emerges quite naturally when you include the HSP of cholesterol
(present at ~20%) into a mix with the typical lipid value. Knowing 1 and 2 and the HSP of the desired
ingredient we can see how likely it is that the ingredient will stay above the skin or partition into the skin.

The third part can be tricky. Suppose you have a lot of smaller molecules that together give an HSP not far
from [17, 8, 8]. These will partition into the skin and start their journey diffusing into the SC. With a low
MW their diffusion will be relatively fast. If the active molecule has a large MW it will diffuse slowly …
and can get left behind, stranded on the surface and/or locked into any other large molecules left on the
surface.

Beyond Franz cells

Because of the multiple interactions of the multiple components in a formulation, it makes no sense to use
Franz cells to analyse a single component coming through the skin. We need to know which parts of the
formulation (including our active) go into the SC. Those with confocal Raman spectrometers can get a
good idea of this, but the signal of the skin itself is complex enough, and changes when the formulation
enters, so it’s optimum only for looking at a few components with strong, distinctive absorption bands. Far
better is to take samples off the surface of the skin at appropriate times and use advanced mass
spectrometry to characterise the composition. You know the quantities at the start, so can normalise the
signals. As time goes on you find out what is staying and what is leaving.

If the component is not on the surface then it’s in the SC. Does that mean it penetrates the SC? Once inside,
what else can it do? It’s not coming out again, so unless it is actively destroyed by the biology in the skin,
it’s going through – maybe slowly, but inevitably.

Although such a technique is not simple in practice, it is so significantly better than the alternatives that it
should become the norm.

114 FST
 SkinCare

What happens next?

The active and maybe some other ingredients have gone into the SC and will inevitably get through to the
other side. The SC is designed to be a barrier. The epidermis is not a barrier – it is a relatively open,
relatively aqueous environment. So unless the active is amazingly hydrophobic, it will continue through the
epidermis and reach the numerous blood vessels in the area. There is no good reason for most actives to
stop once they’ve made it through the SC. If this is OK with the consumers and the regulatory authorities,
that’s fine. Otherwise it’s a problem.

Biology
So far we’ve just used simple solubility and diffusion physics to analyse what’s happening. In this
biological system there are other considerations.

The rule of thumb around 500 Daltons has some obvious exceptions, especially with peptides that can
easily be in the 1000+ range yet disappear rapidly into the skin. This can only be via an active process. The
existence of “transporter systems” in all cells is both well-known and under-appreciated. Cells can’t rely on
molecules passively wandering in and out – they are actively included and excluded via transporter
systems. Whether cosmetics could/should take advantage of transporter systems takes us to the cosmetics/
pharma boundary once more.

The other way for biology to intervene is if the ingredient of interest is actively trapped by a biological sub-
system. If the ingredient is a “key” in a “lock and key” system inside the SC or epidermis then it delivers its
function and goes no further. This is good news in terms of ingredients not penetrating further into the
body, but maybe bad news if this is an effect that could be termed “pharmacological” as, again, we are at
the cosmetics/pharma boundary.

The follicular route

The skin contains plenty of hydrophobic channels, hair follicles, crossing the SC, providing a hydrophobic
flow from inside to out. A suitably hydrophobic formulation might be sufficiently soluble in that
hydrophobic environment to have a chance to diffuse into the skin. But diffusion is generally slower than
the flux through the follicles so without some pumping action in the reverse direction, nothing much
happens. Because each follicle contains a hair, a massaging of the skin causes the hair to move which, in
turn, provides a pumping action. So follicular delivery can be achieved if aided by some modest massaging
of the skin.

Again, this is taking the active to a very-much alive system (the base of the hair follicle) so we have
another cosmetics/pharma boundary question.

Formulating the emulsion

Once you have decided on the type of emulsion (o/w, w/o, macroemulsion, nanoemulsion, microemulsion)
and decided (mostly for cost/marketing reasons) which class of surfactants to use, the first task is to make
sure you are in the right part of surfactant space to have a good chance of creating the right, stable emulsion
with relatively little effort and the minimum amount of surfactant. Remember that the type of emulsion can
make no difference to anything after the ~20min it takes for the water to evaporate, meaning that there is no
longer an emulsion. The type has an effect on the initial look and feel and for 10-15 minutes, so base your
decisions on those timescales.

FST 115
Given that HLB is useless, the rational approach is to use HLD: Surfactancy_HLD. We start with
([Link]

HLD = F(S) − [Link] − α(T − 25) + Cc

Assuming a low content of salts, the dependence on salinity, F(S) can be ignored and, for the moment,
assuming that the temperature T = 25, to calculate HLD we need only to worry about the oiliness of the oil,
its EACN, Equivalent Alkane Carbon Number and the “characteristic” of the surfactant, Cc, which goes
from hydrophilic (large negative values) to hydrophobic (large positive values).

For an o/w emulsion we need HLD < 0, for w/o we need HLD > 0 and for microemulsions and efficient
manufacture of nanoemulsions we need (temporarily for the nanoemulsions) to be in the HLD = 0 zone.

If we are using commercial surfactants, their Cc values should be available from the supplier. Failing that
you can look them up on the Cc web page: [Link]
Given that the Cc of a surfactant blend is the mole-fraction ratio of the individual Cc values, it’s easy to
generate your desired Cc if you start with surfactants in the right range. Don’t try to combine Tweens with
Spans. Although this works in theory, each has a severe tendency to partition into, respectively, the aqueous
and oil phases, rendering the assumptions behind HLD invalid.

The problem is the EACN of your oils. It is not hard to work out the EACN of any reasonable oil, using
values from the EACN web page: [Link] The
volume-weighted average of your oils is the EACN. The problem is that real-world oils are not pure. For
example, octyl myristate might contain small levels of octanol and myristic acid. A triglyceride might
contain small amounts of di- and mono-glycerides. In addition the oils will contain various fragrances with
functional groups such as -OH or -C=O. It turns out that “polar oils” (ones with groups like -OH) exert a
big effect on EACN values. There is no known way to predict the EACN of these real-world formulations.

Fortunately HLD allows us to easily measure the EACN of any mix … and how it might change if we
tweak the fragrance or the oils or if there are batch-to-batch variations.

Knowing the EACN we can now choose the surfactant blend to get us into the desired part of HLD space.
For an o/w emulsion we need -1 < HLD < -0.5 and for w/o we need 0.5 < HLD < 1. Sufficient
emulsification energy will produce an adequate emulsion. For those concerned with taking the product into
production, there are some subtleties about water-to-oil ratios as the batch is created so there is a need to
look out for emulsion inversions: Surfactancy_Emulsion Inversion.

For effortless creation of a fine emulsion, the trick is to take the formulation to HLD = 0. Now the
interfacial tension Surfactancy_Interfacial Tension and Rigidity is super low and emulsion drops form
easily. For a microemulsion that is the end of the story – they are thermodynamically stable and can be
created by simple mixing. For nanoemulsions and stable o/w or w/o emulsions you create the emulsion at
HLD = 0 then change one of the parameters in the HLD equation to place HLD safely into the optimal o/w
or w/o zone.

The classic way to do this, which allows wax ingredients to be melted into the blend, is to heat ethoxylate
formulations to a temperature where the net +0.06(T-25) term takes a negative HLD to 0. At that “phase
inversion temperature” (PIT) the emulsion is created with just standard mixing. Rapid cooling to room
temperature keeps the fine emulsion. For a standard emulsion “rapid” can be slower than for a
nanoemulsion where Ostwald ripening (Surfactancy_Emulsion Ostwald) is a serious problem.

116 FST
 SkinCare

With some consumers wanting “ethoxylate free” formulations (worries about 1,4-dioxane) and/or the use of
greener surfactants such as APGs or polyglycerols where the α term is ~0 and ~-0.01 respectively, the trick
is to start the formulation with the higher Cc component so that HLD is ~0 then quickly add a more
negative Cc component to take the formulation to the desired zone. Similar tricks could be adopted using
part of the oil to start at a lower EACN then raise the EACN to take the HLD to the desired negative value.
These ideas are unfamiliar because of the historical preference for HLB. As HLD is becoming more
accepted, these ideas will become increasingly normal.

The rules of HLD for formulations with plenty of glycerol or other humectants are unknown – the effects
on oils and surfactants will vary. Again, the standard scanning method used to measure EACN effects can
be used for humectants. With a set of test tubes, some standard formulations and a convenient robot, it
doesn’t take long to map the new surfactant space and to formulate accordingly.

What about alternative to surfactants such as “hydrotropes” and “co-surfactants”? This takes us to a world
of marketing speak. “Surfactants” are bad while “hydrotropes” are good. Maybe hydrotropes are not-very-
good surfactants and may be added to “boost” some property of the main surfactant. In that case it’s a co-
surfactant. Other co-surfactants are things like “linkers” that go to the head/tail interface and potentially
increase efficiency by making the surfactant head and tail longer, with an extended influence. Sometimes
these hydrotropes and co-surfactants are simple alcohols like ethanol or octanol.

In general these extra ingredients with ill-defined names and functions have been added by those lost in
surfactant space and in need of a tweak via a relatively simple ingredient. The problem, as discussed below,
is that every extra ingredient added to solve some specific problem ends up as a permanent part of future
variants of that formulation – because future formulators don’t want to risk removing something they
assume has some vital, but unknown-to-them function.

One of the beauties of the HLD approach is that with an approach based on core scans for the main
ingredients and sub-scans for the effects of extra ingredients, it is possible to record that, for example, the
addition of X shifted HLD too far in one direction so Y had to be added to bring it back. A future
formulator, not needing X, can revert to the original HLD and work out what happens when they add
ingredient A and whether they need B to fix the problem that A creates.

Formulating for surfactant efficiency

The ideal surfactant would be low cost, green, kind to users’ skin, tuned to the desired part of HLD space
… and be highly efficient in producing emulsions. At its simplest, efficiency, assuming 100% partitioning
to the emulsion interface, is the ratio of head area to MW, as you can explore in the emulsion surface area
app: [Link] If you have a large head area for a
small MW, a small amount will cover a large area of emulsion.

Once you get into microemulsions, efficiency is more related to the de Gennes ξ parameter used in the
more sophisticated HLD-NAC theory: [Link]
For this edition of FST HLD-NAC is not discussed, but the apps and the description in Practical Surfactants
[Link] are readily available.

Formulating for feel

We don’t know how to formulate for a perfect “feel” of the product – it is a subtle balance of effects best
judged by trained tactile experts.

FST 117
We can eliminate a popular factor – surface tension. This isn’t just because surface tensions of most
cosmetic oils (excluding silicones) cover such a narrow range – at most a factor of 2. And it isn’t just
because surface energy forces are factors of 100s or 1000s too small compared to the shear stresses during
rubbing. It’s not even because the formula for calculating the stresses in Flow_Couette Flow does not
contain surface tension or surface energy (it assumes the no-slip boundary condition). It’s because when
you rub the lotion, the value you’ve carefully measured (oil-air interfacial energy) is irrelevant because
both surfaces of the lotion are in contact with skin.

Silicone oils are different because, for subtle reasons due to the siloxane bonds, the silicones at the skin
surface act as a pseudo-liquid, overcoming the no-slip boundary condition and giving much lower shear
stresses.

If you find a blend of (non-silicone) oils that give a special feel, you can be certain that the effect isn’t due
to surface tension and is unlikely to be due to the viscosity effects discussed next. It probably means some
sort of phase separation, maybe including water that might be in the formulation, leading to some sort of
liquid-liquid boundary that, again, overcomes the no-slip condition that leads to shear stresses in Couette
flow.

But we can formulate to avoid obviously bad feel.

The starting point is rheology, which gives you more understanding than just a few viscosity measurements.

Yield stress. Without the correct yield stress behaviour, Flow_Yield Stress, the product is an instant failure.
Some consumers want instant flow, with no yield stress, others want a “rich” blob that retains its shape until
gently rubbed. Because yield stress is hard to measure objectively (there are at least 6 different ways to
measure it!) the trick is to find a set of samples (your own, or competitor products) that have been judged
by your tactile experts. Now measure their yield stress behaviour as best you can on your specific
equipment. If there is a good correlation between measured values and the experts’ judgement of
flowability then you have a good basis for testing new formulations. If there’s a lack of correlation, have a
go at measuring yield stress via an alternative route.

Flow curve. Next is a flow curve showing the shear-dependent behaviour, Flow_Shear dependent. With
some tweaks, the flow curves can also be used to estimate thixotropy, Flow_Thixotropy, i.e. the time-
dependence of viscosity. There’s a chance that neither of them will correlate much with feel, but will be
relevant to manufacture, pourability, fillability and maybe to the user’s experience of squeezing the product
from a tube or dipping their fingers into a jar. Something as simple as Poiseuille flow calculations for
pumping product through pipes (Flow_Poiseuille Flow) benefits from estimates of the shear rate (flow
velocity/ pipe diameter) and, therefore, the relevant viscosity during pumping. The consumer-relevant
aspects are probably best checked by correlation between flow curve and thixotropy measurements and in-
use evaluations by hands-on experts.

Rubbing in. The rheology of a formulation being rubbed into the skin is too complex for most of us. A 10
µm layer being rubbed at 1 m/s has a shear rate of 1/10-5 = 105/s which is impossible for most practical
devices. This is unfortunate because glitches in the formulation can show up as an unpleasant lack of
smoothness during rubbing. One trick is to place a thin layer between two glass slides then apply a one-off
rapid shear. Under the microscope, formulation failure modes such as forming temporary fibrils from
polymers that could not respond sufficiently quickly (relatively slow relaxation kinetics) can show up,
perhaps under crossed polarizers. It’s not a reliable technique, but it’s simple, and great when it shows up
something unexpected from an otherwise well-behaved formulation.

118 FST
 SkinCare

I once got lucky with this technique in the lab of a famous Italian cosmetics formulator. He
couldn’t understand why an otherwise great formulation didn’t “feel” right. We tried this technique
and could instantly see that one key polymer was stringing under “rubbing in” shear forces. He
was most impressed to have learned so much for so little effort.

Stickiness is obvious to a human and hard to identify with objective measurements. A “probe tack tester”, a
flat-ended cylinder pushed gently into the formulation then withdrawn at controlled speed with measured
force gives some idea of the instantaneous resistance (close to an elastic response) followed by some
degree of resistance via any tendency to form strings. Because any polymers in the system (e.g. thickeners)
have natural relaxation times, the probe test needs to be done at different speeds to see if a formulation that
offers little resistance at low speeds becomes much stringier at (relevant) high speeds, faster than the
polymer relaxation rate.

For “thin” formulations, probe tests will be at a limit of measurability. A more general technique is to
measure G'/G" (Flow_G' and G'' ) over a wide range of temperatures and speeds in your rheometer and to
use time temperature superposition/WLF (Flow_TTS-WLF) to give you the elastic (G') and plastic (G")
responses over a super-large speed range (outside the real measurement speed of the machine itself). As
above, there is little chance of you directly going from measured values to formulation insights. Instead the
idea is to compare the curves of “good” and “bad” formulations and gain insights into, say, speed onset of
high G" behaviour and (by implication or from your rheology software) insights into relaxation rates of
components in your formulation.

Time-dependent feel. All the previous measures take place on the full formulation. Although aqueous
formulations lose their water within 20-30 min, the instant feel dominates perceptions. But consumers take
note of how the formulation changes feel at least for a few minutes, so it’s a good idea to check for
stickiness measured on formulations with known water loss. The G'/G" rheology is probably the most
suitable technique with, as always, no obvious prior knowledge of what measured behaviour is desirable, so
a correlation with subjective feel data are the best to be hoped for.

Formulating for being microbe free

This is an area of scientific and regulatory complexity. But some principles are clear. If your formulation
contains water then microbes can grow. So by law you need to stop this growth via a package of
preservatives. By definition this means molecules that discourage or kill living organisms. So by definition,
this means a risk that they are harmful to skin cells or skin biome. So we have the nightmare scenario for
marketing – they must assure users that everything in the product is wonderful, chemical free and
nourishing (indeed, microbes might find these to be super-nutritious), whilst admitting that it contains
chemicals that kill or inhibit natural organisms.

Given that “natural” does not necessarily mean “good” or “safe”, marketing can’t hide behind the use of
natural preservatives. Many parabens are natural, benzoic acid and benzoates are natural but they are
demonised as “chemicals”. Some “essential oils” such as thymol are otherwise known as phenols, and
phenols are “chemicals” too.

The scientific advice is to go with a minimum package of approved, effective preservatives. But because in
these areas the science is of little importance to marketing and consumers, the formulator has to juggle
whatever are the latest fads from the latest internet scare.

FST 119
Formulating for other claims
The chemicals for removing dark spots, exfoliating the skin and other specialised effects are beyond the
scope of the FST.

For those who wish to claim that the formulation contains the soothing extracts of an exotic plant, contains
vitamins, antioxidants, UV blockers (see Sun Screens) it is likely that levels are small enough to do little
other than modify the EACN in a manner that is easily fixable by adjusting the surfactant ratio to obtain the
desired Cc and, therefore, HLD.

There are three real problems:

1. Permeation through the SC into the living skin. If the formulation happens to encourage soothing
extract to get through the SC, while safety tests on the soothing extract did not lead to permeation,
then maybe something will go wrong – for example a small percentage of irritated customers.
2. Unforeseen interactions. If the soothing extract happens to be a wonderful stimulant for the
growth of some microbe then the previously effective preservative package no longer works. If the
antioxidant reacts with the UV blocker in some odd way, maybe there’s a malodorous side product
that your end user wouldn’t appreciate.
3. Each extra ingredient might require some extra extra ingredient to avoid a long-term problem.
That’s OK. But when the product needs a marketing upgrade to provide yet another claim, it is
unlikely that anyone can remember which part of the formulation was required to solve a problem
of a no-longer-needed ingredient. So formulations become ever more complex.

There comes a point in this over-complication of formulations where the formulator has to fight back:
return to a basic formulation with the essentials then carefully add the marketing claim ingredients looking
out early for unwanted interactions. Marketing will object that you are too slow, but ultimately this back-to-
basics approach saves time and money. With luck, marketing can see that the new formulation, with fewer
ingredients can be marketed as “pure beauty” or whatever wording they care to use. It’s win-win.

120 FST
 Soaps and Washing

Soaps and Washing

Links
Cleaning_Boundary removal, Cleaning_Contact Angles, Cleaning_Surfactants, Cleaning_Temperature
effects, Flow_Basic viscosity, Flow_Yield Stress, Fragrance_Activity Coefficients, Fragrance_Vapour
Pressure, Surfactancy_Anti-foaming, Surfactancy_CMC and Langmuir, Surfactancy_Coacervation,
Surfactancy_Dynamic Surface Tension, Surfactancy_Foam Drainage, Surfactancy_Foam Ostwald,
Surfactancy_Foam Rheology, Surfactancy_Foaming, Surfactancy_HLD, Thickeners_Wormlike Micelles

The formulator has two challenges. The first is to create a soap that is effective for washing (note
that we are not discussing clothes washing, covered in Laundry Liquids ). That is trivially easy.
The second is to create a soap that meets the ever-changing requirements of marketing. That is
mostly easy, but has its challenges.

How to clean
Most dirt on our skin is easily removed with water and some effective rubbing with our hands, a fine cloth
or, if necessary, a brush. It is easy to remove because most of the time we aren’t “dirty” – we merely have
some bits of dirt, sweat (which is mostly non-greasy), and maybe the dust and pollutants from a day
walking around a city.

The water drastically reduces adhesion between particles and the surface (Particles_Basic behaviour) and
the rubbing motion overcomes the no-slip boundary condition (Cleaning_Boundary removal).

A small amount of any surface active agent will wrap around generally hydrophobic particles and help
them to wash away more easily. Here we meet the ambiguity of the word “soap”. It has two meanings:

1. The sodium or potassium salt of a long-chain fatty acid mix, e.g. sodium “palmitate”, the solid
soaps;
2. A surfactant solution based on classics such as SLS or SLES, the liquid soaps.

Solid soaps
Whether you want to call a bar of the first type “natural” or not, it is some mixture of excess alkali and the
glycerol produced as a by-product of the process of heating the starting triglyceride oil with the caustic.
One key advantage of the generally high pH soap is that it helps to destroy the dead skin cells on the
surface, leaving a shiny newer skin and taking any strongly-adhered dirt with it on the dead cells. Adding
some phenol to the soap (“carbolic soap”) helped with this – providing an effective, anti-bacterial but
obviously harsh cleaning product.

The form factor of this type of soap – a solid bar – used to be perceived as a disadvantage, hence the
creation of liquid soaps. Now it is seen as a low-waste, “green” form factor.

It turns out to be relatively difficult to add marketing ingredients to solid soaps – they tend to ooze out over
time. So soaps are relatively unexciting products, requiring claims such as being “French milled” to
encourage us to buy them.

The downside that the calcium & magnesium salts of the fatty acids are insoluble, giving rise to soap scum
helped swing the market towards the second type of soap.

Bars are relatively inconvenient in use – where do they go when you are washing (soap trays are of varying
utility), and how do they stay dry when not being used or transported (wrapped in beeswax?) in a wash
bag? The fact that a bar of soap lasts a long time shows that we can get adequate cleaning from small
amounts of surfactant. With a perception of a lower carbon footprint and of being “natural”, there is some
fight back against the liquid soaps that dominate the market.

Liquid soaps
Whether or not liquid soaps are more convenient and more effective, there is one overwhelming advantage
if you are in marketing: you can add all sorts of ingredients to impress the users. Fragrances, “natural oils”
(an odd ingredient when the soap is intended to remove natural oils from the skin), vitamins, antioxidants
can all be added (as we shall see, these can impose some formulation challenges) to give the impression
that the cleaning will be, somehow, better. Add an impressive volume of foam, and the consumer is
convinced that the product must be good.

The main surfactant ingredients could easily be selected for cleaning efficacy, especially for oil removal.
Applying HLD theory Surfactancy_HLD to optimise for Eötvös number (Surfactancy_Roll-up and Eötvös
Number) could give excellent cleaning. But this is precisely not what is required. A really good cleaner
would remove all the natural oils from the skin and destroy the skin biome. Instead the need is for a
guaranteed poor-quality surfactant, in the hydrophilic part of HLD space and for this SLS and SLES are
ideal. As these happen to give good foaming (Surfactancy_Foaming) under normal washing circumstances,
and are low cost because they are used so much, it is not surprising that they remain the core ingredients.

Scientifically, SLS is proven to have a strong, and still surprising, destructive effect on the skin if left in
contact at high concentrations for long times. That is, even more destructive than the already destructive
effects of leaving high concentrations of surfactants on the skin for a long time. This bit of science of SLS
has led to the demand for “no sulfate” surfactants which are conveniently provided by the isethionates and
taurates which are scientifically not sulfates (they lack one oxygen) but which behave more-or-less the
same so are convenient drop-in replacements.

The alternative ethylene oxide non-ionic class don’t have the reputation of being harsh, but now have the
problem that hyper-low levels of 1,4-dioxane need to be removed which is inconvenient and expensive. In
any case, consumers often choose “ethoxylate-free” surfactants along with “sulfate-free”.

The alkyl polygycosides (APGs) and the polyglycerols are, presumably, making steady in-roads, though
those used to the “feel” of the previous standard formulations are less happy with these … along with the
problem that these formulations foam rather differently (arguable “different” rather than “worse”, but
consumers again note differences) and are not easily thickened with sodium chloride.

Real cleaning
In the rare cases that we are really dirty, given that soaps are designed more for effect than for efficacy, we
have two choices.

1. Carefully use a scientifically proven cleaning system – a laundry detergent, Laundry Liquids.

122 FST
 Soaps and Washing

These are especially good for oils as they use the right types of relatively hydrophobic surfactant.
It is interesting to note that these work very well at levels around 200ppm, compared to a typical
20,000ppm of a typical liquid soap.
2. Use lots of scrubbing with particles, brushes, sponges, pads

In both cases, the skin will be temporarily damaged, so make sure it’s fully rinsed if you’ve used a good
detergent and take care of yourself if you’ve used the rigorous scrubbing.

Luxury feel
It must have been a wonderful day for marketing when the formulators showed that they could produce
exactly the same formulation at the same cost, but with a higher viscosity that made it look luxurious. The
trick was to add a few % of NaCl which created Thickeners_Wormlike Micelles which provided the
magical combination of being “rich” at low shear (they have a significant Flow_Yield Stress) but under
shear handle more normally so aren’t too difficult to apply.

This simple advance has come at a huge cost to the formulation community. Wormlike micelles are
complex, and their structure/behaviour is easily changed. As is well-known, addition of slightly too much
NaCl tips their behaviour away from having a high yield stress. But if for the past year you have had a great
formulation with a fragrance package X, when marketing require fragrance package Y, the wormlike magic
fails … and there is not a serious body of science that allows formulators to tweak things to restore the
viscosity. The same applies to the oils, vitamins, antioxidants that marketing wish to be added.

At a conference, a spontaneous sub-group formed during a coffee break to swap stories about the
huge waste of time, resources and energy caused by small changes in a formulation or the
manufacturing process that tipped the wormlike micelles into an unhappy state. Sadly, between us
there were no solutions to the problem – even though it is huge for the whole industry. The root
cause solution, selling soaps without this unnecessary viscosity effect, was not even discussed.

It also means that sulfate-free and ethoxylate-free formulations, demanded by consumers, are hard to
produce because APGs, for example, are unaffected by NaCl. Vast amounts of formulation effort have gone
into producing products with an effect, viscosity & yield stress, that is only a marketing requirement, not a
genuine consumer benefit.

Of course there are opportunities to use Thickeners_Associative Thickeners. But it is sad that we have to
add extra chemicals to an already over-complex formulation merely to add a marketing feature.

Luxury foam
Foam has no role in skin cleaning. Any surfactant tied up in a bubble is not at the skin surface doing
something. So foam is a bug, not a feature. But because it is hard to make a surfactant formulation that
doesn’t foam, the standard marketing approach is to convert a bug into a feature. Most consumers would be
offended by a non-foaming soap.

In standard lab foaming tests, which bear little relationship to real foam generation during washing,
combinations of CAPB with SLES give somewhat more foam that SLES alone. Hence we see SLES/
CAPB formulations as a common base formulation to which the other marketing attributes are added.

FST 123
The science of foaming (Surfactancy_Foaming) shows that true foamability is not strongly dependent on
the surfactant – you can get great foam from just about anything if you put in the right amount of shear
energy. So a not-very-good foamer can produce large volumes of foam if, for example, you use a Japanese
shaving foam net. Why anyone wants a lot of shaving foam (which provides no actual shaving benefit) is a
different question, the point is that you can get lots of foam even from something that happens to fail
“standard” foam tests. The common shower net creates lots of foam by the same high-shear mechanism.

It is said that foam offers two advantages:

1. It provides a signal that surfactant has been applied to all relevant parts of the body.
2. It helps to spread the otherwise hard-to-spread surfactant formulation

Given that most of the time we are not trying to clean off lots of tough dirt, and that the benefits of the
surfactant are modest at best, it’s probably more important that all relevant parts of the body are wetted and
wiped (the true cleaning process) than guaranteeing that surfactant has reached everywhere.

The “spreading” argument is necessary because the surfactant formulations are high viscosity and,
therefore, more difficult to spread. The marketing feature of luxury viscosity becomes a bug which has to
be fixed with the other marketing feature.

Destroying the skin biome

The human skin has evolved to look after itself rather well without the need for highly viscous, highly
foaming, vitamin-containing body washes. The array of microbes on the surface are actively taking care
that the skin barrier remains nicely balanced and that harmful microbes don’t get a chance to build up their
own rich biome community. By regularly attacking our own skin biome, we are making it hard for the
biome to look after us.

Recognising the damage caused by aggressive soap formulations, marketing require the formulators to add
extra oils to replace those removed from the skin. Even if these oils are “natural” they aren’t the complex
set our skin provided after millions of years of natural selection.

So now the next demand on formulators is to add pre-biotics, pro-biotics or maybe biotics. There is a nice
irony here because one super-strict demand on the soap formulations is that they are essentially microbe
free.

Destroying the microbes

A solid soap bar is a hostile environment to microbes. Indeed, as mentioned above, a high pH soap bar can
be nicely destructive to the skin surface, enabling it to clean more effectively, with obvious collateral
damage to the biome and the risk of harming (“drying out”) the skin itself. But at least your soap bar
doesn’t come with any extra pathogens.

The surfactants, oils, vitamins, pre-biotics etc. added to a liquid soap are a feast for microbes. Even if each
bottle of soap was fully sterilized during manufacture, once it is opened, microbes can enter and over weeks
or months the soap would be smelly and dangerous as the microbes took over.

So the liquid soaps need preservatives. By definition this means molecules that discourage or kill living
organisms. So by definition, this means a risk that they are harmful to some organisms on (our biome) or in
humans. So we have the nightmare scenario for marketing – they must assure users that everything in the

124 FST
 Soaps and Washing

soap is wonderful, chemical free, nourishing and beautiful, whilst admitting that it contains chemicals that
kill or inhibit natural organisms. It is a legal requirement that soaps demonstrate the ability to stop microbes
from proliferating.

Given that “natural” does not necessarily mean “good” or “safe”, marketing can’t hide behind the use of
natural preservatives. Many parabens are natural, benzoic acid and benzoates are natural but they are
demonised as “chemicals”.

The scientific advice is to go with a minimum package of approved, effective preservatives. But because in
these areas the science is of little importance to marketing and consumers, the formulator has to juggle
whatever are the latest fads from the latest internet scare.

Soaps to save the planet

The science of a “save the planet” soap is super easy:

• Use half as much

• Half the time
• Of the simplest-possible formulation

This is a 4x reduction in your carbon footprint, a 4x reduction in cost, with far fewer unnecessary chemicals
washed down the drain, with your skin biome less under attack, and with no downside to your general
cleanliness. Sweat itself is odourless and contains very little oil – so just a warm shower and gentle rubbing
removes it. The smell of sweat comes from the skin biome (especially a disturbed one) converting
odourless sweat molecules into odorous ones, such as hexenoic acids which are easily washed away with
water.

This is not what your management and marketing teams want to hear. But sooner or later, consumers will
catch up with greenwash from those selling large amounts of unnecessary chemicals. The laundry detergent
industry cut product volumes at least 20x along with the carbon footprint of the washing process thanks to
cold water formulations. Interestingly, further reduction of their carbon footprint requires dealing with the
relatively high footprint of the unnecessary fragrances that consumers demand. The footprint of natural
fragrances happens to be larger than their synthetic equivalents, so detergents with less fragrance and fewer
natural fragrance molecules will be better for the planet. These sorts of considerations will eventually reach
the soap community.

If you are going to choose to aim for a 10% reduction in carbon footprint, still encouraging consumers to
use too much of your product, then think hard. Life cycle analysis of bio-based or bio-sourced surfactants
rarely shows them to be significantly greener. For those keen to use biosurfactants, the triple challenge is to
make them with desirable properties, with a low carbon footprint and to make them in the million ton
quantities required to make a meaningful impact on the soap industry. Anything in the 10, 100 or even 1000
ton range is greenwash in terms of saving the planet.

Some smarter formulators are recognising that changing to bio-based surfactants “because they are green”
is not a good way to make profitable products. Instead, they identify some unique property of their
favourite molecule and create a novel product to deliver a genuine consumer benefit.

And if you are going to use a, say, C12-C14 chain, is it best for it to be bio-based or synthetic? Technically,
branched long carbon chains can be better surfactants because they don’t form such compact, liquid
crystalline phases. Because nature does not provide these, if you can reduce surfactant use by 10% through

FST 125
using a branched chain, is that better or worse for the environment?

If, as is clearly a strong trend, much of petrochemical feedstock comes from advanced chemical recycling
(converting to something like pyrolysis oil then putting that into a standard cracker) maybe it’s better to do
this than to debate whether palm oil or coconut oil is more sustainable.

The formulation scientist’s dilemma

The science of soap and washing is so simple that we don’t really need formulation scientists. We could all
agree on an effective general-purpose formulation that gives adequate cleaning, doesn’t mess up the skin
biome, uses sustainable ingredients and effective/safe preservatives, and which is used sparingly by
consumers.

That’s what the science says.

Management and marketing want the opposite – complex formulations with multiple ingredients offering
marketing claims that can out-compete other products, but all based on the lowest-cost ingredients because
price is a premium. The requirement for lowest-cost means that everyone needs to use the same ingredients,
produced at large scale. They want the impossible: to be the same but different.

At some point, reality always wins. Given that the science is clear, something has to change – either
industry first, leading the change, or consumer first with industry scrabbling to catch up.

126 FST
 Solvent-Based Paint

Solvent-Based Paint
Links
Adhesion_Adhesion promoters, Adhesion_Crosslinking, Adhesion_Entanglement,
Adhesion_Intermingling, Coating_Dewetting Theory, Coating_Levelling Theory, Coating_Pinholing
Theory, Dispersions_ODC, Dispersions_PVC and CPVC, Dissolution_Hansen Solubility Parameters,
Evaporation_Basics, Evaporation_Diffusion limited, Evaporation_Marangoni, Evaporation_Temperature
and Antoine Coefficients, Flow_G' and G'', Flow_Oscillatory rheology, Flow_Polymer Viscosity,
Flow_Rotational rheology, Flow_Shear dependent, Flow_Thixotropy, Flow_Yield Stress, Optics_Gloss,
Optics_Scattering and Opacity

You are the specialists in the alkyds, acrylics, urethanes etc. so we are not going to comment on the
chemistry. Instead we focus on the general principles behind all solvent-based paints.

Polymer solubility
Using standard techniques you can readily measure the Dissolution_Hansen Solubility Parameters of your
polymer. The three parameters, δD, δP and δH describe the Dispersion, Polar and Hydrogen bonding
characteristics of your polymer.

You can then choose a solvent with closely matching HSP values, using the 3D Distance metric to give a
number to the quality of the match. The closer the match the “happier” your polymer is in the solvent and
this generally makes your formulation life much easier. Your problem is that a solvent that is good in terms
of HSP match might be bad in terms of cost, safety, odour, evaporation rate etc. If you are restricted to a
single solvent, it can be hard to find an optimum.

Fortunately, the HSP of a blend of solvents is the volume weighted average of the two. This means that it is
possible to take two solvents that have desirable properties, yet a poor HSP match and create a blend where
the match is perfect. Surprisingly this even extends to creating a good solvent from two bad solvents.

The green dot in the centre of the green sphere is at the HSP of
the polymer. Any solvent inside the larger sphere is OK, outside
it is bad. The two blue dots show solvents that are each bad. But
you can see that a 50:50 blend would be exactly in the centre.
The point is not that we recommend making good solvents out
of bad ones, but to emphasise the potential of creating great
solvent blends from individual solvents that, by themselves, are
not as good as desired.

The Flow_Polymer Viscosity behaviour can be captured in the

app:
[Link]

We want the maximum possible polymer for the minimum solvent, but as the app shows, the viscosity
increases rapidly with concentration. As discussed in the Polymer Viscosity chapter, the rules for this
depend on Mc the critical entanglement molecular weight. This is a complex phenomenon which means
that relatively small changes to your polymer can lead to relatively large changes in viscosity at higher
concentrations. This should mean that polymer suppliers (including internal suppliers) should be confident
about formulating in terms of Mc – choosing polymer MW to match the desired entanglement phenomena.
Sadly, although polymer physicists have known for decades that Mc is super-important, manufacturers
generally don’t know, or choose not to reveal, the Mc values of their polymers. If you are a small-volume
customer, there’s nothing you can do. If you are from a megacorp, demand Mc values from suppliers –
maybe they’ll listen to you.

Solvent evaporation
As you find in Evaporation_Basics up to a point, the rate of evaporation of the solvent is governed by the
vapour pressure of the solvent and the airflow. You can see this in a simple app:

[Link]

The numbers you see after Acetone are its MW, MVol and the first two of the three Antoine coefficients,
Evaporation_Temperature and Antoine Coefficients, provided with each solvent. The rate of evaporation is
proportional to the vapour pressure (calculated from the Antoine coefficients and temperature) and to the
square root of air velocity – paint dries very slowly in a room with no air flow!

Solvent blend evaporation

There is a key scientific advantage to using a solvent blend, and a significant disadvantage. First, the
advantage (in addition to the cost, safety, green advantages opened up by solvent blends):

128 FST
 Solvent-Based Paint

[Link]

Here we have chosen to dissolve polycarbonate using a blend of 70% heptane and 30% cyclohexanone –
[not because they are green but because we get a nice graph to tell the story]. At the start, the solvent blend
has an HSP Distance (Ra) of 7.7, making the mix a relatively poor solvent, but OK for the start of the
process where the concentration is relatively low. Because heptane evaporates faster than cyclohexanone
(their Relative Evaporation Rates are 390:29) by the time we are down to 20% of the original total amount
of solvent, we have mostly cyclohexanone which has an Ra of just over 3. This means that in the difficult
diffusion-limited part of the drying process (discussed soon), the chances are that the polymer is “happier”
and more open. If this coating is a gloss paint then the surface will be smooth because imperfections have a
longer time to flow out.

If the relative evaporation rates had been reversed then the good solvent would have left early and the
polymer would have crashed out early.

But now the problem of these solvent blends. The surface tension, σ, of heptane is ~20 mN/m while for
cyclohexanone it is ~25.5 mN/m. If there are any local variations in surface composition then there will be
local surface tension gradients, δς, … and these can rapidly result in Evaporation_Marangoni effects.

The local variations drive flows which spontaneously

form into patterns from subtle through to full hexagons.
A lot of “orange peel” in paints and coatings is due to
these Marangoni patterns.

If you have Marangoni problems you can either fix the

relative evaporation rates of your solvents, or choose
solvents with closely matched surface tensions. Or, if you look at the relevant equations, a thin coating and
a large viscosity is enough to stop the effect.

In terms of additives, any component that moves rapidly to the surface and swamps the surface tension
effects of the solvents will do the job, because there will be no surface tension gradient to drive surface
flows. People say that you need “low surface tension” to level out defects. This is simply untrue; levelling
is driven by surface tension so the higher the better. Instead molecules that partition to the surface to swamp
Marangoni effects have to be low surface energy, otherwise they don’t partition. The end result is the same,

FST 129
but the science is very different. Silicones are especially good at “levelling” (in reality, destroying
Marangoni) because they partition so well and have a low, temperature-independent, surface tension that
swamps other effects. The downsides of silicones are equally well known, so these “levelling agents” have
to be used with caution.

Diffusion-limited drying
At some point, all drying becomes limited not by the evaporation but by the rate at which solvent molecules
can reach the surface. As discussed in full, Evaporation_Diffusion limited, there are only three things you
can do:

1. Use a solvent that keeps the coating open for as long as possible, as discussed above.
2. Heat the system, because diffusion coefficients increase exponentially with temperature.
3. Delay the curing of your polymer system till the solvent has mostly gone, because the crosslinking
will decrease the diffusion coefficient.

Controlling the pigments

Your specific pigments are chosen to give hiding power (Optics_Scattering and Opacity ), to provide cheap
bulk that also forms the core of a tough paint, and to provide colour. Although we naturally work with
weight %, what really matters is volume fraction, φ. If your pigments are well-dispersed then simple
viscosity theory shows Dispersions_Rheology (Low shear) that you can add a lot of pigment with only a
modest effect on viscosity:

[Link]

For spherical particles, up to about 45% nothing much happens. The problems then take off, with the rate of
increase depending on the point at which the particles have no ability to move past each other, the close
packing limit. Things get more complicated when particles are elliptical and when your dispersant isn’t
ideal. The Dispersions_Rheology (High shear) discussion takes you through those issues.

More pragmatically, you need to calculate the Pigment Volume Concentration (PVC) and its critical value,
the CPVC, Dispersions_PVC and CPVC:

130 FST
 Solvent-Based Paint

[Link]

As you get close to the CPVC, paint properties such as coating density, gloss or water resistance become
susceptible to small changes in the formulation so you have to work harder on your pigment and its
dispersant to be able to push the limits.

A good dispersant makes life easier, so finding the Dispersions_ODC

Optimal Dispersant Concentration is necessary in both senses: a good
dispersant has the lowest optimal concentration and the smallest
interparticle forces (lowest viscosity) at that concentration. It is a good
strategy to get the dispersant optimized on the pigment before it gets
into the paint. Adding the dispersant to the paint is likely to lead to
disappointing results because a bad dispersant already on a pigment is
surprisingly hard to displace with a better one.

Controlling the rheology

The formulation team need to embrace rheology. There seems to be a lot of it, Flow_G' and G'',
Flow_Oscillatory rheology, Flow_Polymer Viscosity, Flow_Rotational rheology, Flow_Shear dependent,
Flow_Thixotropy, Flow_Yield Stress but it really comes down to flow curves in rotational rheology
(looking at viscosity as a function of shear rate – both with increasing and decreasing rates) and, less
popularly but more fundamentally, via oscillatory rheology and G' and G".

The Abbott Guide to Rheology at [Link]

is an attempt to demystify rheology. It really is unfortunate that something so essential is often
seen as some deep magic knowable only to rheologists. Knowing how to get good data from a
rheometer is indeed tricky, which is why we need rheologists. But knowing why we want the data
and knowing what to do with it, is not at all hard.

A good specific example is in controlling levelling and sag. You want easy flow to allow
Coating_Levelling Theory yet you want a very high low shear viscosity to resist sag – flowing down a
vertical surface.

An app lets you think through these issues:

FST 131
[Link]

We examine the flow curve over a very small range of shear rates, γ̇, e.g. up to 5/s. The data can be fitted to
a Herschel-Bulkley law:

τ0 + Kγ̇n
η= γ̇

Here τ0 is the yield stress. As it happens, given day-to-day accuracy of measurements of real paints, there’s
little advantage trying to extract a yield stress value and you can simply fit the data to a power law:

η = Kγ̇n − 1
Because shear stress is simply viscosity times shear rate, we can plot the same data in terms of shear stress,
the right-hand curve. The reason is to see that typical stresses are in the 1-5 Pa range which, if you do a
calculation based on a typical paint thickness and density is the sort of driving force inducing sag. From
this you can even estimate the ASI (Anti-Sag Index) used in the paint industry; it is the largest thickness
(alas, in US mils) of a line which doesn’t sag down to the next thinnest line below.

What is interesting about the paper (quoted on the app page) that links the power law fit to the ASI is that
they can get just as good a predictive value from measuring G' via oscillatory rheology. But it’s not just
“the” G'. Instead the paint is sheared for a time then the oscillatory measurement is started. Over time the
paint structures itself (discussed next) and achieves a relatively large G'. It is the short-time G' (not
surprisingly) that links to the tendency to sag.

A theme running through FST is that G'/G" measurements are underutilized because of their perceived
difficulties. With modern rheometers being so user-friendly, this oscillatory approach to understanding the
paint structure seems more insightful than more conventional and difficult low shear rate rotational
viscosity measurements.

132 FST
 Solvent-Based Paint

Paint structuring
The rheological effects of the polymers and well-dispersed particles are a core necessity of a paint. To get
the subtle details right, such as the correct power law behaviour for anti-sag with Emulsion Paint is
relatively easy via Thickeners_Associative Thickeners. There are fewer fancy options in the solvent-based
formulation world, so attention to the subtleties of polymers and pigments is a necessity.

One key trick is to perform rheology analyses not just of your standard paint formulation but also on
formulations with less solvent. This lets you see how quickly (usually good) or slowly (usually bad) the
paint gains structure as the solvent evaporates. A paint with a higher starting structure but a less-steep rise
in structure on drying may well be inferior to one with a lower structure but a steeper rise.

Avoiding particle settling

[Link]

Stokes law tells us that the time for a particle of radius r and density ρp to fall a distance h in a medium of
viscosity η and density ρl (and therefore density different Δρ = ρp- ρl) is given by:

hη
t=
2.18gΔρr2

So less dense particles or a higher-density solution, higher viscosities and smaller particles help keep
particles in suspension. An extra factor is the volume fraction φ of particles. The higher φ the more the
particles self-crowd each other, slowing the speed. The Richardson and Zaki equation tells us that velocity
is reduced by (1-φ)5.65.

Elliptical particles fall a bit slower, via an equation shown in the app.

Crosslinking and adhesion

The build-up of a crosslinked network is difficult to simulate numerically, but, as described in
Adhesion_Crosslinking the chances of converting a large fraction of the reactive groups in low-
functionality systems is much higher than the same number of groups in a high-functionality system. The
difficulty is that highly crosslinked systems restrict the motion of the remaining reactive groups so they are
unable to reach another reactive site. Unreacted crosslinking groups risk the danger of undesirable reactions
over time (e.g. with water and oxygen), which in turn lead to deterioration of the paint.

Putting it another way, the Tg of a highly crosslinked system is higher, so main-chain segmental motion is

FST 133
restricted as soon as the system reaches this Tg. As is obvious, raising the temperature allows more reaction
… till the Tg rises to this new temperature.

An idealised view of this is found at:

[Link]

A paint, therefore, is a pragmatic compromise between crosslink density for general strength/resistance and
the sort of temperatures or times allowed to reach the desired density.

If the chemistry crosslinking the paint also allows a modest amount of reaction onto functional groups on
the surface of the coating (e.g. reacting onto surface -OH groups) then there is the necessary
Adhesion_Entanglement across the interface to give a robust bond to the surface.

134 FST
 Sun Screens

Sun Screens
Links
Diffusion_Diffusion into Skin and Hair, Dispersions_Rheology (High shear), Dispersions_Rheology (Low
shear), Dissolution_Hansen Solubility Parameters, Dissolution_Ideal solubility, Evaporation_Basics,
Evaporation_Humectants, Flow_Basic viscosity, Flow_Yield Stress, Optics_Scattering and Opacity,
Surfactancy_Emulsification, Thickeners_Associative Thickeners, Thickeners_Polymeric Thickeners

A good sunscreen is hard to achieve. It is easy to have lots of UV-blocking chemicals or particles,
it is easy to make a water-and-sweat-resistant product, it is easy to make a super-safe product, it is
easy to make an aesthetically pleasing product. But those desirable properties provide many
conflicts and trade-offs.

SPF
Whatever your formulation, the product you sell has to have a known, validated Sun Protection Factor, SPF.
The app makes it easy to explore the options for creating whatever SPF you are aiming for. In principle you
can use any number of UV absorbers; the app lets you use 2 as a good-enough basis for exploring.

[Link]

The SPF is the ratio of the integrated amount of potentially damaging UV (this turns out to be the red line)
divided by the amount that gets through your filter, the blue Teff(λ) line which has been greatly magnified
so you can see what it’s doing. So the SPF is the ratio of integrals, I1 and I2 which are provided in the
output. The CW is the Critical Wavelength which helps describe the range of filtering capability, you also
get the ratio of UV-A and UV-B absorption and, finally, the SUI, the Spectral Uniformity Index. The
equations and deeper explanations are provided in the app.
Choice of absorbers/scatterers
There are plenty of small molecules with strong absorption in the appropriate wavelengths. Dissolving
them in some pleasant cream used to be easy, and you would have an adequate sun block. The number of
such molecules that you can use is diminishing rapidly, and the number of cosmetically-approved oils is
also diminishing so it’s hard to get the required concentrations. But even if you can dissolve them and the
molecules themselves are entirely safe sitting on the surface, there are multiple problems:

• When they absorb UV they can react and create problematic chemicals that, at the very least, are
irritants.
• By reacting they will (generally) lose their UV absorbance so their efficacy wears off over time in
the sunshine.
• In water they can dissolve/disperse and potentially damage the pool or marine environment
• Being small molecules in a skin-friendly cream, they can migrate through the stratum corneum of
the skin (see SkinCare) and once through that, there’s nothing much to stop them getting into the
rest of the skin and into your bloodstream.

The problem of UV reactivity and loss of absorption can only be solved via a clever choice of absorbing
molecule or by adding co-molecules that take away the UV energy from the absorbent, before it reacts.

For a fuller understanding of diffusion through the skin, see Diffusion_Diffusion into Skin and Hair. It is a
polite fiction in cosmetics that chemicals only go in to the dead stratum corneum and magically stop there,
but the reality is very different.

Given that the number of viable small-molecule UV absorbers is diminishing rapidly, and given the
problem of marine pollution, there are two logical alternatives. First, make the small molecules rather large,
such as Bemotrizinol, which is also water insoluble. Second, make them into a polymeric form. They
certainly won’t migrate through the skin and their relative lack of mobility decreases the chance of UV-
created molecules directly irritating the skin surface. With the right polymer, they won’t wash off into the
water.

The other approach is to use particles as scatterers and/or absorbers. For optimal scattering,
Optics_Scattering and Opacity, you need relatively large particles with a high refractive index difference
from your cream. Particles like ZnO (n~2) and TiO2 (n~2.5) do this very well. They are in themselves safe
(the TiO2 surface is treated to ensure that it doesn’t generate reactive oxygen species) and because they are
large (in order to scatter) there are no issues with nanoparticles. Formulating the cream so that it doesn’t
easily wash off means that these formulations are ideal sun screens in every respect other than aesthetics.
People generally don’t like going around smothered in white gunk.

The same particles formulated to nanoparticle dimensions no longer scatter but have plenty of UV
absorption. Formulating a stable nanodispersion is tricky, but good performance from such strong
absorbers, that don’t degrade in the sunlight, is easy to achieve. But now we swap scares of chemicals for
scares of nanoparticles. The Australian approach to nano-sunscreen safety has a lot to be recommended:
“Sure, there are theoretical small risks from nanoparticles, but compared to the high risk of skin cancer, it’s
a no brainer.”

Formulating the base cream

Some formulations are basically the UV blockers themselves dissolved in a volatile solvent such as ethanol.

136 FST
 Sun Screens

A typical (old-fashioned) mix would be Homosalate, Octocrylene, Ethylhexyl Salicylate and Avobenzone
(or Oxybenzone). Add a bit of polymer, maybe a siloxane, some pleasant fragrance and a few marketing-
friendly natural extracts and you have a classic, effective, sunscreen. Because absorption is a combination
of extinction coefficient and path-length, the concentrations of active ingredients have to be high because
the final layer is thin.

You need to understand the solubility of any solid component in any of the liquid components. For this
Dissolution_Hansen Solubility Parameters are especially helpful because you can readily calculate the
parameters of the initial solvent blend (a volume-weighted average of the individual components) and can
see what happens when volatile components (such as ethanol) leave. What you don’t want is sudden
crystallisation of a solid component (becoming “gritty” on the skin) because it finds itself in an alien
solubility environment.

The alternative is to apply a typical 10µm coating of some pleasant base cream with a lower concentration
of the actives or, similarly, a cream-based dispersion of particles. Because there is plenty of warm sunshine
to evaporate the water, the cream is likely to be an o/w emulsion. So choose your oil by whatever
marketing, cost, safety, sustainability or sensory criteria you like and formulate the emulsion rationally.
Although there are many ways to perform Surfactancy_Emulsification, doing it rationally via
Surfactancy_HLD theory will let you use minimum surfactant for maximum efficiency and will allow you
to explore alternatives to the standard ethoxylates and ionics that arouse passions around health, safety and
natural/sustainability issues.

The rheological requirements are the big challenge. First you need to get the Flow_Basic viscosity right.
Then a luxurious cream should have a significant Flow_Yield Stress so that a blob on the hand just sits
there. But it needs to be highly Flow_Shear dependent so that it flows easily under the shear of the fingers
rubbing it onto the skin. If the final coating is thin, it probably doesn’t need to show Flow_Thixotropy, but
it’s probably worth doing a thixotropy scan to see if this is a feature or a bug of your formulation.

You have to decide whether to use Thickeners_Associative Thickeners or Thickeners_Polymeric

Thickeners. Polymeric thickeners have relatively modest shear-thinning properties so their properties
change less during application and use. The polymers can be chosen for their other properties such as water
resistance. The associative thickeners lose their viscosity very easily with shear, and can restore their
viscosity and yield stress very quickly, so they are good during the moments of application.

Particle Dispersions
You need to find a great dispersant for your particles, ideal using the
Optimal Dispersant Concentration method: Dispersions_ODC

The all-important rheological behaviour will be a combination of simple

Dispersions_Rheology (Low shear) behaviour then, beyond a typical
volume fraction (usually 0.28) you need the Dispersions_Rheology
(High shear) behaviour. Whether you have desirable or undesirable yield
stress and flow under shear depends partly on ideal particle rheology
(which is what the apps mostly describe) and partly effects (fractal
dimensions, N-clusters) owing to poor dispersions. The ODC approach should minimize those and make
life easier.

FST 137
Surface Cleaning
Links
Cleaning_Bleaching and Oxidation, Cleaning_Boundary removal, Cleaning_Contact Angles,
Diffusion_Basic Diffusion, Diffusion_Diffusion Coefficients, Dissolution_Dissolution Kinetics,
Dissolution_Dissolution Rheology, Evaporation_Basics, Flow_Basic viscosity, Fragrance_Activity
Coefficients, Fragrance_Vapour Pressure, Surfactancy_CMC and Langmuir, Surfactancy_Coacervation,
Surfactancy_Dynamic Surface Tension

Designing an effective surface cleaner is easy. Designing an “easy wipe” surface cleaner is
somewhat harder. Designing an effective, easy wipe surface cleaner is fighting the laws of physics,
so is hard.

An effective surface cleaner

To remove dirt from a surface you need to get a flow of the
cleaning fluid (that high velocity V in the diagram) to reach the dirt
along with sufficient mechanical energy to remove the dirt into the
bulk fluid, without damaging the surface. There are two barriers to
this, the first one of which is shown in the diagram – zero flow at
the surface:

1. The No Slip Boundary Condition (Cleaning_Boundary removal) tells us that the fluid velocity at
the surface is zero, so there is no flow of fresh cleaning agent.
2. We need mechanical energy at the sub-µm level, yet cloths and brushes are, of necessity, on the
sub-mm to mm level.

Partial answers to the second problem are:

• Microfibre cloths which have the necessary strength of large fibres, along with many sub-fibres
that greatly increase the mechanical energy across the surface.
• Particulate systems where the particles are given sufficient energy (inertia and particle-particle
contact) to hit the surface and provide cleaning action.

If you aren’t in a hurry and can allow time for cleaning chemicals to diffuse to the surface and for the
dissolved dirt to diffuse away then you can use a powerful cleaning formulation such as a laundry detergent
(Laundry Liquids) which contains the optimised surfactants plus lipases to attack the oils, proteases to
attack protein stains, lipases for any starches and bleaches for hard-to-remove stains. It will contain
chelators such as EDTA or a greener alternative to remove stubborn calcium and magnesium salt residues.
This chemical combination is sufficiently potent to be effective, while not likely to endanger any typical
surface to be cleaned.

There are two big problems with any effective cleaning formulations:

1. Human skin contains many of the elements we wish to remove from a dirty surface. So a good
 Surface Cleaning

cleaning formulation requires the user to wear gloves – a big practical barrier for an every-day
product.
2. If the formulation dries out, all those chemicals leave a nasty-looking residue. So the cleaning
formulation has to be wiped away with clean water. This is more work, and hardness in the water
can end up as unsightly residue on the surface.

There is one obvious fact linked to the second problem, yet seldom addressed by users or formulators. Any
formulation that’s more than, say, 1 µm away from the surface is of no use – it’s just excess wasted
chemicals that require a lot of effort to remove. An ideal cleaning system would deposit just 1 μm of
formulation onto the surface with enough physical motion to enable adequate mixing to avoid long
diffusion times. The Easy-wipe formulations described next achieve the thin layer by being delivered as a
fine spray. But as this is a dilute formulation with little intention of providing cleaning power, it is just thin
and insufficient for removing tough dirts. A spray that can handle the higher viscosity of a potent cleaning
solution would enable a formidable type of cleaning system.

I was seriously unimpressed by a new spray cleaning product I was asked to help improve. They
showed me their standard cleaning test, which their product passed with no problem. I said that
plain water & a microfibre cloth would do the same thing. “OK, prove it!”. They set up the test,
gave me a water spray and a microfibre cloth. I realised that the cloth would be swamped by the
large amount of gunk, so I did a quick wipe with a paper towel, then with the microfibre cloth.
They accused me of cheating by using the paper towel, but I’d proved the physics. Happily, work
on the project was soon abandoned, freeing up resources to work on better projects.

Easy-wipe formulations
Light dirt is easily removed with (deionized) water (see the Adhesion part of Particles_Basic behaviour),
helped by a low level of surfactant to wrap around oily particles to give them a hydrophilic surface, and not
so much to reduce surface tension and decrease the standard measured contact angle Cleaning_Contact
Angles but to provide a low receding contact angle so the film of water does not bead up.

As has been commonly observed, using a microfibre cloth significantly increases the cleaning efficacy of
these formulations, though you don’t use easy-wipe cleaners for removing significant dirt.

As we are not requiring the surfactant to do anything very interesting, we can choose the lowest-cost
surfactant which also has whatever marketing claims are desired: “natural”, “biobased”, “sulfate free” etc.
However, as discussed in the next section, surfactants that “dry out” are likely to leave an undesirable
visible residue.

As these are likely to be spray formulations to give a broad, light coverage, and because the light loading of
surfactant isn’t going to change the spray characteristics very much, the focus should be on the design of
the spray system itself. As the Flow_Atomization chapter shows, there is distressingly little good science to
help with this.

My wife and I once used a large mirror to test out a few different easy-wipe sprays. The winner
was obvious – the design of their spray nozzle was superior, delivering a fine, even mist that wiped

FST 139
 away easily. [Later discussions with an industry expert confirmed that the spray device is at least
as important as the formulation.] One of the “good” formulations showed a problem some days
later. A visible residue appeared. Puzzled, we re-tested at a different area and a few days later the
same thing happened. One explanation is provided below.

Easy-wipe, effective formulations

Although “effective” cannot mean powerful cleaning, it can mean a surprising level of efficacy over the
longer term. The trick is to include some sort of coacervate system (Surfactancy_Coacervation) which
involves cationic polymers along with anionic surfactants. The coacervate can’t help with the cleaning
itself (how could it?). Instead it precipitates as a nano-thick layer on the surface. As is well known (see Hair
Conditioners) the cationics are relatively hygroscopic so the surface remains hydrated. This has two
benefits:

1. As the formulation stays semi-liquid, it remains invisible, solving the previously-mentioned

drying-out problem. [In the failed mirror test, I assume that the formulation held on to the water
for a day or two, but eventually dried out. Later discussions with an expert confirmed that this is
the case]
2. Any oil/dirt reaching the surface now sits on a hydrated layer with low adhesion. So the next
treatment with the easy-wipe formulation renders the surface beautifully clean once more.

How do you know that there is a nm layer of a coacervate on the surface? While ellipsometry can tell you
there’s something super-thin on the surface, how do you know what it is? Amazingly, although ATR-FTIR
has a natural depth resolution of a few µm, with some standard interfacial tricks, you can get a good
spectrum of the coacervate sitting at the interface. Those who are really keen can use AFM-FTIR. Those
interested in such methods can read more in Analytical Techniques.

140 FST
 Toothpastes

Toothpastes
Links
Cleaning_Boundary removal, Flow_Oscillatory rheology, Flow_Particle Viscosity, Flow_Polymer
Viscosity, Flow_Rotational rheology, Flow_Shear dependent, Flow_Yield Stress, Particles_Size
distribution, Surfactancy_Foaming

A toothpaste is a necessary compromise – too good at cleaning and it damages the teeth, too poor
and the teeth suffer. The key properties of a toothpaste come from the particles – they control the
cleaning function as well as the way the toothpaste handles as a squeezable paste. Apart from the
proven efficacy of the fluoride, other ingredients play a minor role in dental health, though are vital
for consumer perception and marketing claims.

To a large extent, cleaning is about overcoming the problem of Cleaning_Boundary removal. The No-Slip
Boundary condition says that any cleaning fluid has zero velocity at the surface … which is the place where
you most want the cleaning fluid to be flowing.

A toothbrush overcomes this problem so there is effective cleaning power even without a toothpaste.
Unfortunately, the total contact area between brush bristles and the teeth is very small, so cleaning is
inefficient.

A paste of particles allows the mechanical energy of the brush to impact a wider area, so cleaning is more
effective.

A stiff brush and a paste of small, sharp particles provides outstanding cleaning … but also erodes the
enamel of the teeth. Given that enamel does not replace itself, this is catastrophic. A soft brush and a paste
of soft, round particles leaves the enamel undamaged, but is also poor at removing residue on the teeth.

So cleaning teeth is about a medium hard brush (the overwhelming majority of brushes sold) and a paste
with a balance of particle sizes and hardnesses.

Optimal particle size distribution

Typically a toothpaste mixes together different silicas, carbonates/bicarbonates and calcium phosphates on
the basis of fashion, cost, marketing claims and technical performance. Focussing only on the latter, what is
the optimal Particles_Size distribution? As the link explains, knowing what your distribution is and
extracting numbers to capture it is a hard combination to achieve. For our purposes, smaller is generally
better as you get more surface area to do the cleaning, and oversized particles feel “gritty” and unpleasant
when brushing. Because of fears of nanoparticles, life is easier if the distribution has nothing below 100nm.

In terms of formulating, you want the maximum amount of particle to pad out your toothpaste. If all
particles were the same size then you could not get much beyond 60% volume fraction because of the close
packing limit. As discussed in Flow_Particle Viscosity, one way around this is to have a multi-modal size
distribution where smaller particles fill the voids between larger one. We also see that in terms of having a
tractable viscosity, spherical particles are better than elliptical or rod-shaped ones.
Particle dispersion
Any supplier of particles to the toothpaste industry knows that the particles should be readily dispersed in
water. Any of us could buy such particles, mix them with a little water into a thick paste and have a viable
toothpaste. The rising popularity of toothpaste “tablets” made from particles compacted together show that
cleaning teeth is not a super-sophisticated activity

The extra ingredients such as a bit of surfactant to remove oils and create foam, a humectant
(Evaporation_Humectants ) such as glycerol to stop the toothpaste drying out underneath the cap, a bit of
rheology modifier polymer, the fluoride and extras such as nitrates and the minty flavour we generally seem
to like, make little difference to creating an adequate paste.

So where is the challenge?

A blob of toothpaste sitting on a brush should just stay there. So it needs a high Flow_Yield Stress. Yet
when we squeeze the tube it should flow out easily, so it needs Flow_Shear dependent behaviour which,
happily, comes naturally from particle-filled systems. But not too easily. There are strict tests that hold a
tube of toothpaste with opening down, timing how long it takes for the contents to flow out – too fast or too
slow and the batch is rejected.

The challenge is that relatively small changes to particle sizes, shapes, dispersion, surfactant, humectant,
concentration can make subtle changes to any one of the critical rheological properties.

How, then, does the formulator formulate? A common theme in FST is the need for “smart mapping”. You
cannot know precisely how a change in particle size. or change of surfactant, or changes to your production
process, or change to rheology modifier will affect the several competing rheological needs. But by
regularly measuring the core rheological properties, especially via Flow_Oscillatory rheology, you gain a
map of how the competing requirements change. With a reasonable map, understanding how to fix one
problem (e.g. a change in yield strength) without causing another (e.g. difficulty to squeeze out of the tube)
becomes possible.

It was once famously said that the only significant ingredient in toothpaste is the fluoride. So how
can manufacturers claim that their special formulation is objectively better thanks to another key
ingredient? The trick is to do the measurements under idealised test conditions which might show a
1% improvement in some desirable tooth property. Although this 1% improvement would not be
detectable by any customer and, in any case, toothpaste isn’t used in the same idealised way, the
claim is valid, and consumers are persuaded to buy the “improved formulation”.

Foaming
Consumers instinctively feel that foaming in cleaners is a “good thing”. But given that teeth cleaning is
about delivering adequate cleaning force to the tooth surface via brush and particles, foaming plays no part.
If a surfactant is required to help remove oily residues from the teeth, then, again, foaming does not help –
it just ties up surfactant molecules uselessly around air bubbles. If, as seems likely, foaming is there as a
signal to the user that they’ve used an adequate amount of toothpaste, and, additionally, as a feel-good
marketing symbol, the question is how to get the right amount of Surfactancy_Foaming. When you delve
into that link you will find that there are far too many myths about foamability and that standard tests are

142 FST
 Toothpastes

near useless for your specific problem. By all means formulate for foam, but do it rationally.

Why are there so many different toothpastes?

Marketing. Given that most of the cleaning comes from the brush and the particles, other than
extreme whitening (lots of sharp particles, so lots of enamel erosion) there’s little practical
difference across the confusing array of “different” toothpastes with remarkably similar ingredients
on the box.

If any of them was especially better, we’d all be buying it. And we’re not.

How do you make a rational choice? The ones at eye level on the supermarket shelf are the
unnecessarily expensive ones. Just look on the bottom shelf for any “normal” fluoride toothpaste at
a more sensible price and it will work fine.

FST 143
Water-based Adhesive
Links
Adhesion_Adhesion promoters, Adhesion_Butt, Adhesion_Crack resistance, Adhesion_Crosslinking,
Adhesion_Entanglement, Adhesion_Intermingling, Adhesion_JKR, Adhesion_Peel, Adhesion_Shear,
Adhesion_Surface energy Interactions, Adhesion_Testing, Diffusion_Basic Diffusion,
Diffusion_Concentration-Dependent Diffusion, Diffusion_Diffusion Coefficients, Dispersions_DLVO,
Dispersions_Rheology (High shear), Dispersions_Rheology (Low shear), Dispersions_Settling,
Dispersions_Zeta potential, Dissolution_Hansen Solubility Parameters, Evaporation_Basics,
Evaporation_Diffusion limited, Flow_G' and G'', Flow_Oscillatory rheology, Flow_Polymer Viscosity,
Flow_Rotational rheology, Flow_Shear dependent, Flow_Stefans Squeeze, Flow_TTS-WLF,
Flow_Thixotropy, Flow_Yield Stress, Humidity_Water mechanical isotherm, Humidity_Water vapor
isotherm, Particles_Sintering, Surfactancy_Dynamic Surface Tension, Surfactancy_Emulsification,
Surfactancy_Emulsion Coalescence, Surfactancy_Emulsion Creaming and Flocculation,
Surfactancy_Emulsion Ostwald, Surfactancy_HLD

Here we ignore all the Adhesive aspects of an adhesive and focus on those special features that
arise because the adhesive is water-based.

Soluble adhesives
There is little to discuss – the physics are the same as standard solvent-based adhesives. Their obvious
downside, susceptibility to water from the atmosphere or from direct contact make them of little interest to
the FST community, except for those working with young children where the ability to clean up with water
after some adhesive use is a real advantage.

From dispersed to insoluble

For simplicity we will call the dispersed blobs of adhesive polymer emulsion drops, even those that aren’t
strictly emulsions. By doing this we can use much of the world of emulsion science without bothering too
much about the fine details.

The problem at the heart of these emulsion systems is that the individual drops much be infinitely stable
when dilute, yet should form a continuous film of polymer once the water has (largely) disappeared. Each
property on its own is not a problem – getting both right is the challenge.

A stable dispersion
For these aqueous systems, Dispersions_DLVO theory is adequate.
 Water-based Adhesive

[Link]

The particles can be stabilised via:

• Charges, typically via an anionic surfactant. If the zeta potential ([Link]

practical-solubility/[Link]) is somewhere over 30 mV then the particle has a good chance of
being stable.
• Steric effects, a long hydrophilic (typically ethoxylate) chain sticking out.

Arguably, charge-charge repulsion is the stronger stabilization method because steric repulsion can be
changed into attraction (bridging or depletion flocculation) via extraneous polymers. The downside is that
those water-loving ionic groups are not desirable in the final adhesive.

The dispersion also has to be stable against settling or creaming:

[Link]

Stokes law tells us that the time for an emulsion drop of radius r and density ρo to rise a distance h in water
of viscosity η and density ρw (and therefore density different Δρ = ρw- ρo) is given by:

hη
t=
2.18gΔρr2

In the screen shot, a classic oil emulsion is assumed, hence the low density. Tuning an adhesive emulsion
drop to have a density of 1 would eliminate issues of creaming or settling. The volume fraction input, φ,
allows the use of the Richardson-Zaki formula where the velocity is reduced by (1-φ)5.65. A higher volume
fraction also increases the viscosity, as discussed next.

FST 145
Viscosity and rheology

[Link]

Up to a volume fraction, φ, of emulsion over 60%, the viscosity increases, depending on the model and on
the internal viscosity of the droplet, by about 40x. In this region the close packing limit for rigid spheres is
being reached and the precise increase to higher φ values is too complex to model. In any case, these
models are for smooth spheres with no attractive forces between them. As you try to go to a higher volume
fraction to make the adhesive viable, we lose any good theory and have to rely on rheology to compare/
contrast formulations.

For the most information with the least effort, checking out Flow_G' and G'' is effective. The balance of
elastic (G') and lossy (G") shear offers insights not so much into any specific formulation, but into trends
occurring with changes of concentration or ingredients.

Straightforward flow curves Flow_Shear dependent and maybe up/down curves for thixotropy
Flow_Thixotropy will provide information relevant to handling the adhesive during production, filling and
application. If the low shear performance (e.g. avoidance of sagging if applied to a vertical surface) is
important then the Levelling & Sagging app at [Link]
Levelling-&-[Link] is useful.

Film formation
It is helpful to introduce a neutral, and slightly unusual, word into the discussions: sintering. There are
many terms used to describe the joining of blobs that are packed close together, but they lack a coherent
theory to capture the key parameters that influence the phenomenon. The Rumpf equation for sintering
describes the key elements behind the joining of particles, and allows us to identify how changes in our
system can affect film formation for better or worse.

146 FST
 Water-based Adhesive

[Link]

The Rumpf equation tells you how overlap x (in the diagram you see 2x) changes over time t for particles
with radius r and “viscosity” η and surface energy γ. The equation includes any applied external force, F,
which we can assume to be 0.

( xr )
2
=
t
η (0.8 γ
r + 0.4
F
πr2 )
The use of x/r in the formula is deliberate because we’re generally interested in the fractional sintering – a
value of x on its own doesn’t convey much.

There are a few interesting aspects to this equation:

• Contrary to instincts, a low surface energy, γ, reduces the rate of sintering. This isn’t surprising
because the process is driven by surface energy. The surfactants necessary to disperse the blobs
now become unhelpful. However, the surfactant effect on surface energy is likely to be no more
than a factor of 2, so it’s not worth worrying about.
• Smaller particles fuse faster. Unfortunately they are also harder to keep separate in the original
formulation, so there are trade-offs
• We need to attend to the viscosity of our particles. In addition to this being a difficult thing to
measure, the variation of viscosity with water or concentration of “coalescing solvent” is an added
complexity.

The apps associated with the app above are all related to food science. But physics is universal and the idea
that the Tg changes dramatically with moisture content which in turn is governed by the water-vapour
isotherm means that there is a logical chain for investigation. Your adhesive is designed to dry out, but what
is the water-vapour isotherm – how dry is dry? And what is the effect of small amounts of water on the Tg?
If film formation is complete long before the water has evaporated then these questions are irrelevant. The
important thing is to know if/when film formation happens.

If the particle Tg is not affected by the water (an advantage for long term performance) then either the Tg
has to be low enough for the film forming temperature to be close to room temperature. Or you have to use
a coalescing solvent. The solvent must have a lower volatility than water so its concentration increases
sufficiently to lower the Tg and decrease η. It must also have a good Dissolution_Hansen Solubility
Parameters compatibility with the polymer.

Thickeners
In theory, the rheology of an aqueous adhesive can be controlled by additional Thickeners_Associative

FST 147
Thickeners and Thickeners_Polymeric Thickeners. There is a trade-off between the tunability of these
systems and their interference with the intrinsic properties of the adhesive.

Where does the water go?

If you do some calculations using the Concentration Dependent Diffusion app at
[Link] and make generous assumptions that
the diffusion coefficient of the wet adhesive is 10-7 cm²/s and that of the dried adhesive is 10-9 cm²/s then
in 1000 min you can remove 90% of the water from a blob of adhesive 200µm wide used to stick together
two impermeable adherends. If we wanted to stick something that was 2cm square that’s a 100x increase in
length so a 104 increase in time, taking us to 107 min, nearly 20 years.

So we have a choice:

1. Pre-dry the adhesive on each surface and hope they act as a cold-seal adhesive, like a PSA but
somewhat tougher because the surfaces don’t need to (and probably don’t) stick to anything other
than the other surface. Clearly this isn’t super-strong, but because Adhesion is a Property of the
System, if you think of the system clearly, this might be strong enough.
2. Only stick onto permeable surfaces such as wood, leather or paper. Although we say that in such
joints we need to allow the glue to “dry”, really the water is diffusing/absorbing through/into the
material hopefully finding ways out into the atmosphere eventually.

148 FST
 - The Science -

- The Science -
You can explore the FST via the science categories. We have chosen to classify them one way. You might
have chosen a different classification, but because you create your own path through the book, the exact
choice of science topics is not so critical. What is important is that you actively explore to find the
scientific ideas and resources that help you formulate more effectively.

One of the certainties in formulation life is that you can find help and inspiration in unexpected places. Like
all explorers, sometimes it’s necessary to be outside your comfort zone.
Absorbency Capillarity
Links

The distance travelled by a liquid into the pores of a coating depends on:

• The radius of the pores, r: larger is faster

• The surface tension of the liquid, σ: larger is faster
• The contact angle of the liquid with the solid, θ: smaller is faster
• The viscosity of the liquid, η: smaller is faster

The key equation for calculating this is the Washburn equation

The current app most relevant is:

[Link]

This provides the Washburn equation which is the distance d travelled in time t of a fluid into pores of
radius r for a fluid with surface tension σ, contact angle θ and viscosity η:

d= √ rσcosθt
2η

Instead of d, the app describes the volume absorbed if a drop is applied.

 Absorbency Porosity and Darcy

Absorbency Porosity and Darcy

Links
Inkjet Inks, Microencapsulation

If we have a powder or fibrous material (non-wovens, paper) then it can absorb liquid via capillary
forces. The standard method for calculating the flow is Darcy’s law. The standard law is not so
useful, so here we calculate the progress of the liquid into the medium with a more practical
version.

Basic Darcy’s Law

We have a porous medium and apply a pressure ΔP (maybe a real pressure, or maybe a capillary pressure)
across a distance L to help a liquid of viscosity η flow into the pores. The flow per unit area, q, is given by:

kΔP
q= μL

The parameter, k, is the permeability. If you have a flow meter and can apply different pressures across
some known lengths then you can determine k.

But the chances are that you aren’t interested in that sort of setup.

Absorbent Darcy flow

We have a porous medium and want to see how the liquid flows into it over time, drawn in by capillary
forces. The app shows an example:
[Link]

The particles or fibres making up the medium have a diameter D. The porosity (fraction of open space) is ε,
the liquid viscosity is η and surface tension is σ, and there is a contact angle of θ. Over your chosen tmax
you can see how the flow advances.

To calculate the flow we create two porosity-dependent parameters S and T

2
S = − 4(1 − ε) − (1 − ε) − 3 − 2ln 1 − ε ( )
2
1 − (1 − ε )
T = ln (1 − ε) + 2
1 + (1 − ε )

The permeability k is calculated as:

D2 ST
k= − 16(1 − ε) (S + T)

δz
The rate of flow in the z direction into the medium δt is calculated via Darcy’s law and k, but it’s
convenient to eliminate k and just use the input parameters:

δz Dσcos (θ) ST
δt = − 8ηz (S + T )

1
Because of the z dependence, the flow slows down over time (the infinite flow when z=0 is accommodated
by a fudge in the code). Other than that, the general trend is obvious: faster flows with larger particles/
fibres, larger surface tension, smaller contact angle and smaller viscosity. The porosity effect is a bit
complicated – larger porosities allow faster flow, but the flow has more porosity to fill so that gives a
slower advancing front.

Washburn Tubes
The fact that the rate of flow in a powder depends on the contact angle gets people excited about the
possibility of measuring powder surface energy by monitoring the take-up rate of different solvents into the
powder packed within a tube. The Darcy law flow under pure capillarity can be seen as classic Washburn
capillarity, Absorbency_Capillarity, so the tubes used in these surface energy measurements are called
Washburn tubes. Via a chain of logic, it can be argued that the porosity effects can cancel out between the
different solvents and with some arithmetic the surface energy can be calculated.

It is important, therefore, to pass on the advice from a distinguished expert in powder surface energies:
“The best thing to do with Washburn tubes is to find an empty drawer in a remote corner of the lab, place
the tubes at the back of the drawer, lock the drawer and throw away the key”. He deeply regrets the time
spent trying to get meaningful results from a technique full of artefacts.

152 FST
 Absorbency Swelling

Absorbency Swelling
Links

With absorbency of a liquid into a swelling medium we usually want a high speed of absorption
and a large amount absorbed. This requires good solubility compatibility between liquid and
sorbent.

But we also need to handle the medium (e.g. for disposal) once all the liquid is absorbed. This
involves a key trade-off between a high number of crosslinks to provide mechanical strength with a
low number to provide sorption capacity.

Flory-Rehner is the standard model for thinking through these issues.

Although apps are supposed to make things easy, this balance of solubility, swelling and crosslinks is tricky
to grasp. Let’s see the app then try to make sense of what’s going on.

[Link]

The X-axis of the left-hand graph is the swelling ratio, VSwollen/VOriginal. The Y-axis shows Mc, the
molecular weight of the average chain between crosslinks. For a high swelling ratio, obviously you must
have a large Mc value.

In the right-hand graph we have the swelling ratio on the Y-axis and see the crosslink density, v, that will
allow a given amount of swelling.

Why two plots? Because the system is complicated and we often come to the problem with different sets of
data. We might know the crosslink density from the amount of crosslinker in the system. Or we might have
some swelling data and want to determine the crosslink density or Mc. By using the mouse on either of the
graphs you can generally find what you need.

The swelling behaviour depends on the MW of the unlinked polymer chain and the Flory-Huggins χ
parameter with the solvent and the solvent’s molar volume, as described in Dissolution_Flory-Huggins. The
χ parameter controls the amount of swelling of the polymer itself – a smaller χ giving more swelling.

The swelling behaviour is analysed by the Flory-Rehner equation:

( )( )
V M φ2
− [ln[1 − φ2] + φ2 + χφ22] = ρ2 M1c 1 − 2 M2c φ20.333 − 2

This isn’t an intuitive equation, but it is what it is. V1 is the molar volume of the solvent, φ2 is the volume
Vswollen 1
fraction of polymer in the swollen gel or, to put it another way, V
original
= φ2 .

To relate to the crosslink density, v, we have the formula:

2ρ2
v= φMc

Here φ is the average # crosslinks per crosslink chain. My

interpretation of this is shown in the diagram.

How do we use this?

For some communities, such as the rubber industry, Flory-Rehner is built in to their ASTM standards for
measuring crosslink density. For some academics it’s a nice way to formalise their careful experiments. Can
you use it in your own formulation work? Well, let’s ask a different question – is there a better way to
capture the tradeoffs between solubility (χ), MW, crosslink density and solvent type/size? It’s a genuinely
difficult topic and maybe Flory-Rehner is better than the non-alternatives.

154 FST
 Adhesion Adhesion promoters

Adhesion Adhesion promoters

Links
Lipstick, Mascara, Water-based Adhesive, Adhesive, Solvent-Based Paint, Emulsion Paint

Adhesion promoters exists to provide Adhesion_Entanglement across the interface. Some help to
create physical entanglements between polymers. Others provide chemical entanglements. We
don’t much care which it is; instead we care that it’s not too little (of course) and not too much.

Promoting Physical Entanglement

Solvent-based adhesives create physical entanglement by mutual solubility across the interface, encouraged
by mutual Dissolution_Hansen Solubility Parameters compatibility of polymers and solvents.

Adhesion to polymers such as polyethylene (PE) or polypropylene (PP) or polyethylene terephthalate

(PET) is difficult because they are each sufficiently crystalline that solvents at room temperature don’t open
up the polymer, so entanglement can’t take place.

For PE and PP, heating above their melting point allows excellent entanglement, that’s heat sealing (see
[Link] The melting point of PET is
impractically high, so heat sealing has to take place via co-extruded amorphous polyesters on the surface.

There are solvents that can open up PE and PP – for years chlorinated alkanes were used as adhesion
promoters. But these are now unpopular. Some phenols are good at opening up PET but again are not
practical for most uses.

A very nice way to create an amorphous PET surface is to hit it with an excimer laser or a xenon flash (see
[Link] This raises the temperature of the top
few nm by 100s of degrees, rendering it amorphous. It then cools rapidly and remains in an amorphous
state. Solvent-based adhesion and heat sealing are both straightforward on such surfaces.

Flame, corona or plasma treatment of PE, PP and PET does three things:

1. It opens up the crystalline structure, allowing physical entanglement.

2. It adds some functionality that might allow chemical entanglement
3. It increases the surface energy. But this is irrelevant. A 10-20% increase in surface energy cannot
be the cause of a 100x increase in adhesion.

Pure PP cannot be usefully corona treated – the surface is totally destroyed. That’s why “PP” films contain
copolymers at their surface. These can readily be treated.
Chemical Entanglement
Because entanglement is three crossings, it doesn’t take much
to get one surface entangled with a polymer which itself is
nicely entangled.

In the diagram, some functional group is on the surface (e.g. an

-OH or -NH2 group) and this reacts into the polymer, for
example via a maleic acid co-polymer group or into a
crosslinking systems such as epoxy or polyurethane.

Because so many surfaces contain usable -OH groups, a generic class of adhesion promoters are the X-
TES, the triethoxysilanes (or trimethoxy silanes). The silane portion reacts onto the -OH groups which
might be on bricks/cements, metal (oxide) surfaces, wood, paper... The X can be -NH2 (in APTES), an
acrylate, an epoxy or more.

The trick with adhesion promoters is to not try too hard. We’ve all added more molecules, provided more
heat, increased the UV power to get stronger adhesion … only to see adhesion fall dramatically. When we
try too hard, the interface is both too brittle and not sufficiently entangled into the rest of the system:

Great promotion from a bad polymer

Corona treat some PE or PP. A few % of -OH, -C=O and -CO2H groups are produced. In this case, the -OH
groups are irrelevant.

Coat the surface with a highly branched PEI, polyethylene imine, a polymer with no desirable mechanical
properties. It’s coated from aqueous solution so there is zero chance of any physical entanglement:

A few of the -NH2 groups react with the -C=O and -CO2H groups to form
an entangled network with the PE/PP. Now coat this surface with an
epoxy, a polyurethane or a UV acrylate. Some of the remaining -NH2
groups react into the coating, so the whole system is entangled and
adhesion is strong.

This works wonderfully subject to one strict condition: the PEI coating must be thin, preferably 10-20nm,
and no more than 50nm. Why? Because it is such a weak polymer that a thicker layer will lead to cohesive
failure within the PEI … and because it’s highly water soluble so will attract humidity to do bad things at
the interface.

Yet again in adhesion, less is more.

Your own chemical entanglement

Life is easy if you can use one of these standard adhesion promotion tricks. But you can’t simply add “an

156 FST
 Adhesion Adhesion promoters

adhesion promoter” – they aren’t magic molecules. You have to be specific about what you are attaching to
what. For your own system you might have to use some thought. To stick to steel, starting with a phosphate
group is always a good idea, as is the case when sticking to teeth. To stick to gold, sulfur is a great choice.
What’s at the other end depends on the specifics of your system. If it’s a classic urethane, epoxy or UV
coating that subsequently cures then an amine is a good idea. Similarly, if it’s a curing unsaturated system
then some sort of double bond will be fine. If there really is no chemical functionality, then the chain
sticking out from your surface will need to be long enough to entangle with the other surface. We don’t care
how we get entanglement, we just need to use our knowledge of the system to ensure enough entanglement
across the interface.

If you can make your promoter polydentate or a chelate, that helps with long-term adhesion. The three
silane sites on X-TES cannot react to give a tripod, but having three sites allows for backup if one fails.
Although you can get good adhesion with an acrylic acid, maleic acid is preferred because there is a backup
acid in situ.

Just be intelligent with your choice and remember that too much of a good thing is bad.

FST 157
Adhesion Butt
Links
Water-based Adhesive, Adhesive

A pure butt joint provides strong adhesion. The problem is that a small shift in the applied angle
changes it from a butt joint to a Adhesion_Shear joint with a large Archimedes lever effect that
will destroy the bond.

Less is more
When an adhesive package tells you to use small amounts of the glue that’s a big clue about the science of
butt joints. Manufacturers make money by selling more not less, but they want happy customers and a thick
layer of adhesive makes for a weak butt joint.

With our adhesive of modulus E we get off to a good start because the modulus
relevant to adhesion in butt mode is the bulk modulus K which depends on the
Poisson ratio ν which is by how much the adhesive shrinks in the horizontal
direction when pulled in the vertical direction:

3E
K= 1 − 2ν

For a typical adhesive, ν ~ 0.33 so K is ~ 8x larger. For a pure rubber adhesive, ν = 0.5 so K is infinite!

The force needed to break the bond for a disk of radius a with an adhesive thickness of d is:

F = πa2 √ KGd
G might be the naïve surface energy W or, more realistically, some plastic deformation value 100s of times
larger.

There are three key points:

1. You want a strong adhesive (of course) but not too strong/brittle otherwise G might be small
through lack of dissipation.
2. You need the thinnest possible layer of adhesive
3. This only works when F is perfectly vertical

The reason for (3) is that adhesion is a property of the system. This idealised large F comes because the
forces are distributed equally around the bond. As soon as F is at an angle, you get large concentrations of
forces at certain points, and once these concentrated forces are large enough (with Archimedean lever
arms), a crack will start and the whole bond will break.
 Adhesion Butt

Stefan’s squeeze law

Because a super-thin adhesive layer is so advantageous, we need to
know how to obtain one. It seems obvious that you just put a drop of
adhesive in the centre of the joint and squeeze hard. But Flow_Stefans
Squeeze law gets in the way. For a given force, the speed at which the
thickness, H, decreases goes as H³ and as the radius, R, of the drop
expands, speed decreases by R4. This combination means that the force to obtain a thin layer is impossible
to achieve in practice.

[Link]

Adverts for superglue show butt joints created between two super-smooth steel cylinders via application of
10 small drops instead of 1 large drop. If we say, naively, that this reduces the Stefan’s law R by a factor of,
say, 5, then the force needed to create a thin drop decreases by 54, 625x. That is a big extra safety margin if,
as per the adverts, you want to bungee jump off a bridge, supported by those two glued cylinders.

While writing my book Sticking Together I found that I needed to refer to Stefan at some point …
then at another, then another… I’d not realised that it’s such an important part of adhesion. It
subsequently turned out that the question of how many drops you use, in which geometry, is a huge
question for the adhesives industry. There seems to be no appable algorithm about the big question
which is whether a bubble will get trapped. The only app I could manage is
[Link] which has plenty of formulae
and discussion for those who are interested.

FST 159
Adhesion Crack resistance
Links
Water-based Adhesive, Adhesive

Adhesion is about stopping a crack growing under an applied force. So let’s see what we can do to
resist crack growth.

Dissipation
It’s quite simple. If the force you are applying to the interface merely
has to break a few bonds straight ahead of it, it will carry on doing
so. If you have a piece of “strong” glass you can easily break it by
introducing a small crack onto the surface. As you bend around the
crack, the glass fractures where the crack energy is focussed … and
continues to crack.

If the force is sent away into the bulk, then there is no longer enough energy to keep the crack going.
Chewing gum is hard to remove not because it is strong but because it is weak enough to absorb the crack
energy, dissipating it as viscous heat as it flows. Car windshields include a soft polymer, polyvinyl butyral,
because the soft, weak, polymer dissipates the stresses away from the “strong” glass.

Because adhesion is a property of the system, we can easily make glass more resistant to a crack – simply
heat it close to its melting point. Similarly, chewing gum is brittle and easy to remove if it is cooled to
liquid nitrogen temperatures.

And because of Flow_TTS-WLF Temperature-Time Superposition we can say that room temperature glass
under a very slow stress will resist the crack and that a super-fast force will remove the chewing gum as its
polymer molecules have no time to move and absorb the crack energy.

In between these two extremes of glass and chewing gum (or, more scientifically, Pressure Sensitive
Adhesives, PSA) the best way to ensure dissipation across an interface is to create
Adhesion_Entanglement, so that attempts to move one bit of a polymer chain will be dissipated across
many nm of chains, just as trying to pull hard on a tangled piece of string makes things worse by spreading
the load across many pieces.

Relaxation times
In order for the adhesive to dissipate the crack energy, the polymer chains need to move. Another way of
thinking about temperature and time is via the relaxation times of the polymer. When we do rheology we
tend to focus on Flow_G' and G''. One of the wonders of rheology is that you can interconvert between
different measurements to see the system in a different way. When you start to think of your system in
terms of relaxation times, new ways to build in crack resistance suggest themselves. See the
[Link] Interconversions app for an
introduction to this fascinating way to expand your knowledge of your system.
 Adhesion Crack resistance

Griffith Cracks
The other way to reduce cracking is by not providing your own sources of
cracks through voids or junk. Instead of over-focussing on the adhesive,
paying attention to removal of defects can be a great strategy. The basic
idea is simple but powerful. The Griffith crack law says that the stress σ at
which a crack yields depends on a modulus E (stronger is better), a factor
G which we’ll discuss in a moment and the size of the defect r:

σ= √ EGr
If you reduce the average size of your defects by a factor of 4 then you double the strength of your joint. It
can be hard to double the strength of an adhesive, so it can often be a good idea to find ways to, say,
remove bubbles from your adhesive system.

In naïve adhesion theory, G is the surface energy. In the real world (the Irwin modification of Griffith) G
can be 1000s of times larger because it represents plastic deformation around the crack, in other words
dissipation of the crack energy over a nm scale.

You tell people about Griffith Cracks and they say “Yeah, yeah” and ignore you – it’s far too
boring compared to focussing on fancy adhesion formulation science. But I’ve had several cases
where it turned out that some pesky dirt or bubbles have, indeed, been the cause of unreliable
adhesion, fixed by relatively simple filtration and de-aeration steps.

FST 161
Adhesion Crosslinking
Links
Water-based Adhesive, Adhesive, Solvent-Based Paint, Emulsion Paint

If, on average, we have X-functional components in our system with a Y% fraction of reactive
bonds, it would be good to know, after a given time, how much of the Y% has been converted into
permanent bonds and how much of the system is comprised of 1-, 2- or more functionalities.

Although we can usually answer the Y% question, and although the questions are inter-related, the
functionality question is mostly obscure.

Degree of cure
A UV curing app provides a good example of how complex crosslinking can be.

[Link]

We have a UV curing system with a UV power, a level of photoinitiator and propagation and termination
rates. In the 4 screen shots we just change the acrylate from mono- up to tetra-functional. The yellow %
conversion line shows that each increase in functionality decreases the conversion over time – from 100%
for the mono-functional down to 28% for the tetra-functional. So although our extra functionality can
deliver benefits in strong crosslinking, they are undermined by ending up with ~70% of the groups being
unreacted.

It's always important to monitor the degree of cure via FTIR, confocal Raman, ultrasound reflectometry
(which, interestingly, gives you the rheology, Flow_G' and G'') or calorimetry. But why is the degree of
cure so limited?
 Adhesion Crosslinking

Rising Tg
If your system is currently above the glass transition temperature, Tg, then monomers, oligomers and
polymer chains are free to move. But at a certain degree of cure, the Tg will have risen so you are now just
below it – and free motion stops. We can see this in an app:

[Link]

In this particular example, an 8% mole fraction of crosslinks is enough to raise the Tg to above room
temperature. To get to 20% you need to raise the temperature to 65°C.

And that is the answer to the crosslinking crisis described in the previous section. Carrying out the curing at
a higher temperature would give you more conversion in a given time because:

1. A 10°C rise might typically double the speed of cure

2. You remain above the Tg so the reactive species can continue to move to find a reactive site.

Distribution of functionalities

[Link] and the similar

[Link] for UV crosslinking

Miller-Macosko theory is not for the faint-hearted. But if you have ever wanted to know about how the
overall MW changes (in the example up to infinity at the gel point of 45.8%) and how the di-, tri- and tetra-

FST 163
functional group concentrations change with conversion, then the app shows you how to do it. At 70%
conversion there are 12, 9 and 8% respectively of 2, 3 and 4 functional links.

Swellability
One reason for creating a crosslinked system is to reduce the potential for swelling by solvents or other
small molecules (e.g. plasticisers) in the system. The shorter the distance between crosslinks, the lower the
potential for swelling. By measuring the degree of swelling if you stop the curing reaction at different
stages, you can get a good idea of this distance between crosslinks. The standard approach is to use Flory-
Rehner theory, which is also discussed in Absorbency_Swelling:

[Link]

The equation, which is not especially obvious or intuitive, is shown in the app. The graphs represent two
ways of plotting the data from the calculations. Suppose you find at some point that there is 4x swelling.
From the graph on the left you see that there is a molecular weight of ~1500 between links or, from the
right-hand graph that the crosslink density, v = 0.4 mol/l.

Although any single measurement may not be too insightful (is 1500 large or small, is 0.4 mol/l good or
bad?), measurements during the curing process as well as measurements comparing formulations will lead
to some deep insights into your crosslinking system.

Mechanical properties
By following the cure via rheology, Flow_G' and G'', using the full capabilities of Time Temperature
Superposition, Flow_TTS-WLF, you will get to see how successful you are at establishing a balance
between raw strength and resistance to brittle fracture. Even if the mechanical properties look good at room
temperature, you need to know if you are close to becoming too weak if used under higher temperatures or
too brittle at low temperatures or, equivalently, prone to creep over long timescales or prone to brittle
failure from high-speed impacts.

It is common to get super-high crosslink densities with epoxies by curing them above 100°C. They
can then become desirably strong and undesirably brittle. The fix for the brittleness is to put core-

164 FST
 Adhesion Crosslinking

shell rubber particles into the adhesive. The rubber core absorbs the crack energy, while the shell
ensures that the rubber is linked into the epoxy matrix.

Entanglement with the adherends

If the system is not entangled across the interface, the adhesion will be poor. Unless the adherends have
functional groups at the right (low) concentration and of suitable reactivity into your system, you need to
select appropriate Adhesion_Adhesion promoters that will give the necessary cross-interface entanglement.

FST 165
Adhesion Entanglement
Links
Water-based Adhesive, Adhesive, Solvent-Based Paint, Emulsion Paint

If you try to disentangle a mess of string you find that the harder you try, the more it resists. Your
energy gets dissipated across many sub-strands and they each help stop the main strand from being
pulled out. That’s exactly how strong adhesion is achieved … by dissipating crack energy across
many individual polymer loops.

Entanglement = 3 Crossings

The dotted line is a real or virtual interface. In the left-hand image, the polymer chain of interest crosses it 3
times. This means that it is entangled. If you pull up on the big arrow at the top left, the lower loop gets
tangled with the other loops.

In the right-hand image, the chain has been cut so it now only crosses 2 times. When you pull up, the loop
is merely intermingled and can be pulled out.

For polymer-polymer adhesion, you just need to get sufficient lengths of each polymer tangled up with the
other, lots of three crossings, and that’s it.

But how do you entangle a polymer with, say, an aluminium surface?

You make sure that the polymer chain has a low level of
reactive groups that can link on to some reactive sites on
the metal surface. Now the two systems are entangled. An
example is maleic acid co-polymers of polyethylene.

You can do the same thing with a crosslinking polymer.

 Adhesion Entanglement

The groups on the metal surface might be something like

APTES (aminopropyl triethoxy silane) which allow the
amine group to participate in a classic epoxy or urethane
crosslinking. Now the tangles are via chemical links (the
red Xs) rather than the physical crossings in the previous
image. In terms of entanglement there is no difference
between these cases. To break the tangle you have to break
a main chain bond or a crosslink bond, basically the same
degree of difficulty. Of course a crosslinked system is less
resistant to long-term creep, but at the core, a tangle is a
tangle, whether it is physical or chemical.

Too much of a good thing …

Many of us imagine that if we add adhesion promoters or functionalise the polymers to react more strongly
with a surface that we would get stronger and stronger adhesion. But as is universally found, as you
increase the functionality adhesion goes up, up, up … then crashes down. Too much of a good thing is bad.

Too few bonds are, indeed, not good enough. But with too many bonds you have a brittle interface which is
also not properly integrated into the bulk of the structure.

The famous experiments from Prof Richard Wool showed

this effect. The image (reproduced with his kind
permission) shows that 0.5% adhesion promoter doubled
the adhesion, 1% tripled it, but 1.25% halved it. The
interface was now brittle and the crack ran along the thin
layer of adhesive at the interface, allowing him to call it
“cohesive failure”, though I would argue that it should just
be called “brittle failure”. This paper was an especially
clear example of, and explanation for, a phenomenon many
of us had observed but had never explained.

Like many formulators I had often noticed that trying too hard to get adhesion resulted in failure.
This made no sense. While struggling to understand adhesion science I was encouraged to read the
papers of Prof Richard Wool. I scan-read one that seemed of no interest and closed down the pdf.
But something made me re-open and re-read it. It was the one with the image above. A decade of
confusion disappeared by looking at one image. It is a powerful demonstration that Adhesion is a
Property of the System.

FST 167
Adhesion Intermingling
Links
Water-based Adhesive, Adhesive, Solvent-Based Paint, Emulsion Paint

If you can get some polymer chains across an interface, you automatically get at least 10x the level
of adhesion obtained by Adhesion_Surface energy Interactions. This small step towards strong
adhesion is called intermingling.

The adhesion, G, from polymer chains of an interfacial density Σ (number per unit area), with chain length
N monomer units and a friction coefficient U is given by a formula from the great de Gennes:

G = ΣNU
In the image, N is 9, we have 16 chains in 16 nm², so Σ
= 1/nm² and U is typically 3.5 J/m². Adding in
Avogadro’s number and changing to m² from nm² we
find that G ~ 400 mJ/m², a factor of 10 greater than a
typical surface energy of 40 mJ/m². If a typical
monomer length is 0.2 nm, then this 10x improvement
comes from only 1.8nm of cross-interface polymer.
How hard is it to get 1.8nm of intermingling? The
Helfand formula tells us, as discussed next.

The famous Prof Richard Wool investigated intermingled or, as he called it, “nail” adhesion using
planks of wood and nails. Enlisting the help of the university carpenter he compared the de Gennes
theory for polymers with his data for planks held together with different lengths (n), nails/length
(Σ) and nail friction (U). The beauty of science is that the de Genne formula worked well at both
the macro- and nano-scale.

But it’s impossible!

You sometimes see claims that this sort of beneficial cross-interface intermingling is impossible “because
most polymers are immiscible”. While it is surprisingly true that most polymers are immiscible, it is an
irrelevant fact. As was shown a long time ago by Helfand, you can get plenty of intermingling if the
polymers aren’t too incompatible:
 Adhesion Intermingling

[Link]

The Helfand formula is trivial. For a Kuhn segment length (something like 5 monomer units) of b and a
polymer-polymer χ parameter, the distance d of intermingling is given by:

b
d= √6χ

Getting a few nm is no problem, so intermingling boosts to adhesion are common. This almost certainly
happens even with some PSA systems. An acrylic PSA has no problem intermingling (maybe after a bit of
time) with an acrylic surface to give a significant boost to the overall adhesion.

Estimating the χ parameter is simple via the Distance between the Dissolution_Hansen Solubility
Parameters of the two polymers.

How do you get the boost? Helfand is thermodynamics, getting the boost involves kinetics. Heat, solvent
and time help increase intermingling, crystallinity makes it almost impossible. The inability to adhere to
polyethylene is not due to any difficulty in finding Helfand-enabled polymers. It’s due to the crystallinity of
the surface, with HDPE being harder to stick to than LDPE.

FST 169
Adhesion JKR
Links
Water-based Adhesive, Adhesive

JKR theory is one of the cornerstones of adhesion science. Because it focusses on pure surface
energy then on its own it’s not all that useful (see Adhesion_Surface energy Interactions) but once
you start using it to measure hysteresis, a world of understanding opens up.

[Link]

The app image shows a JKR experiment that has just started. A smooth rubber ball of radius 20mm and
modulus of 5 MPa is placed next to a smooth surface, with zero force applied. You’ll notice that the ball is
squashed, i.e. some extra force has pulled it into contact. That force is surface energy. Via some means the
contact length, a, is measured. You then apply some modest extra forces and measure how a changes. By
plotting a versus force you get a curve like the one in blue (the Herz curve is a curve with no surface
energy)
 Adhesion JKR

[Link]

Notice that you have to go to a negative force (pulling) to get the ball to unstick itself.

The JKR (Johnson, Kendall, Roberts) formula for a’s dependence on radius R, modulus E*, surface energy
γ and force F is:

a3 =
3R
4E * (F + 6γπR + √12γπRF + (6γπR)2)
This famous formula was created in a pub in the city of Derby where Kendall was working on
adhesion of train wheels to rail tracks. Johnson was a professor at Cambridge and Roberts was in
Johnson’s lab researching windscreen wiper rubbers on glass. It’s a lovely example of apparently
mundane industrial problems producing one of the great formulae of adhesion science.

Unlike classic methods (Zissman plots, Owens-Wendt…) the JKR approach gives a real measure of surface
energy. But as surface energies aren’t so interesting for adhesion, why do we care?

Now look at this Loading curve (increasing force) and Unloading curve (after some time at a constant high
force.

There is a large hysteresis, i.e. the unloading curve corresponds to a higher “surface energy” than the

FST 171
loading curve. What’s going on?

This is the first sign of real adhesion. Something is happening across the interface. For example, polymer
chains might start migrating (in this case by 2nm) across the interface. As you try to separate the surfaces,
these chains have to slide past each other and that extra friction force creates a higher net adhesion force.

Much of real adhesion is due to this cross-interface movement of molecular chains. Small migrations give
“intermingling” which adds frictional forces (Adhesion_Intermingling). Large migrations lead to
“entanglement” which gives strong adhesion (Adhesion_Entanglement).

172 FST
 Adhesion Peel

Adhesion Peel
Links
Water-based Adhesive, Adhesive, PSA

Peel seems simple and obvious – but because adhesion is a property of the system, it is surprisingly
complex.

It’s complicated
The app, although designed for peel of PSA, captures the essence of a typical peel test.

[Link]

The adhesive has lifted off at the right-hand edge and we are looking at stresses ahead of the peel front,
hence the -ve x values. As this is a 90° peel there are no shear stresses, so the blue line is flat.

At the right-hand edge, the tensile stress is maximum, as expected. At the left-hand edge, the stress is 0,
again as expected. What is surprising is that at 0.25mm the stress passes through 0 and there is a
compressive peak at 0.5mm. In the Adhesion_Shear chapter you see that not only does a “shear” test
produce (and fail because of) peel forces, it shows the same compressive zone.

It's non-intuitive
When thinking about adhesion we tend to focus on the adhesive. But if you change the modulus of the
“carrier” (the backing tape for a PSA, this would be the adherend in a conventional peel joint), you find that
everything changes. An increased modulus spreads the whole peel zone so the maximum tensile and
compressive stresses decrease – the bond is stronger.
Now increase the modulus of the adhesive layer. This should surely make the bond stronger. Yes and no.
Yes, the material is stronger, but the tensile stresses are now more concentrated so there’s an increased
probability of passing a critical strain.

Unlike the Adhesion_Butt joint where a thinner adhesive is mandatory, a thicker adhesive reduces the
concentration of forces, which might improve adhesion.

Those who do 3D printing have a classic peel problem. They need light adhesion of the object to
the base plate so things stay stable during printing. They then need to remove the object. Pulling
straight up is butt-joint mode, with strong adhesion. Peel would be easy if the peel could start, but
the object can’t be distorted into a proto-peel. That’s why some objects have a thin rim around the
base which can be pulled up to start peel.

Why so vague?
Missing from the app is any reference to the strength of the interface, whatever that means. For PSAs there
is only 40 mJ/m² of surface energy, but massive dissipation within the adhesive itself (not captured in the
app) provides the typical 400 J/m² adhesion found in practice. For conventional adhesives, dissipation
through entanglement across the interface is not something that can be specified.

So we have some precise calculations about tensile stress distributions that depend on the bulk aspects of
the joint, and handwaving about dissipation and entanglement for the interfacial portion.

If you happen to know the yield stress σyield of your adhesive then you can at least get an idea of what peel
force would lead to yield failure. What you don’t know is whether the interface will fail before then.

It’s not just peel. The app for shear has many parameters, but none of them are to do with the interfacial
capabilities of the adhesive.

And it’s not just these apps. Standard finite element analysis is merely a more precise way to calculate the
stresses. Working out whether dissipation across the interface will keep the bond together is not part of
standard FEA packages.

One famous example in the literature measured the peel adhesion of a polyimide (PI) film to an aluminium
(Al) strip. Holding down the PI and pulling the Al gave one value for the peel strength, holding down the
Al and pulling the PI gave another. After some more sophisticated analyses they concluded from their high-
quality analysis that the “true” peel strength was either 5, 300, 625 or 850 J/m².

There are no simple answers to adhesion problems. The key is to think through those aspects of the system
for which there are good theories and use the theory to optimize those parts of the adhesive system that are
under your intelligent control.

If the backing stretches

The peel test is supposed to measure the work going into disrupting the adhesive. But if some of the work
goes into stretching the backing tape then the peel measurement is, well, what is it? It’s a measure of the
work needed to peel that material, and if that includes stretching the backing, maybe that’s fine. But if you

174 FST
 Adhesion Peel

want to know something about the adhesive itself then having the backing effect isn’t helpful. So
“therefore” you need a stronger backing tape in order to measure the “true” peel.

The scare quote are there to remind you that, as we’ve seen above, the backing, even when it doesn’t
stretch, has a significant effect on the measured peel value. And because “adhesion is a property of the
system”, the “true” peel is an uncertain value.

All the same, using a stretchy backing film, unless it’s for some positive reason, is not a good idea. To get
an idea if your backing is strong enough, this app, [Link]
[Link], gives you an estimate of whether your current backing (modulus, thickness) is strong enough.

I was asked to help on a tricky adhesive problem and we had discussed their test results that made
no sense. So I asked to see the test setup. We walked into the lab and I could see their test machine
across the room … and before we’d made another step into the room I said “There’s your
problem”. Even from that distance I could see that their backing tape was massively distorted. No
wonder their results made no sense. As this weak backing was a feature of the product, to be able
to measure something like the real adhesion the fix was easy – stick some standard adhesive tape
to the back, so the backing was now strong enough.

FST 175
Adhesion Shear
Links
Water-based Adhesive, Adhesive, PSA

The Lap Shear Joint is a beautiful example of why

adhesion is a property of the system. Contrary to
intuition, the amount of lap, L, (above a certain
minimum) makes no difference, and the joints normally fail in peel. And the thickness of the
adherend, h1, has a big effect, even though it’s nothing to do with the adhesive.

Below we shall see that the standard “shear test” for PSAs is another example of a test that doesn’t
do what it claims.

A very bad test

[Link]

If you don’t have access to finite element analysis, then the Goland-Reisner equation for a lap shear joint is
a rather good approximation. The key lesson from this is that the lap-shear test is so fundamentally flawed
that it should seldom be used. Unfortunately, it remains widely used so we have to see why it is so bad and
why the quoted values from the test are misleading.

The app shows the stresses from the right-hand edge to the centre of the bond. So with L set to 10mm we
see 5mm, with the other 5mm being identical. Now let’s see why the test is so bad:

1. We see is that the shear stresses are close to 0 for 2 of the 5 mm. So most of the joint isn’t doing
anything helpful. Academic tests where they deliberately remove adhesive from a joint show that
you can take away 60% of the adhesive with no impact on the force needed to break the bond.
 Adhesion Shear

2. The blue peel stresses go negative (in compression) for a few mm inside the bond (something
discussed in the Adhesion_Peel chapter), but at the edge they rise sharply and exceed the shear
stresses. In other words, these shear tests usually fail in peel. A beautiful example of this is with
carbon composites. They fail “standard” lap shear tests because although their shear strength
(along the fibres) is huge, the peel strength (fibre-to-fibre) is low. For these you have to do the less
convenient but more scientific double-lap shear test.
3. Those who have a lap L, a width b and a force to failure of F claim that the bond has a strength,
F
σ= Lb MPa. This is nonsense. If (above a minimum) they doubled L, F would remain almost
unchanged, but they would have to quote a value that is halved.
4. If you decrease the thickness or modulus of the adherend, the peel and shear stresses increase, so
the bond will fail earlier. You are not just testing the adhesive, you are testing the system.
5. A well-intentioned increase in the modulus of the adhesive will make the adhesive inherently
stronger, but it also concentrates the peel and shear stresses. It’s not necessarily the case that
“stronger = better”.
6. Whether you are using Goland-Reisner or full finite element analysis, there is no way to include
the dissipation of crack energy across the interface, no way to work out the effect of entanglement.
This means that if you are trying to understand the “real” adhesion at the interface, the lap-shear
test is of no help other than distinguishing between cohesive failure (the adhesive was weaker than
the interface) and adhesive failure (the interface was weaker).

Why are there peel forces in a lap-shear joint?

The necessary offset of the two forces across the adhesive

produce a peel bending moment.

If your real-world system works in lap-shear mode, then by all

means use the test. If you are trying to understand the behaviour
of an adhesive in shear mode, the test is not fit for purpose.

The cockroach of adhesion tests

The lap shear test is used extensively. Because it tells you surprisingly little about real adhesion,
why is it still around? Because everyone uses it. A famous centre of adhesion science calls it the
cockroach of adhesion tests: “It’s everywhere, it’s completely useless and no matter how hard you
try you can’t get rid of it.”

Roughening the surface

See Mechanical_Surface Roughness for a fuller discussion of this topic.

The literature of the lap-shear test is especially full of tests on the effect of roughening the surface.
“Everyone” knows that roughening the surface increases the surface area and therefore enhances adhesion
via surface energy, while also providing mechanical interlocking between peaks and valleys. If you go to
the Surface Profile Explorer app, [Link]
[Link] and check out the extra surface area, and the peaks and valleys from a rough surface such as
this:

FST 177
 you find that there is no significant increase in surface area, and
in any case, surface energy is 1000x too small to be relevant. And
there is zero chance of peaks and valleys interlocking. It is all an
illusion from the mismatch of scales. This same surface with the
x-axis scale expanded to be similar to the y-axis (but still a factor
of 10 different) is a gentle landscape:

We can all agree that roughening the surface can remove junk and
maybe provide some fresh chemical functionality for reaction into
the adhesive. But you can also get a clean surface, and the same
enhancements, without roughening.

If you shot blast a steel surface you can get re-entrant surfaces so
that the liquid adhesive is trapped inside. There is good evidence
that you can get ~2x more adhesion compared to the same (cleaned) steel without the surface deformations.
But this is a rare example of cause-and-effect backed up by good data. You cannot take this special case and
generalise to other surfaces.

By all means, if roughening is the best way to get a clean surface, go ahead and use it. Just don’t claim that
the extra adhesion is due to extra surface area and/or mechanical interlocking.

The PSA Shear test

The classic, official PSA shear test shown in the diagram is only a test for
passing the shear test. It has very little relationship with the shear
properties of the PSA for a good reason – it often fails in peel, as we saw
with the lap joint. So the formula shown in the image is not applicable,
much to the puzzlement of people who think that the shear test provides
information about shear behaviour.

Just as carbon composites need a non-standard test to get a better idea of

shear failure, if you want shear behaviour for PSAs, there are at least two
tests (see [Link]
[Link]) that can do a better job at revealing the true shear behaviour.

178 FST
 Adhesion Surface Energy Interactions

Adhesion Surface Energy Interactions

Links
Lipstick, Mascara, Water-based Adhesive, Adhesive

Most people think that surface energy is important for adhesion. In fact it’s 3-4 orders of
magnitude too small to be of any significance. A typical strong adhesive bond might be 400 N/m
while a typical surface energy is 40 mN/m.

If you are a gecko then surface energy is all you’ve got. But remember, a gecko does not want
strong adhesion – it has to easily break adhesion each time it moves its feet.

Adhesion is a property of the system

We can show that this is true in a setup where the only adhesion is surface energy between two super-
smooth pieces of rubber:

The app does the calculations for you:

[Link]

In the case of peel, the force needed to break the bond is 1 mN. Pulling the same system apart under shear
requires 1 N – 1000x times more even though the same surface energy is involved. Pulling apart as a butt
joint requires 32 N, 32,000 times more.

More interesting is that although (of course) the peel force depends only on the surface energy, the shear
and butt joints depend on the modulus and thickness of the polymer, neither of which is an “adhesive”
property.

Standard 3D printing using “fused filaments”, i.e. squirting hot polymer, requires enough adhesion
to the substrate to keep the object from moving around under the forces of the extruded filaments
(the results of too little adhesion are not pretty) but small-enough adhesion that the object is easy to
remove. Pulling the object straight up requires overcoming impossibly large butt joint adhesion.
Peeling it allows easy release, but it’s hard to start the peel in the first place. So it is common to
print “peelable” structures on the base of the object to allow the peeling process to get under way.

Remember that the above example is for super-smooth rubbers. As

soon as you have even micro-roughness, the adhesion falls much
further. The gecko gets around this problem via layers of compliance.
The toes can bend to accommodate large (5mm) changes in the surface,
then the next level is flexible at the 50 µm scale, then the final level is
flexible in the nm level. But, as mentioned above, gecko adhesion must
be weak, otherwise the gecko would not be able to move. There are
times when you might want a gecko-style adhesion, and there are ways
to approximate the highly compliant gecko interactions, but generally
we want much stronger, more resilient adhesion.

For PSA (pressure sensitive adhesives) the adhesive is deliberately

weak so that it can spontaneously flow into full surface contact. For PSA the contact area is pure surface
energy, yet they are 1000x stronger for other reasons. Adhesion is a property of the system.

But if the surface doesn’t wet, you’ll get no adhesion!

We all know that a drop of water does not wet polyethylene. Yet if you put another piece of polyethylene on
top (see the Sticking Together video: [Link]
watch?v=w1dsiMOvdRc&list=PLLnAFJxOjzZuXlacXKsuBSNk3AH-NuuIr&index=15) the water
spontaneously wets both surfaces. The equations for this are well known, and in fact you get full wetting of
water even between two PTFE surfaces. So all adhesives wet between all surfaces. Of course many
adhesives don’t stick to polyethylene or PTFE, but this is nothing to do with wetting.

If surface energy is so obviously irrelevant to adhesion, why do so many people teach that surface
energy is the important attribute? When the key influencers in any field assert that X is important,
it is very hard for non-experts to question the orthodoxy. If surface energy doesn’t seem to work
for your problem, it’s obviously your fault, not the fault of the orthodoxy. Fortunately the newer
generation of formulators have learned that surface energy is a few orders of magnitude too small
to be relevant, so the old story is fading fast.

Measuring surface energy

180 FST
 Adhesion Surface Energy Interactions

Although everyone obsesses about surface energy, measuring it is surprisingly difficult. The convenient
fiction is that we use the Young equation which tells us that θ depends on the surface tension of the liquid
σL (easily measured), the solid surface energy σS which is what interests us and σSL the solid-liquid surface
interaction about which we are very vague:

σS = σSL + σLcos (θ)

This leads to lots of work measuring contact angles with water, diiodomethane and formamide and
calculations via Owens-Wendt, Wu or Lewis Acid/Base algorithms (see the app:
[Link] Or it involves Zismann plots where
the 0° intercept of a line of contact angles versus surface tensions is supposed to give you a surface energy.

The scientifically correct method, which requires a more sophisticated apparatus, is via the JKR method
(Adhesion_JKR). It is amusing to compare real JKR values with the various attempts using contact angles.
The table from the Real-SE app shows that not only do none of the methods give reliable values, you can’t
even find an objective “least bad” method.

[Link]

The acronyms for the polymers and techniques are described in the app.

Those who insist that corona or plasma treatment works “because of higher surface energy” should note the
columns PE and C-PE, where C means corona-treated. The C-PE column is doubly interesting.

• The standard methods for measuring the surface energy are wildly wrong compared to the real
JKR value, so people are drawing conclusions from erroneous values.
• PET usually has a higher surface energy than corona-treated PE and yet nothing sticks to untreated
PET.

If corona/plasma is your preferred way to remove contamination from the surface, and if surface energy is
good at showing the removal, that’s great. Adhesion is a property of the system and any system with junk
that hasn’t been removed is unlikely to give good adhesion. Focus on junk removal, not surface energy.

For those tempted to measure the surface energies of powders under the mistaken impression that (a) the
measurements are meaningful and (b) the measurements will help with adhesion, the caution on the use of
Washburn Tubes in Absorbency_Porosity and Darcy is worth reading.

Adhesibility

FST 181
 The more productive idea of “adhesibility” is described in an app,
[Link]
[Link], that invokes a nanoindenter or JKR-style
machine for measuring the hysteresis effects that illustrate the
potential for adhesion. The initial jump in the loading curve gives
the pure surface energy, and the subsequent cycle during loading
and unloading gives you a lot of information about modulus and
the dissipation that is a key part of adhesion.

A phone call from an irritated reader of attacks on surface energy came up with the question: “OK,
so if surface energy isn’t important, how else can we measure adhesibility of a surface?”. What an
interesting word, and what an interesting challenge. That’s how the adhesibility app was created. I
found that although the loading/unloading diagram was “well-known”, it was not used in the
adhesion science world. Hopefully this will change.

At the time of writing, an “adhesibility” device is under development by a major equipment supplier. More
will be added to this chapter once the device is being widely used.

182 FST
 Adhesion Testing

Adhesion Testing
Links
Water-based Adhesive, Adhesive, PSA

Everyone wants a test for the “real” adhesion in their system so they can optimise the adhesive.
They won’t find one. Let’s see why.

Adhesion is a property of the system

The three core types of test are good examples of why it is
impossible to test for the “true” adhesion of a system. Let’s see in
turn why standard adhesion tests in each of these modes do not,
and cannot, give the “true” adhesion:

• Adhesion_Peel. The stresses ahead of the peel front can extend by several mm and even go into
compression. The shape and magnitude of these stresses depend not just on the adhesive but on
the adherend, its modulus and thickness. A weak adhesive, a PSA, can show great strength
because the peel forces are dissipated over many μm of adhesive. For a strong adhesive, success
depends on the dissipation of crack energy across a few nm, something that we hardly know how
to analyse and test for.
• Adhesion_Shear. Famously the lap-shear joint does not depend (much) on the overlap, and mostly
fails in peel, not shear. Again, the distribution of forces depends as much on thickness and
modulus of the adherend as on the adhesive. And, as with the peel joint, the dissipation of crack
energy across the (hopefully) entangled interface is something about which we have little
information.
• Adhesion_Butt. Now the strength of the joint depends on the bulk modulus of the adhesive, which
in turn depends on the Poisson ratio, something we generally know little about. It also depends
inversely on the square root of the adhesive thickness, so the strength depends a lot on our ability
to create a super-thin, even layer without (say) bubble defects that would be the source of Griffith
crack failure (see Adhesion_Crack resistance). It is surprisingly hard to squeeze a drop of adhesive
to produce a thin layer, see Flow_Stefans Squeeze, so the strength of a butt joint can depend more
on some basic laws of fluid flow than on the adhesive itself. And as soon as the butt force is not
perfectly vertical we start to get failure in shear or peel.

What about fancier tests? Professionals tend to like crack tests. At least, unlike the tests above, they start
with a known crack in a known position. But look at how many crack tests there are:
 Which should you choose? Is 3-Point bend worse
than 4-Point. Is a Topple beam better than a
Wedge? The answer is that they are all so
complex, and so indirectly connected to the
“true” adhesion that it’s hard to like any of them.
If, for example, you look at the app for the
4-point bend,

[Link]

you find from the equations that the journey from data to “adhesion” is complex. The test is used in the
world of multi-layer electronic devices for which it may be suitable. For the rest of us it is not a good
investment of time and energy.

And cracks are complicated. Even if you start with a known setup:

Depending on a number of parameters, including Dundurs α (see

[Link] for an
explanation of Dundurs α and how it affects other cracking situations),
the crack might continue where it is, move higher or lower in the current
layer, head to the interface or even jump across the interface and
propagate in the top layer. It’s complicated.

ASTM-style tests
Many of us carry out tests according to ASTM (or other standards such as ISO) because our customers
demand that we pass them. That’s fine, life’s like that so you have to go along. But just because it is an
ASTM test doesn’t mean that it’s of any use for your system. Carbon composite aircraft structures fail an
ASTM standard test for aircraft structures. This isn’t because carbon composites are bad for aircraft
(they’re great) or because ASTM created a stupid test. It’s because the ASTM experts at the time made the
pragmatic (and correct) decision that for existing aluminium aircraft structures, their lap-shear test was a
good pass/fail criterion where lives really matter. A footnote in the standard points out that it is really a peel
test and, as it turns out, carbon composites easily fail in peel. ASTM (and the aircraft industry) know that
the correct shear test for carbon composites is the double-lap test.

The problem isn’t with ASTM. It is with the general adhesion industry assuming that a test designed for
one purpose could provide insights into totally different types of adhesion issues.

184 FST
 Adhesion Testing

Note, too, that as the story below shows, you can’t mindlessly apply an ASTM test. Subtle details of your
product might cause problems unforeseen even by the ASTM experts.

The advice, therefore, is to spend the minimum time doing the ASTM-style tests required for your industry
and diverting your resources to tests that are more insightful for your specific adhesion issue. Passing those
tests will automatically help you to pass the ASTM tests.

Despite the company gathering lots of ASTM test data on a PSA, their formulation efforts were
getting nowhere. In such circumstances, my instinct is to blame the test so I asked to see it being
performed. They took me to the lab and left me with the technician, who was very good and
performed the test exactly as per the standard. As the test proceeded it was clear that their product
was doing something odd which meant the results were meaningless. I pointed this out to the
technician who agreed: “I tried to tell them, but I’m only a technician”. I asked for the test to be re-
set and when it was ready I got the senior managers to look at the test. They, too, instantly saw the
problem and without too much effort an alternative method was worked out.

Your own tests

You are the expert in the adhesion challenges in your system. So what really affects adhesion?

• Is it chemical reactivity across and interface? Well, test for chemical functionality and reactivity.
• Is it performance in cold conditions? Test for Flow_G' and G'' and, to make sure you can cope
with high-speed shocks (if relevant) extend your rheology via Flow_TTS-WLF.
• Is the problem ageing performance? It’s often something to do with water, so do tests that look for,
say, water sorption then tweak the formulation to reduce the sorption at high humidities.
• Is it a challenge of getting enough dissipation while resisting long-term creep? This needs some
more challenging rheology. But far better doing rheology than meaningless adhesion tests.

Then there are some other considerations:

• Do you need to explore a wide formulation space before optimizing? Find a quick, good-enough
test (maybe one that your robot can do) that helps you decide between hopeful and hopeless
formulations. You often don’t have the time to do detailed ASTM-style tests on a broad range of
samples, so you do too few tests within a narrow formulation zone and can miss superior options.
• Do the tests require live judgement of what’s going on? This is the 21st century so set up a camera
to record every test. Most of the time you don’t need to re-run the video, but when something
interesting (good or bad) happens, having the video is vital.
• If you are using any form of camera/video, remember to put a lot of thought into good lighting and
good lens positioning. It is easy to lose key information which could easily have been recorded
with a little more thought.
• Find tests that most closely relate to the most likely end-use failure modes. Sometimes this needs
imagination.

One customer complained of poor adhesion, while all other customers were happy. It turned out
that adhesion failure occurred during a slitting operation and their very bad set-up was causing

FST 185
 extraordinary stresses that caused adhesion failure. In this case the solution to an “adhesion”
problem was to help them set up their slitting machine correctly. But if this had been an industry-
wide problem, we would have had to devise a test to mimic these extraordinary fast and extreme
stresses, something that no ASTM test could mimic.

Those specific examples may not be relevant to you. The idea is to encourage you to think through the
issues then use the tools within the FST to find solutions.

186 FST
 Cleaning Bleaching and Oxidation

Cleaning Bleaching and Oxidation

Links
Laundry Liquids, Surface Cleaning

It’s easy to clean and protect using bleach. The problem is how to ensure that there is not too little
activity and not too much, doing it for long enough but not too long.

Chlorine or peroxide
Although some specialist reducing bleaches such as thiourea dioxide are available, here we discuss the
classic oxidizers, chlorine, delivered in liquid form as hypochlorite, and peroxide delivered in liquid form
as hydrogen peroxide or in solid form as a “per” salt or (surprisingly perhaps) via a combination of
enzymes and glucose.

In terms of killing microbes or reaching difficult stains, the chlorine bleaches win out because chlorine has
a higher redox potential, and the gas is a small molecule that can diffuse into hard-to-reach and/or
hydrophobic places, which hydrogen peroxide as a hydrophilic solute can’t so readily reach. That
advantage comes with the side-effect of chlorine readily reaching the nose of the user which is a useful
signal that some powerful cleaning is going on, while also being a warning of the presence of a highly toxic
gas.

The extra power of chlorine comes with the extra downside of it being more likely to lead to damage to the
item being cleaned and (from accidental splashes) to the clothes of those using it.

For general purpose stain removal, if we assume that the stain is only on the surface (even if that surface is
within a fabric via capillary flow) then the surface-only action of the peroxide is an advantage as it can still
reach the stain while not being able to penetrate deeper into, say, a coloured hydrophobic fabric to bleach
out the colour or react with and weaken the fibres. This advantage is less clear for cotton with its more
hydrophilic structure.

Sodium/calcium hypochlorite, sodium dichloroisocyanurate

Reacting chlorine with sodium hydroxide solutions gives liquid sodium hypochlorite, the classic bleach. It
has to be kept at high pH as hypochloric acid breaks down rapidly to chlorine.

Reacting chlorine with slightly wet calcium hydroxide (lime) gives “bleaching powder” as a mix of calcium
chloride, remaining lime and calcium hypochlorite. It is a convenient solid form of bleach.

Sodium dichloroisocyanurate is another convenient solid form of chlorine bleach.

Sodium perborate and percarbonate

It is a convenient fiction that sodium perborate tetrahydrate or the more efficient monohydrate are what
they say they are. Instead they are a composed of the 1,4-diboratetroxane di-anion, a fact we can
conveniently ignore. It is also a convenient fiction that sodium percarbonate is a percarbonate. Instead it’s
sodium carbonate with hydrogen peroxide locked into the crystal lattice, again a fact we can ignore.

Stabilising the unstable

We use hypochlorite and peroxide because they are thermodynamically unstable molecules able to react
with relatively hard-to-remove microbes and stains. How, then, can we keep them stable up to the moment
of use?

Sodium hypochlorite is mostly stable enough in a sealed container to keep the chlorine from escaping, as
long as the pH is high enough and the extra ingredients (see below) are chosen wisely. The literature shows
no problem with hypochlorite solutions remaining unchanged for at least 1 year.

The per-salts are highly sensitive to transition metal ions such as iron, so it is important to use low
transition metal ingredients and avoid contamination in manufacture and to add chelators such as EDTA or
a modern greener equivalent to the formulation.

The dry per-salts are relatively stable, though percarbonate needs extra protection on the outside of the
particle via sodium silicate or sulfate. Liquid formulations are tougher to stabilise, and need to be relatively
acidic and shielded from UV light and heat. For detergents that work best at alkaline pH, there is a difficult
contradiction. Again, the liquid formulations need a wise choice of the extra ingredients around them.

Making them more reactive

The per-salts bleach well at temperatures above 40°C but require a boost to be effective at ambient
temperatures. Although it is possible to provide smart transition metal (manganese) catalysts, the standard
technique is to add an activator that converts the per-salt into a peracid. In Europe the default activator is
tetraacetylethylenediamine (TAED) which reacts with the per-salt to generate 2 equivalents of peroxyacetic
acid. In the US the default is sodium nonanoyloxybenzene sulfonate (NOBS) which reacts to create
peroxynananoic acid. The peracids are, respectively, hydrophilic and hydrophobic, so TAED tends to be
better for hydrophilic stains and NOBS for hydrophobic ones.

Famously, one manufacturer introduced a manganese catalyst that boosted the cleaning power of
their laundry products. This gave a strong competitive advantage, so the competitors were quick to
do tests which proved that this formulation was too powerful and that fabrics were getting
damaged by the bleach. Whatever the truth of the situation, that specific product was withdrawn
and a consensus level of catalysts was achieved throughout the industry. The manufacturers of the
manganese catalyst were left with large stocks, so we tried to see if it could help in our specific
technical bleaching application. Unfortunately it didn’t. We tried a more imaginative formulation
with a different catalyst system. This gave astonishing cleaning but was so dangerously reactive
that it was unusable. Getting the right balance is very hard.

A reader kindly informed me that “ [There are] other applications for the same catalyst and new
ones based on iron, including products for dishwashing, bleaching of paper pulp, and siccatives
for alkyd paints.”

188 FST
 Cleaning Bleaching and Oxidation

Creating peroxide with enzymes

By adding glucose to a formulation containing glucose oxidase, hydrogen peroxide is produced. It is an
attractive option in terms of using only natural ingredients, but it has obvious formulation difficulties in
practice.

Bleach-stable surfactants
A thickened chlorine bleach is useful for its ability to cling to surfaces such as toilet bowls to provide
longer-lasting action. Surprisingly (as they say in patents) many standard surfactants have no problem
surviving the high pH, highly oxidative environment of a bleach and a bleach can be thickened simply with
classic alkyl ether sulfates (such as SLES), though more complex blends of anionics with amine-oxide,
quaternary ammonium or betaine surfactants are also common. The fact that such a wide range of
surfactant types can be used shows that bleaches aren’t all-powerful destroyers of complex chemicals.
Using relatively low-cost surfactants as thickeners also allows the claim of extra cleaning power plus the
production of foam which consumers associate (wrongly) with cleaning – though clever foam bleach
formulations that cling to the walls of facilities such as dairies are a smart way to provide the longer-term
activity to remove tough microbes.

For those who want to thicken with polymers, the sodium salts of polyacrylic acids can usually survive
long enough, though some radical stabilizers such as benzoic acid derivatives might be required.

Partitioning
Many of the contradictions between the various ingredients in a cleaning formulation for laundry or
dishwashers can be solved by placing ingredients within different parts of a “pod”. So a strong peroxide
bleach can be kept at acid pH, while other compartments in the pod can contain enough base to produce an
overall basic environment for the washing, along with any transition metal catalysts or activators required
to increase the bleaching power, especially for cold washes.

FST 189
Cleaning Boundary removal
Links
Dishwashing Liquids, Hair Shampoos, Soaps and Washing, Laundry Liquids, Surface Cleaning,
Toothpastes, Lipstick, Mascara

The exact place where we want our cleaning fluid to interact with the dirt is the place where the
“no-slip boundary condition” tells us that the fluid velocity is zero. Unless we do something about
this, we really can’t do much cleaning.

No-slip
In Flow_Couette Flow we see that as we drag some
fluid along (using a cloth, brush or just some pressure
flow) at a high velocity v, the velocity of the flow is zero
at the stationary surface. This is the “no-slip boundary
condition” which is a fundamental law of physics
verified down to the nm scale.

This automatically means that dirt is not being swept from the surface and that fresh cleaning fluid is not
being provided to help attack the dirt.

There are three ways to overcome the problem:

• Create lots of turbulent flow, applying lots of energy with your hands, or by using high-powered
jets like those in commercial dish washers
• Add plenty of particles that can transmit energy right down to the surface. For spherical particles,
there is a “percolation threshold” (see [Link]
[Link]) which for spheres is 28% volume fraction, so you need a lot of particles to
guarantee that energy applied above reaches down to the surface
• Use a microfibre cloth. The main structure of the cloth transmits forces down to the surface, while
the microfibres coming from each main fibre have a bigger chance of contacting the surface and
inducing cleaning.

At a time when the use of microfibre cloths was new and they were being marketed as good at
cleaning, my wife, a non-scientist, suggested I should use them for a product development
challenge. A proper lab experiment on a known-to-be-difficult real-world scenario was set up to
compare a standard cloth with the microfibre one. The control cloth, along with some smart
chemical formulation, cleaned fairly well so then it was time to try the microfibre one. We were
stunned. It was no contest – the microfibre cloth was instantly and obviously better – we’d never
seen cleaning like it. The hunt for an explanation (overcoming the no-slip boundary condition)
took years while the decision to change to microfibre cloths was instant.
 Cleaning Boundary removal

Cleaning pipes
There are two problems with trying to clean out a pipe full of one fluid
by pumping another fluid through it. The first, this time showing
Flow_Poiseuille Flow is that the velocity of the fluid at the surface is still
zero:

A more severe, and less well-known problem arises if you try to pump a
low viscosity cleaning fluid (green) through a pipe with a higher
viscosity contaminant. Thanks to the Rayleigh-Taylor instability, the low
viscosity fluid simply punches through the centre of the pipe, providing
little cleaning towards the edge.

The problem is reduced if the interfacial tension between the phases is super-low, (see Surfactancy_Critical
Capillary Number for an analogous situation in oil recovery). If you can’t match the surface tensions of the
two phases then the trick is to add something to the cleaning fluid which increases its viscosity but which is
also an acceptable “contaminant” once the pipe has been cleaned.

FST 191
Cleaning Contact Angles
Links
Dishwashing Liquids, Hair Shampoos, Soaps and Washing, Laundry Liquids, Surface Cleaning

There is an unhealthy obsession with contact angles as being important for cleaning. Often their
role is minor or even irrelevant. Let’s see why.

Contact Angle measurement

The story goes that we put a drop of cleaning liquid onto a surface
and measure the contact angle, θ. Pure water on a typical surface
will have a contact angle in the range of 40-60°. A water
containing Cleaning_Surfactants might have a contact angle in the
range of 20-40° and “therefore” cleans better because it wets more.

The relevant theory, the Young equation tells us that θ depends on the surface tension of the liquid σL
(easily measured), the solid surface energy σS which is what interests us and σSL the solid-liquid surface
interaction about which we are very vague:

σS = σSL + σLcos (θ)

As discussed in Adhesion_Surface energy Interactions there are many problems getting anything useful
from this apparently simple equation because of the uncertainties of σS and σSL.

There are good reasons why it is too simplistic to think that we can go from measured contact angle to
cleaning efficiency.

Surface energy forces are very weak. We can easily cover a surface with liquid by adding some extra
Uη
energy such as a spray or applying a cloth. The capillary number, Ca = σ , represents the balance of
inertial forces, velocity times viscosity, compared to surface tension. We can vary U and η by orders of
magnitude, while σ varies at most by a factor of 2. It is very easy to overwhelm surface tension effects.

Any liquid between any two surfaces (even water between Teflon) will
fully wet those surfaces, because although the liquid (in this example)
has a 90° contact angle so doesn’t wet the surface, when there is a
second surface, for example a cleaning cloth, it is happier to contact the surface than be in air. You can see
this effect on my Sticking Together YouTube channel, using coloured water and some polyethylene:
[Link]
NuuIr&index=15.

The measured contact angle is very susceptible to slight contaminants. If you really want to know the
contact angle of a specific surface, you first have to clean it thoroughly with, say, acetone, alcohol and DI
water. If a slight contaminant (1 molecule thick) can significantly change a contact angle, maybe we should
 Cleaning Contact Angles

focus on contamination rather than contact angle.

Our problem often isn’t one of wetting the surface but of dewetting. If
our cleaning fluid contains any “contaminant” such as a surfactant
where the hydrophobic portion sticks to the surface and the
hydrophilic head pokes up into the liquid, the receding contact angle might be much lower than the
equilibrium contact angle that is typically measured. Because, as noted in the previous paragraph, contact
angles are highly dependent on contamination, just about anything in a liquid might alter the surface
enough to make a measurement of “the” contact angle meaningless.

A lot of the things we care about, such as what happens as drops of

water dry on a dish or whether we get the coffee-ring stains involve
effects and timescales not covered by classic contact angle
measurements. For example, if two drops happened to be a bit too far apart, they will dry as two separate
drops. But if the same drops were a bit closer, they would coalesce and dry as a single drop:

Why these reminders of why contact angle is generally useless? I had, for many years, used fancy
contact angle measurements each time there was a problem of bad adhesion in a product. After one
such session where again there was no correlation between contact angle and product performance,
I realised that not once in years of measuring contact angles had they helped me solve a problem.
We had been measuring them because everyone else measured them. By banning future use of the
technique (unless there was a very good reason) we stopped wasting time on useless hypotheses,
forcing us to find the real causes much faster.

What is contact angle good for?

For most purposes, the human eye looking at a drop of liquid on a surface will tell you whether a surface
will easily be wet by that liquid or not. Such a test is also useful if you need a quick check as to whether
one liquid is similar to another or whether one surface is similar to another.

If your real concern is about the ability to clean the surface then Cleaning_Bleaching and Oxidation,
Cleaning_Boundary removal, Cleaning_Enzymes, Cleaning_Solvents, Cleaning_Surfactants, and
Cleaning_Temperature effects are likely to be much more important.

Getting into tight spaces

Flow of cleaning fluid into small spaces, pores, fibres, fabrics is governed by capillary forces,
Absorbency_Porosity and Darcy. When we look at the core equation:

δz Dσcos (θ) ST
δt = − 8ηz S+T

the two parameters of relevance are σ, the surface tension and θ the contact angle. Adding surfactants to
reduce the contact angle has the downside of reducing σ. And because the contact angle works via cosine
term, the effect of changing contact angle is relatively small: cos(30) = 0.87 while cos(60) = 0.5, so a
halving of the contact angle increases flow by ~70%. But if to achieve the 30° contact angle, the surface
tension had to be reduced from 60 mN/m to 30 mN/m that’s a halving of the speed so the overall effect is a
13% reduction in flow.

FST 193
The trade-off becomes positive for difficult-to wet surfaces with contact angles greater than 65° and by 80°
the net benefit is a factor of 2.5.

If you are interested in predicting contact angles of liquids on polymers, using Dissolution_Hansen
Solubility Parameters an app can be found on the Hansen-Solubility site: [Link]
[Link]/HSP-science/Surface_Energy.php.

194 FST
 Cleaning Enzymes

Cleaning Enzymes
Links
Laundry Liquids

Enzymes are now so tough and reliable that they are regularly used in cleaning applications.

An enzyme for every occasion

Although we can use powerful general-purpose cleaners such as surfactants, caustics and bleaches, there
are times where precision is required. To clean protein stains we can use proteases, fat stains via lipases,
starches via amylases. Others, such as cellulases aren’t for cleaning clothes but for keeping cotton fabrics
looking nice by removing cellulose microfibres. We can even create hydrogen peroxide for stain removal
using glucose oxidases if we add glucose to the cleaning fluid.

Getting enzymes tough enough to be used in real-world cleaning used to require finding extremophiles in
places like the hot springs of Yellowstone. Now we can engineer them by synthesising promising DNA
sequences that combine efficient active sites with 3D structures (checked using AlphaFold) that confer
stability to the cleaning environment.

Many years ago my then PhD supervisor, an expert on enzymes, recounted a phone call he’d just
had, to see if he could help enzymes work in washing machines at the normal (at that time) high
temperatures. We laughed at the idea that enzymes would ever be used in laundry detergents. Yes,
we were wrong. Prediction is hard!

Or we can add specific crosslinks to stop them unfolding, or immobilise them onto a surface to keep them
away from harm.

Naturally aggressive
Because humans are made of proteins, fats and starches, the enzymes used in cleaning products are equally
capable of causing damage to our skin or (e.g. from powder detergents) lungs. So a key requirement for
these enzymes is reliable removal from the clothes during the rinse cycles. Those consumers who are
nervous of biological detergents might like to reflect that rather than miss out on the proven efficacy of
enzymes, they should at the same time decrease the amount of detergent per wash (most of us use far too
much) and increase the number of rinse cycles.

Formulators working regularly with these enzyme formulations need check-ups at agreed intervals for
enzyme-related problems.
Cleaning Solvents
Links
Lipstick, Mascara

Although solvents are thought of as “bad”, sometimes water, even with surfactants and bleach,
isn’t good enough. So how to select the least bad solvent?

The right combination

The solvent has to simultaneously meet a set of requirements:

1. Efficacy as a solvent;
2. Correct evaporation (not too fast, not too slow);
3. Acceptable environmental impact.

Let’s look at each in turn.

Solubility match
You need to know where the thing you are trying to clean is in solubility space. You then need to find
solvents that are sufficiently close in solubility space to be able to dissolve the problem material, while
being sufficiently far from other things in the system (e.g. plastics) that you don’t want to be touched by the
solvent.

The standard way to know all this is via Dissolution_Hansen Solubility Parameters. Because they show
solubility space in three dimensions, δD (Dispersion), δP (Polar) and δH (Hydrogen bonding) we need a 3D
viewer:
 Cleaning Solvents

[Link]

As the image shows, we can quickly see where we want to be, i.e. close to the “bad” stuff, and where we
don’t want to be, close to the “good” stuff. Using the mouse to see all the right solvents (those near the bad
stuff) we can get a list of solvents that will work well. We can then check for their other properties before
selecting one.

Right evaporation
Solvents that evaporate too fast have two problems:

1. They disappear before they’ve cleaned;

2. They will have a low flashpoint that makes them a fire hazard.

Solvents that evaporate too slowly will have plenty of time to act but:

1. Might have time to attack stuff you don’t want to damage;

2. They are difficult to remove, their odour lingers too long.

By knowing about Evaporation_Temperature and Antoine Coefficients we get a good idea for their vapour
pressure at the desired cleaning temperature, and by combining that with an estimate of the local air
velocity we can use Evaporation_Basics to estimate the evaporation time.

Can we use our solvent?

From the Safety Data Sheet we can find if our prospective solvents are harmful to us and to the
environment and can then decide if the risks are manageable with proper PPE and containment.

Very often the 3-step analysis proves discouraging. There is no single solvent which can achieve all our
goals. There are two things we can do about this.

FST 197
 1. Smart solvent blends
2. Microemulsions

Let’s look at each in turn.

Smart solvent blends

If you take an X:Y blend of two solvents, the HSP of the
blend is the weighted average of each of the 3 parameters.
When you look at the diagram, you see that a 70:30 blend of
two bad solvents creates a new solvent very near our
solubility target:

In other words, two bad solvents can create a good solvent.

This is a very powerful idea. Now we can choose solvents
for their green qualities, for their odour, safety, evaporation
rates … and combine them to achieve the desired solubility.
The ability to work through issues like this in 3D space is a
powerful feature of the HSP approach.

Microemulsions
One obvious way to deliver a solvent in a gentler fashion is as an emulsion in water. But there are at least
two things wrong with this idea:

1. If the surfactant system isn’t good enough, the emulsion can separate (Surfactancy_Emulsion
Coalescence and Surfactancy_Emulsion Creaming and Flocculation which makes the system
inconvenient and ineffective.
2. If the surfactant system is very good, the solvent is locked inside the surfactant shell, so gets little
chance to clean.

The smart thing to do is to use Surfactancy_HLD to produce a microemulsion with a super-low

Surfactancy_Interfacial Tension and Rigidity. A microemulsion is thermodynamically stable, so it is easy to
mix (no high-shear required) and never separates. It is also characterised by a super-low interfacial tension
which means that when the solvent/water hit the surface to be cleaned, the solvent is fully accessible.

A refinement of these microemulsion cleaners is to optimise the formulation to make them fully dilutable.
This requires some understanding of microemulsion Surfactancy_Phase Diagrams, but if you do it right
then you can ship a concentrated cleaner which can be diluted with water with no risk of phase separation.

198 FST
 Cleaning Surfactants

Cleaning Surfactants
Links
Dishwashing Liquids, Hair Shampoos, Soaps and Washing, Laundry Liquids, Lipstick, Mascara

Lowering the surface tension of the water is the least important aspect of cleaning with surfactants.
Providing feel-good foam is key to selling surfactant systems but is irrelevant to cleaning. Let’s
sort out what is and isn’t important.

Removing oils and fats

Most forms of dirt don’t get significant cleaning benefits from surfactants – you can’t speed up dissolving
jam, removing a stain or loosening some dried-on starch via surface active agents. As explained in
Cleaning_Contact Angles, which in turn refers to Absorbency_Porosity and Darcy, changing the contact
angle on the surface is rarely directly significant in terms of wetting, because we can force wetting through
the action of a spray or cloth.

For getting into porous surfaces which have a contact angle greater than 65° with water, a surfactant will
speed things up – by a factor of 2.5 for an 80° contact angle.

Most of the time, however, we need the surfactant to remove oils and fats. And now we have a real
contradiction between consumer perception and scientific cleaning.

Standard dishwashing and personal care formulations are full of highly foaming, hydrophilic surfactants
like SLES and CAPB. They give a good impression of the surfactants doing something.

But as we can find from Surfactancy_Roll-up and Eötvös Number we really need relatively hydrophobic
surfactants that produce a low interfacial tension with the oils and fats, allowing them to roll up away from
the surface and become engulfed as a microemulsion drop. Using Surfactancy_HLD, such surfactant
packages are easy to formulate and perform their cleaning function much better. They are ideal for the
industrial market where real cleaning is a requirement. But no one will buy them for the consumer market
because they don’t behave as users have been trained to expect.

Surface cleaners
Surface cleaners are interesting because the users don’t want showy foam, they want good, instant cleaning
(so the more hydrophobic surfactants should be used) plus a low surface tension and low advancing contact
angle that helps spray drops to spread and coalesce plus a low receding contact angle to avoid de-wetting
and the coffee-ring effects that might appear if isolated drops dry out.

This low receding contact angle trick provides a further benefit that’s real but not so obvious to the user. By
adding some cationic polymers to complement the anionic surfactants, it is possible to create nm-scale thin
films of “coacervates” – complex polymer/surfactant phases, see Surfactancy_Coacervation. These stick
well to the surface and are humectants (Evaporation_Humectants) which attract atmospheric water and
keep the surface hydrophilic. This means that oily dirt cannot adhere so well to the surface and is easily
wiped away. To the user, the cleaner seems to provide “stay clean” or “EZ Clean” properties.

Specifying a surfactant (blend)

For the purposes of digitalisation of surfactant formulation we might expect to require some key parameters
to allow us to make a rational choice of the surfactants to add.

The key parameters are “price” and “perception by the internet”. However green or scientifically excellent,
your Sales team won’t let you use an expensive ingredient. And Marketing will be insistent that you are,
say, “sulfate free” or “zero ethylene oxides” or “palm oil friendly”.

Whether a surfactant is bio-based or not should be irrelevant. What matters is its PCF (Product Carbon
Footprint) or some similar LCA (Life Cycle Analysis) to tell us the real impact on the environment. It is far
from obvious that bio-based surfactants are superior.

Scientific factors like CMC (Surfactancy_CMC and Langmuir) or Krafft temperature or cloud point
([Link] play almost no role. CMCs are ill-
defined for industrial surfactants and play little role (if they did, we would not be using ionics with CMCs
1000x higher than non-ioincs). Although Krafft point is important (you can’t use the surfactant if it’s not
soluble in the formulation), all the affordable surfactants easily meet the requirement – otherwise they
wouldn’t be produced in the large quantities that make them affordable.

But everyone teaches CMC in surfactant classes

Lots of formulators are sent on surfactant science classes and come back with knowledge of
worthless HLB and a detailed examination of CMC. The reason is that CMC is “sciencey” – it can
be measured, and it has phrases like “Langmuir Isotherm”, so if you’re a teacher, what’s not to
like?

Yet everyone who comes back from such classes finds that CMC is irrelevant to their formulation
challenges so assume that they are just unlucky. It’s not unlucky, it’s bad teaching to focus on
things that are “sciencey” rather than things that are usable.

Similarly, although there is a tenuous link between Critical Packing Parameter and surfactant phases
(Surfactancy_CPP and phases) for pure surfactants, CPP is of little help for a real formulation.

If you are aiming for efficacy over Marketing (and this is the case for laundry detergents) then you need the
Cc values of your surfactants to do the relevant calculations for Surfactancy_HLD, leading, as above, to
efficient roll-up cleaning of oils.

Viscosity
Marketing have convinced users that viscous cleaning products are “rich” and “luxurious”. Apart from
times when a clingy cleaning fluid is required, viscosity is unnecessary for cleaning. It costs nothing to
create within standard formulations, using sodium chloride to create Thickeners_Wormlike Micelles. But
those wanting to use alternative surfactants hit the problem that they don’t thicken in the same way, and
alternative thickeners such as polymers have obvious downsides.

200 FST
 Cleaning Surfactants

Foamability
There are few occasions (foamed bleaches for disinfecting places like dairies are an example) where foam
adds anything to cleaning. All the surfactant molecules in the foam are, by definition, not on the surface
doing something useful.

If you really want foaming for Marketing reasons then there are two broad choices.

1. Use classic, standard foaming tests which tend to confirm that standard systems like SLES/CAPB
are the best. This is unfortunate if you have a surfactant system that is superior in other ways but
foams poorly in the standard test.
2. Use the relatively new understanding that nearly all surfactants foam well if you create the foam
properly: the right amount of trapped air, broken down into small bubbles via shear. A simple
device such as a Japanese shaving net can produce awesome foam in formulations that fail the
standard tests. A CAF (Compressed Air Foam, like a traditional shaving foam canister) also ends
up with naturally small bubbles with a good, stable foam. With some imagination perhaps you can
find a method that is simple, green, cost effective and appreciated by your customers.

Smart dosing
There is a simple way to reduce the carbon footprint of surfactant use: get the user to use half the amount,
half the time. That’s a 4x reduction in carbon footprint. “Half the time” can mean “Half the damage to the
skin or scalp biome” (see SkinCare and Soaps and Washing). “Half the dose” means mostly education of
the user that surfactants can only do so much to remove (usually) minor amounts of oils and fats, so
throwing most of it down the drain is not a good use of their money or good for the environment. Helping
them to dose the correct, small, amount via some smart dosing system would establish good habits.

The obvious downside that this means selling less product is partly true. But laundry detergents have seen a
dramatic reduction in the amounts used (and the temperature required for them to work efficiently) and the
industry is still alive.

The comparison with laundry is sobering. A modern, effective, laundry wash uses ~200ppm of surfactant.
A typical shampoo or shower gel dose, to clean far less, uses 20000 ppm, i.e. 100x more. There is much
room for improvement.

FST 201
Cleaning Temperature effects
Links
Dishwashing Liquids, Hair Shampoos, Soaps and Washing, Laundry Liquids, Lipstick, Mascara

We all use Arrhenius as a rule of thumb for temperature effects. It’s usually good enough.

The famous factor of 2 for 10°C

To get molecules to react (e.g. Cleaning_Bleaching and Oxidation) or to interact, requires thermal energy.
Where R is the universal gas constant of 8.31 J/Kmole and T is the temperature in Kelvin then the
Arrhenius equation tells us that a typical reaction proceeds at a rate, k, given by a constant, A, and an
activation energy Ea:

Ea
− RT
k = Ae
At T=300K, RT ~2.5 kJ/mole. A typical value for Ea is 50 kJ/mole. Doing the arithmetic shows that k
changes by a factor of 2 for each 10°C rise in temperature.

A factor of 2 doesn’t sound very exciting. But a 30min wash at 40°C becomes a 1hr wash at 30°C and a 2hr
wash at 20°C. Users won’t adopt lower temperature washes with such time penalties. So formulations have
to be smarter to win back those factors of 2. You can’t change Arrhenius, so explore the FST for smarter
ways to formulate around the factor of 2 issue.
 Coating Dewetting Theory

Coating Dewetting Theory

Links
Solvent-Based Paint, Emulsion Paint

Everyone adds “wetting agents” to help coatings to, well, wet. But for everything other than inkjet,
the processes in coating/printing machines fully wet the surface with forces orders of magnitude
more powerful than surface tension.

The additives, therefore, aren’t helping with wetting. Instead they are helping to avoid dewetting.

Rapid dewetting should be normal

[Link]

We’ve just printed a 20μm high dot, 500µm radius onto a surface with a 40° contact angle. The viscosity is
a respectable 100cP. According to the app, within about 160 ms, the dot has dewetted to form a 100µm high
drop with radius 300µm.

If you play with the settings you find that no practical printing technique, with any reasonable contact angle
of ink, would ever print a dot that was a close approximation to the original size. Dewetting is fast and
unavoidable.

The elegant paper (referenced in the app) describing and experimentally validating dewetting
theory was based on large drops of silicone fluids on controlled surfaces. Having validated the app
with the large drops, it was expected that because small printed drops of more normal fluids with
smaller contact angles don’t dewet, it must be a feature of their size and contact angle. So it was a
 considerable surprise to find the theory predicting that printing was essentially impossible. Such a
clear negative result tells you either that the theory is wrong (it’s not) or that something else is
going on. That led to the following section.

Measuring contact angles

Everyone measures contact angles (Cleaning_Contact Angles), because that’s what everyone does.
Typically you syringe a tiny drop onto the surface and follow the advancing contact angle with a camera
and automatically fit the final shape to a curve that gives the contact angle. As everyone knows, the key to
getting a reliable value is to carefully clean the surface. Any contamination would invalidate the reading.

But this is entirely pointless. When we coat/print we, by definition, “contaminate” the surface. As
mentioned above, we do this under stresses far larger than surface tensions so there’s no problem getting
the shape/size that we want. Our equilibrium contact angle is meaningless because any retraction of the
drop is over a surface that has been contaminated by the ink/coating. If that contamination gives a fully
wetting substrate, e.g. because we have added a wetting agent that changes the surface, then the drop won’t
dewet back to whatever the equilibrium contact angle is supposed to be.

If we routinely measured receding contact angles then we’d choose a wetting agent that rapidly
contaminated the surface under the drop and gave a small, maybe near 0 receding contact angle. This would
mean that the drop is effectively pinned at its current position – it doesn’t want to advance because its
advancing contact angle would be larger, and it can’t contract because its receding angle is small, meaning
that it would return to the pinning spot.

The reason we measure the irrelevant equilibrium contact angle is because everyone else does, and it’s easy
to do. Receding contact angles are notoriously tricky and unreliable because, well, the surface has been
contaminated.

Note that the wetting agent is not achieving its desirable effects by changing the surface tension of the
coating fluid, i.e. it’s interface with the air. It works by migrating to the fluid/solid surface, changing the
solid/air and solid/fluid surface energy components that each figure in the Young contact angle equation.

So how should you solve dewetting problems? The wrong approach is to choose an additive
because marketing literature tells you that XYZ is a superb dewetting additive. The right approach
is to imagine a molecule with a head that will instantly lock onto your solid surface and a tail that
will strongly attract your solvent – and which will partition rapidly to the solid/liquid interface.
You are the expert on your surface and solution. Once you know what your surface/solvent
combination needs, finding something with a good chance of success becomes a rational process.

Pinholes
We often add wetting agents to avoid pinholes. The theory of avoiding pinholes is described in
Coating_Pinholing Theory, but key to this is that the thickness of the coating and the diameter of the
original pinhole (very often from a bubble) are usually more important than the contact angle used in the
theory.

204 FST
 Coating Dewetting Theory

There’s a fudge in that theory. For the pinhole to close, the advancing (or, as an approximation,
equilibrium) contact angle is important. For it to open wider the receding contact angle is the one to use. As
the advancing contact angle is the more pessimistic value, and because the theory emphasises that thin
coatings need greater attention to cleanliness and bubble reduction, it’s good to leave the focus on the
advancing angle.

Thin film instabilities

If you have a layer in the region of 100nm you can
have spontaneous dewetting. Below 100nm (top 3
images) you tend to get “spinodal” dewetting where
you get sudden, isolated droplets. Above 100nm
(lower 3 images) you tend to get “nucleation and
growth” where holes appear at some defects on the
surface. The holes then grow into some sort of
patterned structure. If you can provide some specific
defects in a specific pattern then you can make a
patterned nanostructure. In both cases, the individual
blobs can undergo Ostwald ripening, so that smaller
drops get smaller, larger get larger, till you have just a few large drops.

The thermodynamic stability/instability depends on surface energies, Hamaker constants and charges;
equations are available but we tend not to know the values to put into apps. The kinetic stability depends on
the mobility of the material at the given temperature, humidity or solvent environment. So whether you get
your desired stability or desired nanostructure is not easy to predict.

Those two paragraphs give a taste of a large, controversial area that I don’t properly understand.
Obviously the effects are somehow linked to Coating_Pinholing Theory but the core theory of dewetting
says that any film of thickness less than the critical dewetting thickness hc (depending on surface tension γ,
density ρ and gravity, g) will spontaneously break into drops. The problem is that this thickness is typically
in the mm range, not the nm range shown here and the µm range typical of pinholes in coatings. I’ve not
been able to make sense of this.

hc = 2 √ ( )sin ( )
γ
ρg
θ
2

FST 205
Coating Levelling Theory
Links
Solvent-Based Paint, Emulsion Paint

If you have some coating lines from a brush or roller, you can measure that they have an average
spacing of λ. The coating has a thickness h, a viscosity η and a surface tension σ.

You can easily work out the time for the lines to disappear. Although people focus on the viscosity
and surface tension, they only have a linear effect on levelling. Much more important are the h3
and λ4 dependencies.

[Link]

A coating will often contain some sort of line defect visible

during creation (coating head, brush, roller) but invisible
once dry. As the wet coatweight is reduced the same wet
defect now shows up in the dry coating. Why? Because the
defect has not had time to level out. The reasons for this are
not obvious until Orchard levelling theory is invoked.

The levelling time for a defect of wavelength λ in a coating of thickness h with viscosity η and surface
tension σ to level by 1/e, te, is given by:

te =
(λ)
3η 2π

σh3

Or in terms of height at time t:

ht = h0exp ()
t
te
 Coating Levelling Theory

It immediately becomes clear that relatively small decreases in thickness can give a large increase in
levelling time because of the h3 term. On the positive side, increasing the sharpness (decreasing λ) of the
defect greatly decreases levelling time. That’s why those who buy cheap paint brushes with coarse bristles
(large λ) have to put on more paint (larger h) than those who buy better brushes with finer bristles.

The app includes corrections for the fact that viscosity of a drying formulation increases with time.

Useless “levelling agents”

Because some “levelling agents” work for some types of defects, it is assumed that you can solve these
linear defects with such agents. At best they lower the surface tension which, from the equation, increases
the levelling time. The confusion arises because these agents address defects such as orange peel not by
lowering the surface tension but by swamping the surface with a uniform surface tension that obliterates the
real problem which is due to Evaporation_Marangoni effects caused by self-ordering surface tension
gradients.

The origins of lines

Although the brush has some effect on line spacing, many of the lines with brushes and with rollers are due
to the ribbing defect that has a complex dependence on speed, viscosity and gap. Maybe we’ll have an app
about it sometime.

Rheology of levelling & sag

It’s well known that if you increase the viscosity of a paint, the tendency to sag goes down (you have a
higher ASI, Anti-Sag Index) but the levelling performance gets worse. There is no simple “viscosity”
metric than can resolve this contradiction. It needs more sophisticated measures such a Power Law fluid fit
as discussed in the Levelling & Sag app:

FST 207
[Link]

You can also use G' and G" values as discussed in Flow_G' and G''.

208 FST
 Coating Pinholing Theory

Coating Pinholing Theory

Links
Solvent-Based Paint, Emulsion Paint

Everyone worries about surface tension σ and contact angle θ when they see pinholes. But there
are two key factors that are less talked about:

1. The diameter of the initial pinhole, d, and, therefore, what caused it

2. The thickness, h, of the coating

You will see from the app that large pinholes are inevitable if d is large and h is small. So, focus on
small d’s and thicker h’s if you want to reduce pinholes. Yes, a lower θ helps, but not as much as
controlling d and h.

[Link]

A coating will often contain a little pinhole. Whether it spontaneously

grows (outer arrows) or self-heals (inner arrows) depends on three factors:

1. contact angle
2. coating thickness
3. hole diameter.

While everyone focusses on contact angle, there is little understanding that you can transform a pinhole-
free coating to one full of holes merely by reducing the coating thickness.

The theory shows why you must keep all 3 factors in mind.

A pinhole of diameter d in a coating of thickness h and contact angle θ will grow if:

h
d < 2(1 − cosθ)

The speed at which it grows depends also on the viscosity η and surface tension σ is given by:

σθ3
v= η
We can now see why thinner coatings are more likely to show pinholes. If a pinhole-free coating is halved
h
in thickness then d can fall below the critical value. The only cure is to have a lower contact angle (better
wetting) or a cleaner, bubble-free coating with less initial pinholes. As the (true) story below shows,
eliminating the source of defects fixes the pinhole problem. In that case, the problem was an unexpected
source of dirt, in many other cases (a fact confirmed by talking to experts around the world) the cause is
bubbles, often from an unexpected source – see Dissolution_Gases.

I was working on a coating line where senior managers were paranoid about the use of WD40 near
the machine as this was guaranteed to create “repellencies”, defects caused by oil droplets During
an outbreak of these defects, doubts about them being caused by oil droplets were sufficient for me
to get permission for a tiny “psst” of WD40 to see how long such repellencies would last.
Unfortunately my tiny “psst” was a big blast covering about 1 m² of coating. But instead of hours
waiting for the repellencies to go away, the coating fixed itself in a few metres, and the oil
repellency hypothesis was killed. It turned out that the continuing outbreak of defects was due to
people like me going into the coating environment to find the cause of the defects. By banning
entry for a short time, the problem went away.

The original source of dirt? I’d insisted on cleaning the machine before the run. This kicked up a
lot of dirt which hadn’t had a chance to settle.

We all make mistakes. The key is to learn from them.

The right contact angle

There’s a fudge in the theory. For the pinhole to close, the advancing (or, as an approximation, equilibrium)
contact angle is important. For it to open wider the receding contact angle is the one to use. As the
advancing contact angle is the more pessimistic value, and because the theory emphasises that thin coatings
need greater attention to cleanliness and bubble reduction, it’s good to leave the focus on the advancing
angle.

210 FST
 Diffusion Barrier Properties

Diffusion Barrier Properties

Links
Fragrances, Lipstick, Mascara

Everyone wants barrier films for, say, oxygen and water. They must be low cost, flexible,
impermeable. The ideal would be a single polymer, or, at least, something that can be made in a
single pass. And in general they should be transparent, colourless and fully recyclable. Getting a
good barrier is easy, getting one that meets the contradictory demands is hard.

Barrier properties
If you find yourself confused by WVTR (MVTR) and OTR (Water Vapour/Moisture and Oxygen
Transmission Rates) then you are like everyone else. The WVTR/OTR world is a disaster zone of bad units
and bad measurements. This is all discussed separately: Diffusion_Permeation OTR and WVTR.

For the rest of this chapter we assume that you want low values for both, plus low values for, say
Fragrances and flavour molecules.

Aluminium
We aren’t going to run out of aluminium, it’s highly recyclable and as a thin foil it is a perfect barrier. As a
vacuum coated film on polymer its barrier properties depend mostly on the cleanliness of the coating line.
This is because pinholes have an outsized effect on the barrier properties. Take some time to explore the
subtleties of the 3 models for pinholes, [Link] but
if we take the Coverage model (which is a bit optimistic) then the sad fact is that 1% pinholes means that
you improve the barrier only by a factor of 100 compared to the many orders of magnitude improvement
from a perfect film.

One of the quirks of vacuum coating is that you can’t fix pinholes by just delivering more metal. More
metal mostly increases cost, with a surprisingly small effect on % pinholes – it’s all to do with how metal
atoms flow on the fresh surface.

A clean roll of polymer, an in-situ cleaner to remove the last remnants of dust, and a good protocol for
cleaning the vacuum chamber between runs are all necessary to get good barriers.

The core problems with Al coated barriers is that you can’t see the contents and these “metal” packages
have been demonized by the public as not being recyclable.

AlOx/SiOx
In theory, and in lab samples, these are ideal barriers. A few nm of some oxide (it doesn’t have to be
perfect, hence the “x” in the name) of Al or Si vacuum coated on clean film and in a clean machine (see the
previous section) provide a transparent, excellent barrier to both water and oxygen. Because the layer is
thin, it is sufficiently flexible to be wound into a roll and processed into a final product.
The problems are obvious – these oxide barriers are brittle and easily scratched. So the barrier properties of
commercial rolls of film are surprisingly modest and some glitch during a production run of the end
product can introduce barrier-breaking stresses that would not be detectable; so products with poor barriers
can be (and have been) shipped.

A factor that makes these barriers especially prone to cracking under modest stresses is that the strains
“focus” on the layer via Dundurs α parameter in fracture mechanics. An app that shows the significance of
this effect is at [Link] It shows that the problems
arise due to a big mismatch between the modulus of the substrate (say 1 GPa) and that of the oxide coating
(say 50 GPa). If you can do something to increase the substrate modulus (adding a thin layer of 2 GPa) that
does a lot to decrease the severity of the Dundurs α effect.

Biopolymers
If you take almost any natural polymer, starch or protein, you can produce a film with surprisingly good
oxygen barrier properties, for reasons discussed below.

The obvious problem is that the water barrier properties are generally poor, and highly %RH dependent. At
high %RH even the oxygen barrier properties get worse.

Various chemical tricks can reduce the %RH dependence, but then the polymer is becoming less natural.

It isn’t good enough to have good intentions and a friendly biopolymer. Being green means not
wasting precious resources, but the predictable failure of most biopolymer barriers has meant much
wasted resource.

PE, PP
The standard food packaging films based on polyethylene or polypropylene (PE/PP, for our purposes
indistinguishable) are wonderful water and oxygen barriers. But that’s only because they are multi-layered
with one or more thin layers of EVOH created by co-extrusion.

The fact that they are not pure polymers is very annoying to those who demand that food packaging should
be easily recycled. So why do evil packaging film suppliers insist on contaminating the recycling stream?
Because pure PE/PP films are useless oxygen barriers. [Some new grades are more oriented/crystalline with
somewhat better barrier properties, but it remains a challenge.]

The reason is simple. HSP Dissolution_Hansen Solubility Parameters are used throughout the FST because
they deliver so much power with so little effort. The key is that if the Distance between the HSP of, say, a
polymer and a solute is large then they are incompatible, a small Distance means they are compatible.

It is obvious that the HSP Distance between water and PE/PP is large. So water does not partition into the
polymer. And, because (Diffusion_Basic Diffusion) the rate of diffusion depends on the concentration
gradient, a low level of water in the surface of the polymer means a low rate of diffusion.

It happens that the HSP of O2 is relatively close to the HSP of PE/PP, so O2 partitions readily into the
polymer, there is a large concentration gradient, so diffusion is rapid.

212 FST
 Diffusion Barrier Properties

Now let’s put a thin layer of EVOH into the middle of the PE/PP film. EVOH is a poor water barrier (the
HSP Distance to water is obviously much lower) but its Distance from O2 is large, so there is only a low
concentration that can partition into the EVOH, so a small diffusion. It is remarkable that even 1µm of
EVOH provides a near perfect barrier to O2 inside a co-extruded film. In practice the thickness is larger
and there is more than one layer. This is because it is very hard to extrude a pinhole-free 1µm EVOH
barrier, so going to 2µm and having multiple layers is a cost penalty worth paying in order to obtain the
astonishingly good barrier properties from apparently cheap and simple plastic films.

The reason the biopolymers have good O2 barriers is the same – their HSP Distance is large. And, of
course, their distance to H2O is smaller, so they are poor water barriers.

It’s well known that PE/PP film is a poor barrier for common fragrance molecules like limonene. Again, the
HSP Distance is small. The EVOH multilayer packaging films provide, not surprisingly, excellent barrier
properties for a wide range of fragrances and flavours.

PET
The discussion of PE/PP implies that any polymer will be permeable to something if the HSP Distance is
small. The exception would be a polymer that has a very low value for the diffusion coefficient, D. Even if
there is reasonable partition to set up a concentration gradient, if D is small you won’t get too much
permeation.

PET is used because, when oriented, (both as PET film and as PET bottles) the polymer is relatively dense
and crystalline while remaining crystal clear. This high density means that there is little “free volume” for
molecules to diffuse through. While PE/PP D values are in the 10-9 cm²/s range, for PET we are in the
10-13 range. Four orders of magnitude is a lot. And that’s why we use a lot of PET.

The thickness of barrier is only linear in its effect on permeation. Usually you cannot solve a diffusion
problem by increasing thickness. However, when (for strength, not barrier, reasons) we go to a PET bottle
with a thickness of 500µm compared to a packaging film of 50µm that’s an extra order of magnitude. So
even a relatively small, mobile molecule like CO2 has a five orders of magnitude barrier, allowing us to
store fizzy drinks for a long time.

Clay barriers
Everyone loves the idea of taking a cheap, natural clay and creating a
tortuous barrier layer. For a thickness L, a molecule has to travel a distance
LB to cross the barrier. The more clay we can pack in, the more tortuous
the path and the better the barrier. The problem is that clays, when well-
dispersed, add considerable viscosity to the dispersion so there is always a
trade-off between the amount you would like to add and the amount you
can add.

To create such a barrier we must have a high aspect ratio, l/d, for the
particle, and clays are ideal for that provided you can separate them into
individual layers. That is harder than it sounds and there are many failed
attempts to provide good clay barriers.

FST 213
 While writing a chapter on the use of HSP for understanding so-called green polymer PLA, it was
heartbreaking to see how many PhDs had been wasted on trying to add clays to the polymer. A few
seconds thought showed that the dispersants they were using were guaranteed to fail – which they
did. The one commercially-available dispersant that, from HSP, would give great dispersion
happened to react with the PLA during processing. A few minutes thought could have saved years
of failed research.

Finally, you need the order parameter, S to be close to 1 to maximize the barrier effect. It is not intuitive
that the order parameter go from -½ to 1, but that is the scale that everyone uses:

Now we can calculate the relative permeation compared to no nanoclay:

[Link]

There are some disagreements between models, but you can see that by, say, 15% high aspect-ratio clay,
nicely oriented, you have a great barrier. That 15% is volume fraction – depending on relative densities its
weight fraction is higher.

If these barriers are so good, why don’t we see more of them.

• Because they are based on nanoclays, there are issues of nanosafety. Many would argue that these
are non-issues, but the regulatory infrastructure defines them as nano, so you need to apply nano
rules.
• Because clays (and therefore the barriers) are often yellow.
• Because dispersing the clays down to individual plates, required for a high aspect-ratio, is very
hard. And such dispersions can be highly viscous.
• Orientation during production might not be perfect.
• Because these tortuous barriers can be deceptive. For short-term measurements, the barriers seem
amazing. But once the water or oxygen has traced the tortuous path, you “only” have the increase

214 FST
 Diffusion Barrier Properties

in path length. This might be a factor of 100, which is great, but isn’t the factor of 1000s that they
appear to be when you measure them before breakthrough via the tortuosity.

Pinholes in seams
If, see Adhesive, we use a heat seal to create a seam in a package, then much of our barrier efforts will be
wasted if, through carelessness or dirt, a pinhole goes through the seam, allowing oxygen and water a
simpler way to get inside. The Leakage app shows that even a 25 µm pinhole can make your OTR
irrelevant:

[Link]

The theory is simple, the flux, Q, through the pinhole depends on the diffusion coefficient in air, D, the
pressure difference ΔP across the pinhole (0.21 atm for oxygen), the length of the seam, l, through which
the air passes and the radius of the pinhole, r.

πr2
Q = DΔP l

Q is in units of Pa.m²/s and, as so often with barriers, the difficulty is making sure you get all the units right
in order to compare the various fluxes. In this case, the 25 µm pinhole has no effect on the WVTR but is
almost 4x larger than the OTR through the packaging film.

FST 215
Diffusion Basic Diffusion
Links
SkinCare, Surface Cleaning, Lipstick, Mascara, Water-based Adhesive, Adhesive, Microencapsulation

You can calculate how much of a molecule diffuses into (or out of) a coating by knowing three
things:

1. The thickness of the coating;

2. The maximum solubility of the molecule in the coating;
3. The diffusion coefficient of the molecule inside the coating.

#1 is easy to measure, #2 needs a bit of experimentation (or can be predicted via

Dissolution_Hansen Solubility Parameters) and #3 can be estimated via its molecular weight if
there is a known value for a reference molecule.

With a bit of training it becomes easy to use the relevant apps to understand and calculate
diffusion.

The basics of diffusion are easy to understand.

[Link]

The formula to calculate diffusion is simplicity itself and is called Fick's First Law. It
says that the amount of material per unit area that will flow in a given time (the Flux, F)
over a given distance x depends on a constant, the Diffusion Coefficient D and the
ΔC
change of concentration ΔC over that distance (the concentration gradient), i.e. x , or
δC
δx if you want to be fancy. Therefore:

ΔC δC
F=D x = D δx

Note that purists add a negative sign on the right-hand side. Fick's Second Law of Diffusion is not quite so
intuitive. It tells us that the change of concentration with time is the change of flux with distance. In other
words:

δC δF δ2C
δt = D δx = D
δ2x
 Diffusion Basic Diffusion

Although these two laws are simple, they are differential equations (for example, the concentration gradient
changes with the flux which depends on the concentration gradient) and the solutions in most real cases
(where D also depends on concentration) are not straightforward. Therefore numerical simulations are
required to capture what really happens. Fortunately, we can explore some simpler cases which happen to
be very relevant in practical terms. Once we're happy with the basics we can then look at more general
cases, with the app doing all the hard work of carrying out the numerical simulations.

Steady state diffusion across a barrier

We already know the answer to how much chemical will diffuse across a barrier of thickness L if the
concentration is C at one side and 0 at the other. We have a concentration gradient (C-0)/L, a diffusion
coefficient D, so the flux F is:

DC
F= L

The real problem is not the formula, but getting inputs and outputs in the right units. D is usually expressed
in terms of cm²/s and has values of 10-5 for self diffusion in liquids, thru 10-7 for diffusion through rubbery
polymers up to 10-12 for typical polymers and 10-15 for very close-packed polymers. If you have a D value
in m²/s multiply it by 104 or, the same thing, subtract 4 from the exponent. 10-13 m²/s is the same as 10-9
cm²/s.

From time to time I try to convince myself to switch to m²/s, but we live in the real world and
cm²/s are so common and comfortable that they are the default in the relevant apps.

Concentrations can then be expressed in the familiar terms of g/cm³. Thicknesses are typically in µm but
have to be converted to cm for the calculation. Fluxes are, therefore, in g/cm²/s. A complication is that it's
often easier to think of solubility in terms of volume fraction. Assuming, for simplicity, that all densities are
1g/cc, a volume fraction of 1 equals a concentration of 1g/cc. This is convenient. It is intuitive to think of a
polymer containing a 0.1 fraction of a solvent (i.e. 10%) and this simply becomes 0.1g/cc for the
calculations. Adjusting for relative densities is a trivial task.

Half Time

In addition to the Flux, the time, t½, to reach half the equilibrium amount (either by absorption from empty
L2
or desorption from equilibrium) is given by t½ = 0.049 D . This equation follows from the fact that
absorption or desorption goes as the square root of time. It is often used to calculate the diffusion
coefficient from experimental half-time measurements.

Now we can see what happens as we change our key parameters. Note that because of the limitations of
app interfaces the diffusion coefficient is entered as a value between 1 and 9.9 and an exponent from
10-5 to 10-15.

Thickness
Because of the plane of symmetry in the middle of a free film, L = 2 when modeling absorption or
desorption where mass is transported through both surfaces. When modelling permeation, which is one-
sided exposure, then L=Thickness.

FST 217
Real data
Although it can be difficult to get diffusion data, there are many cases where it’s both easy and useful. A
typical example is where you need to know the equilibrium concentration (which controls ΔC) and the
diffusion coefficient. Let’s look at a typical absorption experiment using the Fickian diffusion app:

[Link]

A 50 µm film is weighed then dipped into a solvent. After, say, 5 min, it is pulled out, quickly dried and
weighed, before being put in and removed after 10 min (total), then 20, 40 up to 180min. From the
increases in weight we can work out a time versus % max concentration curve. In the app we enter L as 25
μm (Thickness/2 because solvent comes from both sides) and by sliding sliders we find a good match to the
experimental curve if D is 1.10-9 cm²/s. From the asymptotic weight gain we also know the equilibrium
solubility. With one experiment we have fully characterised the system.

Of course, it’s not as easy as that. What sized sample is used, what thickness chosen, how do you reliably
wipe dry for each measurement, how do you weigh accurately? Because I’ve done this many times, in
different labs with different abilities to weigh accurately, and with different polymers covering a range of D
values, I can say that it’s not as difficult as it sounds. Within 1hr you can easily work out many of the issues
and if you are lucky you can have some good values in less than a day. If you have a difficult polymer, it
might need some time to get the right sample thickness and some overnight measurements to get the
equilibrium value.

If you cannot fit your diffusion data to the standard Fickian curve, the chances are that you have the
interesting case of concentration-dependent diffusion. This is discussed in its own chapter,
Diffusion_Concentration-Dependent Diffusion.

Alternative measurement techniques are available. For example you can attach your sample to an FTIR
ATR crystal and wait to see how long it takes for solvent on the other side to become detected and link that
time to the “breakthrough” time calculated in the non-“blocked” mode in the app, in this case 5.3 min:

218 FST
 Diffusion Basic Diffusion

If
you

want a good enough estimate of D for a molecule through a common polymer then you can use the
Diffusion Coefficients app at [Link] The
equations used have been developed for the food and pharma industries where they realise it is unrealistic
for everyone to measure D values for all their molecules.

FST 219
Diffusion Concentration-Dependent Diffusion
Links
Water-based Adhesive, Pharma Formulation, Microencapsulation

Standard Fickian diffusion theory works fine for many cases – but if the liquid is significantly
soluble in the polymer, diffusion coefficients get much larger, with some expected and unexpected
results.

2 orders of magnitude
The diffusion coefficients described in Diffusion_Basic Diffusion and in Diffusion_Diffusion Coefficients
are constant. Some polymers have large constants, say, 10-8 cm²/s and others might be lower, say 10-10. The
difference is due to the amount of “free volume” in the polymer. To diffuse, a molecule must have
somewhere to move to, which means some free space that opens up when, say, the polymer chain wiggles.
Compact polymers and those below their Tg have little free volume.

Now let some solvent diffuse into the polymer. Because the concentration is a bit higher, the polymer is
slightly swollen and there is more free volume – so the solvent can diffuse in faster, the concentration gets
higher, the diffusion coefficient gets larger. It can easily increase by 2 orders of magnitude.

[Link]

Here we have a 40µm film of polymer. Because the “Blocked” option is checked, solvent coming in doesn’t
exit from the other side. Exposing a 40µm sample on both sides is equivalent to 20µm blocked. After 1hr
the solvent has just reached the middle. The diffusion coefficient, D, is concentration independent at 10-10.
 Diffusion Concentration-Dependent Diffusion

Now let’s make it concentration dependent so at higher concentration D = 10-8.

Now after 1hr the solvent has almost saturated the polymer. This is what you might expect when you think
about it.

So what is the unexpected consequence?

Asymmetry
In 1 hour, dipped in solvent, the film gets full of solvent. If we take it out and the solvent is free to
evaporate from the surface (so the concentration on the outside is 0) we might expect the solvent to empty
in 1hr. But that’s not what happens:

FST 221
We are now in Desorption mode. After 1 hour, the concentration near the middle is still 50% and the total
remaining (right hand graph) is 40%.

How can this asymmetry arise. Why can it fill to 100% in 1 hour but only lose 60% via desorption?

The answer is that the solvent has to escape via the surface with its low concentration of solvent and,
therefore, a low diffusion coefficient. Absorbing the solvent is like having a tap wide open on the outside,
desorption is via the same tap allowing only a drip of fluid to escape.

Many of us have applied coatings to polymers using solvents that can diffuse into the substrate. We dry our
coatings and think that we have no solvent left. But it might have rapidly diffused in by a few µm and will
take a long time to escape. We might be lucky and never notice – or it might cause a problem in subsequent
processes for us or our customers.

I learned about this desorption phenomenon the hard way. We had coated a product using
cyclohexanone, CHX, as the solvent. Obviously we dried it well; CHX has a pungent odour. The
product was cut into packs of sheets, sealed in polyethylene bags, and shipped abroad. When
opened after some weeks there was an overpowering odour of CHX. Our solvent had rapidly
diffused in and we had been unaware that it would slowly diffuse out during storage and get
trapped inside the polyethylene which, for HSP reasons, is a good barrier.

Those who print multilayers of ink sometimes find that 4 layers adhere well, while 5 layers show poor
adhesion. The instinct is to blame one of the colours and to change the print order to try to fix the problem.
The real reason for failure is a slow build-up of solvent in the substrate; each printed/dried colour leaves a
little extra solvent in the top µm of the substrate. At, say, the 5th layer, the ink/substrate interface is
sufficiently weakened by the solvent and adhesion fails.

222 FST
 Diffusion Diffusion Coefficients

Diffusion Diffusion Coefficients

Links
SkinCare, Surface Cleaning, Water-based Adhesive, Microencapsulation

To calculate diffusion we need to know the diffusion coefficients of our materials. Where can we
find them, and how do they depend on temperature and on the MW of the molecule diffusion?

And if we can’t find them, how can we measure them?

A good-enough guide
Because lots of us need diffusion coefficients, D, to calculate diffusion for purposes such as regulatory
compliance, and because regulators know it is unreasonable to expect everyone to know D values for every
molecule in every polymer at every temperature, there are some semi-official formulae available which, if
you use them, allow you to show that you are putting in reasonable effort.

[Link]

You enter the MW of your molecule, the temperature of interest and your chosen polymer. D is calculated
for you.

The formula used is:

D = 104eA − C1MW
0.667
/
+ C2MW − C3 T

Where C1 = 0.135, C2 = 0.003 and C3 = 10454 and A = A*-τ/T and T is in °K. Those two constants, A*
and τ are polymer specific.

0.667
The app has some alternative equations should you be interested. The exponential dependence on MW
2
( 100 )
is one common form but the way it is used here gives an effective MW dependence which is a
commonly-used approximation, using 100 as a convenient value for typical common solvents such as
toluene.

Measure your own

Although there are many fancy academic ways to measure diffusion coefficients, it’s not so hard to get a
good-enough value via some simple experiments.

The easiest method is to get a thin sample, say 50µm, weighing a few grams, and to place it in some solvent
that you know is compatible, but not too compatible, with the polymer. At increasing intervals take it out,
wipe it dry, weigh and return to the solvent. Finally, leave it overnight and remove/wipe/weigh to get the
equilibrium absorption value. Using the [Link]
Fickian diffusion app in “blocked” mode, slide the various sliders (remember to set the thickness to half
your sample thickness as solvent is coming in both sides) till your data match the experimental values. If
the solvent is too good then you will get a poor match. Swap over to the [Link]
practical-solubility/[Link] Concentration Dependent app and play around with values till you get
an OK fit.

This sounds crude, but it is remarkably effective. You generally don’t need D values to high accuracy, and
the advantage of the simple method is that you can do it with various solvents to check on dependence on
molecular size.

The trick is to get a sample thin enough to allow you to reach equilibrium in a reasonable time, with a
weight matched to whatever balance you have so that weight increases can be measured accurately. Before
carrying out the detailed work you need two preliminary experiments:

1. The simple-sounding “wipe it dry” needs a bit of practice, so before committing to a proper
experiment, try dipping the weighed sample into the solvent, removing it immediately and testing
that your wipe/dry procedure leads to no significant weight increase.
2. The “compatible but not too compatible” solvent can readily be found by setting up, say, 4
different solvents, placing a sample in each and leaving overnight. A solvent that has seriously
damaged the sample is obviously unsuitable, and any solvent with too small a weight increase is
equally unsuitable.

This approach of doing “scouting” experiments is highly recommended; it is a small investment of time to
avoid large wastes of time when attempting unfamiliar experiments.

A $multimillion contract depended on proving that a certain hydrophobic material would not
absorb more than a certain % water over a long time. Doing the test would have required 3 months
of dunking the sample in water – time they didn’t have. Nor did they have the diffusion
coefficients to calculate the effect. I suggested they made a thin sample (their real object was thick)
and doing the dunking, removing, drying, weighing and re-dunking technique to measure D. To
their astonishment after a couple of days they had a reasonable value. Refining their technique, a
few days later they had a reliable value, could prove to themselves and their client that their
material met the spec, and they won the contract. Not bad for a few days of simple weighing of
thin samples of their material.

If you have a convenient ATR-FTIR spectrometer then you can place your ATR crystal onto the polymer
which has the solvent trapped on the other side. Keep scanning till you start to see the peaks from your
solvent. This gives you a breakthrough time which you can find in the Fickian app. There is some
uncertainty about how much solvent constitutes “breakthrough” and how sensitive your spectrometer is to
that solvent, but again you can get surprisingly good values for relatively little effort.

If you want to test for breakthrough using some sort of solvent detector (e.g. a simple mass spec “sniffer”)
then the biggest difficulty is getting a good seal between the sample and whatever is separating the solvent
from your sniffer. There is a long history of false high D values due to inadequate seals, especially for

224 FST
 Diffusion Diffusion Coefficients

polymers with very small D values.

FST 225
Diffusion Diffusion into Skin and Hair
Links
Hair Conditioners, Sun Screens, SkinCare, Lipstick, Mascara, Pharma Formulation

The dual (and sometimes conflicting) demands of beauty and safety require an understanding of
how chemicals diffuse into skin and hair. The combination of solubility and diffusion science in
the FST allows us to make more informed formulation decisions.

Skin diffusion
A paper that happened to be technically correct but deeply deceptive convinced the skin community that
permeation was controlled by LogP, the water/octanol partition coefficient. This is provably wrong. The
real correlation is with molecular size which, in turn, is the single most useful predictor for LogP. And the
correlation was with a partition ratio from water not with what matters which is the flux from whatever
medium is being applied.

The true flux dependency is simple and obvious:

• Solubility of the ingredient in the medium being applied to the skin.

• Partition between the medium and the skin.
• Molecular size correlation with diffusion coefficient.
• Skin thickness (highly variable for different parts of the body).
• Skin diffusion coefficient (probably ~ 10-9 cm²/s but depends on quality and moisture level of the
skin.

With the use of Dissolution_Hansen Solubility Parameters you can readily work out correlations with the
first two and with a diffusion modeller Diffusion_Basic Diffusion and Diffusion_Diffusion Coefficients
you can work out the final piece.

Because skin diffusion is relatively slow, the one ingredient you can ignore is water. There is a vast amount
of work (using, especially, Franz cells) measuring permeation from aqueous solutions over long periods
such as 24hr. This is all a complete waste of effort. The water in a typical skin formulation is lost within ~
20 minutes. This means that all experiments that look at other aspects of aqueous delivery, such as the type
of emulsion (w/o, o/w, microemulsion) are meaningless because after 20 minutes there is no emulsion
because the water has evaporated.

When you do the modelling described above it is, therefore, important to model the % of ingredients on the
skin. A lovely example is a skin formulation that delivered none of the desired API (active pharmaceutical
ingredient) through the skin. It was delivered as 70% water emulsion containing 3% of a surfactant such as
Tween. When the water disappeared, the 3% surfactant was now 10% of the formulation and because there
was no water, it was no longer surface active. It was merely a chemical. In this case, a chemical with a lot
of polyethylene oxide(PEO) chains, and one so large (above the typical 500 Dalton MW limit) that it does
not permeate through the skin. The API happened to have a great HSP match for PEO, so it remained
happily on the surface, trapped in the surfactant.
 Diffusion Diffusion into Skin and Hair

Another aspect of HSP relevant to skin permeation is the fact that two bad solvents can make a good one –
their HSP end up as an average which might be a good match for a chemical or for skin. A classic example
was the observation made some years ago that although propylene glycol, PG and “essential oils” like
limonene were relatively ineffective at aiding skin permeation, a mix was quite potent. When you do the
calculation, the HSP is a good match for the HSP of skin.

HSP of skin?
If we think of the skin as a polymer, and use standard HSP techniques for measuring its compatibility with
various solvents, you can work out that it behaves like a polymer with an HSP of something like [17, 8, 8].
People get angry at this value. After all, they “know” that skin is permeated through the lipid layer and
lipids are hydrophobic. But the lipid layer contains 20+% of cholesterol which easily takes the HSP up into
the [17.7, 8, 8] range. In any case, the idea that chemicals permeate only via the lipid layer has become less
and less tenable.

If you look up the HSP of dimethyl isosorbide (DMI), you find that it’s [17.7, 7.1, 7.5]. If you take the
naïve HSP prediction as true, you would expect DMI to be great at penetrating skin. This is indeed the case.
In fact it can be too good. If you happen to use DMI as a solvent for an API with a large MW and therefore
a slow rate of diffusion, the DMI can penetrate the skin faster than the API. At some point the API on the
surface has too little DMI and crystallises out. When the crystal form happens to be needle-like, the result
can be very painful.

I first came across DMI in a non-skin context. I worked out its HSP values and saw that it would
be superb as a skin permeation enhancer. I mentioned this to someone in the skin care industry
who, to my delight, confirmed that it had been found to be green, safe and an excellent permeation
enhancer.

Hair
Because healthy hair has a strong armour of high-quality keratin, diffusion into the hair itself is minimal. It
takes heat, chemicals and aggressive solvents to open up the keratin to allow molecules (such as dyes) to
migrate inside.

The outer layer of high-quality fatty acids (either free or bonded onto the keratin) such as 18-MEA
(18-methyleicosanoic acid) provides a truly hydrophobic layer which ideally would be mostly left alone.

As hair grows, the younger portions near the roots are in good shape while the older portions, near the tips,
have sustained more damage so are more open to diffusion.

Add to this the assaults from surfactants, conditioners, silicones, quats, hair dryers, bleaches and dyes, it
becomes less and less clear what “hair” is.

Until we know more, we can basically say that under normal circumstances diffusion into hair is, and
should be, difficult and that after extensive human interventions, diffusion is probably easy and undesirable.

FST 227
Diffusion Permeation OTR and WVTR
Links
Lipstick, Mascara

Everyone gets OTR (Oxygen Transmission Rate) and WVTR (Water Vapour Transmission Rate, or
M for Moisture) values wrong. Literature values are wrong, internal publications are wrong, values
provided by testing facilities are wrong. It’s a mess. This chapter does its best to sort out the mess,
with a life-saving app being at its heart.

We’re all wrong and confused

When a leading expert in permeation admits in a paper that she and her colleagues have spotted errors in
their own values, and in books, journals and databases, you know that the situation is bad. The wonderful
paper from Prof Kay Cooksey and colleagues at Clemson University inspired the apps at
[Link] which have been used 10’s of
thousands of times by people as confused as everyone else.

It is so easy to make an error, and even the app admits that a couple of errors had crept it which were
spotted by smart users so they could be fixed.

Even when we’re not wrong, we are often confused. So, take heart. You are not alone in finding this topic
hard.

Permeability versus Transmission Rate

The first confusion is between Permeability, P and Transmission Rate, TR. Another layer of confusion is
that people talk of “permeation” which might mean either P or TR.

Here’s the difference:

• P is the amount that will go through a unit area of a unit thickness of a barrier in unit time and is
usually defined for a single material.
• TR is the amount that will go through a unit area in unit time of your specific piece of barrier film.
It can be the value for a given thickness of a single material or for multiple layers of different
materials.

Calculating TR values
Let’s say that we have 3 layers of materials. We know the TR values of each of the 3 materials (i.e. at the
specific thicknesses used in our 3-layer barrier), TR1, TR2 and TR3 so we can calculate the overall TR via:

1
TR = 1 1 1
+ +
TR1 TR2 TR3
 Diffusion Permeation OTR and WVTR

If we know the thicknesses Lx of the individual layers and we know their P values then TR is given by:

1
TR = L1 L2 L3
+ +
P1 P2 P3

The app does the calculations for you, with other variants that are needed from time to time.

But it’s complicated

If (and that’s a big if) you know that the OTR was measured with air on one side of the barrier, then you
have only 20% O2. If you know that P was measured and calculated with respect to 100% O2 then you
have a factor of 5 difference if you are mixing and matching TR and P values.

Similarly, if WVTR was measured at 50% RH you might (or might not) have a factor of 2 correction for P.
Why “might not”? Because if someone has a barrier that is very sensitive to water, they might want to
quote P “at 50% RH” and, rightly, not including the factor of 2 correction because at 100% RH it might be
a factor of 10 higher.

There is nothing the app can do to help with this mess. You just have to read the relevant literature or report
carefully.

The units of P
This is another nightmare.

[Link]

If a paper quotes O2 permeability in kg.m/m².[Link] (that’s 5 units – mass, length, area, time, pressure) and
you need it in cm³.µm/cm².[Link] then you might have a problem knowing how to convert them. The app
does it for you. If the original value was 1 then the converted value is 6.6.1017.

Note that if you convert from a mass unit (here kg) to a volume unit (here cm³) you need the MW of the
gas, here 32 because it’s O2.

If everyone used clear 5-unit units then conversion is tedious but do-able via the app. But someone looking
at cm³.µm/cm².[Link] might say “Oh, we can cancel the cm” and give the units as cm.µm/[Link]. If their
length unit had been cm then instead of cm³.cm/cm².[Link] they might give you cm²/[Link], which is
even more confusing.

The sad thing is that the “real” units for P are, surprisingly, seconds. In the above example P was 0.0312s.
No one quotes P in s, but if they did, our nightmare of conversions would go away.

FST 229
However, this wouldn’t fix the problem of users recording P based on 1 atm of air rather than 1 atm of O2
(the factor of 5 error), and confusing 100% RH with 1 atm of water which, at 25°C is an error of 760/25 ~
30. It’s surprisingly common to find values that are out by a factor of 30.

I enjoy helping people who email me with formulation issues. But when I see “permeation” in the
Subject I open the email with dread. It means opening my own app, reminding myself of the
complexities, then tracking down the chain of errors (often in literature values) that has caused the
problem in the email. What encourages me to provide an answer is that the knowledge of the
problems and their solutions, via the app, is steadily spreading. I now get fewer requests for help
because the wider community is better able to help itself.

Different T and %RH

WVTR values measured at a given T and %RH can, in principle, be converted to other T and %RH values:

It’s complicated and works only for those barriers where P doesn’t change with %RH. But if someone
quotes a WVTR at 38°C/90% RH and you need to know what it would be at 25°C/50% RH, then with a bit
of thought and effort you can get there.

It’s a mess
Wouldn’t it be nice if a magic app could sort out all these problems. It would, but the problems are so deep
and the data “out there” so poor and confusing that there is no substitute for sitting down and going step-
by-step through whichever problem you are facing.

You will find values that are out by factors of 30 or 106, you will find wrong units or the right units
cancelled out (as in the cm examples above), or just about any mixture of errors you can imagine. If you go
slowly, if you imagine the sorts of errors that you might make, if you think through the traps of unit
conversion, atmospheres, volumes etc. then you will find why the values you find in a paper make no sense
and then reach a reasonable conclusion about the values their experiments have really measured.

That’s the best you can do. Take inspiration from the experts who admitted their own failings and do your
best to work around errors from others and to check that your own work isn’t confusing TR with P, isn’t
cancelling out units, and isn’t wrong by factors of 30 or 106.

230 FST
 Dispersions DLVO

Dispersions DLVO
Links
Water-based Adhesive, Emulsion Paint, Emulsion Polymers, Inkjet Inks

DLVO theory is often seen as the theory of dispersion. But it’s surprisingly limited other than for
charged species in water where it is useful. All relevant formulae can be found when you use the
DLVO app.

Attraction and Repulsion

We have two particles of radius r separated by a distance h and having a
surface of charge φ:

They have an attraction due to van der Waals forces which depend on the
Hamaker constant, A12, between the particles and, strongly, on r.

They also have a repulsion (“like charges repel”) that depends on the charge
and the dielectric constant and concentration of other ions in the water.

The charge is surprisingly ill-defined, for reasons discussed in Dispersions_Zeta potential.

DLVO (from Derjaguin and Landau, Verwey and Overbeek) theory lets us calculate the relative attraction
and repulsion forces:

[Link]
The y-axis is the interparticle energy in units of kT: positive is repulsive, negative attractive. The light-blue
dotted line is the Vh curve, (H for Hamaker) which shows the rapidly increasing van der Waals attractions
as the particles get very close. The light-yellow dotted line is the Vd curve (D for Debye, the theory used
for the electrostatic repulsion) which shows increasing repulsion.

The solid green line is the net result. The repulsion peaks at ~28 kT then van der Waals takes over. But as
long as the particles don’t have enough energy to overcome the 28 kT, they will stay apart.

Controlling the barrier

Because the attraction depends on the Hamaker constant, which is typically 10-20 J, you might think it a
good idea to reduce it. Unfortunately, most of us most of the time have no idea what the real value is and if
we did there’s little we can do to change it. There is however one trick. If you have, say, polystyrene
particles in, say, a mix of decalin and tetralin, it is possible to adjust the refractive index of the solvent
blend to exactly match that of the polystyrene … and A12 goes to zero. [Strictly, you also have to match the
dielectric constants, but the effect is usually negligible]. Some elegant experiments looking into particle
behaviour unperturbed by attractions have done just that. But then we’re not in water, and getting an exact
match to the low RI of water is rather a challenge.

For charge repulsion if the barrier is over 20kT (some say 30kT) then the dispersion is likely to be stable.
Unfortunately, the barrier size decreases as particle radius decreases, so smaller particles are more difficult
to stabilize. The barrier depends on the Debye length which is confusingly called k-1. This is because the
−kh
effective voltage near a particle falls off with distance h as e . Because we want the exponential to be
-1
without units, and h is a distance we talk of k as the Debye length, the inverse of the Debye exponential
factor.

For a given particle you can generally change its charge by changing the pH. Particles might be -ve at high
pH and +ve at low pH, passing through zero charge at the isoelectric point. So the first trick for stability is
to control the pH to remain in a high charge region.

If you have to add extra salts to the formulation for other reasons, the repulsive forces decrease. The ions
shield the mutual repulsive charges. The effect is controlled by the ionic strength which means that cations
like Ca2+ or anions like SO42- are especially troublesome because ionic strength depends on charge number
squared, so if you are OK with a certain concentration of NaCl, you might have problems with the same
molar concentration of CaCl2 or Na2SO4. The app calculates the ionic strength as you change the charge
numbers, and you see the effect on the charge barrier.

The app shows the simple theory. There are some subtleties not captured. For example, if you have two
types of particles, one with a zeta potential of, say, -5mV and one with -50mV, there can be an attraction at
low ionic strength, repulsion at medium concentration and attraction at high ionic strengths.

It is possible to accidentally quench the particle charge. For example, a small amount of cationic surfactant
can be attracted to a negative particle, leading to a near-zero net charge and near-instant clumping of the
particles.

Finally, and importantly, charge stabilization is effective only in water and a few other solvents with a high
dielectric constant. As you move to more typical dielectric constants, the repulsive force decreases rapidly,
giving little effective stability. In these cases you need steric stabilization.

232 FST
 Dispersions DLVO

DLVO as a religion

Famous colloid scientist Prof Barry Ninham has referred to DLVO has a “religion”. As he was
friends with both D and O this was not a personal attack on the authors. Instead it was an attack on
a lazy community that had taken a rough-and-ready, impressive-for-its-time theory and elevated it
to the theory of colloid science. It is a recurrent theme of FST that once-useful theories take on a
life beyond their (small) zone of utility to become unarguable orthodoxy. As a community we
should always take the best from any theory but be actively alert to its faults. This especially
applies to the FST. Please send in your critiques of its faults.

Steric stabilization
Although steric stabilization is not a part of DLVO, it is usually included under the DLVO heading. It
works in aqueous systems but, unlike charge stabilization, also works well in non-aqueous systems, so is a
more general technique.

We have polymers sticking out from the surface by a distance δ. Turning off
the charge in the app and adding δ = 1nm layer of polymer results in a very
strong particle-particle repulsion at a distance of 2nm.

This is steric stabilization and it works very well … until it suddenly doesn’t.

The key factor is the Dissolution_Flory-Huggins χ parameter, a measure

of how much/little a polymer prefers itself to its solvent environment.
When this is less than 0.5, steric stabilization is effective. As you change
the solvent environment and χ approaches 0.5 you see no obvious
problem, so you change the solvent a little more … until suddenly you
get flocculation.

FST 233
 With χ > 0.5 the relevant term changes from repulsive to attractive and the
formulation is ruined. This sudden switch from stable to unstable has surprised
many formulators who have only made a “small change” to fix some other
issue. For those who know about how Dissolution_Hansen Solubility
Parameters relates to the Flory-Huggins χ value, any formulation changes can
be checked to see if they are likely to tip things into the wrong part of solubility space.

Although the basic idea works well, such a simplistic approach cannot cope with the issue of extra polymer
in the solution – something we frequently need to provide. This can give rise to extra stabilization or to
destabilization via depletion or bridging flocculation. Although there are theories for depletion flocculation
([Link] they aren’t really of any use.

Here we have Bridging Flocculation, which is surprisingly hard to explain

compared to depletion flocculation, especially because beyond a certain polymer
concentration, stability is re-established. However, the powerful
Dispersions_Scheutjens-Fleer technique copes with stabilization and de-
stabilization phenomena and should be part of the toolkit of anyone working with particles and polymers.

The fact that a Scheutjens-Fleer app exists is an example of academia at its best. When I
“discovered” it, I immediately saw its power and usefulness so wanted to write an app … then
found that the theory was way beyond my capabilities. An email correspondence with Prof Frans
Leermakers resulted in his kind permission to use his powerful SF Box engine on my server, with
my front end simply sending parameters and plotting the results.

234 FST
 Dispersions ODC

Dispersions ODC
Links
Lipstick, Mascara, Solvent-Based Paint, Inkjet Inks

Many formulators try a few dispersants in their full formulation and go with whichever works. A
better approach is to find an Optimal Dispersant Concentration, which means both an optimal
dispersant and an optimally low concentration required.

Optimal Dispersant Concentration

We add our pigment/particles to our solvent/water and (ideally in real time)
measure the high viscosity due to the pigments being ill dispersed. We then
slowly add a dispersant, measuring the fall in viscosity. After a while, the
viscosity increases as the excess dispersant is probably a polymer, adding its
own viscosity.

For the blue dispersant, the fall in viscosity wasn’t large, and it needed a lot
of dispersant to reach a minimum viscosity with the maximum of
dispersion. That its ODC.

For the red dispersant, the fall in viscosity is large and you need rather little to get a great dispersion. Its
ODC is smaller and the results are better, so it’s the dispersant of choice.

Of course these curves are idealised and measuring the viscosity in real time is tricky, but by finding the
closest possible equivalent in one’s own lab, it’s an effective way to screen multiple dispersants and find the
ODC values (concentration and viscosity reduction) that meet your needs, along with cost, greenness etc.

The idea of an ODC was new to a major corporation working on a challenging pigment-based
product using a range of pigments. They were using a general dispersant and had not realised that a
dispersant optimal for each pigment would give them better performance without the compromises
they were enduring. Having reviewed the amount of work needed to find the ODC and deciding
that they didn’t have the time/resources to do it internally or to contract outside, they decided to
see if they could purchase their pigments already optimised for their specific system. They found a
supplier and not only did the formulations work better, it was cheaper to buy in easy-dispersing
pigments rather than spending time and energy forcing their older ones into the formulation.

Optimal by calculation
With some basic thinking, we can work out an ODC for a rational choice of dispersant. Rational means that
it has a head that likes to be strongly attached to the particle and a tail that will stick out into the solvent and
provide the desired steric stabilization.
[Link]

Here we are adding a dispersant with a head area of 90 Ų and MW of 450 to a 1000 nm particle. We’ve
chosen 2.5 mg of the dispersant per 1 g of particle. We can calculate the surface area of the particle (4πr²)
and the total head area of the dispersant (g/Mwt to get moles, then x Avogadro x Area) so we find that we
have a 100% coverage and a stable dispersion.

Of course we generally don’t know the exact MW of our dispersant (suppliers provide less information than
we might like) nor do we know the effective head area. But we can either choose to rely totally on the
suppliers’ promises that their dispersants are wonderful, or try to change the debate by using calculations
such as this.

Why don’t suppliers give us the information we need? Suppliers often claim that customers don’t
ask for the values, so there’s no need to supply them. Customers often claim that it’s not worth
asking because suppliers never provide them. My view is that the core values could easily be
provided, to mutual scientific benefit. The problem is that corporations prefer to sell magic rather
than science. I’m not making this up. I’ve had heated arguments with marketing departments of
megacorps who are terrified of their customers finding out that their wonder product is no different
from everyone else’s.

What is not included in the calculation is the chances of the dispersant wanting to stay on the particle. This
requires some basic understanding of solubility and χ parameters, both captured in the
Dispersions_Scheutjens-Fleer approach to improving our understanding of dispersants.

If you’ve never heard of Scheutjens-Fleer theory, that’s not a surprise. Nor had I. The colloid science
community has focussed on relatively useless theories while largely ignoring the one theory that is both
powerful and usable. They had the excuse that the theory was hard to use – but that excuse has vanished
now there’s an app version plus a more powerful variant within the HSPiP software package.

236 FST
 Dispersions PVC and CPVC

Dispersions PVC and CPVC

Links
Lipstick, Mascara, Solvent-Based Paint

The Pigment Volume Concentration, PVC, and the Critical PVC are important formulation factors.
If you exceed the CPVC the coating quality gets worse. Because they are volume dependent but
pigments tend to get measured by weight you need to know the densities ρ of the pigments and the
medium.

You also need to know the Oil Absorption, OA, which is the amount of a standard oil that the dry
pigment can absorb. This is often a value given by the supplier.

[Link]

We often need a lot of pigment in our coating or ink once all the solvent used to coat/print has disappeared.
The problem is that pigment particles, when close packed, have lots of empty space between them. If we do
not fill that space with binder then the properties are compromised: gloss, rub resistance, adhesion etc. will
be lower than desired.

For each pigment there is a Critical Pigment Volume Concentration, CPVC, above which there is not
enough binder present to fill all the voids (assuming that our binder is able to flow into and fill all available
spaces). Pigment suppliers give us the density, ρ of the pigment and a value which tells us the amount of
void space, expressed as the Oil Absorption, OA, which is the amount of a standard oil that the dry pigment
can absorb. Because this oil has a density of 0.935 the CPVC is calculated as:
 1
CPVC = OA
1 + ρ 93.5

Now that we know the CPVC, all we need to know is our actual pigment volume concentration, the PVC.
Typically we have a weight of dry pigment, then one or two extra "components", C1 and C2 (binders,
fillers etc.) which are typically supplied as solutions. To get the PVC we need to convert our weight of
pigment into volume (that's easy, it's g/ρ), then we have to take our g of component solution, the density of
the solution, the % solids of the solution and the density of the solvent, from which the volume of the
component can be calculated. Once we know all the volumes, the PVC is easy to calculate. If it exceeds the
CPVC then it is shown up with a red background.

The complication for the app is that with 3 possible components, you have to adjust them individually so
that the Total is 100g. If you accidentally miss getting close to 100%, the Total is shown up with a red
background.

Various properties change with PVC and you can determine the CPVC from seeing where there is a sudden
change. In the app the density is used to illustrate how you might go about this. As the CPVC is exceeded,
the density decreases because of the air.

My acquaintance with CPVC is rather recent. It seemed laughable to me that such an important
aspect of formulation depends on “oil absorption” values measured by ancient techniques.
However, I couldn’t find anything better, and everyone seems to be happy using them. If you (a)
find CPVC calculations laughable and (b) have a better alternative, please let me know.

A handy Wt%, Vol% app

Although it’s trivial to convert weights of different particles, ingredients and solvents into wt% and vol%,
users asked for a handy app to do just that, not focussing on CPVC. So here it is:

[Link]

For user convenience, weights are in arbitrary units going from 0 to 100. As long as all components are
using the same units and weight scale, the app calculates the appropriate % values from whatever the total
weight happens to be (so it doesn’t have to be 100). The densities of some well-known standard pigments/
particles/fillers are provided for convenience.

238 FST
 Dispersions Rheology (High shear)

Dispersions Rheology (High shear)

Links
Sun Screens, Lipstick, Mascara,Toothpastes, Water-based Adhesive

With serious concentrations of particles we see a lot of complex rheological behaviours, not
captured in the idealised Dispersions_Rheology (Low shear) chapter. Unfortunately there are lots
of ways that particles influence rheology so the app, and the chapter, are relatively complex.

Basic high shear properties

[Link]

Ignoring all other settings, we see that the relative viscosity for a volume fraction φ = 0.28 is constant over
the wide shear rate. For higher volume fractions, the viscosity gets much higher at low shear then decreases
rapidly with shear. The formulae behind these curves are complex and you can read them in the app. But
the take-home message is that although it is hard to handle high φ formulations at low shear, they aren’t a
problem at higher shear. The particles tend to line up like cars on a motorway and can flow along with little
problem.

The viscosity approaches infinite at a critical φ* which we can set via φm to 0.63 in these screen shots.
Why do we use φ* instead of φm? That will become clear later.

Aspect ratio
Now let’s make our particles somewhat elliptical, with an aspect ratio (length/thickness) of 4.3.
The low-shear viscosity (you can use your mouse to read off the values) almost doubles. That’s because the
percolation fraction, φc has decreased. What does this mean? As shown in [Link]
practical-coatings/[Link], the chances of there being a continuous random network from one side
of a container to another is near zero till φ reaches the percolation threshold where there’s now a high
chance of such a chain being formed. Above the percolation threshold, viscosity is higher because the
stresses at one side can be transmitted across to the other side via the percolation chain. Shear breaks up
these chains, so viscosity is now shear rate dependent.

The percolation threshold depends strongly on the aspect ratio. For a sphere it’s 28%, for our 4.3 aspect
ratio it’s 18%. So now the 28% formulation is no longer constant viscosity because percolation effects are
strong.

The key lesson is that relatively small changes in the shape of your particles can have a significant effect on
the rheology.

I was once asked to solve a weird particle formulation problem. I was using the app live in the
discussions … and nothing was making sense. In desperation I changed the Aspect Ratio slider and
the problem appeared in the graph data. There were some moments of silence. “Ah, we forgot to
tell you that this specific particle is elliptical while all the others are round, we’d not thought it
worth mentioning”. There are plenty of problems that can’t be solved by playing with sliders in an
app, but it’s remarkable how often an app acts as a catalyst for finding a good solution.

Bimodal distribution
As you reach a high φ you are approaching the critical packing fraction where you can’t fit any more
particles. That’s why viscosities take off strongly above 0.5. The close-packed spheres have lots of space
between them which can be filled by smaller particles with no real impact on viscosity. So we often choose
bimodal distributions (you can explore these via [Link]
[Link]) to be able to pack in more particles:

240 FST
 Dispersions Rheology (High shear)

If we have a 5:1 size ratio and a fraction, k = 0.33, of the smaller particles then the critical packing fraction
φ* rises from 0.63 to 0.68. The low shear viscosity of the 0.54 sample is now 1/3 of the original.

Real particles
Against our intuitions, the above calculations do not involve particle size. We know that in the real world,
smaller particles tend to give more problems of self-association. There seems to be no app-able formula for
this, so we have to use the two sliders N and Fractal Dimension d to get some idea of what might be
causing higher viscosities.

We have the same spherical particles, but now φ* has changed from φm = 0.63 to 0.59. This means that at,
say, φ = 0.54 the particles are much closer to the critical parameter so the viscosity is more than 10x larger.

This has happened because on average there are 1.8 particles in some sort of association. That is a vague
term because it also requires the Fractal Dimension. Perfectly random particles in 3D space have a fractal
dimension of 3. If somehow aligned and confined to a plane, their dimension would be 2. Some modest
alignment takes the dimension down somewhat from 3, in this case to 2.7. There’s clearly some clustering
to reduce the dimensionality; not a lot, but enough to change φm to this new φ* given by:

FST 241
 3
φ * = φmN 1 − d
It is frustrating that this formula is likely to contain a lot of truth while being mostly unusable as we know
neither N nor d. Maybe someone will come up with something more usable. Until then, we just have to
make impressionistic use of it and/or use any nugget of extra information available from our formulation to
better understand what is going on.

Other effects
There are other options discussed in the app. They at least indicate other possibilities for observed
behaviour and if your own formulation shows some peculiarities you can return to the app to see if the
relevant phenomena have some explanation.

Yield stress
By definition, yields stress (see Flow_Yield Stress) is nothing to do with high shear. But it was convenient
to include it in this app because the YODEL model from Flatt and Bowen on which it is based uses mostly
the same parameters. The yield stress τ is given by

2
E φ (φ − φ c )
τ=k ()
r φ (φ − φ )
m m

E is the attractive energy between particles, hopefully small, and r is the radius of the particle, showing, as
intuition suggests, that smaller particles will give a higher yield stress.

Again it is unlikely that you will know E, but at least you get some sense of what influences it. Because φc
is part of the equation, clearly higher aspect ratios (lower φc) will give higher yield stresses.

It is important in academia to have a good acronym for your fancy theory. Is it a coincidence that
the acronym YODEL comes from Swiss researchers? I had a chance to ask them directly and, no,
it’s not a coincidence…

242 FST
 Dispersions Rheology (Low shear)

Dispersions Rheology (Low shear)

Links
Sun Screens, Lipstick, Mascara, Water-based Adhesive, Inkjet Inks

The low shear viscosity behaviour of particle formulations is described by a number of theories
which give similar results up to a volume fraction φ of ~ 0.5. As the particles get very crowded, the
theories diverge, but at that stage it is likely that your particles will be showing other behaviours,
described in the Dispersions_Rheology (High shear) chapter.

It’s relative

[Link]

In this example, the solution without particles had a viscosity of 10 cP. At a volume fraction of 0.3 the
viscosity has increased to ~ 35 cP because the particles bashing into each other dissipate energy. At 0.5 the
viscosity is over 100 cP, a 10x increase.

At first this seems to indicate that if our particles are spherical and well-behaved, we don’t have to worry
about viscous effects till we get closer to the close packing limits around 0.6. For most of us, the difference
between 10 cP and 100 cP is not so worrying.

But now increase the solution viscosity to 20 cP. That’s only an extra 10 cP so you might think that the
viscosity at φ = 0.5 will increase from 100 to 110 cP. But that’s not what happens. Try it yourself. You see
that it goes from 100 to 200 cP. The effects of the particles are relative, so although 10 to 20 cP is only a 10
cP increase, it’s a doubling of that low viscosity. So the particle viscosity is also doubled.

The fact that an insignificant change to a base formulation viscosity can lead to a large change in the final
viscosity is both bad and good news. It’s bad because some necessary new additive to the starting
formulation can give an insignificant increase to the base formulation which is dramatic for the final
formulation. It’s good because if you are hitting problems with your formulation and can’t go to a lower φ,
then some modest consideration of your starting viscosity might solve the problem.

I once got asked at short notice to solve a formulation crisis involving a particulate coating. After
much effort it became clear that the formulation was failing because its viscosity was slightly too
high. Unfortunately, simple fixes such as altering the particle concentration were unacceptable. I’d
only recently programmed the app and had never properly believed that a 1 cP shift from 10 cP to
9 cP in the base viscosity could give a 10% reduction in the final formulation. It turned out to be
easy to make this reduction, keeping everything else constant … and the problem was fixed.

The formulae
The calculated viscosity is always relative to η0, the viscosity of the bulk phase.

The app covers particles from oil emulsions through soft particles to hard particles. The ratio of viscosity of
the particle to the medium is k which for solid particles is “very large” which means > 10000 which is all
that’s required for the formulae.

You can read about the monstrous Yaron, Gal-Or equation and the complexities of the Pal formula in the
app. The very low φ formula from Taylor is included not because it’s of use but because it’s often quoted. It
is basically the Einstein formula with a correction for k which, for large k comes out as 2.5φ:

( 5k + 2
η = η0 1 + φ 2(k + 1) )
The most common formula is Dougherty-Krieger:

φ −2.5φm
η = η0 1 −( φm )
The value of φm, the critical packing fraction is assumed to be 0.68 in the app. Others say it should be 0.58
which gives a much more dramatic take-off at higher φ values. The point is to not worry too much about
the formula but to get the general idea that bad things will start to happen above φ = 0.5, which can be
purely due to the formula but more likely due to particles wanting to flocculate at these high values, unless
you apply high-enough shear to pull them apart. That’s why we have the Dispersions_Rheology (High
shear) chapter.

The number quoted as a critical packing fraction changes in different FST chapters. That’s not
simple carelessness. Different domains confidently quote different φm values and I use values
common to that domain. The different values represent close packing, random packing, random
close packing etc.

Particle size independence

It would make sense that smaller particles, at the same volume fraction, would show higher viscosities
because they interact more strongly. The absence of a radius input therefore seems shocking. But these

244 FST
 Dispersions Rheology (Low shear)

theories are all neutral about particle size.

To put it another way, if you see a particle size dependence in your formulations (and almost certainly you
will) you know that other things are going on. The point of the app is to set a baseline expectation. If your
formulation is close to this then you have the good news that your particles are well-dispersed.
Congratulations. If the formulation shows a big divergence with φ and/or particle size then you
automatically know that particle-particle interactions are significant – so you can focus your formulation
energies on improving the dispersant system.

FST 245
Dispersions Scheutjens-Fleer
Links

The most powerful and useful dispersion theory describing how particles and polymers interact
was developed in the 1970s by Scheutjens and Fleer. But outside academia it has hardly been used.
Now it’s available in an app, there’s no excuse not to use it.

Deep solubility theory

It is unfashionable to apply solubility theory to particles (see Dispersions_Solubility) but the powerful
capabilities of Scheutjens-Fleer theory show that fashion is wrong.

The theory starts with the well-known fact (Dissolution_Flory-Huggins) that the Flory-Huggins χ
parameter based on simple lattice theory ([Link]
tells us that a polymer in a good solvent (χ < 0.5) is extended while in a bad solvent (χ > 0.5) is curled up
on itself because the polymer-polymer segment interactions and solvent-solvent interactons are stronger
than polymer-solvent interactions.

Using exactly the same logic you can get a χ parameter for particle-solvent interactions and for particle-
polymer-solvent interactions.

Now put your particles, solvent and polymer into a virtual box and make some guess about where they
might be. For example, the guess might have the polymer as mostly free, mostly as trains, mostly as loops
or mostly as tails:

Via lattice theory, all the interactions can be calculated via the
various χ parameters. This gives an energy value for this
specific configuration. Now make some guesses about how to
move some of the polymer chains. Re-do the calculation and if
the energy is lower, carry on with those sorts of moves … and
so on till you get a Self-consistent Field result (SF for self-
consistent field theory).

Now you know what the polymer is doing. Let’s see the results:
 Dispersions Scheutjens-Fleer

[Link]

What we see is that most of the polymer (it is 100 “lattice units” long, which is about 30nm) is concentrated
in the first 2 lattice units (0.6nm) near the surface. The blue line shows that these are mostly trains and
loops. The tails, the things we usually draw as dominating the situation, just about poke out to 6 units,
1.8nm. Finally, towards the right-hand side of the graph we see the free polymer.

Now replace this polymer with a simple di-block, 50 units each of A and B, so the same length as the
previous system:

The A block has a high χ with respect to the solvent, so, not surprisingly, we find it (blue curve) near the
particle. The B block (red curve) extends into the solvent. There is basically no free polymer, which means

FST 247
this will be an efficient dispersing agent.

Unlike the other apps, the engine powering these calculations is sitting on the server. It is called SFBox,
developed over the years at U Wageningen and generously supplied by Prof F Leermakers.

Some top scientists choose not to waste their time talking to people like me who can’t just read
their published papers and translate them into glorious apps. There are, however, a rare few who
have great theories combined with the patience/understanding to realise that not everyone knows
as much about their area as they do. A number of the apps used in FST exist only because of such
exceptional academics. The SF app is one of them.

You can explore other options, play with χ values and build your intuition about particles and polymers.

Particle stability
The full power of SF theory comes from calculation of particle-particle interactions involving polymers.
You can set up scenarios such as the above, calculate, first, the configuration of polymers around the
particles, then calculate whether those configurations are stable against flocculation.

Unfortunately those calculations are too complex for a web app so are only available conveniently within
the HSPiP software package.

Using the same Di-block conditions, it is obvious from the interparticle calculation shown here that there is
a very high barrier extending out to 15 lattice units. This would be an excellent dispersing agent.

248 FST
 Dispersions Scheutjens-Fleer

Change the χ value for polymer B to greater than 0.5 and we see that the
particles are going to strongly flocculate. The B polymer chains now like
each other more than they like the solvent, so they give the bridging
flocculation mentioned in the Dispersions_DLVO chapter. This is a more
powerful way to look at these problems than the simplistic theories we
have previously been offered. You can explore not just the χ parameter but
the effects of relative sizes of the A and B blocks, the effects of free
polymers, compare simple di-blocks to brush polymer equivalents, and
much more.

The future for SF theory

Now that SF theory is available both as an app and as a powerful feature of HSPiP there are few excuses
for staying with handwaving “dispersion” or “colloid” science arguments. SF theory isn’t easy, but the
basic ideas of interactions via sets of χ parameters are not hard to grasp. The more of us who use it
routinely, the more we will demand relevant parameters from our suppliers, the academic community and
ourselves. It is very much a theory that should be in the toolkit of many formulators.

I had been entirely unaware of SF theory and found it by accident. It is tragic that everyone knows
DLVO, which is so limited, and so few know SF theory which is so powerful. An obvious theme of
the FST is that much of what we do in formulation work is based on theories that are totally
inadequate, and which should be kicked out to be replaced by better ones.

FST 249
Dispersions Settling
Links
Lipstick, Mascara, Water-based Adhesive, Microencapsulation

If your particles are too large (perhaps from agglomeration) and are too dense for a liquid of too
low a viscosity, they will settle, with a velocity governed by Stokes law, with some suitable
modifications.

Once they settle they might be easily re-dispersed or might, because they are close together, cohere
permanently. Either way, avoiding settling is a good idea.

Stokes law

[Link]

The velocity v at which particles settle depends on the radius r, the viscosity η and the density difference
between particle and water, Δρ. In addition, the Richardson & Zaki dependence on volume fraction φ is
included:

2gr2Δρ 5.65
v= 9η (1 − φ)

The time taken to settle a height h is given by:

h
t= v

If you are using a centrifuge of radius R spinning at an RPM of ω then you substitute g in the equation by
grel:

grel = g(1 + 0.00118Rω)

Although it is incorrect to say that some particles are too small to settle because of Brownian motion
around kT, there is a useful heuristic which says that the particles won’t be observed to settle if the Gravity
Number, Gr < 1:
 Dispersions Settling

0.768πr44Δρg
Gr = 3kT

The reason it is incorrect to talk of Brownian inhibition of settling is covered in the next section.

One final surprise about settling. If liquid of height H is tilted in a tube of width b by an angle θ then the
Boycott velocity (he discovered the effect in the 1920’s) is given by:

( H
vb = v 1 + b sinθ )
This significantly speeds up settling, as Boycott discovered when attempting to settle out red blood cells in
small test tubes. For example, a 5 cm liquid in a tube 1 cm wide tilted at 30° increases settling by 3.5x, as
you can confirm with the sliders in the app.

Like most people, I’d never heard of the Boycott effect. After a post about an improvement to a
settling app, someone commented that it would be good to add the Boycott effect. It was great fun
to learn of something completely new, to find the theory and to share it with many others for whom
it was also new.

Gravitational Sedimentation
Everything, not just particles but molecules and polymers, can settle over time if there is a density
difference. We know this from the common practice of putting biological samples into a centrifuge (large
DNA or proteins) or ultracentrifuge (e.g. smaller proteins).

Instead of aiming for “settled” or “not settled” which is the typical focus of Stokes sedimentation, we can
look for a concentration gradient of settled material.

If we just use gravity, then there is an algorithm from the 1920’s by Mason and Weaver which allows us to
calculate the concentration profile from top to bottom of the tube. Solving the algorithm turns out to be
very hard, but the Werts group at Ecole normale supérieure de Rennes found a way to implement the
algorithm across all cases of densities and sizes.

FST 251
[Link]

The image shows the relative concentration profile (yellow) going from small (~0.35x) near the top to high
(~2.5x) near the bottom with a middle section that retains the original concentration. This happens to be
after 4.2 days and this sample would take 62 days to reach the equilibrium profile shown in blue.

There are a few reasons to be interested in this app:

• The myth that “dispersions settle but solutions don’t settle” is firmly busted.
• There’s a difference between particles settling slowly like this and particles flocculating and then
(because they are much bigger) crashing out.
• The “Too small to settle” heuristic from the previous section is wrong in theory but OK in
practice. These settling profiles over days require samples to be kept undisturbed and under good
temperature control as convection currents could easily undo the settling.
• You can visualise the Boltzmann distribution!

Yes, the Wert group, who kindly provided this image,

showed that the concentration gradient of these gold
nanoparticles at equilibrium matches the Boltzmann
distribution and because the particles are strongly
coloured, you can easily see the distribution.

252 FST
 Dispersions Solubility

Dispersions Solubility
Links

Dispersions can’t be “soluble” – they’re dispersions! But because there is no agreed definition of
“dispersion” we can’t use “dispersion theory”. However, there is an agreed statistical
thermodynamics definition of “soluble” and nanoparticles happily fit that definition. And it turns
out that if you use a pragmatic solubility theory, you can do all sorts of smart things with particle
formulations.

Particles can be soluble

Sure, many particles are insoluble in many solvents, but the same can be said of many crystalline organic
chemicals, of polymers and biomolecules. Just as you need the right solvent for those molecules we think
of as soluble, you need the right solvent for particles.

Take samples containing the same small amount of particles and put them into 10 different solvents in
centrifuge tubes. After a bit of dispersion, centrifuge the tubes. In some solvents, the particles will rapidly
go to the bottom. In others, they are slow to sediment. Correcting for viscosity and relative density, you can
produce a set of relative sedimentation times, RST. Those with a long RST are “happy” in the solvent.
Those with a short RST are “unhappy”.

Do the same thing with the particles in NMR tubes and place them not into a fancy
NMR spectrometer, but a table-top device (image courtesy of Mageleka) to
measure T1/T2 relaxation rates.

Those solvents that interact strongly with the surface, and which are “good”
solvents show a big reduction in relaxation times compared to “bad” solvents
which only weakly interact. These are not subtle effects requiring complex
analyses – you get measurements in seconds.

Or just shake up your formulation with 10 different solvents and use your own judgement to see if the
particles are “happy” or “unhappy”.

Whichever technique you use, the results are a set of good and bad solvents. Each technique has its
strengths and weaknesses (and equipment costs), but they all lead to the same conclusion: that you can
apply solubility theory to the particles and, therefore, formulate intelligently.

Independently of these pragmatic efforts, fundamental work by Shimizu and coworkers demonstrated, via
assumption-free statistical thermodynamics, that “solubility” is indeed applicable to nanoparticles. Yes, for
obvious reasons they are less soluble than small molecules or medium-sized polymers, but they are still
soluble. There is no other scientifically coherent way to talk about their behaviour in solvents.

HSP of particles
The chapter on Dissolution_Hansen Solubility Parameters shows that you can characterise a polymer or
particle via the three HSP values, δD, dispersion, δP, polar and δH, H-bonding. You do this by measuring a
sphere that contains the good solvents and excludes the bad solvents. The best solvent for the polymer or
particle is at the centre of the sphere. The further from the centre, the worse the solvent. This Distance is a
measure of compatibility. You can calculate the Distance from any other solvent, polymer or particle and
therefore have a good idea of the compatibility or incompatibility.

Here, for example, is a sphere of a specific carbon black, CB. This used values of RST from a LUMiSizer
but similar results could have been obtained via NMR relaxation:

Its HSP are [17.2, 8.9, 13.1], meaning that it has quite high Dispersion
(typical for CBs) is medium Polar and quite high H-bonding. If you
happen to know that this batch of CB is great for making your Li
batteries and the next batch came in at [17.1, 9.3, 12.9] you could be
confident to use the next batch. But if it came in at [17.9, 3.2, 4.8]
(measured values for a “hydrophobic” CB) you would know not to use
it.

The correct HSP values will match nicely to the HSP of your chosen
solvents and the polymers and other ingredients you are using.
Sometimes you want to encourage phase separation during coating, so
you must ensure significant differences in the HSP of the phases you
need to separate.

Without these solubility-based measurements, it is extraordinarily hard to characterise a batch of CB (or

any other particle) in terms that interest you. How else can you know its compatibility with your solvents
and other ingredients?

You cannot get these insights via common analytical techniques such as AFM or SEM. Sorption techniques
such as IGC give information about particles surrounded by low concentrations of solvent in the inert
chromatographic gas; they cannot provide information about bulk solubility. Of course these techniques
provide valuable other information, but in terms of formulation, having direct solubility measurements is
important.

Not long after creating HSP for polymers, Dr Hansen was working in a paint company. He thought
it worth trying to apply his solubility theory to paint pigments. If it worked, he’d have a powerful
formulation tool, if it failed he’d learn something interesting about pigments. It worked. That was
in the 1970s. It is only in the past decade that formulators started to realise that “dispersion
science” offered little of practical value while solubility theory via HSP was a powerful
methodology that delivers results.

Surface dependence
The theory is entirely neutral about the surface of the particle. Are the HSP values dominated by the
general particle surface, specific “high energy sites” on the surface, polymers, contaminants, dispersants?
The theory doesn’t care. It pragmatically finds that this sample behaves with these solubility parameters, so
if you want to formulate successfully, use those parameters. If you think that a new batch of the particles is
different, just re-measure the good/bad solvents and calculate a new sphere.

254 FST
 Dispersions Solubility

When we add dispersants to the particles, what happens to solubility theory?

Again, the theory is neutral. If you deliberately change the particles via, say, a new dispersant, re-measuring
the HSP tells you a lot about the success or failure of your change to get you into the solubility/stability
zone you require.

The most powerful theory that takes into account the dispersants we tend to add to our particles is
Dispersions_Scheutjens-Fleer which has been around since the 1970s, but little used because it was hard to
implement. It is based on lattice theory, Flory-Huggins χ values and, therefore, solubility theory.

But particles are dispersed, not soluble!

Some people, and this might include you, refuse on principle to accept that particles can be analyzed via
solubility theory. Particles are dispersed, therefore one should use dispersion theories. But what theories are
those? Apart from Dispersions_DLVO which is really only useful in water (where HSP is weak), there is
very little usable dispersion theory out there. Plenty of handwaving and good intentions. But very little
usable theory. The conventional extension of DLVO to steric effects (described in the DLVO app) is OK,
but cannot cope with effects of free or added polymer. These effects, plus effects such as depletion
flocculation are better analyzed via Scheutjens-Fleer theory than via the relatively unhelpful Asakura &
Oosawa model which is often presented as a key element of dispersion or colloidal science:
[Link]

Happily, the use of HSP for nanoparticles has reached critical mass. There are now committees working on
ASTM-style standards for measuring the HSP of CB for battery manufacture, and “round robin” groups
exist comparing HSP values obtained by different techniques (e.g. visual tests, centrifuge tests, NMR
relaxation). These efforts to standardise tests and work out the advantages/disadvantages of the various
techniques are arriving just as nanoparticle suppliers are receiving demands from their customers for the
HSP of the particles.

So, yes, nanoparticles are soluble, and solubility theory is a powerful tool for nanoparticle formulators.

FST 255
Dispersions Zeta potential
Links
Water-based Adhesive, Emulsion Paint, Emulsion Polymers, Inkjet Inks

Classic DLVO theory Dispersions_DLVO tells us that in water, particles can be stabilized by
charge-charge repulsion. To know what these charges might be, we measure the zeta potential. As
we shall see, it’s a slippery concept and a wise rule is to never measure a zeta potential but to
measure a set of them.

Real charge and measured charge

[Link]

We have a particle, green, with a “real” charge of 61 mV, shown by the red ring. The first thing a charge
does is attract its counterion, blue, so the net charge must be less than 61mV. But that loose, blue ring
attracts red ions and so forth till we reach ζ, the boundary between charges fixed to the particle and the
general background solution. The measured voltage at this point is “the zeta potential”, in this case 44 mV
Along the way we pass through S, the Stern layer (52 mV) which some classify as the end of the original
double layer, though there is much confusion about the relation between S and ζ.

If you increase the molar concentration of background salts then the Debye length, k-1, decreases and the
potential away from the particle (including the zeta potential) decreases.

If you put these particles and their background salts into a zeta measuring device (one such device is called
a Zetasizer) then from the light scattering behaviour under an applied voltage gradient, “the” zeta potential
is measured. Exactly what it is and what it means is less important than the fact that its sign and its general
magnitude tells us a lot about the particle. The sign can typically switch from negative at high pH to
positive at low pH, passing through the isoelectric point of zero charge and, therefore, the most perilous
state in terms of charge stabilization. The magnitude can be roughly classed as “small” (less than 10 mV),
unlikely to show any charge stability, “medium” (up to 25 mV), with borderline stability and large (40 mV
is a comfortable value) for a “stable” system.
 Dispersions Zeta potential

Never measure a zeta potential

Because of the messy nature of “the” zeta potential, surprisingly small changes in the formulation can result
in surprisingly large changes in the potential. The “real” surface can change dramatically with small
changes to the surface chemistry. Different counterions might have very different interactions within the
Stern layer. Small levels of contaminants (such as cationic surfactants interacting with anionic surfaces) can
have devastating effects. And, of course, pH changes might take you close to the dangerous isoelectric
point.

That’s why measuring a zeta potential is so dangerous. The trick is to measure the potential over a range of
likely conditions and possible contaminants. You will find a safe formulation zone and also know what
things must be strongly avoided and therefore checked for before adding to the formulation.

We were using a dispersion stabilised by a large (negative) zeta potential; it had always worked
well. But we encountered some other problem and a special surfactant was suggested to fix the
problem. Like all surfactants it just had a name, say, XY95. Only small amounts were required, so
there were few risks of side-effects. On adding this tiny amount of XY95, the formulation crashed
out. Digging into the surfactant datasheet we found that it was cationic. Even the small amounts
that locked on to the anionic shell of the dispersion were enough to change the effective zeta
potential to near zero, and the formulation failed.

And if you can’t find a stable zeta potential zone, then swap your effort over to steric stabilization as
described in the Dispersions_DLVO chapter or even better in the chapter on Dispersions_Scheutjens-Fleer.

FST 257
Dissolution Crystallization
Links
Pharma Formulation

Crystallization is both vital and frustrating. It’s vital for obtaining pure compounds and frustrating
because things don’t crystallize the way we want, when we want.

It hasn’t helped that for decades people have tried to use CNT (Crystal Nucleation Theory) despite
everyone agreeing that it’s unfit for purpose. Some brief suggestions for a better approach are
included here.

Solubility versus Temperature

For easy, high efficiency, crystallization you want a nice solubility curve – a large solubility at a
temperature OK for handling your chosen solvent, and a low solubility at whatever temperature (say 20°C
or 0°C) you can conveniently crystallize and filter.

Via a Differential Scanning Calorimeter (DSC) it’s not hard to measure the MPt and ΔHf, the melting point
and enthalpy of fusion. If you use Dissolution_Ideal solubility theory you can then find how solubility in a
perfect solvent changes over your temperature range. If, via COSMO-RS or Dissolution_Hansen Solubility
Parameters you have an idea of the activity coefficient, γ, of the solute in any real solvent, then you can get
a basic estimate of the sort of curve you can expect:

[Link]

In this example, with a relatively low MPt = 100°C, then over the range 10 - 60°C there’s a 5:1 solubility
ratio, but with a relatively large quantity of 200 mg/ml remaining in solution.
 Dissolution Crystallization

If you change the MPt to 200°C then you get only a 2:1 ratio, with 150 mg/ml remaining.

This information doesn’t tell you if you can get good crystals – instead it’s giving you an idea of the
viability of the crystallization process given your molecule and the sorts of solvents (with their respective γ
values) and temperature ranges that you might be prepared to use.

Real solubilities depend on more than the ideal solubility effect. The causes are subtle and it needs
powerful tools such as COSMO-RS to model them. If you don’t have access to such a package, use the
ideal solubility calculations as a starting point.

Zones

Forgetting about units and focussing on principles, in this graphic we see that as T decreases from 40° to
33° you reach the solubility limit of this molecule. If you stayed there all day, nothing would happen.
Indeed if you went to slightly lower you’d still be in the “Dead” zone and nothing would happen.

If, for whatever reason, some seeds (junk, old crystals, scratches on the walls …) are present then at 30°
you might start to see some crystals – this is the Secondary Nucleation Threshold, SNT.

If everything were perfectly clean you might have to go to 27° to get anything to happen – you’d been
through the Metastable Zone Width of 6°.

If you were impatient and cooled rapidly to 25° you might just get some oil to appear.

This story is highly idealized and contains some confused and confusing ideas, but it’s the standard story so
we have to tell it.

Metastability
Metastability leading to sudden (Spinodal) oiling out can be described via a straightforward theory:

FST 259
[Link]

You define a free energy landscape with 4 sliders (P1…P4) and you slide X to learn what the free energy
landscape means. The app is of no practical use – it’s there for whoever wants to understand that
metastability and oiling out are rational, even though highly undesirable for those who just want nice
crystals.

What the app stresses, however, is that when something oils out, it’s not a puzzlingly liquid version of the
solute. What oils out is a fixed (low) concentration of solvent in solute, in a bath of fixed (low)
concentration of solute in solvent.

CNT
For those who want to learn about Crystal Nucleation Theory, the app [Link]
practical-solubility/[Link] describes it. Because (as it turns out) most practical crystallizations take place
via seeding, CNT is completely irrelevant. The parameters that are endlessly measured and cited help no
one. It’s much more important to understand and control the seeding phenomenon.

Crystallization Map
Far better, instead, to think in terms of a crystallization map where CNT is a relatively rare (it works for
sodium chloride!) phenomenon. The following 4 images are from [Link]
solubility/[Link] which contains a set of 12 diagrams that cover a wider range of
crystallization options

260 FST
 Dissolution Crystallization

CNT itself says that a minimum viable crystal must first form, then everything happens after that. Loose
CNT says that you can get loose clusters of molecules that can tip over into being crystals.

Heteronucleation encourages a loose cluster to form which in turn can change to a smaller crystal then
grow to a larger one.

Finally, those loose clusters can become bigger and bigger till you get the liquid-liquid phase separation of
a solvent-in-solute cluster in a sea of solute-in-solvent.

There are many reasons for preferring the cluster-based view over CNT.

• The assumptions behind CNT almost never apply

• The experiments to measure CNT values are tedious and unhelpful
• Clusters explain many otherwise weird phenomena such as sudden crystallization, selective
enantiomorphs, memory phenomena around the solubility limit
• Clusters are a natural part of Dissolution_Kirkwood-Buff which naturally explains multiple
solubility and solubilization phenomena
• Using cluster-sensitive techniques such as scattering will provide more molecular insight than
CNT experiments.

FST 261
Polymorphs

[Link]

It is annoyingly the case that many molecules can crystallize out as different crystalline shapes/sizes called
polymorphs. Each polymorph has its own MPt and ΔHf so they have different solubility curves, as shown
in the app.

Which form crystallizes out is a complex mix of solubility curves and metastable zone widths.
Disentangling these effects can be challenging, so you need an app to make sense of what can happen:

[Link]

262 FST
 Dissolution Crystallization

In this example, if you cool to the relative temperature shown in red, you are likely to get polymorph I. But
depending on the relative “skew” of the curves and the widths of the metastable zones, different outcomes
are possible. Use the app to explore the other possibilities.

The statement above that CNT is usually irrelevant because crystallization usually happens via seeds is
very much supported by the (true) stories of rogue polymorphs. A pharma company which has patented and
certified one specific polymorph which reliably crystallizes in their facilities world wide … finds one day
that a different polymorph appears at one site which can then never produce the previous one. Through
inter-site travel, the new polymorph spreads and the company is forced to re-register the new polymorph
for safety and efficacy – at vast costs. If a competitor happens to have patented this newer polymorph then
life is very difficult. It is hard to know how polymorph seeds can travel and disrupt old processes, but it is
impossible to explain by those who insist on using CNT.

FST 263
Dissolution Dissolution Kinetics
Links
Surface Cleaning, Pharma Formulation

Getting stuff to dissolve when you want, as fast as you want, can be frustrating. Here we cover
some scenarios for which reasonable models exist. Take what you can from them to help with your
own dissolution challenges.

Tablets and crystals

Dissolution of solids is conveniently modelled using Noyes-Whitney theory, that has been around for over a
century. The app is for the specific case of a tablet, but the general principles are the same for a crystal.

[Link]

We have a tablet of radius 5mm, thickness 1mm. At a density of 0.8 g/cc that’s 63mg. We are dissolving it
in 100ml of water so its maximum concentration is 0.63 mg/ml, well below the saturated limit, Csat, of 10
mg/ml. The solute has a diffusion coefficient of 5e-6 cm²/s (you can estimate it from the MW and viscosity
of the solution) and we follow the process for 180 min where the concentration levels out at the 0.63 mg/ml
limit – the tablet has fully dissolved.

Everything is known in advance except for the thickness of the Diffusion Layer between the saturated
solution at the surface of the tablet and the bulk solution. Something like 50µm is a reasonable starting
point for an estimate. In reality you find h by fitting the experimental dissolution curve.
 Dissolution Dissolution Kinetics

The theory is easy. For a surface area S, the rate of change of concentration, C, with time, t, depends on D,
S, h and Csat:

δC DS
δt = h (Csat − C)
This makes intuitive sense. Diffusion (per unit area) is simply (see Diffusion_Basic Diffusion) the diffusion
coefficient times the concentration gradient which is the difference between the concentration at the surface
and the bulk, divided by the thickness of the diffusion layer. The larger S, the greater the area for diffusions
and the faster the rate.

But, of course, S decreases as the tablet dissolves and C increases; so for both reasons the rate of
dissolution decreases over time. The app takes care of the details of S changing and C increasing.

Although in this example the dissolution ends long before the saturated concentration, the solubility of the
solute makes a big difference to the rate; try changing Csat to see the effect. This is because of the
assumption that the concentration at the surface is Csat so a larger value means a larger concentration
gradient.

Conventionally, h is controlled by stirring. As a rule of thumb, if you increase the stirring rate by a factor of
2, h decreases by a factor of √2.

Dissolving a powder
We all know that if you add a powder (especially in the kitchen) to a solvent (water) with reasonable
stirring, it can dissolve nicely. But if you add the same amount of solvent to the powder you can end up
with a sticky mess that refuses to dissolve. In the world of technical polymers such as PVdF it is well-
known that if you try to dissolve the PVdF powder in a good solvent (such as NMP) it is almost impossible
– you get a sticky mess. If you disperse it first in a poor solvent (such as a ketone) and then add the good
solvent, it’s easy to dissolve.

If your powder, of radius R, doesn’t form lumps, if the polymer is nicely soluble, if diffusion into the
polymer is fast and if the polymer is not too entangled, then life is easy. Speed of dissolution is proportional
to 1/R and to RPM, the rate of stirring.

From those simple rules, it seems obvious that finer powders will dissolve faster. But there’s a problem.
The gap between powder particles decreases as R decreases. Suppose that on contact with the solvent the
surface of the powder absorbs the solvent and expands rapidly to form a gel. If the particles are closer than
the swelling expansion then they almost instantly create a lump held together by the gelled state. If we
assume that the lump has a dimension on the mm scale rather than the 0.1mm scale then we instantly have a
10x longer dissolution time. Only if dissolution of the gel is faster than its expansion can we avoid this
problem.

We can explore this in the app:

FST 265
[Link]

Here we have a 150μm powder which is a polymer that’s somewhat entangled as MW/Mc = 2, where Mc is
the Critical Entanglement MW, i.e. the MW above which the polymer chains are guaranteed to be tangled
and therefore, for example, increase viscosity and decrease dissolution rates (it’s hard to pull away from the
tangles) (see Adhesion_Entanglement). You see a fairly rapid dissolution in the first 10 min but then things
slow down. That’s because we have a medium Gel Speed, meaning that the polymer swells quite quickly
and you end up with a fraction of free powder, fPowder = 0.33 which dissolves quickly and the remaining
0.67 fraction are lumps > 1mm so they dissolve slowly. The factor k is calculated from the RPM (higher k
for higher value) and MW/Mc (lower k for higher value) and confirms what you see in the graph, the
higher the value of k the faster the dissolution.

What is interesting is that although smaller powder particles dissolve more quickly (more relative surface
area), if they can form rapid gels then grinding them to a smaller size can increase dissolution times
because the gels form faster than the dissolution rate.

How do you know your “Gel Speed” and MW/Mc? The papers quoted in the app discuss various ways to
calculated Gel Speed and they assume that you know both your MW and Mc. Sadly, most of the time most
of us don’t known Mc, which is largely a fault of the suppliers of our polymers who could easily measure
the values of the products they sell.

Those familiar with entanglement theory will know of “reptation”, the snake-like movement of polymer
chains trapped in the tangles. We might imagine that the rate of pulling away the polymer from the powder
surface would depend on the reptation timescale but interestingly it’s more about the timescales of polymer
chains coming out of tubes, which can be very long.

There is one well-known trick for solving the gelling problem. If you can add a non-polymeric co-powder
in sufficient quantity and small size to cover a significant fraction of the polymer particles, then they
interfere with the instant gelation, dissolve easily and allow the polymer to dissolve without problems.
Those who like sweet cocoa drinks know this – the cocoa powder is covered with fine sugar crystals.
Without the sugar, the cocoa becomes a gelled mess.

266 FST
 Dissolution Dissolution Kinetics

Polymer dissolution
Now we have a slab or film of polymer. To get it to dissolve we must first ensure that it’s soluble in the
solvent! A good HSP match Dissolution_Hansen Solubility Parameters ensures this.

Next the solvent has to diffuse into the polymer. At first this will depend on “the” diffusion coefficient, D,
Diffusion_Basic Diffusion which varies strongly from polymer to polymer depending on the amount of
available “free volume” through which the solvent can diffuse. To be above Tg and to have low
crystallinity gives a large D with the opposite also being true.

But as solvent starts to penetrate, the amount of free volume increases so D increases. This is the
phenomenon of Diffusion_Concentration-Dependent Diffusion. Once we get a solvent concentration
(everything is expressed in volume fraction, φ) of something like φ=0.2 we are in a gel state with rapid
diffusion. The polymer swells strongly which slightly slows down diffusion (the solvent has longer to
travel) and greatly increases the difficulty of writing an app.

But even at a high φ we have generally tangled polymers in a viscous state so we haven’t swept it away –
it’s not yet “dissolved”. So the issues of MW and Mc, discussed above, are relevant here too. A high MW
polymer with a low critical entanglement MW will resist dissolution even though it’s well on its way to
being dissolved. So the speed of flow from our mixing system will have a significant impact on the
dissolution rate.

So far, a meaningful app has been difficult to provide. Instead, the general view of the process is something
like the following:

We start at (relative) time 0 at (relative) thickness 1. As

the solvent enters, the gel line, the point at which the
polymer transitions from glassy to gel-like, moves down
into the film. At the same time, the overall polymer
swells so the surface rises, in this case by 30%, before
dissolution erodes the swollen polymer. Once the gel
phase has gone, the final dissolution is fast.

The exact shapes of these curves depend on, for

example, entanglement. A more entangled polymer takes
much longer to move from the surface into the solution, so the relative swelling is larger. The velocity of
the solvent above the polymer has an effect, with a higher velocity removing swollen polymer sooner. The
solvent quality also makes a difference as a poorer solvent will struggle to swell the polymer quickly so
dissolution will be slower.

FST 267
Dissolution Dissolution Rheology
Links
Surface Cleaning, Pharma Formulation

Rheology affects dissolution, and dissolving stuff affects rheology. For the effects on dissolution,
see Dissolution_Dissolution Kinetics. Here we look at how dissolving stuff affects the rheology of
the final product.

Polymers and tangles

[Link]

In general we need high concentrations of our chosen polymer for our formulation and we invariably find
that the viscosity at our required concentration is just a bit too high. Or, more often, if we have a great
product with the right viscosity, someone wants to use less water or solvent and our viscosity becomes too
high.

A great way to reduce viscosity while keeping a high % solids is to use a lower MW. Some quick checks
might show that the performance is unchanged, but usually a higher MW was chosen for a good reason.
Polymers have an Mc, critical entanglement MW. Below Mc the polymer chains are too short to tangle.
This means that their viscosity in solution is lower (good) but that their mechanical performance (e.g. as an
adhesive) is worse. If your MW is, say, 10x Mc you have great mechanical properties but now the solution
is too viscous because of all the tangles.

Traditional polymer science considers super-low concentrations of polymers. They are nice theories but
 Dissolution Dissolution Rheology

irrelevant to our needs. For example, they like to work at concentrations, C, below C* which is when
average polymer chains start touching each other. They certainly don’t like to work above Ce which is
when the polymer starts to be entangled and where we are likely to be working. This means that we have
lots of good, but useless theory, and, above Ce we have very little. The app, therefore, is an attempt to
capture some of the key themes above Ce.

The shape of a polymer

[Link]

As you change the Flory-Huggins χ parameter from 0 (completely compatible) to 0.5 (borderline, the θ
solvent) the average polymer chain changes from nicely expanded to curled up in a ball. The app lets you
look at low concentrations so you can explore the shapes with any χ value. But because we need high
concentrations that are impossible when χ is large we need only confirm the obvious, that a happy polymer
is nicely spread out, which means that its viscosity will be higher. There’s nothing we can do about that,
given that we need the high concentrations for our application.

However, we need a dilute-polymer value at χ = 0.5, which is the radius of gyration, Rg, for a nominal 40K
polymer. We use it to estimate the effective size for our actual MW and at our actual χ value. Rg is routinely
measured by laser scattering. To make the estimate we need to know how many monomer units we have in
our 40K polymer.

Knowing Mc

It should be a law that all suppliers of polymers tell us the Mc value. They aren’t too hard to measure if you
have (as a supplier should have) a reasonable polymer lab and can measure, say, melt viscosities at different
MW.

FST 269
 Below Mc the viscosity increases linearly with MW,
above it goes to the power of 3.4, a strange number
with a good theoretical justification:

Other methods, such

as [Link]
rheology/L-M_Mc.php are available.

The Polymer Viscosity app gives a list of Mc values

for many typical polymers.

The polymer suppliers need to be persuaded to provide Mc values. It’s up to you, the formulation
community to demand that they provide you with the right science, so you can choose the right
product.

Putting it all together

As you increase concentration beyond Ce viscosity increases rapidly. If you decrease Mc and/or increase
MW, the viscosity increases faster as you have more tangles. As you increase polymer/solvent
compatibility by reducing χ, viscosity also increases as the polymer chains expand into the solvent.

Because you might not know Rg, an estimator is provided using input values you either know or can
estimate easily.

Shear dependence
Obviously, As discussed in Flow_Shear dependent, the viscosity will tend to decrease with shear rate:

[Link]

270 FST
 Dissolution Dissolution Rheology

It would be good if we could link the science of low-shear viscosity to the shear-dependent behaviour, but
given that no good theories exist for the low-shear viscosity of polymers, we have to make guesstimates.
The more tangles, the greater the fall in viscosity when they untangle, and the higher the shear rate at which
that will happen. That’s little help, but there seems to be nothing better available.

FST 271
Dissolution Flory-Huggins
Links
SkinCare, Adhesive, Pharma Formulation, Microencapsulation

The solubility of polymers is usually described via the surprisingly simple assumptions behind
Flory-Huggins theory. The χ parameter at the heart of the theory is then used across a wide range
of formulation issues. Why do we use such a simple theory? Because it works well-enough, and we
don’t have any practical alternative.

Lattice theory
If we start with the absurd idea that molecules in solution can be described as being points on a lattice, we
can use rather simple ideas to work out the entropy, ΔS, and enthalpy, ΔH, of the system from which we get
ΔG, the free energy. A toy system shows us how:

[Link]

In the screenshot we have a 50:50 mix of red and blue spheres. The energy of interaction of 1-1 is δ1 = 14
and for 2-2 it’s δ2 = 16. We aren’t worrying about units but these might be kJ/mole. Because 1 prefers 1 and
2 prefers 2, the interaction 1-2 is √ (δ1 δ2) ~ 15 is de-stabilizing.

Because the spheres are randomly mixed there is a significant entropy gain of -3.12. The enthalpic balance
of likes and dislikes is +2, so ΔG =TΔS + ΔH is negative, so the system is happy.

Now make the spheres more unlike each other. Reduce δ1 to 13 and increase δ2 to
17.1. The entropic gain does not change but the enthalpic loss is larger, so ΔG
becomes sufficiently positive (this is inexact, as mentioned below what matters is
the 2nd derivative…) that the system phase separates.

Moving away from this toy system to a polymer where the spheres on the lattice
 Dissolution Flory-Huggins

are joined together, and where we use a value χ which goes from 0 (polymer and solvent like each other as
much as they like themselves) to 0.5 (bordering on instability) to 1 (they very much dislike each other) we
can calculate the entropic and enthalpic terms via the Flory-Huggins equation where Δµ is the chemical
potential φ1 is the volume fraction of solvent, φ2 is the volume fraction of polymer and x is
MWPolymer/MWSolvent which is a large number:

[
Δµ = RT ln[φ1] + φ2 1 − [ 1
x ] + χφ22]
The first two terms are the entropy ones, and the last term, with the χ is the enthalpic one. It’s not at all easy
to see how it plays out. There are two reasons for this:

1. There is only a small range of conditions where the outcomes hangs in the balance.
2. It’s not the sign of Δµ that’s important but, rather, its 2nd derivative.

That’s why we need an app:

[Link]

We have a polymer ~40K MW and a solvent of 100. The Flory-Huggins χ parameter is greater than 0.5 yet
the test tube shows a single phase. The chemical potential is flat and just barely negative before, driven by
entropy, it becomes strongly negative.

Now increase the polymer MW to 45K:

It’s not clear what’s happening to the

chemical potential, but its 2nd derivative
flips and we get a spinodal
decomposition. This means that there is a
sudden phase separation and you get two
phases. One is a dilute (0.035) solution
of polymer in solvent, the other is a
phase of polymer swollen with solvent.
So although we say that the polymer is
insoluble in the solvent, the solvent is happily soluble in the polymer.

FST 273
 We can revert to our original 40K
polymer and get a spinodal
decomposition in another way. This time
we’ve changed the MW of the solvent.
It’s a universal that (on simple entropic
grounds) larger solvents are less good
than smaller equivalents with the same
(enthalpic) χ value.

It’s a common observation that many green solvents aren’t as effective as we might like. This is often
because they are larger molecules than the ones they are intended to replace.

This hovering state around χ = 0.5 is the “theta solvent” state, where polymers are in a neutral Gaussian coil
rather than expanded in a good solvent or tightly coiled as they phase separate.

[Link]

At a relatively small χ, the polymers are expanded and even overlapping. At a relatively high χ their
solubility is low and they are much smaller:

274 FST
 Dissolution Flory-Huggins

Explanations for the End-to-End radius and Radius of Gyration are provided in the app.

Getting your χ values

A full academic group with access to lots of techniques and time can get a few χ values for a few polymers
with respect to a few solvents. The rest of us have no real choice other than using Dissolution_Hansen
Solubility Parameters. By knowing the HSP of the polymer (measuring it if necessary by testing yes/no
solubility in a range of solvents) and the HSP of any given solvent, you can calculate an HSP Distance, D.
The χ parameter for that value of D and a solvent of molar volume MVol is given by:

MVol.D2
χ= 4RT

This is fine for the intended theoretical situation – an amorphous polymer of high MW. But we often need a
χ value for polymers below their Tg, micro-crystalline (or highly-oriented) polymers, cross-linked
(swellable) polymers such as rubbers or, going well outside this regime, pigments and nanoparticles. Then
we use the radius of the sphere that encompasses the polymer in the practical solubility measurement.

For example, normal racemic polylactic acid, PLA, has a radius of 8 which relates to a normal χ value ~0.5.
A high MW version of l-PLA that has been strongly extrusion oriented might have a radius of 4 because
more distant solvents can’t touch it. So now we define χ ~0.5 at a radius of 4.

I learned this the hard way. Dr Hansen and I were asked to design a green solvent formulation that
was just inside the radius of PLA. We found a great solvent blend which was quickly tested … and
shown to be useless. Only then did we learn that the application used highly-oriented PLA. When
we got it measured, the radius was half the standard value. A reformulation to this value then
worked as required.

Such an approach does not get the approval of purists. But the purists haven’t given us anything we can
use, so as practical formulators we use the tools that are proven to do an adequate job, and use them with
due awareness that they are not as precise as we would like.

FST 275
Dissolution Gases
Links

The solubility of gases is something most of us can ignore most of the time – then you encounter
an unexpected problem caused by an issue you didn’t suspect would be important.

Little to know
The typical solubility, S, of a gas in a solvent in g/kg depends on pressure, P (bar), temperature T (°C) and
MW via a simple formula valid (within a factor of 2) from 0°C to 80°C:

S = P(0.04 − 0.0008T + 0.000007T2)

MW
28

There are obvious exceptions

when you have “reactive” gases.
For CO2 the formation of
H2CO3 gives a solubility 50x
higher than expected. Ammonia
forms ammonium hydroxide,
hydrogen chloride splits into the
acid form etc.

Pressure effects for normal

gases (solubility is proportional
to pressure) look equally dull
until you go scuba diving,
where the solubility contrast
between high and low pressures can be life threatening.

The reason for this chapter is that there is a real issue with the modest temperature dependence, one that is
little known and yet has caused problems for many. And the chapter gives us an opportunity to talk about
champagne…

Coating defects
When people see a pinhole defect (Coating_Pinholing Theory) in their coating they are tempted to call it a
“repellency”. And because names imply causes, they look around for the oil contamination that has caused
the repellency. They generally can’t find any such contamination.

What else could cause it? The answer is a gas bubble. Of course people are aware that mixing and some
aspects of a coating head can create bubbles in a formulation and work hard to eliminate them. But still the
pinholes are there. The app at [Link] explains why:
 Dissolution Gases

There are other inputs and outputs in the app (it also
calculates time for bubbles to rise) but here we focus
on a cause of coating defects. In this example, the
aqueous coating was stored at 20°C where it had time
to equilibrate with air so is a saturated solution. On
moving to the coating machine, its temperature rises to
22°C. Amazingly, the reduced solubility of air, just in
that 2° difference, is enough (if it all came out and all produced the same-sized bubbles) to create ~600
bubbles of 100µm diameter in each ml of solution.

Of course in real life we don’t get all the dissolved air coming out. But the point is that even 1 bubble can
ruin your day and the super-saturated solvent can readily produce that 1 bubble. If, however, the solution
had been stored at 22°C and coated at 20°C, not only would there be no extra bubbles, there’s even a
chance that a rogue bubble from, say, mixing, might dissolve back into the water.

Champagne
A bottle of champagne might contain CO2 at 5 bar, which translates to 11g/kg at 10°C, so ~ 8g/bottle.
When we open it, if 4/5 of it (6g) came out at once, that would be ~3l of CO2, something you wouldn’t
want to happen if you preferred drinking the champagne.

A simple formula shows us why it can’t all come out easily. Suppose the CO2 coming out of the
champagne produces a bubble of radius r and that the surface tension, γ, of champagne is ~ 50mN/m (the
ingredients act as poor surfactants). The pressure, P, inside the bubble is given by:

2γ
P= r

If r = 1µm then P=0.1/10-6 = 100 kPa ~ 1 bar. So a bubble of r~0.2μm has an internal pressure of ~ 5 bar,
and we know the CO2 is dissolved at that pressure.

So we have the problem that as a small amount of CO2 tries to come out of solution, producing a bubble <
0.2 µm, it can’t. And because no small bubble can form, no bigger bubble can occur via diffusion of the gas
into the bubble. So fizzy champagne is impossible because nucleation, formation of the proto-bubble, is
impossible.

We know that the puzzle is solved via seeds, e.g. slight defects in the champagne glass.

Although, for fun, we’ve discussed champagne, this nucleation problem is universal. Many manufacturers
of polyurethane foams, knowing that smaller foam cells are better insulating, have tried to seed their
formulations in order to create many smaller bubbles rather than fewer larger bubbles. Although it can
work in the lab, it is hard to scale up. Closer to champagne, some foams are made by dissolving scCO2 into
a polymer and then releasing the pressure suddenly. Because the pressure difference is so large, even super-
small bubbles are able to form so the seeding problem goes away.

FST 277
Dissolution Hansen Solubility Parameters
Links
Deodorant Sticks, Sun Screens, SkinCare, Water-based Adhesive, Adhesive, Solvent-Based Paint,
Emulsion Paint, Pharma Formulation, Microencapsulation

Many solubility issues, e.g. replacing one solvent blend with another, can be done efficiently using
HSP – Hansen Solubility Parameters. These are 3 parameters for any solvent, polymer or pigment/
particle that capture the essentials of their interactions:

1. δD is the Dispersion component – related to RI, polarizability, van der Waals forces
2. δP is the Polar component – related to dipole moment
3. δH is the Hydrogen-bonding component – usually obvious from the molecular structure.

The HSP of a solvent blend is the volume-weighted average of the individual components

The compatibility of, say, a solvent (blend) with a polymer depends on the Distance, D, between
the HSP values:

2 2 2
D2 = 4(δD1 − δD2) + (δP1 − δP2) + (δH1 − δH2)

Because we know the HSP of most solvents and because we can know or measure the HSP of our
other components, we can easily formulate by, for example, ensuring the smallest Distance
between components we want to remain compatible.

Although there are plenty of HSP-related apps at [Link]

[Link], the science is best explored using the HSPiP (Hansen Solubility Parameters in Practice)
package of software, datasets and eBook created by Prof Steven Abbott, Dr Charles Hansen and Dr
YAMAMOTO Hiroshi.
 Dissolution Hansen Solubility Parameters

The image is included because it captures so much useful information which we can explore.

The Table

With the table at the top left you get an idea of the δD, δP and δH parameters for some familiar solvents.
Hexane has (of course) no polar or H-bonding and is held together as a liquid via dispersion forces of 14.9
(the units are the obscure MPa½). We often neglect these because they seem dull, but they are what holds
most things together, and they should never be underestimated. Cyclohexane has the same 6 carbons, but its
δD is 16.8. We can’t see benzene in the table but it is 18.4. As we go to increasingly “strong” solvents, δD
increases. These parameters are “cohesive energy density” so the more compact the molecule (benzene’s
density is 33% higher than hexane’s) the higher δD. Another way to think about it is in terms of
polarizability (that’s where van der Waals forces, or, specifically, the London force components, come
from) which is related to refractive index – nbenzene = 1.5, nhexane = 1.37.

We can then look at δP. It seems reasonable that we go from chloroform’s 3.1 through acetone’s 10.4
through DMF’s 13.7 up to acetonitrile’s 18. You would not be able to guess the correct numbers, but the
trend is reasonable, and strongly linked to dipole moment.

Similarly, δH is reasonable. Toluene’s value of 2 is at first surprising, but aromatics are significant H-bond
acceptors. Acetone’s 7 is reasonable for the H-bonding C=O and we go up to ethanol’s 19.4 which is less
than methanol’s 22.3 which is much less than water’s 42.3.

Given that H-bond is a “polarity-related” effect it is sometimes asked if we could just use δD and a super-
δP. But look at acetonitrile. It is highly polar but a weak H-bond acceptor, while ethanol is not especially
polar but is a strong H-bonder. As solvents they behave very differently, and 50+ years of experience have
shown that we really do need all three parameters. Why not four parameters, splitting H-bonding into
Donor and Acceptor? Because no one has ever created a self-consistent and useful scheme using them.
HSPiP includes the option to use a split δH but it is little used.

FST 279
For those who worry that these numbers are just some arbitrary fitting values obtained by Hansen in the
late 60’s, for solvents the square root of the sum of the squares of the three values must equal the enthalpy
of vaporization, the δD values must relate to refractive index and δP to dipole moment. HSP are extensions
of Hildebrand theory which is related to Flory-Huggins and the χ parameter (Dissolution_Flory-Huggins)
which in turn are related to lattice theory. Yes, HSP are not precise, yes, you have to use your formulation
knowledge to get the best out of them (a feature, not a bug) but, no, they aren’t arbitrary.

The Sphere

The “Score” column in the table contains some 0 (“bad” solvents) and 1 (“good” solvents) values. The
scores were found by putting a small amount of the substance of interest (in this case PLA, poly lactic acid)
into each of 27 tubes and adding the same volume of 27 different solvents to each tube, shaking and leaving
for some hours. The scores in this case were found using the human eye. Other ways to distinguish between
good/bad (including numeric measures such as measured solubility or relative sedimentation time in a
centrifuge) achieve the same thing. Using a fitting function, the green sphere has been created with all the
good solvents (in blue) inside and all the bad ones (in red) outside. The small green dot in the middle
represents the HSP of the PLA.

What can we do with such a value? This is where the HSP Distance comes into play. Here’s the distance
formula once again:

2 2 2
D2 = 4(δD1 − δD2) + (δP1 − δP2) + (δH1 − δH2)

If we ignore the factor of 4 (there for technical reasons), D is simply the Pythagorean distance in 3D space.
It is rather obvious that the bad solvents are far away and that some good solvents are closer to the centre
than others. And, yes, those closer to the centre are better solvents. Those at the edge really are borderline.
An increase in MW or crystallinity might make the sphere smaller, so a previously good solvent becomes
bad. This uncertainty at the edge is a feature, not a bug.

The RED column (Relative Energy Difference) shows D/Ra where Ra is the radius of the sphere. By
definition, a RED > 1 means insolubility for that specific substance when measured to a specific criterion
for good solubility. If your criterion for solubility is 0.01% of a 5000 MW polymer, the radius will be much
larger than if your criterion is 10% solubility for the same polymer with a 500,000 MW. HSP are a
pragmatic tool to be applied to your specific problem. The core HSP value does not depend on your
definition. The RED is your way of deciding how close you need to be to that core value in order to obtain
the solubility you desire.

Other options

Because HSP can be used for soluble polymers, swellable (crosslinked) polymers, APIs, nanoparticles (see
Dispersions_Solubility if you disagree) and more, it is unlikely that one fitting procedure (against a specific
objective function) works well for everything, so there are choices of functions, each with their trade-offs.
You can use scores of 1 (best) to 6 (worst) instead of 0/1, you can use numerical data rather than scores.
And you can correct for temperature (HSP values decrease with increasing temperature).

The point of these options is to allow the pragmatic formulator to work in a way that makes sense to them.

280 FST
 Dissolution Hansen Solubility Parameters

Solvent blends
The solvents used in the measurement of HSP values are not chosen because they are nice or green. They
are chosen because they span HSP space. There is no point in testing hexane, heptane and octane, or
ethanol, propanol and butanol. You choose a representative solvent from each class then, in addition to
typical ketones and esters use solvents such as NMP, chloroform, DMF, DMSO that are in unique parts of
HSP space. You (or your robot) are using only a few ml of each solvent in capped tubes, so the safety risks
are manageable. Thanks to these solvents you get a good HSP value. But the fact that pyridine happens to
be a close HSP match to PLA is of no real-world use to you – you aren’t going to use it for a formulation.

So you take your target HSP value (e.g. the one for PLA), assemble a convenient list of common solvents,
then sort them by Distance to find, hopefully, one solvent that meets your requirements for cost, safety,
volatility, odour … Here is such a list using HSPiP’s Solvent Optimizer:

We find that DMI, cyclohexanone, butyl benzoate and 1,3-dioxolane are all reasonable matches to the
“target” HSP of [18.7, 7.7, 7.0] which was calculated for PLA. The problem is that DMI is expensive and
of low volatility (the RER column is Relative Evaporation Rate, with nBuAc = 100), some people don’t
like the odour of cyclohexanone, butyl benzoate is also involatile while dioxolane is too volatile, with a
flash point (you need to scroll the columns to find it) of -6°C which makes it difficult to use safely.

If your solvent list was of a smaller set of green solvents, it is unlikely that you could find a single solvent
that was a good-enough match.

FST 281
 That’s where solvent blends become important. The HSP of an x:y blend of two
solvents is the x:y ratio of their δD, δP and δH values. To make an extreme point,
lets look at the situation where we have two bad solvents, outside, but on opposite
sides of, the sphere. A 50:50 blend would give an HSP right in the centre,
implying that two bad solvents could become a good solvent blend. Although the
idea is anti-intuitive, it turns out to be correct. And if you can make a good
solvent from two bad solvents, even more can you make a blend from two OK-
but-not-good-enough solvents, providing their respective deficiencies (e.g. one’s
δH is too low, the other’s δH is too high) are complementary, allowing an optimum intermediate blend.

When Charles Hansen developed HSP he realised that they predicted that two bad solvents could
make a good one. This made no sense, so he did some experiments with bad solvent pairs to
disprove his theory so he could then find a better one. To his surprise, the blends worked as the
theory predicted. People remain surprised! In tutorial classes, people can find the HSP of nail
polish. Butyl benzoate and benzyl alcohol are two of the test solvents and they are both found to be
bad. From the Sphere fit they can then calculate that a 50:50 blend of the two is an OK solvent.
They are always astonished when they test the blend and find that it works.

In the PLA example, if you wanted to use DBE as a relatively nice solvent, but which has a Distance of 5.4,
you can find (manually or by pushing some buttons) that a 41:59 mix with benzyl benzoate (Distance of 4)
gives a blend with a distance of 2.2. The colour coding shows that the δD values are high (red) and low
(blue) compared to the target, the other way round for δH, with nothing useful in terms of δP.,

Solvent blends provide another capability. It can be very inconvenient to

assemble, say, 15 different solvents. An alternative approach is to carefully
choose 4 solvents and make systematic blends of them. If the first solvent is
known to be good, and the other three are known to be bad, then you might
find good (blue) and bad (red) solubilities if you try the 13 blends, easily
created by some automated system. The resulting sphere won’t be perfect (the
blends cover only a limited range of HSP space) but (via a built-in tool, or
using common sense) you can identify a few key test solvents in the missing
parts of HSP space to refine the fit. This “grid” method is very popular and
especially well suited for robotic systems.

282 FST
 Dissolution Hansen Solubility Parameters

Getting HSP values

We have seen that HSP values for solids (polymers, crystalline solids, nanoparticles) can be measured using
the sphere method. Low volatility semi-solids can be measured using Inverse Gas Chromatography (see
[Link] for an explanation, and read the free
IGC eBook from the same site). But new solvents are hard to measure and if you need the HSP of lots of
molecules, you are unlikely to carry out lots of sphere tests.

You can then use HSPiP to get a reasonable estimate

simply from the SMILES structure. [If you are not
familiar with SMILES, there is an excellent guide on
Wikipedia.] Suppose you are thinking of some new
green ether-alcohol, you can enter its SMILES (in this
example it’s CC(C)OCC(C)OC(CC)COCCO) and get
both its 3D structure and an estimate of its HSP [16,
5.4, 7.1] plus handy estimates of many other properties.
If the properties are not to your liking, you can try
another variant.

If, instead of it being a small molecule, it was some new green

polymer, you can enter the Polymer SMILES:
XCC(C)OCC(C)OC(CC)COCCOX. The Xs are used to indicate
the polymer chain links and you can see them in yellow in the
image. In this case the calculated HSP are [17.4, 2.6, 2.2]. We
have gained δD, as is typical with polymers, and lost both δP and δH because the terminal -OH group has
become another ether link.

Complex formulations
We often find that we need solvents that must be good for some component of a formulation and bad for
another. HSP make it easy to explore the trade-offs.

In this example we have two components that are well-separated in HSP space
(for ease of illustration). Each has one or more good solvents (blue) from the list
of solvents being used. Most of the chosen solvents are bad for both. The line
and blue dot show how you can create solvent blends (there is a slider to allow
this to happen) to find a blend that meets your specific requirements. This ability
to work in 3D to explore multiple components interacting with multiple potential
solvents is a powerful way to create advanced formulations.

Just remember that each copy of HSPiP comes with the advanced neural network
of the person using it. The numbers provided by HSP are not the endpoint. They are the beginning of an
interaction with your own chemical knowledge.

Not just HSPiP

It has been convenient to use HSPiP to illustrate the core science of HSP. You can find many useful apps
and lots more explanations at both [Link] and
[Link] On the Hansen-Solubility site you can download some useful Excel

FST 283
sheets that do many of the basic calculations for you and there is a selection of YouTube tutorials to deepen
your understanding.

The 3D Optimization discussed in the previous section is available as an app:

[Link]

284 FST
 Dissolution Ideal solubility

Dissolution Ideal solubility

Links
Sun Screens, SkinCare, Pharma Formulation, Microencapsulation

However good your solvent, for a crystalline solute you can’t get better than the ideal solubility
which is related to the energy required to break down the crystalline structure. The theory used
here, although “standard”, turns out to be tricky (with unknown parameters) and not wonderfully
accurate. But it’s better than having no theory at all.

[Link]

The next app will be more useful, but this version captures the essence of ideal solubility. For a solute with
a MPt of 150 and a reasonable ΔHf of 30 kJ/mole and the hard-to-measure ΔCp of 134 J/mole/K the
solubility at temperatures up to the MPt are shown. At room temperature the solubility is just below 0.1
mole fraction. If you need 0.2 mole fraction there is nothing you can do, because this is the solubility with
an ideal solvent – one where the activity coefficient = 1. Real solvents won’t be ideal, so solubility will be
less than this.

Of course you could change the definition of “ideal” and if the solute is a base, add an acidic solvent. But
those sorts of tricks are obvious. Let’s stay with this definition of ideal.

Why is solubility limited? Because you first have to “virtually melt” the solute at your chosen temperature.
The higher the MPt and the larger the ΔHf, the harder it is to virtually melt the solute, so the lower the
solubility. If you assume the validity of lattice theory ([Link]
[Link]) then an ideal solvent is one where the solute-solute, solvent-solvent and solute-solvent
interactions are identical. But first they both have to be liquids! That’s why we have to virtually melt the
solute.
A key reason ideal solubility theory isn’t too helpful (in addition to some questionable assumptions) is that
the correction for heat capacity is complicated and most of us don’t know the relevant values.

Irritated by the fact that for 1000s of compounds with known solubility, the ideal solubility calculations
were either too hard or too inaccurate, Yalkowsky introduced his famous approximation. Select the
Yalkowsky option and you get this curve:

This isn’t too far away from the standard ideal equation, but uses only the MPt, Tm to calculate the mole
fraction x:

ln(x) = − 0.023(Tm − T)

The full ideal solubility equation, for those who are interested, is found on the web page.

Non-ideal solubilty

[Link]

286 FST
 Dissolution Ideal solubility

This app plots that data in a more user-friendly format and adds the activity coefficient γ to take into
account real-world solubility effects. It also allows you to plot in more user-friendly mg/ml units.

Where do you get γ values? Professional solubility software such as COSMO-RS readily gives values. Or
you can make good-enough (relative) estimates via Dissolution_Hansen Solubility Parameters.

I’m being polite about Ideal Solubility. It really is a poor intellectual construct that has wasted too
much effort around its crazy assumptions. There is a better way to think about it, but it will be
some time before there’s an app that can bring the superior approach to life. As soon as it’s
available, the alternative approach will be added to the FST.

FST 287
Dissolution Kirkwood-Buff
Links
Pharma Formulation

This chapter is courtesy of Dr Seishi Shimizu at U York, UK.

What does “interaction” mean in solution?

Thinking about a pair of molecules and its interaction

(e.g., hydrogen bonding, charge interactions, and van der
Waals forces) is enough in gas.

But around the pair in solution, there are many other

molecules continuously moving about.

So, we are forced to think about the distribution of

molecules to measure interactions.

The Kirkwood-Buff integral measures the interaction

using the change of concentration from the bulk. Here
you can see that on average there are more of the green
molecules near the solute than are found on average.

Excess numbers and Kirkwood-Buff integrals

• If red-blue interaction is attractive, concentration of green

around red is higher than in the bulk. This difference is the
excess number of green molecules around a red molecule, Nu1.
• If red-green interaction is attractive, concentration of green
around red is higher than in the bulk. This difference is the
excess number of green molecules around a red molecule, Nu2.
• When the interaction is repulsive, Nu1 or Nu2 will become
negative.

There is a limitation to this approach and how this could be overcome

• Blue molecules are more concentrated in bulk than green. This means Nu1 can be larger than Nu2
simply because there are more blue molecules than green. This poses difficulties when we need to
compare red-blue and red-green interactions.
• To compare blue-red and green-red interactions, we need to account for the bulk concentrations,
c1 and c2.
 Dissolution Kirkwood-Buff

Therefore, the measures of interactions are given by the Kirkwood-Buff integrals as

Nu1
Gu1 = c1

Nu2
Gu2 = c2

Experimental determination of the Kirkwood-Buff integrals

• Kirkwood-Buff integrals can be determined from solubility measurements, when combined with
density and activity data.
• Kirkwood-Buff integrals can be estimated from molecular dynamics simulations.
• Thus, Kirkwood-Buff integrals links bulk measurements to molecular distributions and
experiments to simulation.

Specific examples are available at [Link] Many of the

apps allow you to load your own “normal” datasets and analyse them via KB theory.

FST 289
Dissolution Solubilizers and Hydrotropes
Links
Deodorant Sticks, Laundry Liquids, SkinCare, Pharma Formulation, Microencapsulation

Although we have reasonable theories for classical solubility and for oil-water microemulsions,
there is a large area of formulation space where we rely on “solubilizers” to get solutes into water.
Until recently, the explanations for these effects were incoherent. With the recent Kirkwood-Buff
approach we can at least understand the effect of a solubilizer even if we cannot, yet, predict it.

Confusing terminology
Getting something into solution via non-classical solubility has been described by many names.
Hydrotropes, co-surfactants, pre-ouzo effect, “surfactant-free emulsions”, kosmotropes/chaotropes,
entrainers (for scCO2, see below) and solubilizers all appear in the literature. For the rest of this chapter we
shall just call them solubilizers as that’s a handy generic name that describes the purpose, without the extra
baggage associated with other names.

Bad explanations
There are two aspects to bad explanations:

1. They are factually wrong;

2. They don’t help.

The world of solubilizers has suffered from both aspects. Lots of words have been expended on things like
“water structure” as “explanations” for these effects, yet no one has been able to formulate better based on
those wrong and unhelpful approaches.

Typical examples are:

• Water structure. The solubilizer changes the structure of the bulk water, allowing the solute to
dissolve;
• Pseudo-micelles. The water makes the solubilizer form pseudo-micelles which, just like real
micelles, then dissolve the solute;
• Complexes. The solute and solubilizer form specific complexes that, for some reason, aid
solubility.

The water structure hypothesis is especially popular because (a) we know that water is structured and (b)
we can give fancy names such as kosmotrope and chaotrope to molecules on the grounds that they make or
break water structure. Sadly, after a century of the use of these terms, no coherent formulation
understanding has emerged – for the good reason that water structure does not play a role in solubilizer
effects.
 Dissolution Solubilizers and Hydrotropes

A good explanation
We have 3 components: the water, 1, the solubilizer, 2 and the solute u. Just like lattice theory (see
Dissolution_Flory-Huggins and [Link] we can
think of all possible interactions. In this case we have 1-1 (water structure), 1-2, 1-u, 2-2 (pseudo-micelles),
2-u and u-u.

Because we are trying to solubilize something with a low solubility, we know that 1-u interactions are
unfavourable (if they were favourable, u would be soluble) and, at the low concentrations we can obtain, u-
u interactions are irrelevant. So we can focus on the other interactions.

Using assumption-free statistical thermodynamics via Dissolution_Kirkwood-Buff theory we can nicely

compute how these interactions play out in terms of solubility or, rather, rate of change of chemical
potential of u, δμu with respect of the concentration of 2, δc2. The G terms are Kirkwood-Buff Integrals,
KBI, which describe how much two things like being next to each other compared to their statistical
average. So a large positive Gij implies that the molecules are positively attracted to each other:

δμu Gu2 − Gu1

δc2 = − RT
1 + c2(G22 − G21)

Let’s take the equation slowly. For good solubilization we want the rate of change of chemical potential
with concentration to be as large and negative as possible. RT is what it is. Strangely, the concentration of
the solubilizer is on the bottom, which means the higher it is the less negative. Clearly that c2 term can’t
drive the solubilization.

While we’re looking at the bottom, note that a large G22 decreases the effect. G22 is a measure of the
amount of “pseudomicelles” in the system. Contrary to the popular explanation, pseudomicelles make
things worse.

We already know that Gu1 is small because the solute doesn’t much like the water, which is why we’re
adding a solubilizer.

Before we reveal the true answer, look at the role of G11 the water structure term. It’s not in the formula!
Water structure (kosmotrope or chaotrope) has no influence on solubilization, despite a century of opinions
that it was the explanation.

Finally we see the solubilization is driven by Gu2 the tendency towards mutual clustering of the solute and
solubilizer. This is very different from the idea of a “complex” between them. When you plot Gu2 against
c2 (as we shall do in a moment), you see Gu2 rise and fall. We can understand the “rise” part – as you get
more solubilizer there’s a bigger chance of mutual clustering. Why does it fall? Because Gu2 is the
difference of numbers in the vicinity from the statistical average, and as c2 increases, the average increases
so the difference decreases.

If we go to the app that describes typical solubilizer behaviour, we can get lots of information. But the
dominance of the Gu2 curve and the irrelevance or even counter-effect of G22 is also apparent.

FST 291
Extracts from [Link]

The left-hand graph from the app shows the solubility of butyl acetate in water as urea (a well-known
solubilizer) is added. The right-hand graph shows the Gij values. The only large values are from Gu2 (for
KBI, Gu2 = G2u), with all others being irrelevant.

What the equation above cannot explain is the sigmoidal curve characteristic of these solubilizers. That’s a
bit more complicated – it’s the solute-induced solubilizer clustering. Gu2 becomes large because the solute
encourages more solubilizers to gather around it.

The KB approach is assumption-free, not at all hard to implement and gives clarity of explanation
in complicated systems. Yet it’s still rarely used. Many are comfortable using hand-waving non-
explanations and feel threatened by ideas that clearly disprove everything they’ve said for the past
decades. But younger minds (and the brightest of the older generation) tend to adopt approaches
that work, so the KB approach is spreading steadily.

scCO2
Looking at the solubilizer equation you can say “Yeh, just another equation”. But it has a beauty to it in that
it’s attempting to balance all possibilities in a simple form. It would be nice if there was another example of
such an equation from a different world.

As is well-known, supercritical CO2, scCO2, is a potentially interesting solvent because you can release
your solute simply by letting the CO2 disappear when you release the pressure. The problem is that CO2,
liquid or supercritical is a useless solvent – maybe similar to pentane. There’s not much you can dissolve in
pentane. So scCO2 extraction uses “entrainers”, molecules at a low level (a few %) which dramatically
increase the solubility. Water would be the most popular entrainer, and it indeed works for caffeine, if it
were itself more soluble in scCO2. Because of its low solubility, people use molecules like ethanol or
acetone instead.

The entrainer effect has been explained in, you guessed it, 3 potential ways:

• CO2 structure. The entrainer changes the structure of the scCO2, allowing the solute to dissolve;
• Pseudo-micelles. The scCO2 makes the entrainer form pseudo-micelles which, just like real

292 FST
 Dissolution Solubilizers and Hydrotropes

micelles, then dissolve the solute;

• Complexes. The solute and entrainer form specific complexes that, for some reason, aid solubility.

In reality, the equation driving entrainment is (see the app at [Link]

solubility/[Link]):

δμu Gue − Gu1

δce = − RT
1 + ce(Gee − Ge1)

Yes, that’s a copy/paste of the equation above, using e for the entrainer rather than 2 for the solubilizer. The
entrainer effect is dominated by solute-entrainer interactions, entrainer-entrainer interactions reduce the
effect and CO2 “structure” isn’t even included in the equation.

In this case, (see the Simple scCO2 app, [Link]

[Link]) we can analyse the large literature on scCO2 solubilities to work out what constitutes a good
entrainer for a given molecule, i.e. what would create a large Gue? The answer is simple. If the solute has
no potential for H-bonding, there’s no significant entrainer effect from anything. If the solute is capable of
H-bonding, then any entrainer with the right functionality to form H-bonds will help.

As with solubilizers, a simple equation brought the clarity necessary to get to the heart of the entrainment
effect. The H-bonding idea had been proposed some decades earlier, but the scCO2 community expended
its energy exploring the wrong ideas, wasting those decades.

Nearly every scCO2 paper starts with “scCO2 is a green solvent …”, which is fine. But to be green
you must not waste resources. As pointed out by Abbott and Shimizu in a paper and a book
chapter, a few 10’s of smart scCO2 entrainer experiment could have sorted out how and when
scCO2 might be a useful solvent for a given process. Instead, we’ve had 100’s of experiments
using ad hoc approaches, assumptions, analyses and wrong explanations that have wastefully
produced no usable common understanding. It was genuinely painful to go through all those papers
to dig out the nuggets of understanding contained in the Simple scCO2 app.

The KB-χ
The fact that the same beautiful equation describes different processes suggested to Dr Seishi Shimizu a
universality to the KB equation. If we cut out the bits which will seldom be important for these sorts of
problems we end up with:

δμu G
δc2 = − RT 1 + cu2
2G22

We can imagine the ratio of Gu2 to 1 + c2G22 as capturing the same essence as the Flory-Huggins χ value;
the balance of competing attractions. Calling this the KB-χ provides a unifying name for the wide world of
solubilizers.

Pre-ouzo, surfactant-free emulsions and more

Take a solution of anethole or thujone (ouzo and absinthe respectively) in 40% ethanol/water and add a lot

FST 293
of water. It becomes cloudy as the oily flavour molecules become insoluble. But if you add smaller
amounts of water, the solution becomes “sparkly”, there are nano-sized fluctuations near insolubility. If you
do this for something like a fragrance then you have a “pre-ouzo” formulation. If you like to use a different
terminology and, maybe, have less sparkle, then you can make a “surfactant-free emulsion”.

In all these cases you have water, something reasonably water soluble such as an alcohol and your desired
molecule such as a fragrance.

The sparkle show that the molecules are gathering into fluctuating clusters. Another word for Kirkwood-
Buff theory is Fluctuation Theory, and the KB approach is ideal for describing what is going on. The
relevant formulae are more complex because these wild fluctuations are near the borderline of phase
separation (always a risk with these formulations) but once more they do a much better job than the non-
alternatives.

Although these formulations have attractive names, there’s no escaping the fact that you have to add lots of
what amounts to a co-solvent, which diminishes their attractiveness as an alternative to ethanol (though this
is an important consideration in some cultures) or to surfactants. For example, in a fragrance formulation,
unless the added smart, green solubilizer is volatile, it’s left sitting on the skin, which consumers may not
appreciate.

A smart, efficient microemulsion created via Surfactancy_HLD also has a sparkle from nano-sized
scattering.

294 FST
 Evaporation Basics

Evaporation Basics
Links
Sun Screens, SkinCare, Fragrances, Surface Cleaning, Water-based Adhesive, Adhesive, Solvent-Based
Paint, Emulsion Paint, Inkjet Inks

The evaporation rate of solvents depends on:

1. The temperature;
2. The vapour pressure of the solvent at that temperature;
3. The air velocity.

This app contains the Antoine Coefficients (see Evaporation_Temperature and Antoine
Coefficients ) of many common solvents – which let you calculate the vapour pressure at your
chosen temperature. It also include the effect of air flow. With zero air flow there is almost zero
evaporation.

[Link]

Solvent evaporation depends on the (temperature dependent) vapour pressure of the solvent (blend) and on
the air speed across the surface of the solvent. The dependence of the vapour pressure P on temperature T is
governed by the Antoine Equation and its three parameters, A, B and C:

B
log10P = A − C+T

The app has the Antoine Coefficients for the common solvents and the more powerful HSPiP package has
many more.

The dependence on air flow is not obvious. The problem is that diffusion of the solvent vapour is very, very
slow. For realistic evaporation, the vapour has to be swept away by turbulent air flow:
Although those who apply typical coatings don’t realise it, a typical lab air flow over a typical surface
might be 0.25-0.5 m/s, more than enough to give reasonable evaporation rates. Obviously drying with little
air flow is both inefficient and dangerous in terms of solvent vapour build-up. In real dryers, air flows are
much larger.

The equations used in the app have a complicated dependence on Reynolds Number, Schmidt Number and
Sherwood Number from which the “Mass Transfer Coefficient” is calculated. You can read them in the app.
If you have to do calculations without the app and have an evaporation rate at a known velocity, v, the
standard approximation is that rate is proportional to √ v. This tells us why we need so much airflow – to
double the drying speed needs 4x the air velocity.

Those who speak of “IR driers” or even “Laser driers” love to talk about the efficiency with which
their heat energy is absorbed. But heat can’t dry. Drying is removal of molecules and only air flow
can do that. If you want an “IR heater” then that’s fine. If you want an “IR drier” make sure the
device comes with plenty of airflow.

Diffusion limited

At some point, the rate at which solvent leaves the coating becomes limited by the rate at which the solvent
can diffuse out of the coating. This is a complex issue involving not just standard Fickian Diffusion
Diffusion_Basic Diffusion but also Diffusion_Concentration-Dependent Diffusion.

As discussed in the Evaporation_Diffusion limited chapter, if your system becomes diffusion limited, the
velocity of air above the coating is irrelevant to drying speed. The only way to speed up diffusion-limited
drying is to raise the temperature. For a formulator, a key consideration is to keep the coating “open” (e.g. a
high-boiling solvent, a better HSP match) for longer, delaying the transition to diffusion limited.

296 FST
 Evaporation Diffusion limited

Evaporation Diffusion limited

Links
Water-based Adhesive, Adhesive, Solvent-Based Paint

Evaporation by sweeping away air molecules from the surface is the fastest way to remove solvent.
But at some point, the rate at which solvent can diffuse to the surface is slower than the rate at
which the solvent is swept away. Now we’re in the realm of diffusion-limited evaporation.

A simple, unusable equation

There’s a relatively simple equation for the rate of diffusion-limited evaporation, E, from the surface:

−K
2 −K
3
K1
E= h e C e T

This tells us that the E depends on three terms:

1. A constant K1, plus a 1/thickness (h) term as a reminder of the obvious fact that thinner coatings
provide a reduced barrier to diffusion;
2. An exponential term, depending on K2 which gets smaller (slowing evaporation) as the solvent
concentration, C, gets smaller because this is a classic case of Diffusion_Concentration-Dependent
Diffusion. “Smaller” means 2 or 3 orders of magnitude, which is why it is so hard to get rid of the
last residues of solvent;
3. An exponential term, depending on K3 which gets larger (increasing evaporation) as T increases,
which tells us the obvious: to get rid of residual solvent you need to increase T as much as
possible.

In an ideal world we would all know K1, K2, K3 but in practice almost no one knows what they are because
it is too much work to determine them. That’s why there isn’t an app.

Detecting diffusion limited evaporation

The effects of Evaporation_Enthalpy of Vaporization mean that in a hot oven with a good airflow, the
temperature of the drying surface is low. For a water-based coating, the oven can be at 120°C and the
surface will only be in the mid 40’s C. In general, you want maximum airflow, so you put plenty of energy
into your fan system.

As soon as the surface temperature starts to climb you know either that you are fully dry (unlikely, except
for very thin coatings, hence the 1/h term in the equation) or that you are in diffusion-limited mode.
Because of the K3/T term, you now want the oven to be at the maximum reasonable temperature … but fan
speed is now irrelevant, so you can save energy by turning the fans down to whatever minimum flow gives
you the temperature you need.

This simple switch from high to low airflow should be obvious. But this chapter is written because it is all
too common to see final zones of drying systems with maximum fan speed “to drive out the remaining
solvent”. Nothing the fan can do can make the solvent diffuse faster through the coating.

It is also all too common to save on temperature measurement inside the oven, so users often have no idea
when the system switches from evaporation-limited to diffusion-limited. This absence of knowledge makes
it hard to optimise the system for speed and efficiency.

A good example is when drying a temperature-sensitive coating. If it must never go above 60°C it seems
dangerous to set the oven temperature to 120°C. But if during that part of the process the coating is at 47°C
(which is the case for a water-based coating thanks to the large enthalpy of vaporization of water) there is
no reason not to get the most drying in the shortest time. What is important is knowing when diffusion
limited drying kicks in, because then the temperature will rise dangerously high.

A counter example of drying too fast thanks to a poor choice of solvent is connected to the phenomenon of
“skinning”.

Skinning and the myth of “drying from the reverse side”

Suppose you have a mixed solvent system where the combination is OK for the formulation, but where the
better solvent evaporates faster. You might have chosen this fast-evaporating solvent to make drying faster,
but the result is to make things worse. The idea of a blend is a good one, based on Dissolution_Hansen
Solubility Parameters but you really need the worse solvent to evaporate faster. In the case of the better
solvent evaporating faster, the surface layer might be rapidly depleted of good solvent, leading to the
formulation crashing out and becoming diffusion limited. This is “skinning” and is a big barrier to rapid,
high-quality drying. Usually skinning is identifiable too late – when the coating emerges from the oven,
sometimes with blisters if the internal temperature of the coating exceeds the boiling point of the solvents.
With an internal temperature monitor it is easy to see a sudden increase in coating temperature that should
be a sign of drying being nearly complete but common sense will tell you that the onset is far too early.

There are two approaches to solving skinning – one wrong, one right:

1. The wrong way is to try to “dry from the reverse side”. This is a common idea which is literally
meaningless. Drying can only take place from the surface and applying heat from the reverse side
cannot alter this fact. To everyone’s surprise, the thermal conductivity of most reasonable wet
coatings is so large that there is no significant temperature difference across the coating (it might
be large enough to encourage Evaporation_Marangoni but is irrelevant to efficient drying).
2. The right way is to change the solvent blend so that the worst solvents evaporate faster, allowing
the good solvents to keep the formulation “open” for as long as possible. This allows the whole
system to dry faster (even if you end up using solvents with a higher boiling point!), and also leads
to reduced stress in the coating and to higher gloss when you ensure that the very last bit of
solvent to leave is the very best solvent for the formulation.

298 FST
 Evaporation Diffusion limited

[Link]

The example of MEK:DMF and PVdF shows how the “bad” MEK evaporates quickly leaving the “good”
DMF behind, with the starting Ra (a measure of the HSP Distance between solvent blend and polymer) at a
relatively high value because of the 80% MEK, going to a minimum at 26% MEK then reaching a “good”
value of 2 once the MEK has all gone. This strange example exists because as described in
Dissolution_Dissolution Kinetics if you try to dissolve PVdF powder in DMF you get a horrible gel, but if
you add the DMF to a poor dispersion of the PVdF powder in MEK, it dissolves nicely.

FST 299
Evaporation Enthalpy of Vaporization
Links
Fragrances

When we look at Evaporation_Basics, the main concern is the vapour pressure of the solvent and
the airflow trying to sweep it away. It is obvious that higher boiling solvents will evaporate more
slowly than lower boiling ones.

But when we are doing serious drying, using large airflows, we often find that high boiling
solvents disappear surprisingly quickly … compared to water.

Evaporative cooling
The enthalpy (or latent heat) of vaporization or of evaporation (the terms are interchangeable) is in J/mole
or, more practically, J/kg, telling us how much energy is required to change a liquid into a gas. We have a
(literal) feel for the concept because even a warm breeze when we come out of the water after a swim can
make us feel cold because our skin provides the energy needed for the water to evaporate and be swept
away.

If you place an aqueous coating into a drier with a good airflow, even if the air is at 140°C, the coating
itself (assuming it’s not on a thick, solid surface which conducts heat quickly) might only be in the 40-50°C
range. Why? Because if the temperature was higher the water would evaporate faster and soak up more
energy, and if the temperature was lower, the slow evaporation would not require so much energy and the
temperature would rise. Technically the coating is at the “wet bulb temperature”.

A handy approximate formula gives us the answer for water with a BP, Tb = 373°K. It tells us that the
temperature of the coating, Tc is given by:

Tb 0.68
Tc = 137 ( )373 log10 (Tair) − 45
For Tb=373 then we have 137 log10(413)-45 = 313°K = 40°C.

This has positive and negative implications:

• +ve If you have a sensitive aqueous coating that must not exceed, say, 60°C, you can use a hot
oven to dry more quickly while remaining in a safe temperature range
• +ve High boiling solvents with a low enthalpy of vaporization are surprisingly easy to dry in an
oven with a fast airflow. In a reasonable drier, a solvent like NMP with a BP of 204°C evaporates
~2x faster than water because its 550 J/g is so much smaller than water’s 2400 J/g. The coating
dries at ~95°C in the same oven where water is at ~40°C.
• -ve Although water is green, the fact that its enthalpy of vaporization per gram is 3x to 5x higher
than many less green solvents significantly reduces its attractiveness in terms of PCF (Product
Carbon Footprint) unless the energy needed to evaporate the solvent is itself entirely renewable.
 Evaporation Enthalpy of Vaporization

If you need to model the drying process, the science is well-known (see Evaporation_Heat and
Mass Transfer) but tricky to implement. The much-used drying model in the TopCoat package
from RheoLogic is excellent though (a) you have to pay for it and (b) I wrote the model so, just as
with the links to HSPiP in which I also have a financial interest, you need recognise that this
advice is potentially biased.

FST 301
Evaporation Heat and Mass Transfer
Links

When we provide flows of hot air to help evaporate a solvent we are mostly engaging in heat
transfer, supplying (costly) energy, in order to achieve mass transfer, which is moving the solvent
from the coating. Direct calculation of mass transfer is less intuitive than calculation of heat
transfer, but fortunately a trick allows us to jump directly from heat to mass.

Energy in …
We are providing plenty of W of energy via a heated airflow which can be astonishingly ineffective at
delivering the heat to the coating. This is because of the no-slip boundary condition which says that the
velocity of a fluid at the surface (which is where we want a high velocity) is zero. We can improve the
situation by providing turbulent flow, but this needs clever design of delivery of the hot air.

From elementary considerations of Evaporation_Enthalpy of Vaporization we know how much heat we

need to put in to be able to evaporate the solvent.

Mass out …
It might seem simple to combine the W in and the enthalpy of vaporization to give us a mass of evaporating
solvent per unit time.

But it’s not. The problem is the uncertainty of the transfer of heat from the hot airstream to the surface of
the coating (and then into the coating itself via thermal conduction). In some circumstances it is possible to
calculate the HTCF, Heat Transfer Coefficient, but in most cases the system is too complex so, instead, we
have to measure it.

This still leaves the problem of how the HTCF translates to the MTCF, Mass Transfer Coefficient.

The Gutoff method

The pragmatic approach is to use the method developed by Gutoff for his pioneering work on practical
models for industrial dryers.

Given, at any instance, a temperature difference, ΔT, between the dryer temperature and the coating, given
a HTCF, a constant K which we will discuss shortly, a partial pressure, P, of solvent at the coating
temperature, a heat capacity CP and weight WT of the coating+substrate then the change of temperature
δT
with time δt is given by:

HTCF
δT HTCF.ΔT − K
P
δt = C P WT

K contains the tricky conversion between our known HTCF and the required MTCF, the mass transfer
 Evaporation Heat and Mass Transfer

coefficient and is given by:

Le.π.MAirCPAir
K= λMSolv

Here we have the heat capacity of air, CPAir, atmospheric pressure, π = 100 kPa, the molecular weights of
air and of the solvent, MAir and MSolv, the latent heat of vaporisation, λ and the Lewis constant, Le which
is the ratio of thermal diffusivity and mass diffusivity. Calculating Le is a challenge but happily, to a good
approximation, Le = 1 for typical dryers. Happily, too, the heat capacity of air is 1 J/g to the level of
accuracy required. The λ and MSolv values are calculated from the solvent data and the current % of each
solvent.

So, at each position in the oven we know δT/δt. The next position is assigned a temperature of T+ Δt.δT/δt,
where Δt = ΔL/V is the time taken for the web to move by the step length ΔL at web velocity V in m/s.

So far, all we’ve done is changed the temperature of the web. There is a second equation that tells us how
much weight of solvent is lost in that same time step. It uses the same ΔT, the difference between oven and
web temperatures.:

δT
δW HTCF.ΔT − CPWT δt
δt = λ

So we use the calculated δT/δt (modified by heat capacity and total weight, WT of coating+substrate) not
only to increase the web temperature in the first equation but also to calculate the solvent loss in this time
step.

Clearly, implementing all that for a given coating system is not trivial and there isn’t a convenient app for
it. The important point is that drying involves a subtle balance of heat and mass transfer which in turn
depend on solvent parameters plus the weight of coating and substrate.

In the end, for rapid drying of a given coating formulation, all you can do is optimize the HTCF, so let’s see
how to measure it.

The Gutoff approach is tricky to implement. The much-used drying model (grounded in the Gutoff
approach) in the TopCoat package from RheoLogic is excellent though (a) you have to pay for it
and (b) I wrote the model so, just as with the links to HSPiP in which I also have a financial
interest, you need recognise that this advice is potentially biased.

Measuring HTCF
The HTCF is in W/m²K. To measure it you take a modest cuboid of a material with known density and heat
capacity (a lump of Al is perfect) and from its dimensions you know the area, A, exposed to the heat. You
put a thermocouple inside it, ensure it’s equilibrated at room temperature then place it into the dryer and
plot the temperature with time.

From the simplified equation:

FST 303
δT HTCF.ΔT
δt = ACPWT

it is easy to fit the curve with something like Excel. Or you can slide sliders in the app to match your
experimental data:

[Link]

Although the basic setup and data analysis are simple, measuring the HTCF for your specific equipment
will pose some practical problems, so you need some ingenuity. If, for example, you want to measure
HTCF in the various zones of a coating machine, although you might be able to fix your block of Al to the
web, and it might be possible to get it through the machine, how do you measure the temperature? The 21st
century has lots of ingenious, low-cost devices that you can solve problems like this. Be assured that having
the ability to measure the HTCF of every zone, under a range of conditions makes the pain needed to find a
suitable technique very much worthwhile.

304 FST
 Evaporation Humectants

Evaporation Humectants
Links
Hair Conditioners, Sun Screens, SkinCare

It is surprisingly tricky to work out how much of each humectant is needed to keep a required
amount of water within the end product.

Norrish
The first thing we need to know is what % water remains in a humectant in contact with air at a given RH.
The Norrish equation works in terms of mole fractions, x, with xw being water and xh being humectant
with, conveniently, xh = 1 − xw. The equation involves a constant K, known, in principle, for each
humectant:

RH
xw = −Kx2
e h

Knowing the MWts of water and the humectant you can work out the wt % of water in the humectant at
any given RH.

Now assume a small saturated % of water in the system (which might be 0) and assume (the errors are
small) that his is linear with RH. Now provide the wt% of humectant in the system and you get the final
water weight %.

[Link]

In this example we have glycerol at 50% RH. It takes on 21% of water. We’ve added 5% of glycerol so
that’s ~1.05% water from the humectant. There’s also 0.25% from the system itself, 50% of the saturated
value. In total we have 1.3% water.
Assumptions and alternative views
Maybe the assumptions are too simplistic, or maybe we aren’t adding humectants simply to hold
equilibrium amounts of water. Here are some possibilities.

• Somehow the humectant must be nicely integrated into the formulation but behave according to
the Norrish equation to estimate the % water. This sounds unlikely.
• A variation on that assumption is that the humectant doesn’t react into the system. If it reacted
then the humectant effect must disappear.
• That the Norrish equation is symmetrical for absorption and desorption. Typically, desorption
isotherms show higher amounts of water than absorption (hysteresis). Maybe we add humectants
to delay a move to equilibrium moisture content.
• The water sorption into the rest of the formulation is linear with RH up to your stated maximum.
This is unlikely, but errors based on the assumption are small
• The extra water absorbed does good things to the formulation. This seems unlikely, unless we are
adding a lot of humectant to highly water-sensitive formulation. Outside extreme cases, why are
we adding these humectants?
• During drying of a formulation, keeping things “open” helps everything to relax into a high-
quality coating. Maybe (as per the desorption hysteresis) that’s all we’re doing.

What is clear is that there is a great lack of clarity about why humectants are added to the formulation. The
simple assumption that “it’s to increase the water content” does not seem to reflect the multiple
circumstances for which they are used, nor the relatively modest increases predicted from the simple
formula.

Users of the app have requested extra humectants to be added to the list. Given a reliable
humectant Humidity_Water vapor isotherm it’s not hard to fit it to the Norrish equation and add it.
However, even for something as simple as glycerol there are varying isotherms. For example, a
paper giving the isotherm for triethylene glycol gave their isotherm for glycerol and PEG400.
Neither was compatible with the isotherms found in other sources. So TEG has not been added to
the app.

306 FST
 Evaporation Marangoni

Evaporation Marangoni
Links
Solvent-Based Paint

Many coating defects, from odd streaks, through orange peel to full-on hexagon patterns (yes,
these really happen) are explained by the Marangoni effect. By understanding it, fixes can readily
be found. There are active debates about whether this should be called Gibbs, Bénard, Rayleigh, in
various combinations but it is common practice to focus on the general effect and call it Marangoni
rather than worry about subtle distinctions.

[Link]

In the diagram we see that there is a temperature difference ΔT (maybe by evaporative cooling) between
the base of the coating and the surface and that there is a small difference δT between one part of the
surface and a neighbouring part, λ/2 away. That small δT creates a difference in surface tension, σ, (higher
T means smaller surface tension) which depends on δσ/δT which is unique for each solvent. The surface
ripples to accommodate the differences in surface tension.

Such small ripples might well go unnoticed, but they

drive a flow from bottom to top. This too might remain
unnoticed, but competing areas of rising and falling
flow tend to organise themselves spontaneously into
coherent structures, ultimately hexagonal cells, and the
human eye can spot such defects very clearly. Most
“orange peel” is a Marangoni effect – a proto-hexagonal
effect can often be spotted in the semi-random defects.

I was a junior formulation engineer supervising the first coating trial of a fancy new machine.
Looking at the coating end of the machine, everything was looking great … till those at the other
end asked why the coating was covered with hexagons. We had no idea. Going to the wise head of
research, when she heard the word “hexagon” she needed to hear no more. “It’s Marangoni”. And,
after a brief explanation of the effect we could very quickly implement a root-cause cure.
An exactly analogous effect occurs when concentration differences at the surface create surface tension
differences. This is especially common when using ill-matched solvent blends where the two solvents have
very different surface tensions and very different evaporation rates allowing large concentration differences
to build up spontaneously.

To avoid this you can:

• Use a solvent (blend) that shows a very low change of surface tension.
• Match the evaporation rates of the solvents to avoid concentration differences.
• Inhibit the circulation via a high viscosity,
• Use a small thickness which makes it harder for the circulation to build up.

All these factors are combined into a Marangoni number, Ma, and the idea of a Critical Marangoni Number,
typically 80, above which a spontaneous pattern is likely to appear.

We have two effects – the temperature-driven and the concentration-driven, with the former much more
discussed and the latter much more common. The equations for Ma are similar. For the thermal effect:

δσ hΔT
Ma = − δT ηκ

where κ is the thermal diffusivity, i.e. how readily the top surface temperature can equalise with the base.

For the concentration effect:

δσ hΔC
Ma = − δC ηD

where this time, D is the diffusion coefficient, where a larger value equalises concentrations faster.

Although you can look up all the values for your system and get a great calculation of Ma, it’s generally
good enough to think what you can do to decrease the root causes. For example, those using xylene/butanol
mixtures tend to use xylene/2-butanol rather than xylene/1-butanol because the evaporation rates of xylene
and 2-butanol are better matched than with 1-butantol, so ΔC is likely to be smaller. Or you might choose
to check the surface tension values of potential pairs of solvents and choose the pair with the smallest
difference.

Look for the unexpected structure

If you find a coating defect, it can be a good idea to look closely under a microscope at the start or finish of
the defect. We sometimes see streaks that might have multiple causes, but if you see something like a small
hexagon at the start, that’s a good sign that Marangoni is the root cause, so the defects can be fixed by
higher viscosity, thinner coating or a better solvent match.

Attempts to coat a customer’s formulation kept failing because of “streak” defects. The customer
was happy to blame our machine, but there was no obvious reason why the machine was causing
the streaks. Down the microscope, a little hexagon at the start of each streak showed that it was a
Marangoni problem. The customer had chosen a two-solvent blend using ethyl acetate as one of
the components. The fact that it was more volatile than the other component hadn’t seemed a

308 FST
 Evaporation Marangoni

problem to them. Swapping to the less volatile butyl acetate immediately fixed the “streaks”
problem.

FST 309
Evaporation Spray Drying
Links

Because spray drying needed to be described in the Microencapsulation chapter, this is a copy/
paste of that text.

A formulator’s view of spray drying

We have the same problem as with coatings. Spray drying,
mostly from water, is mostly about getting a good atomizer
and good airflows, neither of which is under the formulator’s
control. In this section we don’t worry if we have a solution
or an emulsion. The emulsion aspects are described in the
next section.

There are difficult tensions over which we can have some control.

• Production will always demand higher concentrations (less water to evaporate) and lower
viscosity (easier atomization) which are usually contradictory.
• They also want a rapid transformation on the outside (crust formation) so that particles that touch
each other or the wall don’t stick, without that crust inhibiting rapid evaporation of the water. Such
inhibition can slow down the process or, at worst, lead the water to explode as the internal
temperature exceeds 100°C.
• We want maximum hot air temperature for rapid drying, but we don’t want to destroy the contents
of the particle, such as enzymes or friendly biome microbes.

Let’s try to do something about those contradictions.

Maximum concentration, minimum viscosity

If our water contains particles or (it’s the same thing) emulsion drops, then the same app shows the
viscosity dependence on volume fraction, φ and on the viscosity of the dispersed particle (a few cP for a
typical oil and for a solid particle 10,000 cP is good enough:
 Evaporation Spray Drying

[Link]

This tells us that without particle-particle interactions, just them getting in each other’s way, the viscosity
doesn’t change much from water’s 1cP till about 50% volume fraction when it reaches 10cP and is starting
to grow rapidly.

But be careful. If your “water” starts at 10cP because of some soluble additive, then at 50% volume
fraction the viscosity is now 100cP – the crowding effect is multiplicative, not additive. If you really need a
few % more particles, do whatever it takes (and it’s often relatively easy) to reduce your “water” viscosity
by a few cP. This is an important and little-known trick.

If, however, your particles show significant interactions because, for example, you have a poor dispersant
on solid particles, or because (and here is one warning about nanoclays for extra barrier properties) they
have a high aspect ratio (length/thickness) then viscosities can shoot up.

We can explore these ideas in Flow_Particle Viscosity which includes this app:

[Link]

FST 311
As you play with the sliders (the N and Fractal Dimension sliders describe particle association, there is also
the Aspect Ratio slider) you can see two things:

• It is scarily easy to shoot up to super-high viscosities at low shear.

• But if your pumps are good enough to create high shear in the atomizer nozzle, then maybe you
will be OK.

If you are going to rely on the very strong shear thinning of particle formulations, make sure you have
tested many likely variations in production batch formulations. It is very easy to go from a 1000x low shear
increase in viscosity to 10000x with a small change in dispersant or aspect ratio.

If you are dealing with polymer solutions then you need to be familiar with Flow_Polymer Viscosity as
shown in the app:

[Link]

The core idea is not so much that higher polymer concentrations lead to rapid increase in viscosity (we all
know that), nor that high MW polymers are worse than low MW (again we know that). The key is MW
compared to Mc, the “critical entanglement MW”. You can have two similar polymers, with similar
solubilities and similar MWs, but one of them will be much more viscous. This is because the number of
tangles depends on MW/Mc. If the two polymers have Mc values differing by a factor of 2, then the one
with the lower Mc is twice as tangled so will be more viscous.

In many applications (Adhesion_Entanglement) you want lots of entanglement, so you have a real fight
when increasing concentration and MW. For spray drying you may not need so much entanglement so can
choose either a lower MW or a polymer with a large Mc, so a high MW polymer (desirable for other
reasons) becomes acceptable.

However, if you want a low diffusion coefficient for your encapsulated molecule in the polymer, all other
things being equal, an entangled system leads to lower diffusion. Everything in formulation is a trade-off.

Another way to decrease viscosity of a polymer solution is to decrease its maximum solubility. Compared

312 FST
 Evaporation Spray Drying

to a “happy” polymer with chains reaching out and tangling, the “unhappy” polymer is more coiled up on
itself so isn’t so tangled. The obvious flaw in this approach is that you need a high solubility to get enough
polymer into solution.

Fast drying, all the way to dry

Everyone who dries coatings wants maximum drying speed with minimum formation of a crust that inhibits
further drying. See Evaporation_Basics and Evaporation_Diffusion limited for a more detailed discussion.

The key idea is that in “constant mode”, which is evaporation-limited drying, you are relying on a
combination of heat and mass transfer, readily arranged in a modern drying system. As soon as you become
“diffusion limited”, you can’t dry faster than the solvent gets through the crust, which becomes slower as
the crust thickens, then there is nothing you can do other than increase temperature because diffusion goes
faster at higher temperatures. High temperatures of a coating or a particle are generally not desirable.

So how do we keep our crust “open” enough to keep the drying evaporation limited? There are a few
standard tricks:

• Add a “high boiler” (e.g. propylene glycol) in which your formulation remains at least partially
soluble. This keeps the crust “open” for much longer.
• Use the formulation tricks for highest concentration with minimum viscosity to keep your
formulation “happy” for as long as possible. You don’t care that the shell is now very high
viscosity, you just want to make sure that there is still enough water for diffusion to be rapid and
not rate limiting.
• Add some junk (e.g. rough particles) that stops the crust getting too compact. There’s an obvious
downside if this makes the capsules too leaky, but with some thought you can get a reasonable
compromise. If those particles are sticking out they will also reduce capsule-capsule and capsule-
wall adhesion while the drying is incomplete.

This problem of the crust becoming too good a barrier is especially severe for those trying to use nanoclays
as tortuosity barriers. There are no obviously good workarounds for this.

Keeping cool

How do we spray dry heat-sensitive enzymes or microbes in 140°C (413°K) air, Tgas, without destroying
them? A handy approximate formula (found in the app below) gives us the answer for water with a BP, Tb
= 373°K. It tells us that the temperature of the microcapsule, Tmc is given by:

Tb 0.68
Tmc = 137 ( )
373 log10 (Tair) − 45
For Tb=373 then we have 137 log10(413)-45 = 313°K = 40°C. Yes, our rapidly evaporating water is only at
40°C, so our enzyme or microbe inside the capsule is OK. If, however, we create a barrier crust then the
capsule rapidly approaches 140°C and we have failed. Avoiding diffusion-limited drying really is important
for spray drying.

A spray drying app

For those keen to explore some of the complexities of a pharma spray-drying system, this app gives more
than enough to think about:

FST 313
[Link]

The meaning of the multiple inputs and outputs are described on the app page.

Why only this one app? Because the spray drying community has not provided any other usable models.

314 FST
 Evaporation Temperature and Antoine Coefficients

Evaporation Temperature and Antoine

Coefficients
Links
SkinCare, Fragrances, Solvent-Based Paint

The easiest way to characterise the vapour pressures above a liquid at a given temperature is to use
the Antoine Coefficients. These are merely fitting parameters to experimental curves and large
datasets of values exist.

A fragrant example
Suppose you had a mix of 8 fragrance molecules and wanted to know how the high, medium and low notes
(high, medium and low volatility components) changed over time. The app shows an idealised version:

[Link]

The ethanol, in red, has gone almost instantly. The limonene has all gone by ~14 min, the eucalyptol by
42min and the rest linger longer.

The app uses the Antoine Coefficients for each of the 8 molecules (chosen from a list of 40+) and
calculates the individual vapour pressures, VP, as:
 AB
log10 (VP) = AA − AC + T

The 3 coefficients, AA, AB and AC are combined with the temperature T. The largest collections use
values based on mm/Hg and °C temperatures. If you use values from different tables you have to be careful
of the units.

You often find that different tables contain different constants for the same molecules. If you go to the
trouble of plotting them, you find that the values are indistinguishable. As mentioned above, Antoine
Coefficients are merely fitting constants and within inevitable experimental error, different authors decide
on different constants to achieve the same end result.

It’s not ideal

Real liquid mixtures are non-ideal so the vapour pressures will increase or decrease according to the
activity coefficients at any given % mixture. The activity coefficients of binary mixes can be calculated
with reasonable accuracy using UNIFAC:

[Link]

For example, at 50:50 mole fraction of ethyl acetate and limonene, the ethyl acetate has an activity
coefficient, γ, of 1.27 and limonene 1.16, so ethyl acetate will be more present in the gas phase by a factor
of 1.27/1.16 ~ 10%.

Beyond binary, you need more sophisticated software such as COSMO-RS. No attempt to introduce non-
ideality was made in the previous app.

316 FST
 Flow Atomization

Flow Atomization
Links

This is a very short chapter

No useful theories
The first theory of multi-droplet formation from a jet of fluid, the Rayleigh instability is ~150 years old.
Work from Ohnesorge and others in the 1930s sketched out a landscape of stability and instability in flows
and we have two ways to look at dependencies on dimensionless numbers such as Reynolds and Weber:

[Link]

In the right hand graph of Ohnesorge versus Reynolds we see that as we go from low to high Re we go
from dripping (Rayleigh) to orderly breakup (Sine wave) to disorderly breakup (Waves) and, finally, to
atomization.

Vρl
Because for flow velocity V, characteristic length (nozzle diameter) and viscosity η, Re = η , we reach
the obvious conclusion that higher velocities and lower viscosities help atomization. Viscoelasticity risks
producing strings rather than drops, but again that’s rather obvious.

After that, we are without useful guides. Although formulae exist for some core drop sizes, they offer no
insights into drop size distributions. Various fancy computational techniques exist for modelling
atomization but these are of no practical help to the formulator.

Your choice of atomizer is constrained not by theory but by custom and practice and cost. For a simple
hand spray pump a simple jet nozzle is fine and for spray drying a rotating co-air-flow nozzle is often
required. In general we want to encourage the wave-to-atomization transition by imposing extra
perturbations such as swirling motion or ultrasonics. But these extra effects are even less amenable to
meaningful calculations so there is little the formulator can do to influence the outcomes other than reduce
viscosity and viscoelasticity.

Even something as seemingly simple as a trigger gun spray for pharmaceutical or home use is complex
physics and engineering (and intellectual property). When you start to squeeze, nothing should come out
(otherwise it would be a dribble), then the full spray should emerge … till you stop pressing and the spray
must cut off quickly rather than fail as a dribble.

For “normal” sprays, with “normal” fluids, there are a few rules of thumb. Let’s have a reference drop of
size Dr in pure water with surface tension, σ, 72 mN/m, density ρ of 1 and a viscosity, η of 1 cP, sprayed
with a pressure of Pr. So the new diameter Dn with surface tension σn, density ρn and viscosity ηn is given
by:

Pn −0.3 ρn −0.3 ηn 0.2 σn 0.5

Dn = Dr ( ) ( ) ( ) ( )
Pr 1 1 72

I apologize for this short chapter, but 150 years of research has not resulted in anything else that we can
use.

For many years I’ve known an expert on atomization. As an expert, he can draw on his experience
and use his array of complex computational tools. Every few years I would ask if there are any
useful new theories that could be used by formulators without his specialist knowledge. I asked
again while writing this chapter and the answer is still “no”.

Trigger mechanisms more sophisticated build up pressure before opening then nice spray then quick shut-
off

318 FST
 Flow Basic viscosity

Flow Basic viscosity

Links
Dishwashing Liquids, Hair Conditioners, Hair Shampoos, Sun Screens, Soaps and Washing, Laundry
Liquids, SkinCare, Surface Cleaning, Lipstick, Mascara, Pharma Formulation, Microencapsulation

The basics of viscosity are easy to grasp. We just need to be sure we’re clear about them before
starting on more complex aspects.

Basics of viscosity
We have some fluid fixed at the lower surface and on top
the fluid is moving with velocity V thanks to a stress τ,
which is force/area, N/m² or Pa. The thickness of the fluid
is Y. We can imagine the fluid as individual small layers
(technically called streamlines). The stress, τ, needed to
keep the fluid moving increases with V (obviously) and
decreases with Y – because the bigger the gap between the stationary and fast fluid, the easier it is to move
it. We can therefore say that:

ηV
τ= Y

̇
The constant of proportionality, η, is the viscosity. The V/Y term is the shear rate γ so we write that

̇
τ = ηγ

This equation was derived by Newton in the 17th century so this is “Newtonian viscosity” where η is
constant with shear rate. For many systems, η is shear rate dependent, as discussed here Flow_Shear
dependent.

m/s Pa
The units of shear rate are m which are simply 1/s. So viscosity is 1 / s = Pa.s. For a low viscosity such as
water ~0.001 Pa.s we either say 1 mPa.s or 1 cP where cP are centipoise where 1 Poise is 1/100th of a Pa.s.
It is very easy to get confused between Poise and Pa.s so try to avoid Poise. But cP are fine given that they
are the same as mPa.s.

To get a feel for the different components you can use the app:

[Link]
If viscosity is so easy …
… why is rheology so complicated? Because our systems are complicated.

We have short and long polymers, tangled and un-tangled, we have particles with weak and strong
interactions with themselves and the polymers, we have different aspect ratios … and as we apply shear all
sorts of different things can happen.

So our basic intuitions of viscosity take us only so far. After that we need Flow_Shear dependent and we
need Flow_Oscillatory rheology.

320 FST
 Flow Couette Flow

Flow Couette Flow

Links
Pharma Formulation

If you have a stationary fluid next to a moving wall then the fluid gets dragged along by Couette
flow. The opposite effect, forcing fluid through a stationary gap, is described by Flow_Poiseuille
Flow.

Couette Flow
We have a stationary wall and, at a distance
h, a wall moving with velocity V. It turns
out that the velocity profile is linear, so at a
Vx
distance x from the base, v = h

V
The shear stress, τ, is constant, simply h

Why do we need to know about Couette flow? Because there are times when such flows, for good or bad
reasons, happen in your system. And because the Couette viscometer relies on the effect.

Couette Viscometer
A cylinder of radius a and length l spins at speed ω inside a cylinder of radius b containing
a fluid of depth greater than l. The viscous drag requires a torque τ to overcome it. From
this torque we can calculate the viscosity η via:

4πηlωa2b2
τ=
b2 − a2
Flow Creep and Relaxation
Links

Any material placed under load can show elastic deformation, which reverses when the load is
removed. What we usually don’t want is for the material to creep, to extend steadily under load so
that it’s permanently deformed. Here we see the theory of creep and its partner, relaxation.

Relaxation and creep

[Link]

We have two springs, E1 and E2 which, under load, simply

extend and would return to their original position if the
load is removed. The “dashpot” next to E2 is a device that
slowly extends, irreversibly, under load. A dashpot is a
purely viscous behaviour so doesn’t class as creep.

The right-hand graph shows creep – the strain (%

extension), ε, increases slowly over time under a constant
stress σ0. After 150s the stress is removed and the system partly returns, this is relaxation, then after a
holding time of 50s, the stress is reapplied to give more creep.

The left-hand graph shows relaxation. An initial strain, ε0 is applied which creates an initial stress which,
over time, relaxes. As with the creep, after 150s the strain is removed and the system starts to partially
recover.

It’s well-known that the creep can be described by a Kelvin-Voigt model of E1 and the dashpot in parallel.
While the relaxation can be described by a Maxwell model with E2 and the dashpot in series. Unhelpfully,
 Flow Creep and Relaxation

it is usual to describe either Kelvin-Voigt or Maxwell even though real materials show both types of
behaviour. The app uses a combined model with 5 parameters. Although we have a simple E2, instead of a
simple E1 we have a “strain hardening” spring, in which the modulus gets higher with extension, involving
an extra parameter E11. And instead of a simple dashpot with a single parameter, we have A which controls
the frequency response (the typical dashpot parameter) plus a non-linearity term b.

The relevant equations are all shown in the app.

Complex behaviour
All materials creep and relax over some appropriate timescale. The Deborah Number D:

[Link]

trelax
D= tmeasure

tells you whether your timescale is relevant. As the prophetess Deborah said: “The mountains flowed
before the Lord”, meaning that over a sufficiently large timescale παντα ρες (panta res), everything flows.

So if you measure your creep over 1s, 1min, 1hr, 1day, 1 month or 1 year you might see very different
behaviour. As making measurements over 1 year is a bit tedious, you might want to do them at a higher
temperature and use WLF Flow_TTS-WLF to recalculate at lower temperature, reflecting the fact that a
mountain flows in human timescales if it gets hot enough to be a volcano.

Interconversions
But what happens if you can’t conveniently do a creep test on your rheometer or DMA? Maybe you can
measure relaxation or G'/G". One of the wonders of rheology is that you can interconvert between
measurements.

I had no idea about this. I was watching one of the excellent TA Rheometer webinars when the
speaker said “And of course you can use the software to interconvert between all the modes.” I had
to stop the video and replay it to make sure I’d heard correctly. Science is wonderful.

If you have relaxation data you can convert it to creep etc. An app lets you see that in practice. The app lets
you explore a wide variety of datasets. Your rheometer software should be able to do your conversions for
you.

FST 323
[Link]

It’s common to assume that we should just read the latest reviews and papers to get the best
science. But I simply could not understand how to do these interconversions. A polite request for
help to one professor got the short reply: “My job is to create good maths equations, not to explain
them to people like you”. Finally I went back to the 1980 3rd edition of the 1961 book by Ferry.
There, the explanation was clear and simple.

324 FST
 Flow G' and G''

Flow G' and G''

Links
SkinCare, Lipstick, Mascara, Water-based Adhesive, Adhesive, PSA, Solvent-Based Paint

For some reason, the most useful measures in the whole of rheology, G' and G" create fear and
confusion, especially when you add tanδ. In fact, they are easy, and you get a lot of information
from measurements that your rheometer can routinely make. The app makes it simple to see what’s
going on.

Elastic and Plastic

We know that under many circumstances (e.g. low temperatures and high speeds) materials can act
elastically: as the stress (force per area) increases the strain (fractional extension) increases, and as the
stress returns to zero, so does the strain.

Under other circumstances (e.g. high temperatures and slow speeds) materials can act as purely viscous
materials – as stress increases the strain increases but as the stress returns to zero the strain remains the
same. This is plastic deformation.

Equally obviously, materials under intermediate conditions will show some elastic and some plastic
deformation.

G' is simply the measure of the purely elastic (shear) modulus and G" the measure of the plastic/viscous
G ''
modulus. And tanδ = just tells us whether things are more (tanδ > 1) or less (tanδ < 1) plastic. Let’s
G'
see that in an app:

[Link]

We have an oscillating stimulus, in gold, which produces a response, in blue. The magnitude of the
response is not what concerns us, it’s just some value of 0.5. What’s important is that the response is
exactly following the stimulus, the angular difference between them, δ = 0. This is a purely elastic
behaviour so G' = 0.5, G" = 0 and tanδ = 0.

Now we have a purely viscous sample where the response angle δ is 90° out of phase – meaning that the
peak of the stimulus corresponds to a 0 value in the response, and a 0 of the stimulus corresponds to a
maximum in the response. G' = 0, G" = 0.5 and tanδ is infinite.

Before continuing, why is a viscous response 90° out of phase? At the peak of the stimulus, the velocity of
the oscillating motion is zero – the motion is changing direction. Viscous stress is viscosity times shear rate,
which is zero. At the zero-crossing of the stimulus, the motion is fastest, so the viscous stress is largest.

Now we have the intermediate state where there are equal amounts of G' and G". Their total, G* is still 0.5
but their individual values are 0.35. That’s because G*²=G'²+G''². Now tanδ = 1 as there is an equal
amount of elastic and plastic deformation.

That’s it. You now know what G' and G" are and that tanδ isn’t scary.

[Why is it called tanδ? The G*² equation gives a clue. We’re talking Pythagoras and right-angled triangles,

326 FST
 Flow G' and G''

with δ being the angle, G'' the opposite side, G' the adjacent side and tan=opposite/adjacent.]

Although G' and G" are easy to understand, measuring them requires plenty of thought. The ideal
setup is one where you have a method designed for your sorts of materials by your internal
rheology expert. After training in use of that method (sample preparation, machine settings) you
can then run your own samples.

Seeing what’s going on

[Link]

The specifics of this graph are not what’s important. Instead think of this as the generalised behaviour of a
polymer going from low to high T. At low T it’s a very high modulus, both for G' and G". As it’s warmed it
reaches a rubber plateau where G' remains constant and G" steadily decreases, so tanδ also decreases. Their
values cross at ~-10° in this plot and that is the conventional signal of Tg, the glass transition temperature.
As we exceed Tg, the plastic modulus continues to decrease … until the polymer starts to approach its
softening point. The plastic component starts to increase and crosses at the point when the elastic
component finally fails as the polymer melts. In the end all that remains is the viscous component.

But you can look at the same graph in another way. You can keep T at 25°C and think of the measurements
taking place at high frequencies or short timescales (on the left) and super-low frequencies, long timescales
on the right. Because of Time-Temperature equivalence (see Flow_TTS-WLF), both views (or any
combination) are correct. At a short timescale, a polymer cannot react to a stress, so it behaves as if it was a
frozen lump of polymer measured at a normal timescale … and so forth.

This one graph teaches you a lot about what’s going on inside the polymer, and behaviours such as
brittleness or creep can be readily imagined. That’s the power of this way of looking at materials.

It turns out that although plots of G' and G" are very common, there are two difficulties:

1. There are different ways of plotting them. I like low T to the left, high T to the right, so G' is
highest on the left. Others prefer low frequency on the left, high on the right, so G' is highest on
the right. Sometimes you have to find the legend to tell you which is G' and which is G", and
sometimes tanδ is also plotted, usually with a Y2 axis on the right. You can therefore expect to be
confused when someone shows you a rheology plot. It is entirely acceptable to say “Wait a

FST 327
 minute, I need to orient myself and work out what I’m looking at.”
2. Although the basics aren’t hard, working out what’s happening, and why it’s happening, isn’t so
easy. Again, it is acceptable to say something like “Wait a minute, are we saying that the plastic
component is doing this at this temperature? Wouldn’t that mean that Y is happening?” Your
initial interpretation might be wrong, but you’re starting a conversation that leads to deeper
insights.

It’s not just you who has these problems. We all do. So take your time, ask questions, and enjoy discovering
new insights.

Sometimes you have two samples with very different behaviour and run the rheometer to see what G' and
G" differences there are – and discover that the two samples are nearly identical. This sort of negative result
is as useful as a positive one. If clear differences aren’t visible in low-amplitude oscillations then you need
to look at other parameters to explain the performance differences.

To get the data you use Flow_Oscillatory rheology.

E’ and E’’
So far, our sample has been sheared – maybe in an oscillatory rheometer or a DMA (Dynamic Mechanical
Analyzer) in shear mode.

You can equally use a DMA in tensile mode, and determine the tensile elastic and plastic moduli, E’ and
E’’.

The values can be interconverted by taking into account the Poisson ratio, ν, which is by how much the
material will shrink in one direction if stretched in the other. For a typical polymer ν = 0.3 and for a perfect
rubber ν = 0.5:

E
E = 2G(1 + ν) and G = 2(1 + ν)

Who cares?
Many chapters of this book refer you to this chapter. G' and G" behaviour, the elastic and plastic behaviour,
of our formulations are often key to unlocking puzzles of performance. The unnecessary fear of rheology
and, perhaps, the historical difficulties of using older-style rheometers have caused formulators to avoid the
topic. Although rheology can be complex, and although formulators don’t have to be rheologists,
understanding and using the insights of G' and G" is something formulators should embrace.

A project at a megacorp was going nowhere. Every time they fixed one problem, another one
appeared. This was because there were trade-offs between G' & G" about which they had no idea
because they didn’t seek any rheological help. When, after much persuasion, they went to their
rheologist, the results quickly showed why they’d kept failing and where the sweet spot was likely
to be. After months of well-intention formulation tweaks without guidance from G' & G", within a
week they had a working formulation.

328 FST
 Flow Ohnesorge

Flow Ohnesorge
Links
Inkjet Inks

Physics imposes severe restrictions on our ability to spray or inkjet. For example, there is only a
narrow window for an inkjetable formulation, defined by a few dimensionless numbers, Reynolds,
Weber and Ohnesorge. Inside, you get great drops, outside you don’t. Or if, for example, we want
atomisation then we need to be in a different part of dimensionless space.

The small inkjet window

Given a typical inkjet head, you can’t print a good dot if the viscosity of the ink is greater than ~ 20cP or
less than ~ 5 cP. Your surface tension can’t be too high or too low. This knowledge is encapsulated via 3
dimensionless numbers, shown in the app:

[Link]

When the drop of viscosity η, surface tension σ, velocity V, density ρ and characteristic length l numbers
are:

Vρl
1. Reynolds, the ratio of inertial to viscous forces: Re = η
V2ρl
2. Weber, the ratio of inertial to surface tension forces: We = σ
η
3. Ohnesorge, the tendency for the drop to stay or fly apart: Oh = √lρσ

The left-hand graph shows the relevant regions in terms of We and Re. Typical trade-offs are when the
viscosity is too high to make good drops or, in the other direction, satellite drops are created too easily. The
right-hand graph shows them in terms of Oh and Re. The line between sine-wave break up (what we want
for inkjet) and wave-like breakup is the one defined by Ohnesorge himself.

Atomization
The forces involved in inkjet are modest – as you would expect when you need to make millions of
identical little drops. To get Flow_Atomization requires significantly larger drops and higher velocities,
hence atomizers are much more about big pumps and high powers.

Sadly, if you go to the Atomization chapter you will discover that beyond this simple generalisation, there
is little appable science to help us to spray better.

330 FST
 Flow Oscillatory rheology

Flow Oscillatory rheology

Links
Deodorant Sticks, SkinCare, Toothpastes, Lipstick, Mascara, Water-based Adhesive, Adhesive, PSA,
Solvent-Based Paint, Pharma Formulation

Everyone is comfortable with classic, rotational, viscometry, Flow_Rotational rheology, including

complex phenomena such as Flow_Shear dependent, yet oscillatory rheology, which is so
powerful, is regarded as difficult and obscure – something for rheologists, not formulators.

Well, embrace oscillatory rheology. It’s not difficult and it provides lots of formulation insights for
relatively little work.

Oscillations
We have the liquid trapped between a lower plate and, in this example, a cone. An
oscillatory stress is applied to the cone, and the strain, the amount of movement, is
measured. Or an oscillatory strain is applied, and the stress needed to create that
strain is measured. There are great debates about which type of machine is better, but
the answer is that the best machine is the one you have, and with modern technology
the differences between the two are not so important.

The oscillations give modest strains, maybe just 0.1%-0.5%. If you try for smaller
strains, the signal becomes too weak, and larger strains can take you into the non-
linear region where analysis gets difficult. Experts are happy working with LAOS (Large Amplitude
Oscillatory Shear), but most of us can’t cope with it.

What do we learn?
As described in Flow_G' and G'' the stimulus (either stress or strain) provides a response (strain or stress)
which we can interpret in terms of the elastic response, G' and the plastic (viscous) response G". In the
image we have an equal fraction of each.
It’s great that with just a few seconds of oscillations we can work out the complex behaviour of our
formulation. The fraction of elastic and plastic response tells us a lot about, say, entanglements. But if we
are going to the trouble of setting up these measurements, we should try to extract the maximum value.

Temperature equals Time

A modern rheometer allows us to do these oscillations at a range of temperatures, thanks to a Peltier plate
that might easily go from -20 to +100°C. As you change the temperature, the magnitudes and relative
proportions of elastic and plastic flow will change. This gives you valuable insights into what your
formulation is doing at a fundamental level. At low T the behaviour will be more elastic, as motion
becomes “frozen” and at high T it will be more plastic. The crossing point where G' = G" is typical of the
Tg of the formulation – the glass transition temperature above which large motions become possible.

The rheometer also lets you do these oscillations at a range of frequencies, typically from 0.01 up to 100 /s.
Now you can see the dynamic behaviour of your formulation. Typically it will show plastic behaviour at
low frequency, because the polymers and particles have time to move. At high frequency it will behave in a
more elastic manner as the system doesn’t have time to move internally.

The paragraphs on temperature and on timescale show similar effects. Indeed, TTS (Temperature Time
Superposition) or TTE (Temperature Time Equivalence) (they mean the same thing) is a profound aspect of
physics and can be captured across a wide range of materials and formulations via the WLF (Williams,
Landell, Ferry) formula, Flow_TTS-WLF.

By using WLF to combine sweeps done (intelligently) at a range of temperatures and frequencies, you can
work out the G' and G" behaviour over a wider range of temperatures or frequencies (it’s your choice) than
is possible to measure on any practical rheometer:

332 FST
 Flow Oscillatory rheology

[Link]

The app simulates the behaviour of a classic pressure sensitive adhesive over a range of 240°C and 10-5 to
107 Hz, even though the real measurements were made over a much smaller range.

Variations on a theme
Suppose you apply a paint to a surface with a brush. This is a high shear, high speed process for which
oscillatory rheology is not much use.

But you now want to ensure that any lines in the coating will level out and that paint on a vertical surface
won’t sag, i.e. flow down as imperfect waves in the surface. You can use careful low shear rate rotational
viscosity to pick up things like yield stress (Flow_Yield Stress) and thixotropy (Flow_Thixotropy), see
[Link] but you can equally look at G'
and G" as there is going to be a strong link between levelling/sag and these parameters. But paints take time
to build up their structure after rapid processes, so a conventional G'/G" measurement might be looking at
the wrong behaviour. Instead, you give a strong rotational stress to the paint in the rheometer then switch to
oscillatory mode and watch how G' and G" change over time. A swift recovery to large G' values will
indicate a low tendency to sag, but also a poor levelling behaviour. Slow recovery will encourage levelling
but promote sag. You aim for a compromise.

You can model this recovery process as a way of looking at thixotropy:

FST 333
[Link]

Rotational or Oscillatory?
Historically, formulators have used rotational viscometry. The machines are easier to use and ideas such as
shear thinning and thixotropy can be explored without too much difficult. But this is the 21st century when
equipment is more affordable and usable. So the clear answer to the question of Rotational or Oscillatory is
“Both” or, maybe, “A rheometer because it can do both”. Each mode provide deep insights into your
formulations. If you only have a viscometer, you are missing out on oscillatory insights. With a rheometer
you can gather both types of data and use their relative strengths to get a full picture of what’s going on
inside your formulations.

Oscillatory rheology is not too hard to understand, but there are many subtle details needed to get
good data from a rheometer. The ideal setup is one where you have a method designed for your
specific needs by your internal expert. They know how to answer questions like: Do you use plate-
cone or plate-plate? Do you put sandpaper on a plate to stop sample slippage? What’s the best
strain to choose? What’s the optimum sample thickness? How do I trim off stuff oozing out from
the plates? After training in use of that method (sample preparation, machine settings) you can then
run your own routine samples.

334 FST
 Flow Particle Viscosity

Flow Particle Viscosity

Links
Toothpastes, Microencapsulation

The particles/pigments in formulations have a big effect on viscosity. The weight fraction is not
relevant to these effects, it is the volume fraction, φ.

When φ is low, say up to 40%, the effects on viscosity are not large – unless there are specific
particle-particle or particle-polymer interactions. The effects are described with the Dispersion
Viscosity app which applies to solid particles and to emulsion particles.

At higher φ values the viscosity is dependent both on shear rate and on subtle particle-particle
interactions.

The topics are discussed in their respective Dispersion chapters, with links provided below.

Low-to-Medium φ

[Link]

You can read the full story at Dispersions_Rheology (Low shear)

High φ
[Link]

You can read the full story at Dispersions_Rheology (High shear).

336 FST
 Flow Poiseuille Flow

Flow Poiseuille Flow

Links
SkinCare, Pharma Formulation

When you push a fluid through a narrow gap of size h that you are fighting a 1/h3 (or 1/h4 if it’s a
pipe) dependency on the pressure required. Going from 100µm to 99µm makes little difference,
going from 2 to 1µm increases the back pressure 8x or 16x. If one of the walls is moving then we
have Flow_Couette Flow.

The Poiseuille Equations

The pressure, P, needed to create a flow Q of a fluid with
viscosity η through a pipe of radius h and length L is:

8QηL
P=
πh4

And for a coating flow between a gap of height h

12QηL
P=
h3

In both cases the effects are, as intuition suggests, linear with the required flow, the length and the viscosity,
so doubling any of those merely doubles the pressure required. But halving the radius or gap gives a 16x or
8x increase in required pressure.

The velocity gradient is parabolic; at a distance x from the wall, velocity is proportional to x², i.e. it goes
from zero velocity at the wall (the no slip boundary condition) to a maximum at x = h.

The impacts of Poiseuille are significant.

• If you save a bit of money by buying a 12mm garden hose rather than a 18mm version, your flow
is reduced 5x.
• If [you might like to open Excel and do this calculation yourself] you have a pump pulling a liquid
of viscosity 100cP from a tank down a 3m long 12mm pipe to feed a coater delivering 50gsm
across 1.5m at 20m/m then the pump has to pull ~0.9 bar – which isn’t a problem. If you go up to
25m/m then the pump has to pull ~1.2bar … and your coating is full of air bubbles because the
below-atmospheric pull has caused cavitation. This is a real example that happened to me.
Fortunately, the fittings on the tank and pump allowed engineers to quickly swap over to an 18mm
pipe (back pressure 0.2bar) and the production run proceeded without air bubbles.
• If you are applying a drop of adhesive it at first seems easy to squeeze it out into a wider blob to
fill the gap, but soon you find that it’s impossible to squeeze it thin enough, wide enough. This is
Poiseuille at work but is generally called Stefan’s Squeeze law Flow_Stefans Squeeze.
• Getting blood to flow down small capillaries is basically impossible because of Poiseuille. But our
bodies manage it no problem. That’s because Poiseuille assumes zero flow velocity at the walls
 (see the No Slip Boundary Condition discussion in Cleaning_Boundary removal). A capillary has
a special polymer liquid brush at the wall meaning that the blood proceeds as a plug flow –
uniform velocity – with much lower back pressure needed.

There is a wonderful paper called From Red Cells to Snowboarding: A New Concept for a Train
Track. The train track is one packed with goose down. It is a rare example of academic playfulness
combined with deep technical insight. Novel ways to overcome the no slip boundary condition are
very much needed.

Non-Poiseuille flow – Shear-Thinning fluids

The parabolic shape of Poiseuille flow arises in Newtonian liquids. For shear-thinning fluids the shape is
more complex. If we assume (see [Link] a
Herschel-Bulkley fluid with a yield stress τ0 and a power law K, n dependence on shear rate so that stress τ
is given by:

̇n
τ = τ0 + K γ

Then via a complex set of equations, described in the app, we can work out the velocity profile:

[Link]

Here we have a semi-plug flow. If you set the yield stress to 0 then you find the classic parabola.

The app mentions that the velocity profile can be measured via ultrasonics and that by reverse calculation
you can measure the rheology of a fluid in real time inside a steel pipe in production. Such on-line rheology
is very powerful because subtle changes in formulation which might not be spotted via standard QC can
show up as a batch continues flowing through production.

338 FST
 Flow Polymer Viscosity

Flow Polymer Viscosity

Links
Toothpastes, Water-based Adhesive, Adhesive, Solvent-Based Paint, Emulsion Paint, Emulsion Polymers,
Microencapsulation

You have two problems with polymers:

• Getting enough solubility to have the concentration you require;

• Having a low-enough viscosity at the concentration you require.

The first can be solved via Dissolution_Hansen Solubility Parameters. The second requires
understanding of the interaction between the MW of the polymer and Mc, the Critical
Entanglement MW. The app is a quick guide to the complexities of the topic.

[Link]

The viscosity of a polymer solution depends on its concentration and MW, of course. But the MW on its
MWt
own is meaningless because what really matters is the ratio Mc where Mc is the “critical entanglement
MW”. Some polymers tangle more easily than others (i.e. at a lower value of Mc) so for a given MW the
viscosity is higher because of the greater number of tangles.

The other factor is the “happiness” of the polymer in the solvent as judged by the Flory-Huggins χ
parameter. When it is “unhappy” (χ > 0.5) it is curled up on itself and isn’t so viscous. In a good solvent (χ
< 0.5) it expands so can tangle more, and the viscosity is higher.
The app lets you explore these basic ideas. Some of the inputs are a bit obscure and some of the calculated
values are of nerd interest only, so you’ll need to read the text of the app. There is a list of MC values for a
modest range of polymers, but its reliability and applicability to any specific version of a polymer of
interest is unknown.

A complementary app lets us have a peek at what’s going on inside the solution.

Polymer physics

[Link]

Here we have an idealised set of polymer blobs. On average there is plenty of space between them – their
volume fraction φ2 (the 2 means the solute in the solvent, 1) is 0.02 which is below the critical fraction φ*
where they are guaranteed to be touching … and therefore having the first significant increase in viscosity.
“Significant” in the eyes of polymer physicists because this is the transition into the “dilute” region which,
for them, is considered to be dangerously concentrated and crowded. We can slide φ2 to a larger value, or
we can reduce φ* by making each polymer blob larger. We can do that by:

• Increasing the MW or, in terms of the app, N the number of monomer units;
• Increasing the size of a monomer unit, i.e. using a different polymer;
• Decreasing the average bend angle, θmin, between segments so that the polymer is more stretched
out;
• Decreasing the Flory-Huggins χ to make the polymer happier in the solvent and, therefore, more
stretched out.

340 FST
 Flow Polymer Viscosity

In this screen shot we see overlap … those darker patches. That’s

because N was increased from 1000 to 2700 leading to a decrease of φ*
to 0.019, leaving φ2 at 0.02. The same result could have been achieved
by tweaking any of the other 3 parameters, or some combination of them.
The point is that in theory, and in real life, there are multiple ways that
you can start to get your polymer crowded upon itself and, therefore,
starting to build up viscosity.

Relating back to the earlier app, if, Polymer B has a monomer size b
which happens to be longer, and a φmin which happens to be smaller
than the equivalent Polymer A, even though their Ns are the same, then B will be more easily tangled than
A. This means that B’s Mc is smaller than A’s so for a given MW, B will be more tangled and viscous.

Because all this is standard polymer physics, and polymer physicists tend to give up when they are in these
zones of “high” concentration (0.02 volume fraction!), there is no extra slider that allows you to see what
happens at the sorts of real volume fractions (> 0.1) that are of interest to us.

What this means for your formulations

Our ability to formulate numerically with polymer solutions at the real-world concentrations that interest us
is inhibited by:

• The failure of polymer physics to give us intellectual tools that work in our concentration range of
interest.
• The failure of major polymer suppliers to give us MC values for the polymers we wish to use.
They give us MN, they give us MW but they don’t give us MC. This is a scandal.

What this means is that you have to adopt a do-it-yourself mentality.

Via Dissolution_Hansen Solubility Parameters get an idea of a good and not-so-good pair of solvents for
your polymer.

Create a solution at a level that seems near the solubility limit of each solvent. It doesn’t matter what that
limit is, but you must know the φ values (you need to correct for densities) at the respective limits.

Because it should be a rule to never measure a viscosity, use your rotational rheometer to measure the flow
curves (viscosity versus shear rate) across a comfortable range of shear rates, Flow_Shear dependent.

Now repeat these flow curves for stepwise dilutions of your respective solvents – that’s why we started
with concentrated solution.

If you have samples of two different MWs of your polymer, repeat this exercise with the 2nd MW.

You now have a rich dataset that will allow you to map out the key effects of concentration, solubility and
MW. Doing any of these sets of experiments for the first time is tedious and time-consuming. Not doing
such sets on a regular basis, once you’ve overcome the learning curve, is a sub-optimal approach to
formulation.

One more thing

FST 341
All this only tells you about low-shear viscosity. As we (usually) need the formulation to flow at some
point then we need a good flow curve – plot of viscosity versus shear rate and, as it’s a good idea to think in
this way, how the viscosity depends on the shear stress:

[Link]

Here the viscosity plummets at 10/s which may or may not be what you desire. How do we predict this
behaviour from the polymer-solvent properties? At the time of writing, I don’t know. So you need good
access to a good rotational rheometer to get these data.

342 FST
 Flow Rotational rheology

Flow Rotational rheology

Links
SkinCare, Toothpastes, Lipstick, Mascara, Water-based Adhesive, Adhesive, Solvent-Based Paint, Inkjet
Inks, Pharma Formulation

Rotational viscometry can often be misleading while rotational rheology provides deep insights
into your formulations. When you combine rotational rheology with the oscillatory version,
Flow_Oscillatory rheology then you gain even deeper insights.

Never measure a viscosity…

There are lots of viscometers out there, such as cups ([Link]
practical-rheology/[Link]), and simple rotating discs (“Brookfield”), neither
of which gives a genuine viscosity measurement, then there are scientifically meaningful
Flow_Couette Flow and capillary viscometers. The rheometer shown in the diagram is
working in rotational mode but can also work in oscillatory mode (Flow_Oscillatory
rheology) so having a rheometer capable of doing both is highly recommended.

For formulators the advice is clear: “Never measure a viscosity”, meaning that any single
measurement can be misleading. You can have two formulations with the same, measured, viscosity under
your specific technique, yet they can have totally different rheological properties and, therefore,
performance as formulations. The diagram shows an extreme example. The first is a shear thinning liquid
and the second is a Newtonian liquid. If you measure each of them at 40/s rate you would say that they are
the same – which obviously they are not!

For QC a well-thought-out single viscosity test is acceptable, but you have to be sure that the sorts of errors
that appear in production can’t produce single-viscosity values that fool QC.
… Because it’s complicated
Why are viscosity and rheology so complicated? Because our systems are complicated.

We have short and long polymers, tangled and un-tangled, we have particles with weak and strong
interactions with themselves and the polymers, we have different aspect ratios … and as we apply shear all
sorts of different things can happen.

The most common complication, as implied by the shear arrows in the diagram, is shear-dependent
viscosity: Flow_Shear dependent. Then we have the complications of thixotropy, Flow_Thixotropy. We
have to remember that although thixotropic systems are shear-dependent, there is a big difference between
them. Classic shear-dependence depends only on the shear rate. Thixotropy depends on the time spent at a
given shear rate.

As the diagram also implies, we don’t just have issues of polymers, we have particles as well,
Flow_Particle Viscosity.

At low shear rates we also encounter the phenomenon of Flow_Yield Stress which can be important for
cosmetic creams and foods like mayonnaise, where, in both cases, you don’t want the fluid to flow when
sitting as a blob. In other cases, yield stress can be a nuisance as it stops fluids flowing when we want them
to.

344 FST
 Flow Shear dependent

Flow Shear dependent

Links
Deodorant Sticks, SkinCare, Toothpastes, Water-based Adhesive, Adhesive, Solvent-Based Paint, Pharma
Formulation

We know that formulations shear thin – viscosity gets lower with higher shear. A good viscometer
or, better, a rheometer can measure viscosity at shear rates from low, say 0.1/s to high, 1000/s –
called the Flow Curve. This range covers typical conditions of formulation manufacture and use.
To compare formulations you need to describe the flow curve with a few parameters. The Cross
model is convenient. It has two intuitive parameters:

1. η0 the viscosity at very low shear;

2. ηinf the plateau at high shear .

The curve from high to low is described by two more parameters:

1. n, which describes the “sharpness” of the transition;

2. α, which describes the position for the fall-off of viscosity.

With those 4 parameters for each formulation, you have a good basic description of key
parameters. Other parameters are discussed below.

Why do we have shear-dependent behaviour?

Because our systems are complicated.

We have short and long polymers, tangled and un-tangled, we have particles with weak and strong
interactions with themselves and the polymers, we have different aspect ratios … and as we apply shear all
sorts of different things can happen. The app lets you see lots of the different behaviours.

[Link]

It is instructive to select the non-log option of shear rate to see the real shape of these curves. Log
plots are often deceptive. You can also plot with log shear stress σ on the x-axis, or σ on the y-axis
something that rheologists love to do. These alternative plots are common in the literature so it is
good to get used to seeing them. Much of our fear of, and confusion with, rheology comes from
seeing the same core data plotted in different ways. These are all the same data. Use the app to get
used to seeing the same phenomena in different ways; each view has its merits.

We know that the simple viscosity relationship discussed in Viscosity Basics was derived by Newton and
fluids which obey that law are Newtonian fluids. The cause of viscous drag is familiar to those who drive in
busy traffic. If a driver in a slow lane moves into a faster lane that causes the cars behind to slow down to
accommodate the new arrival.

346 FST
 Flow Shear dependent

For ordinary molecules in liquids their random movement between

streamlines causes the same slowing down (viscous drag). When polymer
molecules are concerned, if one part of the polymer is in one streamline and
another part is in a different streamline then the polymer starts to be stretched,
causing a bigger drag between streamlines. The effect is larger if
concentrations and MWts are large enough to create tangles. However, with
faster streamlines (with V going from lo  hi) the polymer chains tend, on
average, to become more and more aligned with the flow so they cause
proportionally less drag between the streamlines. So higher shear leads to
lower viscosity in many polymer systems. For such non-Newtonian systems
the graph of viscosity versus shear rate, γ, can be conveniently described
using the Cross formula:

η0 − ηinf
η = ηinf + n
1 + (αγ)

where ηinf is the viscosity at (essentially) infinite shear, η0 is the viscosity at zero shear and α and n are
fitting constants.

Systems where the viscosity decreases with shear are pseudoplastic. Shear thinning should not be confused
with Flow_Thixotropy. Although most thixotropic systems are pseudoplastic, not all pseudoplastic systems
are thixotropic!

The same formula handles dilatant systems where the viscosity gets higher with shear. The explanation for
that is a log-jam effect where particles/polymers can't respond (reorientate) quickly enough when bumping
into each other.

The shear thinning behaviour of systems with high particle loadings is described in Flow_Particle
Viscosity.

For a more detailed look at these systems, the Flow Curves app [Link]
rheology/[Link] provides more insight. In particular you can describe the low-shear regime via a
Power Law model, used in the understanding of Coating_Levelling Theory.

FST 347
Flow Stefans Squeeze
Links
Water-based Adhesive, Adhesive

How hard can it be to squeeze a drop of, say, adhesive so that it creates a thin, even layer? It turns
out to be surprisingly hard. This means we have to find workarounds which have their own
problems.

H3 and R4
If you have a phenomenon with a linear or even squared dependence, you are probably OK. If you see a
cubic or quartic dependence, you know you are in trouble.

Here we have a drop of radius R, thickness H and viscosity µ which

we are squeezing with a force F. How quickly does the thickness
reduce and the radius increase? The effects of F and µ are linear, so
follow your intuitions. The problem is, as Stefan showed in 1874,
that the squeeze slows down very rapidly because of the H3 and R4
terms.

δH 3FH3
δx =
3πµR4

Because we are interested in both H and R, the app shows them both:

[Link]

In both graphs we see that getting a thinner, wider drop is easy at first, but things rapidly slow down
 Flow Stefans Squeeze

because of those cubic and quartic effects. When you need, e.g. for a Adhesion_Butt a thin layer of
adhesive, you simply cannot do it with a single drop.

Breaking the problem into multiple drops gives a dramatic improvement. If the individual drops are 1/5 the
radius of the single drop, then the problem is 1/625th of the original.

This, however, leads to a follow-up problem.

Joining the dots

We’ve made great progress in filling the space with

the 7 dots instead of 1 larger dot which would
already have failed. But what about those trapped
bits of air between the dots? Although there’s no app
to deal with that geometry, you can get a good idea
of how easy/hard it will be for the air in the round bubble to diffuse, under pressure, into its surrounds:

[Link]

In practice, it turns out that these multi-drop blobs of adhesive form coherent films most of the time, though
the industrial adhesives community spends lots of time debating whether drops, stripes, stars etc. are
preferred for a specific geometry.

While writing my book Sticking Together I found that I needed to refer to Stefan at some point …
then at another, then another… I’d not realised that it’s such an important part of adhesion, and for
other parts of formulation space. It subsequently turned out that the question of how many drops
you use, in which geometry, is a huge question for the adhesives industry. There seems to be no
appable algorithm about the big question which is whether a bubble will get trapped, Bubble-Gone
was the closest relevant calculation I could find.

FST 349
Flow Thixotropy
Links
SkinCare, Water-based Adhesive, Adhesive, Solvent-Based Paint

Thixotropy must not be confused with shear thinning effects described in Flow_Shear dependent.
Yes, thixotropes shear thin, but they take time to recover their original viscosity when the shear
stops. A thixotropic paint can have an otherwise unusable high low-shear viscosity because when
stirred, its low-shear viscosity is now low to make painting and flow-out of brush/roller marks easy
– but can return to the high viscosity value to avoid sag as described in Coating_Levelling Theory.

Thixotropy confuses everyone because it is hard to get a meaningful measurement.

[Link]

Thixotropy is the time-dependent change in viscosity, different from the shear-rate dependent viscosity
(see Flow_Shear dependent) that is normal for pseudoplastic systems. It is complex to study because each
time you look at it, you are changing it! In addition to this rotational rheometer technique, thixotropy can
also be studied using an oscillatory rheometer, described below.

Clearly thixotropy is important for paints and coatings. Too little and it’s difficult to control the application.
Too much and the formulation is hard to get to flow when you need it.

There are plenty of papers with theoretical treatments of thixotropy hysteresis loops. I've chosen for the app
an approach that is mathematically tractable and theoretically clear. This starts with the Flow_Shear
dependent app, using the Cross model to create the "equilibrium" line which is what you would measure if
your thixotropic fluid had been infinitely well-sheared and had no time to recover. The plot goes up to your
chosen maximum shear rate.
 Flow Thixotropy

The Up line (the flow curve when you increase shear rate) is what you would expect from your thixotropic
liquid which starts at near-zero shear (0.1/s) with a viscosity of η[Link] rate at which it approaches the
equilibrium line depends on three factors.

1. How fast you scan your shear rate - the slower you scan, the longer it takes to get to your
maximum shear rate and, therefore, the sooner (visually, in terms of shear rate) you reach the
equilibrium line.
2. The timescale for breaking up your structure, τbreak. The longer this is, the more thixotropic your
system will be.
3. This is combined with how much τbreak changes with shear rate, expressed as γ̇ power. If this is 1
then the time decreases proportionally to shear rate

In principle, your "down" curve (now the flow curve with decreasing shear rate) will be very different
from both your up curve (of course) and your equilibrium curve. This depends on τbuild, the timescale for
re-building your original structure. This is assumed to be shear-rate independent. You only see significant
effects (i.e. different from the equilibrium line) with fast scans and if τbuild is quite large.

You may be surprised that the curves don't look too much like the ones we always see in simple
explanations. The reason is that there are lots of ways of looking at the data, and lots of "equilibrium"
curves, so you have to first get your equilibrium curve right, which might need one sort of scale (e.g. log-
linear viscosity), then view the thixotropy via a log-log plot with shear stress instead of viscosity. When
testing the app against published papers I often thought the results were wrong, but then found I was using
the wrong scale in the app. Here are the same data plotted with log shear rate but with log and linear
viscosity. Neither is right or wrong, nor is it right or wrong to choose linear shear rates (not shown). The
app lets you choose whichever view gives you the most insight … and ability to compare to the literature.

Thixotropic Area

The area S between the up curve and down curve (though here it's between up and equilibrium) is some sort
of indication of the thixotropy of the system. However, the area has no absolute meaning because it
depends on scan rate and maximum shear [Link] relative area Srel which is S divided by the area under
the up curve (Smax) is said to be rather more independent of the experimental variables. Both values are
calculated for you. When you are happy with this sort of scan, it becomes routine to compare thixotropic
behaviour, probably via Srel.

Another way to look at thixotropy

FST 351
[Link]

If you use Flow_Oscillatory rheology you can analyse thixotropy by first breaking up any thixotropic
structure by a rotational pre-shear, then use gentle oscillations (e.g. 1 Hz) at low strains (e.g. 1%) to see
how long structure takes to recover.

As explained in the app, G' follows a recovery curve starting from the original G0’ via:

G ' = G '0 + Atn

You characterise your formulation via you G0’, the constant A and the power law n, though I have no
experience with how to use those parameters. My uninformed view is that in the long run this would be a
better approach to managing thixotropy, especially when issues such as levelling and sag are also measured
via oscillatory techniques.

352 FST
 Flow TTS-WLF

Flow TTS-WLF
Links
SkinCare, Lipstick, Mascara, Water-based Adhesive, Adhesive, PSA

One of the remarkable laws of physics is that in many situations time and temperature are
equivalent, allowing TTS (Time Temperature Superposition) which can conveniently be carried out
using the WLF (Williams, Landell and Ferry) equation.

WLF allows us to do many formulation tricks, so it’s worth getting to know the simple formula and
how to use it.

Polymer properties
Take a polymer and measure its elastic modulus, E’ and its tendency to move under stress, E’’. See
Flow_G' and G'' for more information, noting that G values are shear moduli while E values are tensile
ones.

How well do the two values you’ve measured, E’ and E’’ characterise the polymer? The answer is
“poorly”. We know that below the polymer’s glass transition temperature, Tg, the properties change,
making it stronger but potentially more brittle; similarly as it reaches its melting point, the polymer gets
weaker and under a tensile load it can flow gently, i.e. it shows creep, Flow_Creep and Relaxation..

But even if we knew these properties at different temperatures, we’d still not characterise the polymer
properly. We know that if we hit the polymer really quickly it can shatter – it’s brittle at those speeds. And
if we leave it under tensile load for a long time, it can creep.

The preceding paragraphs show not only that properties change over temperature and over time, but that the
resulting phenomena are the same – there’s no difference between, say, brittleness viewed by low
temperature measurements at normal speeds or by high speed measurements at normal speed. That’s at the
heart of TTS – if you measure some properties at a given temperature and a given timescale, you can work
out what the properties would be at a different timescale at that temperature or a different temperature at
that timescale.

The equation for doing this transformation is WLF:

C 1 (T − T r )
log (at) = − C 2 + T − Tr

To understand what the mysterious at is we can first see the equation in action:
[Link]

On the left is a collection of sweeps (in this case of G') done at various frequencies from 10-1.25 to 101.75
rad/s and (colour coded) temperatures. On the right are the same data linearised via WLF. You see that they
now cover a virtual timescale range of 10-23 to 10-12.5 rad/s all at the reference temperature of 47°C.

The at value is the amount by which the timescale (or its inverse, frequency) has to be changed to translate
a value at one temperature to produce the same value at the reference temperature.

One of the upsides of understanding WLF is that at a glance you can look at a rheology graph and know
whether it has been TTS corrected. No one can measure G' at 10-23 rad/s, so the value must be a virtual
value via WLF.

The constants C1 and C2 are just fitting constants, but they have some sort of meaning. C1 is the range of
frequencies spanned by the relevant data, in this case 23 orders of magnitude. C2 is the temperature range
over which the properties change by half C1. In the app a more sophisticated variant of WLF is used so the
meanings of C1 and C2 are not so clear.

Gathering and fitting WLF data

Because of the abstract nature of the WLF equation and the non-intuitive at, there is a universal fear of
attempting to gather and fit WLF data. Because the app can load any reasonable WLF dataset (provided as
a .csv file in the format specified on the app page), you might be tempted to simply use the app. And your
rheometer will also have WLF so you don’t even have to think about it – just press a few buttons.

But it is highly recommended to do at least one fit yourself in something like Excel. The simpler WLF app,
[Link] provides a link to an Excel dataset (it’s real
data, but without details of the sample being tested). You can go step-by-step through the columns to see
how the calculations are made. Finding the fitting parameters involved some trial and error (go ahead, try
your own fitting values), but again wasn’t so hard. There is no way I know to fit the parameters using the
Excel Solver.

354 FST
 Flow TTS-WLF

Here’s how that Excel dataset was created. During some work on an unfamiliar rheometer we
couldn’t find the WLF package in the rheometer’s software. In my hotel room that evening I had to
work out how to do the WLF transform in Excel. Although it’s trivial, it seemed very hard. In the
morning I presented the fit – and they also had an overnight email explaining where to find WLF
in their software. Fortunately the results were the same.

That spreadsheet has been downloaded many times from my WLF page because everyone else
finds their first WLF to be as hard as I did.

FST 355
Flow Yield Stress
Links
Deodorant Sticks, Sun Screens, Soaps and Washing, SkinCare, Toothpastes, Lipstick, Mascara, Water-
based Adhesive, Adhesive, Solvent-Based Paint, Pharma Formulation

Although we expect liquids always to flow, even if slowly, some formulations have a “yield
stress”. Small values of stress (e.g. from gravity or surface tension), below the yield stress, produce
no flow. This is perfect for fighting sag as described in Coating_Levelling Theory and for
producing a luxury feel for SkinCare products.

It is difficult to get reliable yield stress values because measurements need to be made with small
forces in delicate rheometers. What’s important is to understand the idea and find a measurement
technique that with modest effort gives reasonable values for those products where yield behaviour
might be important.

For some formulations you want them to flow under all circumstances. For others you want a blob of cream
on your hand, or some paint on a surface to remain in place, resisting the force of gravity. This resistance is
called yield stress. Yet with some modest extra stress (rubbing with a finger, moving with a paint brush)
you want the formulation to flow easily. So the yield stress must be tunable to your specific needs. This
means that you need to measure it.

Unfortunately, yield stress is hard to measure precisely and there are at least 6 ways to measure it. Why are
there so many ways? Well, experts have their different opinions and some even argue that yield stress isn’t
a real phenomenon so there’s nothing to be measured! The app text gives some explanations of the different
techniques. And don’t worry, even if the purists are right and yield stress doesn’t exist, formulators over the
decades have gained deep insights into their systems by assuming that it does exist and that they can
measure it with their choice from the different techniques:
 Flow Yield Stress

[Link]

There are two reasons for showing this app:

1. You will often find the same problem being addressed via yield stress measurements using
different techniques, so it’s good to recognise that these different graphs are all showing exactly
the same yield stress value.
2. If you are having trouble getting good values from your current technique, you might be
encouraged to swap to another one. They each have proven useful to some people in some regions
of formulation space.

So don’t worry too much about whether your technique is giving the definitive values (it probably isn’t);
find whichever technique is easiest and most reproducible on your equipment for your range of
formulations, and find the yield stress values for a variety of products (your own and competitors’), then
find a correlation with good overall performance versus yield stress. You can then tune future formulations
to these optimal values.

These 6 ways were described in an excellent academic paper (cited in the app). I had a chance to
meet the professor behind the work. I accidentally referred to him as a rheologist. “No, I’m not a
rheologist; I’m someone who is passionate about using rheology to solve problems.” I like this
distinction. I hope that FST readers will become more passionate about using rheology to solve
their own problems.

FST 357
Fragrance Activity Coefficients
Links
Deodorant Sticks, Soaps and Washing, Laundry Liquids, Fragrances, Surface Cleaning, Microencapsulation

The evaporation of the components of a complex fragrance under the assumption of ideal mixtures
is described in Fragrance_Vapour Pressure. The activity coefficient effects (non-ideal mixtures)
when “the same” fragrance is provided in different formats (shower gels, candles, hand creams …)
are so significant that “the same” fragrance is often very different.

Activity coefficients
If you measure the vapour pressure of a volatile molecule above an ideal mixture, it will be exactly equal to
the value calculated from its standard vapour pressure and its mole fraction in the mixture. If, however, the
vapour pressure was, say, 1.5x larger than expected, you would say that the activity coefficient of this
molecule at that mole fraction is 1.5.

You can build up an intuition of the sorts of activity coefficients you might expect from similar or
dissimilar molecules via the Binary Mixtures app:

[Link]

Ignore the greyed-out box that says Water-Propanol. This is limonene (1) and ethyl acetate (2) calculated
via the standard UNIFAC method.

The left-hand graph shows that small amounts of limonene (e.g. x1 = 0.1) are unhappy in excess ethyl
acetate, with the mouse readout showing that the activity coefficient, γ1 = 2.06 – there would be 2x the
amount of limonene in the vapour phase than expected from its mole fraction. As you increase the mole
fraction of limonene it is happier, so at x1 = 0.5 γ1 = 1.15, only 15% more than ideal.

Activity coefficients are non-symmetrical. When ethyl acetate has x2 = 0.1, γ2 = 1.71, it’s less unhappy in
limonene than limonene is in ethyl acetate, though at x2 = 0.5, γ2 = 1.27 so it’s more unhappy.
 Fragrance Activity Coefficients

This simple example shows that the chances of calculating the activity coefficients in a complex fragrance
are small. So why are we bothering with this chapter?

The same fragrance in different media

Suppose I came up with an exclusive “Essence of FST” fragrance and wished to provide this wonderful
aroma experience across an FST product range of shower gels, candles and hand creams. Now suppose that
the fragrance included limonene and ethyl acetate. In the candle, the activity coefficient of the alkane
limonene will be ~ 1 while the ethyl acetate will be ~ 2. In the hand cream, which we can assume to contain
plenty of polar molecules, the limonene might be, say, 1.5 while the ethyl acetate will be, say, 1.2.

If we put the same limonene:ethyl acetate ratio into each formulation, the perceived aroma would be ethyl
acetate rich in the candle and limonene rich in the hand cream.

So now you understand the problem faced by major brands when they want to establish a common
fragrance across a range of products. They need a different fragrance formulation for each product.
Sometimes this can be via adjustments of ratios within a single formulation, but for something like a
shower gel it might not be possible to accommodate the super-large activity coefficients for some
components, so the fragrance has to be “re-imagined” – to give the same overall impression without using
the same ingredients.

I once met a “nose”, one of only a few 100 people on the planet with exceptional abilities to
analyse and create fragrances. When I asked her if I could smell some ambergris she dipped a little
strip into her sample and I held it under my nose … and smelled nothing. “You have to wave it
under your nose!” Sure enough I was able to experience the amazing complexity of this
combination of ambroxide and ambrinol.

Head space analysis

Given that calculations of activity coefficients of a 20-component fragrance mixture are impossible and
then calculations on the fragrance in a candle or a hand cream are even more impossible, what can you do
about it?

Head space GC is the least bad option. Comparisons of the fragrance in its normal carrier (the so-called
neutral molecules such as dipropylene glycol), in a candle and in a hand cream would give a good idea of
how key fragrance components shift in relative amount in the vapour phase.

FST 359
Fragrance Barrier Properties
Links
Laundry Liquids, SkinCare, Fragrances, Microencapsulation

If you have a fragrance or flavour formulation you might want it to be released only when and
where you want, so you might want a barrier in place. Unfortunately, for many practical polymer
barriers, different fragrance molecules will diffuse through the barrier at different rates – so
although you might keep the majority of your fragrance, some components might disappear and,
therefore, change the fragrance or, in the case of food & drink, give rise to “flavour scalping”.

Loss via partition

[Link]

At time t = 0 we have a fragrance formulation (the same as used for Fragrance_Vapour Pressure) trapped
within a polymer barrier of known thickness and with known Dissolution_Hansen Solubility Parameters.
The laws of Diffusion_Basic Diffusion tell us that the rate at which any molecule diffuses through depends
on the diffusion coefficient, D which, for simplicity is provided as a single value, D100 for a nominal MW
 Fragrance Barrier Properties

n
= 100 molecule and where individual values are D = D100 ( MW
100
) . In the app, for simplicity n = 1. For
many polymers n = 2 and in some cases there’s a strong exponential dependence. However, fragrance
molecules cover a relatively narrow range of MW values so the value of n is not so important.

More important is that diffusion depends on the concentration gradient, so the higher the concentration
(partition coefficient) of each molecule on the fragrance side, the greater the flux of that molecule.

We estimate the partition coefficient based on the HSP Distance – so the closer the HSP values (you can
find them in the app’s source code) to the selected polymer, the larger the partition coefficient and, for a
similar-sized molecule, a faster loss of that molecule.

At a time when PLA was a relatively new polymer I had to predict its fragrance barrier properties
in a book chapter. It was no problem predicting (correctly) that it’s a good barrier for hydrophobic
limonene and hydrophilic acetic acid. I also predicted that PLA packaging would be no good for
cinnamon-flavoured products as there’s a close HSP match between the HSP of PLA and
cinnamaldehyde.

In the example shown, the polymer is more similar to a typical fragrance molecule than it is to ethanol, so
after ~20hrs what remains is mostly ethanol.

If we have 3 components where red has a small HSP Distance from the barrier
polymer, blue has an intermediate distance and green is very distant, the red
components partition strongly and are lost more quickly than the blue. The
green which hardly partitions into the barrier is the slowest to be lost. The
thickness of the barrier and even the MW effect of the different molecules are
much less important than these partition effects, which can affect diffusion by
factors of 100s or 1000s.

Real world flavour scalping

The image shows an Excel plot of ~20
components of an aqueous drink flavour stored in
an alkane polymer. After a modest storage time
(days, not the months required for product
storage), a GC analysis showed that some
components were 90% lost, others in the 70-80%
range and some not at all. The effects of this
flavour scalping were severe – the flavour was
nothing like the original. Experts in food science
were baffled.

But from the estimated HSP values of the specific molecules and the known HSP of the alkane polymer it
was easy to calculate the expected loss depending on partition coefficient (from HSP Distance) and a
correction for the MW. Although the match isn’t perfect, it was sufficiently good to confirm the root cause
of the problem.

Calculations showed that a swap to a different polymer would have significantly reduced the problem. A

FST 361
multi-layer package might also have worked. But at that stage in the project it was impossible to change the
package … so the product had to be abandoned.

The take-home lesson is that a few simple HSP calculations and a basic knowledge of diffusion science can
avoid some very expensive mistakes.

362 FST
 Fragrance Vapour Pressure

Fragrance Vapour Pressure

Links
Deodorant Sticks, Soaps and Washing, Laundry Liquids, SkinCare, Fragrances, Surface Cleaning,
Microencapsulation

Even a simple fragrance might contain 20 components, a high quality perfume might contain 100.
However many components, there is still the issue that the fragrance must change over time as the
High (or Top) notes (volatile) disappear, the Middle notes (intermediate) take over and finally the
Low notes dominate. Using a simple 8-component mix and choosing from a variety of volatilities,
you can see how things work out over time.

H, M, L
We seem to like our fragrances to hit us quickly with their High notes, then we like things to settle down to
a more stable Medium sensation, while the Low tones linger on as a memory of the fragrance we enjoyed.
One of the many complexities of the perfumer’s art is to ensure that a fragrance that is changing all the time
retains a psychological constancy. How that is done cannot be captured in an app.

Here we look at the key facts of the individual vapour pressures of the pure components, via their
Evaporation_Temperature and Antoine Coefficients. Vapour pressure translates into molar concentrations,
so we need the MW of each component to know the relative masses in the vapour phase. Then we need the
rate at which the components will be removed by air flow. That allows us to work out the instantaneous
removal of vapour components, which changes the amount in the fragrance, so that the whole system
evolves over time.
[Link]

For illustrative purposes, we have a mix with larger amounts of the volatile components (left) leading to
smaller amounts of the least volatile. You see the volatile components disappear rapidly (causing the
strange looking discontinuities in the various curves) and the low volatility components relative % increase
to the ~40:35:25% of Carvone, Linalool and Citronellal after 60min.

The app contains the Antoine Coefficients, AA, AB, AC of 40+ components that you can choose. The
individual vapour pressure is given by:

AB
log10 VP = AA − AC + T

You can find the values in the source code of the app.

When I first had to write a fragrance VP/HSP/Diffusion program I allowed the users to include 20
fragrance molecules – this seemed far more than anyone might need. My client laughed at my
ignorance. “20? 20? A simple fragrance has 40 and more complex ones can easily have 100”. I
quickly changed the code to meet their needs.

Air flows and evaporation

The details of how air flow translates into removal of solvent are described in Evaporation_Basics. The key
message is that the rate of removal is (approximately) proportional to Airflow½, so a perfume on one’s
wrist will evaporate slowly when sitting still (airflow ~ 0.05 m/s) and 4.5x faster when walking with the

364 FST
 Fragrance Vapour Pressure

arm swinging at 1 m/s.

Real formulations
The app assumes that the Fragrance_Activity Coefficients are all equal to 1, i.e. they are an ideal mixture.
For a mixture of fragrance molecules, this assumption is not too bad – they cover a modest range of
chemical functionalities. But fragrances exist in carriers and are placed into things like soaps, candles,
creams … Now the activity coefficients can be higher. Whether the relative activity coefficients are much
different within a sensible formulation (e.g. one designed for candles will use less hydrophilic molecules
than a fragrance designed for a shower gel) seems generally to be unknown.

Those with access to COSMO-RS can obtain relatively accurate activity coefficients at least for the simpler
systems and it can be useful for simulating a candle or cream environment.

FST 365
Gelling Networks and Percolation
Links
Deodorant Sticks

Many gels are formed from a low volume fraction, φ, of self-associating particles. While it’s easy
to see how a lump of particles might appear at the bottom of the tube Surfactancy_Emulsion
Creaming and Flocculation, Separation_Settling and Centrifugation, it’s not obvious how the
whole tube can be gelled via a “percolated network”.

“Particles”
Gelation is a generic phenomenon. When it takes place within a polymeric system (e.g.
Thickeners_Associative Thickeners) we don’t have much difficulty visualising a tangled polymer network.
Here we are interested in generic particles. They can be classic nanoparticles such as silica or nice
polystyrene spheres much studied in colloid science, but they can also be semi-liquid such as blobs of
emulsion polymers. For the purposes of this chapter, their nature is unimportant, because we want to
understand the phenomenon in general. As we shall see, the existence of these gels depends on subtle
particle-particle interactions described by virial coefficients so the precise nature of the particle itself (solid
or semi-liquid) is less important than the particle-particle interactions.

From ergodic to icosahedral via isostatic taking in percolation

It’s sometimes fun to get to use fancy words. But to understand gelling, these 4 words are rather useful,
even though unfamiliar to most of us.

Let’s start with “ergodic”. This simply means that any particle within our
relatively low volume fraction (it’s 15% in this and subsequent diagrams) is able
to reach any part of the container. Although any given particle might take a long
time to explore the whole container, if the system is ergodic it can reach any part.
When we first mix up our particles, we have created an ergodic fluid. Its viscosity
will be a bit higher than the background fluid (see Flow_Particle Viscosity and
[Link]
where you find that if φ = 15% the viscosity is doubled.

To have a gel, our particles must be “isostatic” a technical term from

Maxwell saying that the net force on any particle is zero – so there is a
3D balance of forces between particles. The way to achieve this is via
particles forming an icosahedral network. Given that we don’t have
enough particles to form such a network filling the whole system, it can only be formed as icosahedral
chains as implied in the diagram. This alerts us to the obvious fact that we must have particle-particle
interactions to allow such networks to form.
 Gelling Networks and Percolation

So our particles have to start forming these sub-chains. Here we have the same
15% of particles that have started to create icosahedral sub-chains. [Yes, with x-
ray diffraction you can pick up the signal for these icosahedra]. The viscosity will
be somewhat higher as these sub-chains might interfere with each other’s motion,
but it’s certainly not a gel. Obviously this setup is not very isostatic because all
those free ends aren’t experiencing a balanced force. The shift from “viscous” to
“gel” appears if we leave the system alone, allowing time for ends to meet,
producing long chains.

Finally, the sub-chains have assembled into a percolated structure. This term is
confusing. As explained in Dispersions_Rheology (High shear) and
[Link] when you go
above φ = 28% for spheres (or less for ellipses), there is a (statistical) path of
particle-to-particle contacts across the container and viscosity starts to increase.
This definition of percolation is respectable and much used. For gelation we need
a different definition, equally respectable and much used, which is that there is a
path from any particle to any other particle. In both cases, the container is
spanned and the viscosity is higher but simple particle systems only achieve the 2nd type of percolation at ~
75%. The power (for good or bad) of percolated gels is that this 2nd type is achieved at much lower volume
fractions – because of the particle-particle interactions and the icosahedral chains.

Why should you care?

It’s nice to have fancy words, but do they help us either achieve gelation when we want it, or avoid it when
we don’t want it?

My personal answer to the question is Yes! I spent a miserable few weeks totally confused by
gelation and percolation. Nothing made any sense. Everyone waved their hands but no one
provided a formal view … till I found the right technical language and could construct my own
diagrams to make sense of it all.

To avoid gelation, the trick, obviously, is to have minimal particle-particle interactions, using a good
dispersant (Dispersions_ODC). And to get a good gel, the opposite trick is to avoid large particle-particle
interactions that will lead to lumps of stuff flocculating out of the liquid.

Now imagine that you have a nice, un-gelled dispersion for some application. You check it after 1 hour and
it’s still nice. You come back the next day and it’s a gel. It’s easy to imagine that something new happened
to cause the system to gel. Maybe the pH changed, maybe there’s some subtle chemical reaction at the
surface of the particle. More probably it’s simply that the step-by-step build up of icosahedral chains takes
time, and viscosity changes are relatively minor so you assume that nothing is happening. If you check the
viscosity, the shear of the measurement might well break up the sub-gel chains so you will always confirm
that everything is OK, even though it’s not. We’ll come back to the timescale of full gel formation once
we’ve explored a little AHS theory.

FST 367
What % particle do you need for gelation?
There is a respectable theory called Adhesive Hard Sphere (AHS) which assumes that spheres go from zero
to strong interaction when they touch – a simplistic idea that works well-enough. The strength of the
particle-particle interaction at contact is described by a “reduced virial coefficient”, B*, which defines a
1
Baxter temperature, τ = 4(1 − B * )
. Negative B* values imply particle-particle attraction. We can see how
this attraction affects gelation because the critical percolation threshold, φc is related to τ via:

1 − 2φC + 19φ2C
τ= 2
12(1 − φC)

We can, connecting φ to B* via τ, see the % particle needed for

strong percolation. Although most of us won’t know our B*
values (should you want to do so, you can measure them using
light scattering dependency on φ), it’s interesting to see that the
percolation volume for gelation can readily reach just a few %
with adequate particle-particle interactions – which is either good
if you want gelation with low additions of particulates, or a
problem if you had hoped for a relatively low viscosity particle
dispersion below the 28% threshold for the weaker form of percolation.

Another way to think of the percolation threshold is via the mean cluster size, S. If we have a volume
1
fraction, φ, less than φC, then S = 2. In other words, when φ = φc, S is infinite, every particle is
(
1− φ
C
φ
)
connected to every other particle.

Timescale
As particles assemble into larger clusters, their “relaxation times” (a proxy for their size) increase according
to a t0.57 dependence, a curious value because these are these are fractal clusters rather than pure spheres
which would give a classical square root effect, t0.5. So to double in size takes somewhat under 4x the time;
growth is relatively slow. Once critical clusters have assembled then the network grows in a t1 timescale,
relatively quickly. That’s why these gelations can be confusing. Nothing much happens for a long time,
then obvious gelation seems to come out of nowhere.

It’s even more confusing than that. There is another type of timescale. The beautiful academic work that
resolved those timescales took hours of careful measurement in idealised conditions. We generally want our
gels to form much faster. If the timescale for forming particle-particle bonds is short then a non-optimal gel
will form quickly, without the chance to build up an isostatic network from repeated making/breaking
during diffusion of particles and clumps of particles. We might be lucky and the non-optimal gel works fine
during the product lifetime. Or we might be unlucky and, when subject to some shock, the gel might
collapse.

Why stop now?

Conveniently, our particles assembled into a percolated gel network of those icosahedral strands. Although

368 FST
 Gelling Networks and Percolation

we’ve agreed that having particle interactions that are too strong can lead to a flocculated mess, there’s no
obvious reason why the particles should stop attracting each other. The annoying phenomenon of syneresis
(Gelling_Syneresis) is at least partially due to the particles continuing their self-attraction. The non-optimal
gels might be especially prone to sudden syneresis. Although syneresis is hard to understand and control,
this network chapter provides a good basis for thinking through the issues.

The fact that these gels have no reason to stop is a feature, not a bug. In general we make gels because they
are easy to break when we require – a skin cream easily spread or a soft spoonful of yoghurt. It is a feature
with downsides, which is why we have the syneresis chapter, but it’s still a feature.

Gel dynamics
These gels are fragile in both directions. They can carry on clumping, as in the previous paragraph, or be
destroyed by shear. The classic tests are via Yield Stress measurements, Flow_Yield Stress which can be
via Flow_Rotational rheology or via Flow_Oscillatory rheology. By going rotational you can also look for
Flow_Thixotropy. By going oscillatory you can use Flow_G' and G'' at different frequencies to compare the
elastic component (G') and the plastic, flow component (G"). The advantage of the frequency sweeps is that
you directly get the dynamics of the particle/gel interactions which will presumably be useful for you in
use.

That last phrase is a reminder that we’re here not to contemplate the theoretical wonders of gels, but to find
ways to create them with the desired characteristics. Let’s take two gels, say yours and a successful product
from a competitor. If preliminary feedback from internal tests is that your gel doesn’t, say, “feel” so good,
what do you do to change it? Given the complexity of gels, there’s no point blindly trying different
formulations hoping that one will hit the competitor’s sweet spot.

Instead we do the smart mapping, discussed in other parts of FST. Any such smart mapping will include
G'/G" measurements at different frequencies. They are easy to do. If, at some reasonable frequency such a 1
Hz your G' is very different from the competitor’s that will probably come at no surprise – the gels will
behave obviously rather differently if you tilt them in a tube. You can probably imagine how to increase
particle-particle interactions if you need to increase G' significantly or decrease them if your gel is too stiff.
But you are a good formulator and probably had things about right – your gel was “the same” as the
competitor’s. That’s when the frequency behaviour becomes important. The “feel” of a gel when rubbed
with a finger is connected with timescale in the 10s or 100s of Hz. When you find that the competitor’s G'
is 1/10 or 10x that of yours at some higher frequency, that gives you a clue about what needs to be fixed.

This is where smart mapping starts to help, assuming you have the habit of measuring these key parameters
on all reasonable samples. You might have tried some different formulation that showed the sort of high
frequency G'/G" found in the competitor’s product. That formulation will have failed for other reasons, but
you now have a technical clue about how to get into that part of formulation space.

Smart digitalization

Senior corporate people love “digitalization” till they survey the landscape of bad measurements of
irrelevant data carried out on under-specified formulations for no reason other than some fad for
measuring X for reasons now forgotten. They spend millions on robots that can create 1000s of
worthless datapoints … and find that ML, AI etc. cannot create understanding from worthless data.

FST 369
Smart digitialization requires the team to think of what physics is likely to correlate strongly to the product
properties. If there is a direct link from physics to property, life is simple. In complex systems like gels, we
know in advance that we don’t have the theories, the apps, the models to provide these direct insights. So
we choose a minimax, the minimum number of types of measurements (e.g. frequency sweeps of G' & G")
likely to give the maximum insights into how to steer a formulation towards success. Because we can’t
know in advance what is “good” and what is “bad”, we need the measurements on a range of known good
and bad products, internal and competitors’, to start building up a picture to feed to our AI.

370 FST
 Gelling Syneresis

Gelling Syneresis
Links
Deodorant Sticks

You have a nice gel, then over time you see a layer of liquid forming on the surface, or, in a
transparent vessel, you see the gel pulling away from the edges, giving a floating blob of denser
gel. Sometimes everything seems fine till something (active or passive) causes the gel to detach
from the walls. The effect is called syneresis (Greek for “coming together”). It is frustratingly hard
to understand and control.

What’s a gel?
The sorts of gels that are relevant to syneresis as described here involve
remarkably few “particles” (for this chapter we’re not too concerned about
whether they are solids or semi-solids) that have assembled into a network. In
Gelling_Networks and Percolation we saw how we could achieve a volume-
spanning fully percolated (all particles have a route to all other particles) with
even 1% volume fraction if the particles had sufficient mutual attraction, though
the one in the diagram is relatively unsubtle as it has 15%. In this diagram, the
particles are contacting the walls at the orange blobs. This may be simple
touching or there may be (there commonly is) extra adhesion forces to the walls. The complication of such
adhesion is a key aspect to the frustration of understanding and fixing syneresis issues.

That chapter had a warning that too little attraction (obviously) stops formation of a gel, while too much
creates a flocculated mess at the bottom of the container. There was also the warning that the particles
might want to continue to self-assemble … and the only way to do that is by removing particles from an
area with average concentration. We can see this happening in two steps

Syneresis

We start with the original diagram on the left. In the middle we see that the particles have become more
concentrated, leaving larger areas of free liquid that may or may not be visible. Whether or not there is
adhesion at the wall, the compaction will lead to some elastic stresses along those parts of the chains that
extend to the wall. Finally, on the right, the stresses pull in the rest of the chain and our gel is now much
more locally concentrated, leaving lots of free liquid, obvious to the eye. This is syneresis. If there had been
adhesion at the walls, it might take more build-up of self-association before overcoming the adhesion. For
example, in a cylindrical container, suddenly spinning it imposes a shear force at the walls and this might
be enough to overcome the adhesion and cause the syneresis to become visible.

For those studying the physics of syneresis, adhesion to the wall is an unwelcome complication. If the walls
are treated to inhibit adhesion, then the syneresis can take place more smoothly. By tuning the fluid to be
density matched with the particles, gravitational effects are removed. By using a temperature sensitive
surface treatment (e.g. pNIPAM which changes its hydrophilic/hydrophobic tendencies sharply at ~32°C)
the particle self-attraction can be tuned, allowing reversible syneresis.

Under these ideal circumstances you can spot that unlike the slow t0.57 or t1 behaviour in assembling the
gel, syneresis is fast, t2 behaviour called “ballistic motion” where the mutual self attractions cause an
efficient directional motion rather than the more random assembly processes. In the non-ideal
circumstances involving wall adhesion, syneresis is fast because of the built-up stresses across the chains.
For those who knife cheese curd, this fast response is a feature, not a bug, as it allows rapid separation of
the cheese solids.

Gravity
Take a tube with coloured, somewhat dense particles and leave it for a while. The particles settle – and no
one is surprised, see Separation_Settling and Centrifugation . Now make the fluid viscous, or increase the
fluid density via some additive. The settling is slower. Now start with particles that form a weak gel. It
settles. Is that settling or is it syneresis? I would call it settling, but there are plenty of papers where this is
called syneresis. Now watch it settle … then suddenly collapse – that probably is syneresis. The literature is
confusing because different types of gels are being analysed from different viewpoints. One take on
gravitational collapse is that the gel has its own internal strains which are resisted by the network … until
gravity causes sufficient movement for local resistance to be overcome locally, triggering a collapse of the
whole structure driven not by gravity but by the fact that the particles are in a lower-energy state when they
are compacted.

Preventing syneresis
The literature on avoiding syneresis is maddeningly vague. Against vague hypotheses of what is causing
syneresis there are vague assertions that the problem is fixed by “increasing osmotic pressure”,
“strengthening the network”, “making thicker strands”, “changing the relaxation behaviour”. If the
experiments aren’t controlling for adhesion to the walls then a “fix” might be nothing to do with the
network and only an increase of wall adhesion. There’s nothing wrong with fixing a specific syneresis
problem by increasing wall adhesion. The problem is fixing it via some presumed network effect, unaware
that it’s a wall effect.

The term “osmotic pressure” is thrown around in syneresis discussions. In some it means the
pressure from mutual attractions of the particles – a terminology that is technically true but rather
unhelpful. In others where the system is a complex gel with small pores, osmotic pressure is a

372 FST
 Gelling Syneresis

stabilising influence if the stuff in the pores (e.g. a polymer) is too large to get out through pore
openings. Others use osmotic pressure as some form of magic that causes syneresis to be reduced
or increased depending on the form of handwaving being used.

Things are worse with gravitational syneresis. Papers have little problem showing that “strengthening” the
network fixes the problem. They might equally say that adding an external viscosity agent or a density
modifier is a fix for syneresis. It’s a fix for the problem (good) but adds to the confusion about syneresis
(bad). If your gravitational syneresis is not far removed from classic sedimentation that’s OK. If it appears
as sudden (unwelcome) event, maybe that’s real syneresis.

How about applying rheology to understanding cause and effect. Again, the results are maddeningly vague.
In addition to the confusions over adhesion and gravity, the large variety of rheological measurements
makes it impossible to build a coherent picture. You can, for example, measure yield stress by at least 6
techniques (see Flow_Yield Stress ), you can measure G' and G" (Flow_G' and G'') at different oscillation
rates and different maximum strains.

Those making gels out of particles large enough to be seen in a microscope can use the particles’ own
motion (microrheology) to measure G'/G" as the gel matures. Not surprisingly, G' increases sharply relative
to G" at the gel point. Good to know in terms of gel formation but not directly applicable to syneresis.

My reading of the results is that you can find some hand waving explanation to fit the results, providing no
reliable methodology for fixing a similar syneresis problem in the future. A high G' (strong elastic
modulus) might be good in some circumstances, yet be an indication of strong interparticle forces that
might collapse the whole gel. Too much of anything is bad.

Yes, we should use rheology to probe our gels – it’s a powerful set of tools for answering questions. But we
need some proper questions to answer.

Let’s try to ask some clearer questions. This means treating the walls of the container to avoid adhesion of
the gel. If this gives instant syneresis then you already know that you either have to rely on wall adhesion in
the product or use the fact that your gel is hopeless as a starting point for re-formulation.

As an example of a clear question, let’s use the φc formula from the network chapter. We can get a gel at a
desirable low value for the critical volume fraction, φc, if there is strong particle-particle interactions. A φc
of 1% is very attractive in terms of efficiency, but that leaves a large amount of empty volume for that small
fraction of highly self-attracted particles to collapse into, with little redundant resistance to collapse from
its super-thin network. Reducing the particle interaction strength (methods to do this in your own system
should be obvious) would lead only to a non-gel, so you have to reduce the strength and increase the
particle concentration to above the new higher φc.

If you can create such systems and you then use rather too much rheology to compare and contrast them,
you might be able to extract a reliable rheological signal that relates to a reasonable hypothesis around φc.

An alternative hypothesis, from the gelling network chapter, is that the balance of timescales for particle
motion and particle bond making/breaking is wrong. By definition, a well-balanced system will allow
particles to move and to make and rearrange bonds in comparable timescales. If bonds are too weak and/or
bondmaking too slow, no gel is formed. If the bonds are strong and formed quickly then the network will be

FST 373
mechanically sub-optimal (not at all isostatic) and will contain a lot of built-in stress, so will be susceptible
to breakage under shock. This might happen with “raspberry” particles, those that are relatively rough with
high inter-particle friction.

The opposite case is rapid bond creation with weak interactions. Now a gel can form nicely thanks to the
rapid movement into an isostatic network. There is little remaining stress so the gel is quite impressive …
till a shock is imparted. If the particle-wall interactions are strong then the resulting gel is sub-optimal
because it is biased towards the walls. Some wall shear will create a disconnect and the resulting release of
stress rapidly causes a “ballistic” collapse of the gel.

There’s one more thing. The discussion so far has assumed that interparticle attractions are a constant.
Syneresis will also appear if some change in temperature, pH, chemistry or partitioning leads to a change in
these attractions that allows the network to re-distribute itself to a lower-energy state.

Conclusion
If you are exasperated at this point, well, decades of syneresis research have not led to reliable toolkit ideas.
The term is too vague, yet each person using it (or avoiding using it) has their own definition which might
make their own analyses and recommendations misleading to others.

All you can do is avoid easy, but misleading stories and do what you can to pinpoint why your system is
not resilient to the shocks and stresses that make syneresis appear. This means a few things:

• Being clear about what particle-particle forces are holding things together.
• Estimating whether they are strong or weak.
• Checking on concentration-dependence to see if you are in an unlucky zone just below a zone with
enough reliable network connections for a comfortable stable zone, or in a zone where there are
too many interactions and the whole assembly collapses in on itself.
• Thinking of likely timescales for bonds making and breaking, on whether there is high “friction”
stopping particles from moving or whether they are relatively fluid and can slide past each other.
• Using simple “osmotic” thinking for systems where there is an “inside” and “outside” in your gel
and where a component, such as a polymer, can keep a high osmotic pressure inside so the liquid
doesn’t want to escape to create syneresis.
• Checking that the syneresis isn’t happening due to some change in the balance of forces due to
pH, chemistry etc.

374 FST
 Heat Conductive

Heat Conductive
Links

Because it is relatively easy to model conductive heat flow through a single layer we first look at
an example of insulation. Then we see what happens over more general multiple layer systems.

A simple equation

[Link]

As a practical example of heat flow and what it entails, we set up an insulation panel of thickness L and
thermal conductivity K. We want to preserve some “good” temperature, TG from some “bad” temperature,
TB, where good might be 20°C in the house and 0°C outside, or 5°C in a “good” fridge and 25°C in the
“bad” room temperature, as in the screenshot. The temperature difference is ΔT

The heat, Q, in W/m², that can flow across the insulation is given by:

KΔT
Q= L

This heat flux is unhelpful without us knowing where the heat is going. Let’s assume we have a material (in
the screenshot it’s water) of thickness h. From its density ρ and heat capacity Cp, we can work out the
δT
temperature increase per second, δt , per W of heat.

δT Q
δt = hρCp

As the temperature rises, ΔT decreases, so Q decreases as does δT and the rate of heating is slower for a
thicker material, with a larger density and heat capacity.
More complex conductive heat flow

[Link]

As we saw before, the heat flow, Q, across a distance h depends on the temperature difference ΔT and on
the thermal conductivity, K:

KΔT
Q= h

In this example we have multiple layers so we have a net Q, Qnet, given by the flows in and out of each
layer:

KinΔTin KoutΔTout
Qnet = hin − hout

The rate of temperature change is given by:

δT KΔT
δt = hρCp

We can simplify the formula to:

δT DΔT
δt = h

376 FST
 Heat Conductive

K
where the thermal diffusivity is D = ρCp . Although both K and D are commonly known, D saves us
having to look up ρ and Cp for the material into which our heat is flowing. It also means that for this
4-layered app we need only 2 values to specify each layer rather than 4. There is a good range of D values
in Wikipedia.

For a multilayer problem, we need to have an infinite source of heat in at the top, at a constant TTop and we
can choose TBelow to be some fixed temperature or (in the app set it to 0 to) make it float as if it was in
contact with low conductivity air. Then given the starting temperature and thermal properties of each layer
δT
we just step through in time applying the formula for δt .

The real problem is showing the results. The screenshot looks very confusing until you start using the
mouse to discover that the graph is rainbow coloured in time – blue is short time, red is long time.

Many years ago we wanted to have a high temperature next to a thermal head and a low
temperature 50 µm away from it. Given that our heat pulses were only in the msec range, we
assumed this was easy, but our experimental results showed high temperatures in the 50 µm zone.
That’s the first time I did a thermal conductivity calculation. The result was so obviously wrong
that I spent hours trying to find the bug … till I realised that it was correct. Heat flow can be very
anti-intuitive.

Thermal contact resistance

The app assumes that the top layer is heated by a source at constant temperature able to deliver whatever Q
the calculation requires. In reality there is always a “thermal contact resistance” between, say, Cu heating
plates and the polymer surface. Interestingly this can be calculated using the app in Mechanical_Friction,
because friction depends on contact and a contact mechanics calculator is part of that chapter. The
“contact” for friction is the same as “contact” for thermal contact resistance.

I had never needed to bother about contact resistance … till I had to model heat sealing,
[Link] I had high-quality data from
the literature and my results showed temperature rises far higher than experimentally observed.
That’s when I learned the need to include thermal contact resistance in such calculations.

FST 377
Heat Convective
Links

Convective heat transfer is harder to model than conductive and radiative, but the app gives some
idea of what is going on for convection in air and in water.

A simple equation
Newton showed us a long time ago that convective cooling in a “still” environment (i.e. we don’t have
forced fluid flow) is easy to understand. The rate of heat loss per unit area, Q, for temperatures TH and TC
( )
is Q = h TH − Tc or, more compactly, Q = hΔT. All we need is the heat transfer coefficient h. We
know that it’s larger for water than for air. But getting numbers for this intuition is hard.

Rayleigh and Prandtl

Convection involves viscosity, η, heat capacity Cp, thermal conductivity, k, density ρ, thermal expansion
coefficient β, thermal diffusivity α and gravity g. Convection also takes place over a length scale L which
is, hopefully, obvious to the problem in hand. The parameters each have complicated dimensions, making
general formulae tricky. It turns out that calculations are simpler if we use dimensionless numbers, and for
convection we need the Rayleigh number, Ra given by:

ρβΔTL3g
Ra = ηα

We also need the rather simpler Prandtl number, Pr given by:

Cpη
Pr = κ

We then combine these for various convective cases. Assuming gentle convection with laminar flow we
have:

Vertical convection:

hV =
k
L
( 0.68 +
(
1+
0.67Ra0.25

( 0.492
Pr )
0.563 0.444
) )
Horizontal convection, Hot Above:

k
hHA = L 0.54Ra0.25

Horizontal convection, Hot Below:

 Heat Convective

k
hHB = L 0.27Ra0.25

App calculation
Because it is far too tedious to find all the required parameters for a specific fluid, and because the
calculations are in any case inexact, and L is not always well-defined, and because we should be using
parameters defined at the film temperature which is some average of the hot surface and cold fluid, the app
simply provides general guidance based on parameters defined at “normal” temperatures for either air or
water.

[Link]

As you would expect, Q decreases from Vertical to Horizontal-Above to Horizontal-Below:

Putting convective Q into context

Compared to conductive heat flows, convective flows are small. A good example is in heat sealing,
[Link] You can get to, say, 120°C in 0.5s in
the clamped jaws, but, as you can find in the app, in 5s of convective cooling the temperature drops only to
90°C. During that long time the softened polymer in the seal is at risk of damage. So you have to provide
rapid convective air flow (not covered in these apps) or provide conductive cooling via direct contact.

FST 379
Heat Radiative
Links

Radiative heat transfer is easy to describe in equations, with the app making it easy to put the
equations into practice.

Wavelength-dependent emission

[Link]

Any object gives off thermal energy over a spread of wavelength, λ. Planck’s law describing these
c
emissions is best described in terms of frequency ν = λ , using the speed of light, c, to do the conversion.
At a temperature T °K, invoking Planck’s constant h and Boltzmann’s constant k we have the intensity, I at
frequency ν given by:

Iν = ε ( )
2hν3
c 2
1
hν
e kT − 1

The emissivity, ε, varies from 1 for a pure radiative black body (which covers most “normal” surfaces) to
something like 0.02 for a shiny metal.

We see how they differ for our two chosen temperatures, with the emission normalised to the highest of the
two curves. The hotter one is in orange, the colder in blue.

Radiative heat flow

The total flux emitted by the each of the two surfaces is also calculated. They come from Stefan-Boltzmann
and invoke Stefan’s constant, σ = 5.67e-8.
 Heat Radiative

Q = εσT4
Because we are often interested in heat flux (usually heat loss) between two (planar) surfaces we can
calculate this, taking into account the separate emissivities as:

Q=
(
σ T41 − T42)
1 / ε1 + 1 / ε2 − 1

This calculation of total flux is done for you in the app. It’s complicated enough, but, as the next paragraph
shows, it can get more complex than that.

If, one day, you happen to need to calculate the radiative heat flux between a hot drum and some coffee
beans roasting in the drum it turns out to be a bit tricky. Q is W/m², but what is the relevant area when
roasting beans? It seemed obvious that the large internal area of the drum would be the relevant value, but
the calculated radiant heating was unrealistically high. It turns out that if you have a cylinder inside a
cylinder or a sphere inside a sphere, the area used is the smaller one, and that you have to add an emissivity
correction based on the relative areas. As the drum is a cylinder, it needed an equivalent cylinder diameter
of the beans being roasted. Knowing the mass of the beans and a typical loose density when being roasted,
the volume of the virtual cylinder could be calculated, giving results that made sense.

Should you find yourself in need of such a calculation, you can find the formula near the end of the Coffee
Drum Roasting app: [Link]

I now know some world-class baristas who are passionate about finding the right science to help
them brew ever-better coffee. Their initial contact was about espresso crema but since then we’ve
worked on many interesting problems, including how to make a cold, non-dairy latte. The coffee
roasting app evolved from what seemed to be a minor discussion about a specific phenomenon in
roasting (“the flick of doom”) and is now a powerful tool challenging much of coffee roasting
convention. It was they who demanded that I add the radiative heat calculations (a difficult
challenge!), as the relative importance versus conduction and convection in roasting is a contested
issue.

FST 381
Humidity Water mechanical isotherm
Links
Water-based Adhesive, Emulsion Paint

We can plot how the moisture absorption of a powder depends on the RH (Relative Humidity, or
aw, water activity); that’s a sorption isotherm. Less common, but of great importance, is the
mechanical isotherm – how the mechanical properties depend on the moisture content.

Tg effects

[Link]

The prime mechanical effect of moisture on the powder material itself is the effect on Tg, the glass
transition temperature. The standard equation is a sort of Fox equation ([Link]
practical-solubility/[Link]) taking into account Tg0, the Tg at 0% water and Tgw, that of
100% water which is -135°C. Where w is the fraction of water then, via a constant K which typically has to
be measured for each material (but is commonly available as tables of values for foodstuffs) then:

(1 − w)Tg + wKTgw
Tg = (1 − w) + wK

What do you do with this knowledge?

Science is universal. I was involved in some intense debates about how best to model the powder
clumping tendency of a household cleaning product. It’s a complex problem. Eventually we found
work (cited in the app) funded by Nestlé covering a similar issue in the processing of food powders
 Humidity Water mechanical isotherm

such as granules of instant coffee. This approach broke the problem into three steps, each of which
could be modelled in an app. Instead of complexity we now had clarity and comparative simplicity.

First you need to know w at a given relative humidity, and you can get that from the Isotherm app:
Humidity_Water vapor isotherm . Then, by knowing Tg you can work out how likely it is that the powder
particles will stick together, and you get that from Particles_Sintering.

This Tg approach to food processing and baking can also be found in the extended work of Louise Slade
and Harry Levine.

FST 383
Humidity Water vapor isotherm
Links
Water-based Adhesive, Emulsion Paint

When a dry sample is exposed to a low water RH (Relative Humidty) (or aw, water activity), after
some time it reaches an equilibrium of water content. Increase the RH and water content goes up.
Plot a curve, and that’s the adsorption isotherm. Now decrease the humidity in steps, that’s the
desorption isotherm. Why measure these isotherms? Because the properties of many formulations
depend on them.

Fit for purpose

[Link]

So we have a curve. The original datapoints aren’t shown (we shall see them later); instead it has been
fitted to some equation which we assume is a good fit. In this case, the equation is the famous GAB model
(Guggenheim-Anderson-de Boer ) using the three parameters M, C, K. It is supposed to be a refinement of
the BET model (Brunauer-Emmett-Teller) but it turns out that each of them makes laughable assumptions
about most relevant sorption isotherms. They assume planar surfaces that get covered with a monolayer of
water than then subsequent layers of water build up onto that monolayer. The fact that people use GAB for
analysing the sorption isotherm of, say, milk powder (that’s the curve in the graph), makes you wonder
where the monolayers might be.

The analysis via BET or GAB is assumed to give you a “surface area” defined by the monolayer coverage,
with other parameters intended to inform you of other, vaguer notions. Again, by noting that GAB is used
on milk powder, which definitely doesn’t have a planar surface and certainly does some absorption as well
as adsorption, we have to ask why people are using an equation based on such poor assumptions..
 Humidity Water vapor isotherm

What is not under debate is whether GAB gives a good fit to many isotherms. The answer is “Yes” for an
interesting reason: GAB is right for the wrong reasons.

For no good reason, the world of Evaporation_Humectants relies on the Norrish equation for fitting the
data. In fact, there are more than 80 published isotherm equations, many of which can give the same quality
of fit to the data, even though they are supposed to be based on different mechanisms. It’s a mess.
Fortunately, we have a way out of the mess.

Same data, correct fit

[Link]

There are the datapoints, fitted to the ABC model. We can fit the data to GAB and find the parameters used
in the previous model, with M called nm, C being called CB and K the same:

FST 385
Why is ABC the correct fit? The clue is in the app name – Isotherms-ST. The ST stands for Statistical
Thermodynamics which is an assumption-free way of analysing the data. The other two plots reveal key
stat therm data directly from the raw data so they aren’t “fitting”, they just say what’s going on. The ABC
values are, indeed, fitted values but (a) they are provided for convenience and (b) the assumptions behind
the fitting are minimal; no absurd ideas of monolayer filling of planar surfaces. ABC doesn’t even
differentiate between absorption and adsorption.

The meaning of the ABC parameters is described in the app and in the papers quoted there. If all you want
is a nicely-fitted curve for other reasons (which we will come to) then it doesn’t matter if you use GAB or
ABC, for the very good reason that GAB (and BET) has the same functional form as ABC. We can directly
(algebraically, it’s not “fitting”) translate GAB (or BET) parameters into real ABC parameters. That’s how
the GAB curve was created in the app.

BET surface area

Those who use the isotherms to work out the surface area of their powder will automatically choose to use
“BET surface area” because this is something that can equally be obtained via N2 at 77°K. However,
nitrogen at 77°K is arguably somewhat different from water at 25°C so maybe we should enquire about the
effective surface area available to water. Here it turns out that for many/most samples typically analysed via
BET/GAB, the values are of dubious value. ABC can provide more reliable values, some of which are,
under specific circumstances, identical to BET – again for the good reason that BET is a sub-set of ABC.
The difference is that ABC provides a measure of how meaningful that surface area is – and it’s often not
very meaningful. Let’s see it with the STSA app, Statistical Thermodynamic Surface Area:

386 FST
 Humidity Water vapor isotherm

[Link]

We are told that the STSA is 971m²/g. We are also told that this is a bogus value. Directly beneath the
STSA is an indicator of where the isotherm lies on the range from “Multilayer”, meaning that the water
builds up on itself long before there is anything like monolayer coverage, over to “Microporous” where the
water is clearly building up inside micropores, giving no meaning to “surface area”. In this case (as is very
common with food/water isotherms) we are firmly in the multi-layer zone, probably because there’s plenty
of absorption of water into the milk powder. So ABC tells us not to use the STSA as an indicator of any
sort of true surface area.

Would a finer milk powder have given a larger STSA? Probably – there is at least some portion of real
surface area in that value. But we can confidently say that pretending to know the surface areas via the
isotherm is not a good idea.

The stat therm basis of the ABC isotherm also allows cooperative isotherms to be modelled
correctly. So instead of 80+ isotherms attempting to capture IUPAC Types I to VI isotherms, there
is one core, correct, theory. Will those working on isotherms jump over to the stat therm approach?
They should, but they won’t. Scientific inertia is powerful. Now, if you are a keen BET, GAB,
Norrish … user, ask yourself if you will change. The assumptions behind them really are nonsense.
The stat therm apps are as user-friendly as possible – you can load your old data and see it afresh.
So give it a go.

Desorption and hysteresis

It is often the case that the desorption isotherm deviates significantly from adsorption, typically with the
sample holding on more strongly to the water. This “hysteresis” can be best analysed via statistical

FST 387
thermodynamics, but at the time of writing, this analysis is not published. As FST is a living book, this
section will be updated.

Temperature effects
Again, stat therm allows us fresh insights into temperature effects in isotherms and, again, at the time of
writing, this analysis is not published. Again, as FST is a living book, this section will be updated.

Using the isotherm for other purposes

Having a numerical value (via any fitting formula) for the % water absorbed at a given RH allows us to
provide at least one answer to the question: “who cares what the isotherm is?”

If we have some milk powder, what we care about is whether the powder will start to self-adhere and
become unusable both for the end user (no one likes opening or using a container full of caked milk
powder) and for the producer. If a silo containing tons of milk powder is allowed to go to too high an RH,
the powder self-sinters (Particles_Sintering) because it has become too weak to resist the pressure inside
the silo. That’s why we have a parallel chapter, Humidity_Water mechanical isotherm, linked to the GAB-
based app that opens this chapter. Naturally, the mechanical isotherm app is itself linked to the sintering
one.

388 FST
 Mechanical Friction

Mechanical Friction
Links

Friction seems so simple but frustratingly it can only be measured, not predicted from first
principles. This is because the true contact area between two surfaces, a key aspect of the frictional
force, is unknown.

The force, F needed to overcome the friction of an object mass m (times gravitational force, g) on a surface
is given by:

F = μmg
Where μ is the friction coefficient. This can be the static coefficient, needed to get an object moving or the
kinetic (or dynamic) coefficient to keep it moving, which is usually lower because less force is needed to
overcome sliding friction. It seems non-intuitive that the contact area does not appear in the equation. The
reason is that whatever causes the frictional force depends on the contact area (so doubling the area would
double the frictional force) and on the mass per unit area, which halves as the contact area doubles. The
effects cancel out – at least for “standard” friction. We see later how the assumptions can break down.

Although μ can be measured using some force sensor dragging a sample along a surface, this isn’t so useful
for measurements in real-world situations.

Instead, you have a relevant weight (its mass or contact area don’t
matter) sitting on your surface and tip the surface till the weight
starts to slide. By measuring the height, h and length, l, you can
calculate tanθ = h/l and this is the static friction coefficient!

h
μ= l = tanθ

It is interesting to take a block with length l, width w and height h, each very different from each other, and
repeat the experiment with different sides in contact. The measured tanθ remains constant however you
arrange the block.

This simple friction coefficient test is surprisingly little known. I’ve used it many times to solve
problems where friction coefficients were required but everyone thought you needed a “real”
machine, which they didn’t have. The first time I used it was to measure the friction coefficient of
cheese on different packaging films. Yes, it’s an important value for those who are frustrated by the
difficulty of sliding a block of cheese back into a packet.

The µ found at onset of motion is the static friction coefficient. If you slightly reduce h till the mass stops
moving, you can find a slightly lower h (and therefore tanθ) and immediately get the smaller kinetic
coefficient. You can stop and start as many times as you wish to check that your h values are reasonably
constant, giving you reliable μ values.

Friction coefficients
If friction was about numbers of atomic interactions across the interface, then a perfectly flat surface, such
as mica, should have a friction coefficient = 1, which it does. If forces between surfaces can be spread/
dissipated over a larger area then the assumptions behind friction theory break down, so rubbers start off at
μ = 1 and can be higher than that. If there is some difficulty in getting atoms to interact because they are
dissimilar then friction coefficients should be lower; indeed, take metals A and B: µA-A and µB-B are
usually higher than µA-B. Because nothing much likes to interact with perfluoro surfaces, friction
coefficients onto Teflon are often < 0.1.

If the surface molecules can fall apart under the frictional forces then µ will be lower. That’s why waxes
work well and, even better, why liquid layers act as lubricants – for as long as they stay trapped between
the surfaces rather than being squeezed out.

That is a lot of hand waving. If you look at large tables of A-B friction coefficients, it is hard to come up
with any more refined understanding, especially as quoted values can themselves be variable.

Contact mechanics
As discussed in Mechanical_Surface Roughness, you should never measure just the amplitude of
roughness, you should equally measure its wavelength.

Strangely, any surface can be made equivalent to one where every

element has an effective radius, Reff, and the RMS wavelength can
be measured as Δq and the RMS amplitude as Rq. There is a
relationship between the three properties, which is useful if your measurement device gives you only two of
them:

Δq =
Rq
√ Reff
From this we can calculate for a given “real” area, A0, the actual contact area, A and from that the relative
contact area ARel. It is quite shocking to see how little of the surface is really in contact:

[Link]

In this example, for a material with a modulus of 1 GPa, the actual contact area for a nominal 100 cm² area
is 7.3e-3 cm², i.e. less than 0.01%.

2F
This area is given by A = where E* is a modulus correct for the Poisson ratio, assumed to be 0.3.
E * Δq

390 FST
 Mechanical Friction

F F
The effective load rather than being σ = A0 becomes σ = A so can easily exceed the hardness of the
material. [Hardness is not the same as modulus: see Mechanical_Hardness.] By reducing the hardness from
σ
0.305 GPa in the example to 0.118, we find that the hardness is exceeded (σrel = σ0 ) and now we get
plastic deformation, meaning that assumptions about friction coefficients are no longer valid.

As mentioned in Heat_Conductive the same calculation allows you to estimate the electrical and/or thermal
contact resistance, calculated as relative Conductivity in the app.

Kinetic friction, stick slip and squeaks

It always take some time for a system to change from static to kinetic friction and back again. A surface
will have some natural waviness, and the device pulling the object isn’t perfectly rigidly coupled. Take this
all together and you get the famous stick-slip condition and squeaks from things like brakes. This is all
calculable, though it needs some heroic degree of knowledge of your system, as the app explains.

[Link]

FST 391
Mechanical Hardness
Links
PSA

Hardness is resistance to permanent marking so is an important parameter. We know that a coating

has a Mechanical_Modulus defined as elastic resistance to small stresses. Hardness measures
resistance to plastic stress, i.e. permanent flow under, say, a scratching load.

You can buy hardness measurement machines, but a simple device lets you measure hardness
anywhere, e.g. inside an oven if you want to know how hardness changes with temperature.

Hardness is a very important parameter that many of us are

unfamiliar with. This is unfortunate because it is of
importance in many situations where we might be tempted to
use the more familiar “modulus”. The definition of hardness
seems to make little sense:
Force
Hardness = Contact Area

However, a simple example helps make it clear. The same tip is pushed with the same force into materials
with the same modulus E1=E2 but with different hardnesses H1 > H2. The tip enters further into 2 and on
withdrawal it is clear that the tip has left a bigger permanent impression on the substrate - i.e. there has
been more plastic flow and less elastic recovery. Finally, we can see that the contact area with the tip is
higher for 2 than for 1, so Force/Contact_Area is less for 2 than 1, so the Hardness is lower for 2. It's as
simple as that!

[Link]

One way to measure hardness is by pushing a Vickers square pyramidal diamond tip into the surface with a
known load, F. You measure the diagonal width, D, of the mark you’ve created and calculate the hardness
 Mechanical Hardness

via:

F
H = 1.85
D2

Depending on whether F is in N and D is in m or F is in kg and D is in mm, you get H in units of GPa or

Vickers units. The app provides both and also tells you the depth of the indentation.

The graph gives you a feel for how accurate your values might be given your chosen load and your likely
errors in the measurement of D.

The problem with Vickers Hardness machines is that they are best suited for seriously hard materials such
as metals. The formulation world tends to avoid measuring hardness because the official alternative to
Vickers seems to be a nanoindenter, described shortly. And such machines can’t conveniently be used in,
say, an oven or at a customer’s facility.

In my experience there are two “good enough” methods for measuring hardness of typical polymers and
coatings, usable in a wide variety of formulation setups.

1. Attach 3 Vickers tips to a plate and, with a suitable added weight place the device on your surface,
wait ~2 minutes then measure the 3 indents after ~1 minute. Those waits allow time for (a) plastic
deformation and (b) elastic recovery. You might choose different times, but these are an acceptable
starting point.
2. Instead of Vickers tips use spherical ball bearings. Measure the radius of the indent and instead of
the Vickers equation use a Brinell sphere test.

The official “Brinell Hardness Number” BHN for a ball of diameter D giving a circular impression of
diameter d, is given by the following, along with the real hardness, H:

2F 4F
BHN = or H =
( 2
πD D − √D − d 2
) πd2

The beauty of these methods is that with 3 tips or balls you have automatic balance in applying your load,
you have 3 datapoints for statistics, and you can make your loads and devices to whatever is needed to give
quick, reliable values. You don’t even need to do accurate calculations – you can have a QC test saying “In
spec is an impression greater than X and less than Y”. The first time I used a Brinell tester I had never
heard of Brinell. I just found a plate, 3 ball bearings and a mass that reliably gave us 3 nice indents for
material we knew to be in-spec, and where some deliberately out-of-spec samples (too hard and too soft)
showed up clear differences. When we developed a different product, we simply created a different plate
with different ball bearings and mass, giving us a similarly quick and easy QC test.

Nanoindenters
A fancy version of a hardness tester is a nanoindenter. This gives lots more information, but is more
expensive and delicate, though you can do other things with a nanoindenter than just measure hardness.

FST 393
[Link]

The nanoindenter gives a full test of elastic and plastic effects and their relevant parameters. A sharp
(Berkovich) tip is pushed into the coating with an increasing force and the displacement is measured. The
force is decreased and the displacement continues to be monitored.

The app is backwards - you input the modulus, E, and hardness, H, (along with a maximum load Pmax and a
Poisson ratio, ν, typically 0.3) and the nanoindenter trace is simulated, and the extra parameters such as
elastic recovery are calculated. The idea is to give a feeling for what a real-world experiment might look
like.

As before, the calculated hardness is based on the definition of:

Force
Hardness = Contact Area

Obviously the assumptions behind the creation of the nanoindenter curve are only approximations. The aim
is to build up your intuitions of how things change, not to do hardness calculations, which your
nanoindenter will happily perform for you, once you can get hold of one.

The calculation assumes a diamond indenter of modulus 1141GPa, ν=0.07. The contact area is calculated
using standard Berkovich parameters.

It was obvious how to make a hard hardcoat – increase the crosslink density and add lots of
nanoparticles. We could increase modulus (see the neat trick Mechanical_Modulus for measuring
modulus of a thin coating) but we couldn’t increase hardness, resistance to scratches. A chance
came up to try out a (then new and exciting) nanoindenter. The expert running the machine could
see that we chemists didn’t understand the difference between modulus and hardness. “You know,
don’t you, that hardness is force divided by contact area?” We admitted that we didn’t know this,
and that the definition made no sense. It was a lightbulb moment. Most of our samples had a high
elastic modulus, but the better ones showed less plastic deformation, so contact area was smaller
and they were harder. Changing our focus to plastic flow led directly to new nanoparticle

394 FST
 Mechanical Hardness

formulations where the dispersant reacted into the matrix, to reduce flow around the particles. For
years we’d heard the word “hardness” and confused it with “modulus”. Cross-discipline
interactions are vital for progress.

FST 395
Mechanical Modulus
Links
PSA

Mechanical Hardness is resistance to permanent marking, via plastic deformation so is an

important parameter. The resistance to elastic deformation is an essential material property called
modulus, defined as Stress/Strain where Stress is the force you apply and Strain is the amount the
sample stretches. It’s good to know both properties as a measure of the “strength” of your materials
and coatings.

Classic modulus measurements of bulk materials use large samples clamped into Instrons. For
coatings, the modulus can be easily measured if you can make a thin sample on some sort of
release coating and do a simple droop measurement!

Classic modulus
You clamp a sample into the jaws of an Instron-like machine and apply a
stress (force per unit area) and measure the strain (% change in length). For a
while the stress is proportional to strain and the gradient, Stress/Strain is the
modulus. Stress is in Pa, strain is unitless, so modulus is in Pa, more usually
MPa or GPa.

Eventually the material starts to deform, the graph becomes non-linear and at
the maximum stress (“yield stress”) the sample yields.

Modulus from Bending

To those of us who don’t have a DMA, it is difficult to measure the modulus of a thin sample directly. This
trick using the bending of a thin sample is super-useful.

[Link]

Take a thin strip of your coating, thickness H. Put it onto the edge
of a sample holder and extend it by length L. Under its own
weight it droops by distance D. If you know its density ρ then you
can calculate its modulus, E:
 Mechanical Modulus

1.3L4ρg
E=
H2D

It’s as simple as that!

Getting a good sample of known H and finding a way to measure D accurately is tricky at first, but your
organisation can quickly find ways to do it routinely.

I’ve used this trick many times on a large variety of samples, from the hardcoats (modulus ~ 2
GPa) mentioned in Mechanical_Hardness to inkjet films in the 10 MPa range. For obscure reasons,
we needed the inkjet modulus at various temperatures so we just did the test in freezers, fridges
and ovens. Not super-accurate, but the results were fit-for-purpose, which is good enough.

FST 397
Mechanical Surface Roughness
Links
Adhesive

Surface roughness is important for the look, “feel” and behaviour of a surface. Modern measuring
devices can give you many parameters, but for many surface effects only two, Rq and λq, the first
an amplitude and the second a wavelength are really important.

With the aid of the app, which lets you explore different surfaces, you will quickly understand
what the important roughness parameters mean.

[Link]

Surface roughness is important for Optics_Gloss, Absorbency_Porosity and Darcy, Mechanical_Friction

and the “feel” of a surface.
 Mechanical Surface Roughness

Most people are familiar with roughness amplitude as measured in

“R” values like Rz, Ra or Rq. But this is only part of the story. Look
at these surfaces. Two pairs have the same amplitude but are
obviously very different. The difference is in the frequency or, more
usually, wavelength. So whenever you measure a surface you must
measure both amplitude (“R” values) and wavelength (“λ”) values.
Only then can you fully understand what is going on. It is
extraordinary that most people extract only amplitude values from
their surface profile devices … partly because many devices don’t provide the wavelength values easily or
at all. The strong advice is to never buy a surface profile machine that cannot provide wavelength data.

The origin of the surface roughness app was desperation. A supplier delivered a new batch of a
product with an in-spec surface roughness but which obviously failed even a casual visual
inspection. I wrote a surface analyzer program to process the raw digital data and added all the
common parameters, including the λ values we’d never heard of. The R values showed no
difference between products, the λ values were very different. That was a long time ago. The
Surface Profile Explorer has been online for many years. But still people are unaware of the need
for both R and λ values.

The app lets you compare/contrast many different types of surfaces and get values calculated for the
standard R, λ and Δ measurements surface machines can provide. There is a description on the app page of
what each parameter means. Modern methods tend to get areal roughnesses rather than linear values, but
the learning from the linear app carries over to the areal values.

You should set a rule that when you measure a surface, always get a digital readout of the surface scan –
don’t just rely on the numbers produced by the machine. Sometimes those numbers are wrong because of a
defect or problem that is obvious to the human eye, but not to the computer doing the calculations.
Necessarily, this means in addition to not buying a machine without wavelength data, don’t buy a machine
that can’t provide you with the raw data.

Again, this is a rule based on desperation. We were looking for differences between some other
surfaces and the standard meters weren’t giving useful information. We’d just acquired a digital
version of the meter and could now look at the raw data. It was instantly obvious that some large
glitches, unrelated to our issue, were distorting the measurements. Removing the glitches allowed
us to get the data we needed to solve our problem. The human eye is often much smarter than any
digital analysis.

FST 399
Optics Colour
Links
SkinCare, Lipstick,

Colour is much more complicated than we might like. The objective measurement of colour is hard
enough, interpreting it is also hard. Then the subjective element is also hard – a perceived colour
depends on the colours around it. Here we provide some essentials.

What we see
The graph shows that our eyes have 3 colour cones that have
response curves to light over 3 wavelength ranges, Long,
Medium and Short. Why nature has selected such an odd,
unequal set of ranges is unclear [there are conflicting theories,
none of which I find convincing] – but it makes colour science
more difficult. (Note that imperfections in the plot and those that
follow are from the standard datasets):

RGB
Humans prefer to think in terms of Red, Green and Blue, but this
means that the RGB elements, include some weird negative
values obtained by human colour-matching tests that involved
back-projection of some red to give the negative values. These
negative values are one source of colour science complexities.

Because of these complexities, we are forced to use XYZ colours.

XYZ
The letters X, Y and Z are deliberately chosen to have no
obvious meaning. They are just a mathematical re-normalization
of RGB. We are even further from real colours, but XYZ
coordinates allow us to specify any human-visible colour
accurately, which is why they are used. The arithmetic behind
them is not especially hard but giving the formulae here won’t
help my purpose which is to take you through the chain of logic
which takes us ever further from those photons hitting the
receptors in our eyes. The next step looks at a real colour with
absorption across the spectrum.
 Optics Colour

A real colour in XYZ

The real colour has some absorbance over the visible wavelength range. We can integrate the amount of
light in the X, Y and Z ranges by multiplying the X, Y and Z curves with the colour curve. This gives us
three values: xv, yv and zv. To quote the colour we add the three values to give Tot and provide the colour
triplet [X=xv/Tot, Y=yv/Tot, Z=zv/Tot].

RGB and CMYK

Although we know that describing additive colours via Red, Green and Blue cannot capture the full colours
of human vision and that the subtractive colours, Cyan, Magenta and Yellow suffer from the RGB problem
plus the limitations of CMY pigments, we still have to work with them. [See the Kubelka-Munk app,
[Link] for some hints of issues with
imperfect pigments]. Adding in BlacK to reduce the amount of C, M & Y for grey levels adds a further
complication.

Many years ago, a colleague and I decided to spend a few hours learning colour theory. How hard
could RGB and CMYK be? Decades later, and now, even with the resources of modern technology,
I still find that I need a few days of full immersion in the science before I can handle a colour
science issue. Everyone agrees that colour science is really, really difficult and definitely cannot be
picked up in a couple of hours.

How white is white?

The amount of light detected by our colour sensor across the wavelength range depends on the colour of the
white light used to illuminate it. D50 light is relatively yellow, with a colour temperature (“black body
temperature”) of 5000°. D65 light is whiter, with a hotter colour temperature of 6500°.

From now on, we assume that your measurements are made with your choice of D50 or D65. It doesn’t
matter which, as long as you specify it. It’s easier if you then stay in one mode or the other, but it’s possible
to interconvert if, say, you’ve done everything in D50 and you need to communicate with someone who
does everything in D65.

FST 401
CIE
[X,Y,Z] triplets are perfect for conveying colour values, but useless
for us understanding them. So there is yet another conversion
formula to take XYZ into something we can work with. We create
X Y
x= X + Y + Z and y = X + Y + Z and by assuming that the
“luminance” of the colour is simply Y we can create the famous 2D
CIE colour space. Colour space images are courtesy of Wikipedia:

What a wonderful diagram, but also how useless. Look at how

much of the colour space is taken up by subtle differences between
greens, and how the blues are crammed into one corner. It is
amazing that such a useless diagram is seen as so fundamental to
colour science.

L*a*b*
So there is yet another conversion of CIE space into a more equal
space using L (luminance) and, again, meaningless letters a and b.
The * designation is an historical feature. This one is shown with a
luminence of 75. You need a different chart for any other luminance.
So although we all use Lab (or a specific version of Lab, there are
many variants), and although it is useful for colour measurements,
it’s not at all useful for humans to be able to choose specific desired
colours. Some say that a* is red-green and b* is blue-green, but this
is unhelpful. Lab is just another attempt to do the impossible –
compact the complexities of colour into a simple set of numbers.
However, Lab is really useful for colour differences.

Colour differences
If you have two colours with values L1, a1, b1 and L2, a2, b2 then you can calculate the colour difference,
ΔE via

ΔE = √ ((L1 − L2)2 + (a1 − a2)2 + (b1 − b2)2)

The convention is that colours differing by less than ~2.5 are not distinguishable by eye.

But, of course, there are slightly different definitions of ΔE depending on different Lab spaces. So you
might choose to quote ΔE00 rather than ΔEab.

402 FST
 Optics Colour

Gamuts
By definition, CIE and L*a*b* cover the range of human
vision. But our computer monitors and printing inks are
limited by the spectrum of the phosphors (emissive displays)
and absorbers (subtractive colours) so we cannot reproduce
them all. We therefore need to be aware of the colour gamut,
the range that we can reproduce. There are a number of
standard gamuts, including sRGB and Adobe RGB.

Note that this is defined with D65 white. A typical gamut for
our laptops is sRGB. From the graph it looks very bad. Surely
we need something like Adobe RGB! But because of the
distortions of the CIE plot, sRGB is good enough for most of
us. And because we are so used to sRGB, we don’t notice that
some shades of, say, green, can’t be produced.

In any case, we are getting more and more detached from real
colours. When you take a photo on your smartphone, the digital processing makes the sky bluer, the grass
greener and the red apple redder – because we prefer reality to look that way.

HSV and HSL

If we have to choose a colour it is hard to use CIE or
L*a*b*. So via yet another transformation we can
create HSV (Hue, Saturation, Value) or HSL (Hue,
Saturation, Lightness) spaces and select colours via
some pseudo-3D chooser:

We can plot the values as cylinders: HLS (left) and

HLV (right). Or as conical chroma plots.

If any of these plots was wonderful then we would all

use it. But they each have their compromises, so we
generally choose whichever is the default in our
graphics software.

What is the real colour we see?

As is well known from many optical illusions, from the famous dress, and has been observed with joy or
frustration by artists, illustrators and amateur home decorators, you can measure an objective colour with
your equipment, confirm that it is this sort of blue but to most people it looks that sort of blue.

Similarly, if you are looking at, say, an apple as you walk from bright sunlight into a room with artificial
lighting, the colour stays constant to your eyes, yet the XYZ coordinates will have changed significantly.

The human vision system is not a camera with an objective measure of the light falling onto pixels. It is a
complex system designed to extract maximum relevant information for minimum expenditure of processing
power. We know that the colour of the apple is constant, so of course the colour doesn’t change as we move

FST 403
between different lighting conditions. Our heuristics for judging a colour surrounded by other colours are
good for most of what we do, and if they get fooled by clever optical illusion setups, that’s fine.

Rainbow colorization
It is common to colorize graphics (I do so in a number of my apps) and common to just use a standard
rainbow like this, from [Link]

As described on that page, this is a truly awful method – it has big perceptual blind spots, it over-
emphasises some areas, it’s bad for people with colour blindness, and it fails to reproduce in greyscale.

Having learned of the evils of rainbow colorization I went to [Link] and the
super-useful datasets provided in many formats, and chose this “least worst rainbow”:

If you use any of my color-coded apps, you’ll hopefully find that rainbow sufficiently familiar, yet also
providing a better range of information.

404 FST
 Optics Gloss

Optics Gloss
Links
Solvent-Based Paint, Emulsion PaintLipstick

Controlled gloss or matt finishes are super important for coatings. The values are measured with a
gloss meter, at different angles. These values depend on two properties of the surface:

1. The roughness amplitude;

2. The roughness wavelength.

The important thing is to have a surface roughness measuring device that can give you both values.
Only then can you begin to understand how to control gloss properly.

[Link]

We all know that the gloss of a coating or surface depends on the roughness of the surface (it also depends
on the scattering from the bulk, see Optics_Scattering and Opacity.) Frustratingly, it seems hard to match
roughness directly to gloss. This is for two reasons.

1. There are numerous “R” measures of "roughness" amplitude, see Mechanical_Surface Roughness,
so it is not obvious which is most likely to correlate to gloss.
2. Another key factor which is often not measured turns out to be crucial: the frequency, or
wavelength or (scientifically) the correlation length of the roughness.

The diagram shows how two surfaces with the same roughness amplitude will be different because they
have different wavelengths:
 The first issue can be addressed easily – out of all the different
amplitude (“R”) measures, gloss correlates strongly with Rms, the
root-mean-square roughness, sometimes called Rq. Note that this is
different from (and generally higher than) the common Ra value
and usually much less than the equally common Rz. You can work
with any convenient R value as they are often strongly correlated,
but it is safer to use Rms as your working value.

The effect of the wavelength of the roughness is easily explained.

Suppose (absurdly) that a given measured roughness undulates over 10s of μm. At any given point the
deviation of the slope of the undulation from zero is small, so the light is reflected as if the undulation were
not there. So a large wavelength or small frequency gives essentially no reduction in gloss. This is because
the gloss meter measures light over a small but significant range of angles of reflection, typically 2.5°, the
"collection angle". You can alter this in the app if you wish, but leave at 2.5° if you don't have any other
data. If, conversely, the same roughness occurs over a nm range then the slopes are very steep so the
angular deviation (scattering) is large.

The basic calculation takes a roughness and a correlation length and computes the gloss value if measured
at 550nmm in the middle of "white light". The graphs show:

• How changing Rms at your given correlation length changes gloss at 3 different wavelengths;
• How changing correlation length at your given Rms changes gloss at those 3 wavelengths.

Hopefully, by understanding how the two parameters interact you will be able to find rational strategies for
reaching and controlling your desired level of gloss or matt.

One question remains. How do you get a correlation length from your surface roughness scan? In the
absence of a proper correlation function analysis, just chose the λa value , again see Mechanical_Surface
Roughness or, even better, the λq value which is based on Rms.

It is sadly the case that many surface roughness devices just output a few "standard" numbers, often
neglecting anything other than Ra and Rz and being generally unaware of frequency/wavelength issues. If
you are seriously interested in surface optics, you cannot work without a proper digital analysis of the data
in terms of Rms, something like λq and, even better, some sort of Fourier analysis from which you can
derive the proper correlation length, or the exponential fit to correlation length which is the value used in
this app.

Accurate calculations of gloss are highly complex. Fortunately there is a simplified equation:

For light of wavelength=λ, surface of Rms=σ, with a measurement collection angle=Δ and a gloss angle=θ
the gloss (assuming an incoming refractive index of 1) is given by:

Gloss = exp −16[ [ πσλ ] cos2θ{1 − 2{ Gλ(a) }sinΔcosθ}]

2

G(a) is a function depending on the correlation length, a, given by:

G(a) = ( Δ2 )atan(Δa)

406 FST
 Optics Scattering and Opacity

Optics Scattering and Opacity

Links
Deodorant Sticks, Sun Screens, SkinCare, Solvent-Based Paint, Emulsion PaintLipstick Mascara

For systems like inks, paints and sunscreens we often want light scattering to provide opacity and
“hiding power”, while for clear coatings we want to avoid scattering. The basic theory provided
here is full of assumptions, but is the least bad option available.

Mie Scattering
If you have well-isolated particles in a uniform medium you can apply Mie scattering theory. This is too
complex for an app, but there is an agreed “good enough” formula we can use. But any formulation with
enough particles to be interesting will not conform to the assumptions behind Mie theory. Despite this,
everyone uses Mie theory as a guide, so here it is:

[Link]

We have some particles of diameter d, in this case 500nm. They have a refractive index, nparticle, of 1.69.
They are in a medium with refractive index, nmedium, of 1.4. We can see the amount of scattering (the
numbers have some meaning, but just think of them as relative values) in two ways:

1. If you are interested in scattering at a standard wavelength, λ, typically 550nm you see how the
scattering depends on particle size.
2. If you are interested in a specific particle size, you can see how scattering depends on wavelength.

When you play with the sliders you find that for a given particle size or wavelength, scattering increases as
the difference in refractive indices increases.
The simplified formula (van Hulst) that is conventionally used requires a parameter p:

( )
nparticle d
p = 4π nmedium −1 2λ

From this we can calculate the scattering Q:

4
Q = 2 − p sin (p) +
4
p2 (1 − cos (p))
If you want good opacity for white light then go for the highest refractive index particle you can find,
typically TiO2, and use the app to find the particle diameter that gives the maximum. For TiO2 it is well
known (and the app confirms it) that particles of 300nm are optimal. If we assume that ZnO’s refractive
index is ~2 then the optimal particle size is ~900nm.

If you want transparent coatings then SiO2 is highly preferred as its refractive index is in the 1.4 range so is
a good match for typical formulations that are also in the 1.4 range.

Scatter and colour

In paints it is common to use lots of TiO2 to provide the opacity and provide the colour via relatively small
amounts of pigments. This works well because popular paint colours tend to be lighter shades, so the
whiteness of the TiO2 is a benefit.

For printed colours such as inkjet, you need all the pigment you can get, so both colour and scattering are
produced by the same pigment. There is a trade-off. Smaller pigment particles provide more colour per unit
mass – the centre of a large pigment particle can absorb no extra light as it’s already been absorbed by the
outer few nm. But smaller particles scatter less. Getting the pigment to the correct trade-off size is part of
the ink designer’s craft.

When the inks are printed onto a scattering background, such as paper, you lose colour intensity because it
gets drowned out by the scattered white light. You can calculate this effect via Kubelka-Munk theory:

[Link]

It’s obvious here that the dye-based ink at 1.1 gsm is very much drowned out by the white background!

408 FST
 Particles Basic behaviour

Particles Basic behaviour

Links

Anyone who handles particles/powders struggles to control them. But after decades of particle
research, we still have few reliable tools to understand their behaviour. The science here is at least
some core basics for the formulation team to agree on.

Adhesion
There are a few core aspects to particle adhesion.

A sphere of radius R and a work of adhesion W (usually assumed to be the “surface energy”, typically 40
mN/m or 40 mJ/m², has an adhesive force Fadh given by:

Fadh = 1.5πRW

If the particle is stuck onto a vertical surface, will gravity pull it off? The gravitational force is:

4
Fg = 3 πR3ρg

and if you do the calculations for typical densities then for anything smaller than ~1mm, gravity is
insignificant.

As is well known, if humidity is high (say above 50%) then you can start to get capillary condensation
around the particle, giving an extra adhesion force. Where the surface tension of the liquid is γL, the contact
angle is θ and the surface/liquid energy is γSL then:

Fc = 4πRγLcosθ + 4πRγSL

We can get the calculations from the app:

[Link]

Those calculations are for smooth surfaces. Add a roughness σ for a particle with a (reduced) modulus E,
then the adhesion is reduced by a factor of AP (Fuller-Tabor’s Adhesion Parameter), which, if larger than
10 essentially means no adhesion:

Eσ
AP = W √ Rσ
There’s also a tendency to clump into macro-particles of diameter D:

[ ]
5 2
2E[2R] 2 3
D = 0.064 W

Again, these are calculated on the second app on the same page:

Engulfment
Sometimes particles can disappear spontaneously into a relatively soft surface. It seems an unfamiliar
phenomenon, but in fact is well known to those who have tried to matt a surface and found that the particles
they add tend not to give the desired rough surface.

If the radius of the particle is smaller than Rcrit then the particle will be engulfed:

7W
Rcrit = E

Cleaning by blowing
Given that the pure surface adhesion force is so small, it should be easy to remove a particle by blowing
over it. Given that the adhesion force is proportional to R, smaller particles should be really easy to
remove! Yet we observe how hard it really is. This is because of the No Slip Boundary Condition,
Cleaning_Boundary removal, which says that the air velocity at the surface is zero and rises relatively
slowly above the surface. A small particle not only gets little air flow, but the removal force goes as R² so is
less for the smaller particles. If you can get the air into a turbulent mode then the removal force increases.

Some convenient equations give us the drag for laminar flow in air of density ρ and viscosity η a distance x
from where the air hits the surface:

Fdrag = 2.6ρR²v √ ηx
For turbulent flow the equation is:

1.44 1.36 0.54

19.8ρ(2R) v η
Fdrag = 0.14
x

The app gives some idea of whether you are likely to remove a particle or not. Whether you are in laminar
2Rv
or turbulent mode depends if Re = η > 5.

410 FST
 Particles Basic behaviour

[Link]

Hoppers
It is well known that near many hoppers across many industries their stands a big hammer. When the
powder in the hopper gets jammed, the hammer is used to bash on the sides. The fact that this is still
common in the 21st century is a sign that the science of powders in hoppers is not as useful as we should
like. So when you read the following, expect to be disappointed. If there was some better, more useful
science, it would be included here. Of course you can always use a big computer model with particle
dynamic calculations, but if you have that sort of computational power and the data to feed into the model,
you won’t be reading this section.

A simple metric for a powder is the angle of repose, θ, obtained

when the powder is poured gently onto a surface.

We can consider it to be equivalent to a measure of µ, the

coefficient of friction, (Mechanical_Friction) where µ = tan(θ)

Obviously you want your hopper to be at an angle greater than θ. Confusingly, hopper angles are defined
with respect to the vertical, so we want the hopper 90-θ to be greater than the angle of repose θ. In addition
we have friction with the wall itself, so we have the “wall friction angle” φ as an extra value. However, the
powder can stick to itself with another angle, δ, which is the “internal angle of friction”. We can combine
all those, with some horrendous algebra, to solve a core problem: what is the minimal opening, D, to a
hopper that allows the powder to fall out rather than bridge?

[Link]

The final factor required for the calculation is the density ρ. Quite what this means is debatable, but the
errors in the calculations are so large (as academic debates confirm) that worrying about ρ is not too

FST 411
important.

This is all OK if the powder is free flowing. But under the weight in the silo the powder might compact. If
you know the powder’s degree of compaction under pressure, you can, in principle, do something about it.
But although the following thought experiment is simple, doing it isn’t:

You trap some powder in a box and

compress it with stress σ1. You
removed the sides of the box and press
till you reach σc at which it cracks. You
just need to do this for a series of σ1
values, but because you need to pack
the box each time people avoid this and
use something like a Jenike box
instead. The app page described the circuitous route needed to extract data.

Surface energy
The unhealthy obsession with surface energies being important for things like adhesion has led to a large
expenditure of time using Washburn tubes (the choice of some) and Inverse Gas Chromatography (the
choice of others). If you are tempted by Washburn tubes, read the caution in Absorbency_Porosity and
Darcy. If you are interested in IGC, read my free eBook [Link]
chromatography/[Link].

That’s it
Are you disappointed that we have nothing more to say? Well, it is really the case that decades of particle
work has given us very little to put into your toolkit. Yes, you can read books with lots of general tips and
tricks, but when it comes to reaching into a toolkit to solve a specific issue, there is very little available. But
if you know something that I don’t, let me know. I would be delighted to expand this chapter.

412 FST
 Particles Porosity

Particles Porosity
Links

The porosity of our particles and powders has a big effect on the overall properties. So how do we
measure porosity?

2.5 Techniques
There are 2.5 standard methods

1. Mercury Intrusion Porosimetry (MIP) for micro- and meso-pores

2. Capillary Flow Porometry (CFP) also for micro- and meso-pores
3. Nitrogen porosimetry for nanopores

Why 2.5? Because #3, nitrogen porosimetry, is so full of assumptions and models that it’s not really a valid
technique – except if you compare values between samples that you have analysed via the same subjective
technique. So here we discuss MIP and CFP.

The techniques work on the principle that the pressure P of a liquid inside a pore of diameter D depends on
the surface tension of the liquid, σ, and the contact angle θ via:

4σcosθ
P= D

[Link]

For MIP, where σ = 480 mN/m and θ = 140° the technique is simple in principle. Evacuate the sample then
apply mercury under pressure, monitoring the volume uptake as the higher pressure increases the number
of pores into which the mercury can flow. The app lets you create a simplified pore size distribution, the
graph on the left, and see, on the right, what you would obtain in a real experiment.

The volume is only “relative” – in real life you would know the volume of particles and the volume of
injected mercury so could get the total pore volume as well as the size distribution.

In the app the pore size is limited to ~ 10nm because below that the pressures become super-high, requiring
more specialised equipment.

The same pore distribution measured with CFP is at first sight similar:

But because the fluorocarbon used has σ = 16 mN/m and θ = 0° the pressure range is lower. And because
the technique works by emptying the filled particles, at the highest pressures, the flow rate continues to rise
as there is no longer any blocking fluid.

Filling the pores

The filling of a porous material is described via Darcy’s law, discussed here: Absorbency_Porosity and
Darcy.

414 FST
 Particles Sintering

Particles Sintering
Links
Water-based Adhesive,Emulsion Paint

We have chosen the relatively neutral word “sintering” to cover many aspects of particles coming
together via extended contact and some applied pressure. This covers food powders, water-based
paints and adhesives and “real” sintering of ceramics. Each application needs some flexibility in
interpreting the specific formula used here; what’s important is to identify and control the key
aspects influencing the effect for good or bad.

From free-flowing to caked

[Link]

We have some 1mm powder particles and there’s a 5 mN force between them, for ~60s. The “viscosity” of
the particles, η0, is 20 MPa.s (that’s Mega, not milli). The glass transition temperature Tg is 40°C and
because we are at 25°C we’re below Tg so our actual viscosity, η, is the same 20 MPa.s

From the Rumpf equation we can work out the % overlap (x/r) between particles after time t, where 2x is
the contact width and r is the radius. It depends on the force F, the surface energy γ and our viscosity η:

2
( xr ) ( γ
= 0.8 r + 0.4 )
F t
πr2 η

Now let’s expose the powder to humidity. It absorbs moisture according to its isotherm, Humidity_Water
vapor isotherm. Because of the water, the Tg is reduced according to Humidity_Water mechanical isotherm,
let’s say it’s fallen from 40 to 20°C. Now our powder is caked:
That reduction in Tg reduced the viscosity from 20 MPa to 0.6 MPa and that was sufficient for sintering to
give a 12% overlap.

That’s the scary thing about sintering – relatively small changes can make a large difference. This works
both ways. The assumption so far is that we want our powder to stay free-flowing. Sometimes you want
particles to sinter sufficiently to give you a stable “green” shape which then can sinter properly by, say,
heating in a furnace.

The sintering mindset

You may have a different word for describing particles getting stuck together. That’s fine. The reason for
this chapter being on “sintering” is that there’s a good equation (Rumpf) that captures the core inputs and is
valid across a wide range of particle issues. Using such a generic term allows you to focus on the generic
science principles before getting sucked into the details of your own subset of issues.

For example, mud-cracking of particle-based formulations is unlikely to happen in formulations that sinter
before the system becomes too dry. A couple of mud-cracking apps are discussed in Emulsion Paint.

Science is universal. I was involved in some intense debates about how best to model the powder
clumping tendency of a household cleaning product. It’s a complex problem. Eventually we found
work (cited in the app) funded by Nestlé covering a similar issue in the processing of food powders
such as granules of instant coffee. This approach broke the problem into three steps, each of which
could be modelled in an app. Instead of complexity we now had clarity and comparative simplicity.

416 FST
 Particles Size Distribution

Particles Size Distribution

Links
Toothpastes

Particle size distributions are difficult to measure reliably, difficult to interpret and the numbers
quoted for a given particle might be inappropriate for the intended purpose. Via the app you can
quickly understand why it is all so confusing … and how to get the numbers you need.

A brief comparison of measurement methods helps you find the least bad way for your particles.

[Link]

We have a deliberately bi-modal distribution with 90% of the particles at 100nm and 10% at 378nm. These
peaks are shown in red, the large one at the small radius and the small one at the large radius. The larger
particles also have a wider distribution. These peaks represent the number of particles measured at each
size.

Now shift to the blue peaks. These show the opposite trend – a small peak at low radius and large peak at
large radius. These peaks show the mass (or volume) of particles at each size. Because mass goes as r³, a
relatively small number of particles represent a large mass.

We can see this in the cumulative curves. The orange number curve shows that most of the particles are
below 150nm, while the cyan mass curve shows that most are above 350nm.

The green lines represent the area equivalent data, i.e. based on r².

To describe the distribution in a single number is unhelpful, but very common. Note that graphs of
distribution are often in radii while quoted numbers are usually in diameters, so there’s a factor of 2 from
what you see in the plot. There are many possibilities. D[1,0] is the number average (mean), D[3,2] is the
area average and D[4,3] the volume average. A common quoted value is D50 which is the value where half
the volume-weighted particles are below that value.

The Steepness value is used especially in the world of mineral particles and is defined (as a percentage
value) as the ratio of D30 & D70, representing the points where the cumulative volume value exceeds 30%
and 70% respectively.

Measuring particle size distributions

A brave European project once asked several well-equipped labs to measure the particle size distributions
of a number of powders in a “round robin” event. The results make depressing reading because there was a
large variation not just between methods but between different groups using the same method. We’ll reveal
the agreed least bad method at the end.

• Direct microscopy: TEM, SEM, Light microscopes

◦ Advantages: “Real” measurements that include shape data.
◦ Disadvantages: Relatively small sample size. Occlusion of particles. Bias towards a
“nice” image
• Sieving
◦ Advantages: A real measurement
◦ Disadvantages: Suitable only for large sizes. Limited numbers of sieves, so crude data
• Light scattering:
◦ Advantages: Works for small spherical particles in high dilution
◦ Disadvantages: Doesn’t work at higher concentrations or with larger particles or with
complex shapes and size distributions. Many assumptions between signal and results, so
size distributions can often be unreliable.
• Analytical centrifugation (with real-time data)
◦ Advantages: Works across a wide range of samples, sizes, distributions. Fewer
assumptions in data interpretation
◦ Disadvantages: Can be fooled by different particle densities and shapes

Although no technique is applicable to every type of particle, and each technique is preferred in different
specific circumstances, the analytical centrifuge had the best combination of versatility and reliability of
reported distributions. Although light scattering has had a reputation for being the default option, its many
flaws for real-world particles suggest that the default option should shift to centrifugation.

418 FST
 Separation Filtration

Separation Filtration
Links

Although there is a well-known filtration equation, there are too many variables in it to make it
useful, other than a guide to what you might do to keep flow rates acceptable.

The filtration equation

This is a variant of the Darcy equation, Absorbency_Porosity and Darcy. We have particles of density ρ,
6
specific surface area S which for spheres of diameter d is given by S = d , at concentration Cs kg/m³,
packing together with a porosity (void fraction) ε. Currently there is a volume V of filtrate already collected
across a filter of area A. We have a constant k = 4.17 (from Carmen-Kozeny) and are applying a pressure P
across the filter cake. The filter itself has a resistance R. We are interested in the increase of filtrate volume
per unit time:

3
δV PA2ε ρ
δt =
(
η kCsV(1 − ε)S2 + RA )
This is mostly intuitive. Higher pressure, larger filter area, higher porosity, lower viscosity, larger particles
(smaller S) and a lower-resistance filter medium speed up filtration. The problem, of course, is that the rate
of filtration decreases because of the 1/V dependency. If you increase the pressure to increase the flow, the
chances are that the porosity will decrease, making matters worse.

And as is well-known, if your Particles_Size distribution includes lots of small particles, these can easily
fill up pores between larger particles, decreasing ε. The fine particles can also start to clog up your filter
medium, increasing R. This leads to a key message of this short chapter: if you need to filter, do what you
can to reduce the fraction of fine particles before you start the filtration.

Spend time with the world of high-quality coffee and you find that fines are a big problem for
optimal brewing which involves filtration through the espresso puck or through filter papers. It is
often noted that coffee pods can give good, strong coffee using maybe only 60% of a typical
espresso dose. The reason is that the roller mills used for the coffee pods can make fine coffee
particles with low amounts of fines. Extraction is efficient without filters blocking. These roller
mills aren’t practical for daily use, and even high-quality burr grinders can’t get the same balance
of small particles with low levels of fines.

Filtering aerosols
It can seem strange to us that a, say, “1μm” filter can take out a high fraction of nanoparticles from the
incoming air.

To understand this we first need the definition of “x μm” then we need to see how a fibre-based material
captures particles.

For a pure filter material with uniform x µm holes the filter size is x µm. But a typical fibre filter used in,
say, aerosol masks doesn’t have “a” hole diameter. Instead one side of the filter is placed into a liquid of
surface tension γ with a contact angle (usually chosen to be near 0) of θ, and a pressure, P, is applied which
is just large enough to create a bubble. From the simple capillary force equation the effective filter size, D,
is:

4γcosθ
D= P

This value is purely a convenient fiction, useful for checking that masks are OK; that D is large enough to
allow breathing, and small enough to provide a good filter.

A well-known graph shows how such a filter captures particles via:

• Sedimentation – the particles are so heavy they fall onto the filter;
• Impaction – the air channel is wide but changes direction so suddenly that the particle carries on
and hits the filter;
• Interception – the particle cannot fit through the small gap in which it finds itself;
• Diffusion – the particle wanders randomly through the air and happens to hit the fibre material.

Obviously missing from this is a line for “electrostatics”. That is because it is really tricky to design, model
and rely on electrostatics which can so easily change with, say, relative humidity.

It’s good to know that when we put on relatively simple masks to handle dusty particles, this combination
of mechanisms does a good job of limiting risk, with each mode being especially good for particles that
other modes might miss. It is the diffusion mechanism that captures most of the small particles from a mask
porous enough to allow us to breath easily.

I tried to find the formulae for producing those graphs from reasonable inputs, but was defeated by their
complexity.

420 FST
 Separation Settling and Centrifugation

Separation Settling and Centrifugation

Links

Given some simple equations and a few basic parameters, knowing for how long you need to let
your particles settle, or how fast to centrifuge your sample to achieve separation is straightforward.

The basic settling and centrifugation equations

[Link]

We have spherical particles of radius r, density ρp in a liquid of density ρl and viscosity η. We want them to
settle a distance of h. Without centrifugation the settling force is 1g. With centrifugation in tubes sitting at a
distance rcent from the centre of rotation of the machine spinning at our chosen RPM we have an extra g
h
force, grel which is 1 with no rotation. The time, t, taken to settle depends on settling velocity v, so t = v.
The velocity is given by:

grelr2
v = 2.18(ρp − )
ρl η

The relative force of gravity is given by:

grel = 1 + 1.118e−3rcentRPM2

Some extra features

If the particles aren’t spherical so their aspect ratio (length/width), A > 1, then they behave as if they were
spherical particles with a larger radius, reff given by:

√ (√A2 − 1)
atan
reff = r
√A2 − 1
If the volume fraction of particles, φ, is large then the particles slow each other down. The standard
5.65
Richardson & Zaki formula (in the Eulerian frame of reference) is a reduction in velocity of (1 − φ) .
If the tubes are held at an angle θ away from the gravitational pull then the Boycott effect takes place such
h
that in a tube of width b, the effective velocity is increased by b sinθ. The effect is surprisingly large and
very real. Boycott first observed it in the 1920s for settling of red blood cells.

Like most people, I’d never heard of the Boycott effect. After a post about an improvement to a
settling app, someone commented that it would be good to add the Boycott effect. It was great fun
to learn of something completely new, to find the theory and to share it with many others for whom
it was also new. It also shows the power of open science – there’s always someone smarter than
you out there from whom you can learn cool new stuff.

422 FST
 Surfactancy Anti-foaming

Surfactancy Anti-foaming
Links
Dishwashing Liquids, Hair Shampoos, Soaps and Washing, Laundry Liquids

A well-known anti-foaming theory is of no real help, and a profound analytical technique seems
too difficult to use in a formulation lab. So anti-foaming (even with silicones) is less about hard
science and more about experience and, sadly, luck.

The core problems

You want some blobs of oil to go to the foam wall and destroy it. If the
blobs of oil are too large, there are too few of them to make a difference.
This, therefore, excludes simple addition of hydrophobic oils such as
mineral oil or silicones which quickly form such blobs. To get small oil
drops you either have to have a not-very-hydrophobic oil, which might
not destroy the foam wall, or you need to emulsify the oil, which means
that it is protected by a surfactant layer so isn’t destructive. Or you can
break up drops via faster agitation, which is likely to induce more of the
foam you are trying to destroy. One way round these contradictions is to
engineer a complex system, which is how modern silicone/silica anti-
foam, discussed below, can be effective, but still with plenty of trial and
error.

There is a further problem. Within a foam, 90+% of the liquid is in the

Plateau borders and the Nodes, so very little of the anti-foam has a
chance to destroy the film walls. This beautiful image, reproduced with
kind permission by Dr Illy, shows how oily particles in an espresso
crema aren’t acting as an anti-foam, they are just sitting in the borders
and nodes.

Those drops of oil sitting in the nodes will do nothing until the liquid
has drained so much that the drop is compressed against the wall and
can then destroy the foam. This is the reason behind the phenomenon of anti-foams seeming to have no
effect for a while, then suddenly the foam disappears. The calculation of the radius of these borders, in the
second row of the Anti-foam app is discussed below.

Although there is a well-known theory around entry, bridging and spreading coefficients calculated in the
top row of the app:
[Link]

it has largely been discredited, because it focusses on oil drops touching the foam wall while the arguments
above point out that this hardly happens. At best it tells you when failure is guaranteed by relative surface
energies, but as failure is normal, even for those predicted to be good, that’s not much help. There is
another reason why it doesn’t work.

The anti-foam app also takes a pessimistic view of anti-foaming. An earlier version had implied
that anti-foam science was useless. I then had the chance for a delightful working lunch with a
world expert on anti-foams. After considerable debate, I returned home and changed the app to say
that the science was “mostly useless”. The expert, with proven success in industry, taught me a lot
about specific issues. My pessimism is about the lack of usable science for us non-experts.

A subsequent discussion with an expert on modern silicone anti-foam systems allowed me to

expand this chapter, but the core problem of lack of usable science remains.

The Entry Barrier

Using their Film Trapping Technique, FTT, the Denkov
group in U Sofia showed that it can be remarkably difficult
for an oil drop to push through an “entry barrier” at the foam
surface:

If we could properly understand the entry barrier, measure it

routinely, and test the barrier with different anti-foam oils we
might be able to make some progress. Unfortunately, the
FTT depends on the considerable skill and knowhow of the
Sofia team and hasn’t yet become a routine test in other labs.

Silicone anti-foams
Throw in some silicone oil … and the chances are that it will not help de-foam. As discussed above, the oil
will just gather as a blob, maybe breaking a few local foam walls but having little impact on the foam in
general. What we need it lots of small droplets of silicone oil – and for that we need good surfactancy. If
the foaming surfactant was a good emulsifier for the silicones, it might be too good – hiding the silicone
from the foam. The short discussion in the Surfactancy_HLD chapter suggests that the only good surfactant
for a silicone oil is a silicone surfactant. If the silicone surfactants (often complex silicone-PEO-PPO
systems) are ignored by the rest of the system (as seems likely) then the silicone oil droplets can be nicely
emulsified by the silicone surfactant, giving a semi-independent system able to deliver the silicone oil to
the foam walls. There is good evidence that this semi-independence is real.

424 FST
 Surfactancy Anti-foaming

One way to control the silicone oil against breaking up too much into drops that are too small and too
shielded to be effective against foam, is to make the “oil” a polymer, maybe a “rubber” from some silicone
crosslinking. Those who use PDMS, which is high MW silicone, generally expect it to be rubbery via
crosslinking. Obviously, highly-crosslinked PDMS could produce excellent, stable, (say) 5 μm silicone
spheres, but these (especially when wrapped around by PEO/PPO chains) will be mostly harmless to the
foam.

The size of the silicone drops/particles has to be optimized for each application, again for reasons that are
unclear. The second row of the app is relevant to the problem of drops being trapped in the Plateau borders.
If your bubbles are small and the foam is dry, the length and radius of the Plateau borders are both small, so
small drops will be able to act. Large bubbles in a wet foam will need very large drops/particles to start to
disrupt the borders.

What about the thickness of the foam wall? It is generally agreed that the capillary pressure sucks water out
within time frames of seconds so they are all effectively sub 100 nm thick. There seems no obvious link to
drop/particle size and the very thin walls that are formed so quickly.

Even with blends of oils, polymers, rubbers and fancy surfactants, the anti-foam can be useless. The final
ingredient is some relatively sharp hydrophobic (silane-coated) silica particles that, using simplistic
language, can puncture the foam wall, allowing the silicone oil to do its thing. Why not use these
hydrophobic spikes without all the silicones? One common explanation is that they are difficult to disperse
in a way that gets them to the foam walls. Another is that even if they pierce the walls, the surfactant can
just flow around it. So the spikes need to bring the silicone along with them to finish off the job they
started. Neither explanation seems satisfactory.

For a functioning anti-foam we need the oil, surfactant and silica all to be together, all being carried to
every part of the foam, all able to destroy the foam walls and not get stuck together as hydrophobic islands
nor get stuck in the Plateau borders and nodes, nor get stuck on the walls of our process equipment to be
removed, with great difficulty, when the deposit gets too thick. How much the famous low surface tension
of the silicone oil plays in this is unknown. We can’t do side-by-side comparisons with mineral oil
equivalents as there are many other differences between successful formulations.

What might go wrong? Everything. The real surfactant might interfere with the silicone surfactant or the
silica particles. Other parts of the formulation, such as, builders, enzymes or bleaches in a detergent, or
fibres and coatings in paper pulp processing, might divert the silicone surfactants in some strange way.
And, as discussed in Fragrances, one fragrance package might be fine while another wrecks the anti-foam
performance by, presumably, partitioning unhelpfully at the PEO/PPO/Silicone boundary and wrecking the
curvature needed to emulsify the silicone oil. Or maybe the fragrance changes the chameleon-like nature of
(flexible) silicones that can happily present a relatively hydrophilic side if the methyl groups choose to fold
inwards (perhaps around the fragrance molecules). A “hydrophilic” blob of silicone oil will no longer be a
threat to the foam wall.

As mentioned above, those skilled in the art, with plenty of experience, can have a feel for how to navigate
around this complex space, and how to find plausible hypotheses for failure in a specific environment, with
plausible fixes for the problem. Yet, as these experts freely admit, their attempts can still fail. It’s even
tougher for the rest of us who don’t have this specialised knowledge.

How might we change this? In the spirit of “smart mapping”, used elsewhere in this book, you can imagine
systematic experiments that could explore simpler regions of anti-foam space which, when combined
against results from real-world experiments, might lead to a more fundamental approach to problem

FST 425
solving.

But there’s one more problem.

Testing an anti-foam
The chapter on Surfactancy_Foaming shows that there aren’t many good rules for deciding which
surfactants are low foam – it depends strongly on the context of how you make the foam. A surfactant
might be useless in one context and create a splendid foam in another. This makes testing for anti-foam
properties especially hard. The only test that really matters is in-use. Putting on lots of washing machines
and testing with different anti-foam formulations is a necessary qualification step, but hopeless for
development. So “all” you need is a relevant foam test. But because the general approach to foaming is a
confused mix of tests that are often poor or irrelevant, there is a lack of rational debate as to which tests
map best onto which real-world anti-foam challenge. Do you use a Ross-Miles, a Foam Analyzer, a kitchen
blender or kitchen mixer, a shake test, some timed sequence in a SITA tester, and so on? It was amusing to
see a video saying that “the” test for foam was ASTM D1173. Yes, this is the test for those interested in
foams analysed according to ASTM D1173 (I could have chosen other standards) but might be, and
probably is, totally irrelevant for your specific foaming issue.

The trick, therefore, is to find a quick and easy test that maps well onto your specific challenge. You can
focus your formulation efforts to get the most information for the least effort. The chances of success in the
field are greater than if you had spent time working on, say, ASTM D1173 if it happens to have near-zero
correlation with the foaming mechanism in your system.

You already have a selection of your own and competitor formulations with known field results with
specific surfactant systems (e.g. the 2016 and 2018 versions of well-known detergents). If your test does a
decent job at distinguishing good from bad then you not only have a fast-track method for future
development, you have a reliable method for testing hypotheses: “If I replace X with Y in the successful
2016 de-foamer formulation, will it make things better or worse, if so, why?”

The fixation with ASTM standards gets in the way of optimising other complex formulations such as
Adhesive. It’s not that ASTM standards are bad (they’re not), but that they are often irrelevant to your
specific problem and tie up too much precious formulation resource. Once you have a quick test that lets
you formulate rapidly against criteria relevant to your problem, the chances are high that the product will
then pass the appropriate ASTM test or, even better, the end use test which is what really matters.

The best anti-foam …

… is to not create bubbles in your process.

Analyse why there is air, how you can avoid or reduce it, and, if you have to have air, how you can avoid
individual bubbles (usually harmless) from reaching the surface, or joining up to form a foam, which
includes trying to avoid larger bubbles breaking up into smaller ones. A single bubble of 500 μm is much
less harmful than the same volume of air in 125 100 μm bubbles, not just because the isolated bubble is less
protected against damage, but because foams from smaller bubbles (see other foam chapters) are
significantly more stable than larger ones.

By shifting the discussion from trying to find the perfect anti-foam (which doesn’t exist) to rational ways to
mitigate the production of foam the chances of a long-term solution to your problems increase significantly.

426 FST
 Surfactancy Anti-foaming

Happily, others have reached the same conclusion. and many have, indeed, found ways to solve their
foaming problems without all the problems they used to have with antifoams.

FST 427
Surfactancy CMC and Langmuir
Links
Dishwashing Liquids, Hair Shampoos, Soaps and Washing, Laundry Liquids, Surface Cleaning

Critical Micelle Concentration, CMC, is featured heavily in surfactant science and yet is of little
interest to the formulator. It is best measured with carefully purified surfactants in pure water, and
yet our formulations are with commercial-grade surfactants and we have lots of other ingredients.

If CMC was important, we’d all use long-chain ethoxylates which have low CMC values. In reality
we use lots of C12 sulfates with high CMC values.

If you measure CMC using a plot of surface tension versus concentration, the other information
from the experiment is more interesting than the CMC.

The Langmuir isotherm

[Link]

When you plot surface tension, γ, versus surfactant concentration, c, the resulting plot can be analysed via
the Langmuir-Szyszkowski plot. It depends on the starting value, γ0 which is 72 mN/m for pure water, on a
surface excess concentration Γm and an equilibrium constant K:

γ = γ0 − RTΓmln(1 + Kc)

The app takes your experimental γc, the CMC and the concentration at which the surface tension has
decreased by 20 mN/m to construct the whole curve and extract the important Γm and K values.

What is significant about 52 mN/m? This is when you approach full surface coverage, an idea that shocks
 Surfactancy CMC and Langmuir

those who think that CMC is the point of full surface coverage. But CMC knows nothing about surfaces.
You can measure CMC via bulk properties such as conductivity or scattering. The relative unimportance of
CMC is shown when you plot the same data on a linear concentration scale:

Now you see that the interesting stuff takes place early on and it is actually hard to spot where the CMC is,
because nothing much changes at the point.

This can all be explained by a pair of beautiful diagrams kindly supplied by Roberta D’Incà Levis.

We see the steady increase of concentration in the bulk, leading eventually to micelle formation, but,
especially clear in the Linear plot, we see that the surface is near saturation a long time before micelle
formation. As a bonus, the Log plot shows where Γ is determined, over the relatively narrow range where
the surface goes from rather crowded to very crowded.

Γm, K and A

Given that the CMC is uninteresting in most respects, if you want to do these titrations with your (pure)
surfactant (curves from commercial grade surfactants offer few deep insights), then learn to interpret the
surface excess concentration Γm, the partition coefficient, K, and the surfactant head area A.

It turns out that values for surface excess don’t differ strongly over most common surfactants. The partition
coefficients follow intuitions – more hydrophobic surfactants have a higher coefficient. If you want a lot of
surface stability from a little surfactant then, arguably, a large head area might help. But it’s usually more

FST 429
complicated than that. A is also used for calculating the Critical Packing Parameter, Surfactancy_CPP and
phases .

If the world of surfactant science had de-emphasised CMC and had focussed, for example, on the Cc values
for Surfactancy_HLD theory, we would all be formulating much more effectively.

γc

For most surfactant uses, γc is of surprisingly little importance. It is measured, carefully, over timescales of
minutes. Most surfactant uses require the surfactant to act in milliseconds or seconds. So for formulators,
Surfactancy_Dynamic Surface Tension is far more important. In a nice irony, one way to ensure a fast
reduction in surface tension is, as that chapter shows, to choose a surfactant with a relatively high CMC.

430 FST
 Surfactancy Coacervation

Surfactancy Coacervation
Links
Hair Shampoos, Soaps and Washing, Surface Cleaning, Pharma Formulation

Coacervation is one of those formulation techniques we take for granted in our respective areas of
expertise, such as Microencapsulation or creating all-in-one Hair Shampoos. It is unique because
we want it for controlled “failure” to keep things in solution. Yet we have no usable theoretical
tools and little practical guidance about how to design our own formulations. So this was a very
short chapter. Thanks to a brilliant lecture by Dr Tobias Halthur of CR Competence AB, the
chapter has been extended to include a summary of all charged/uncharged surfactant/polymer
interactions showing that the coacervate effect is a generalisation of a few key ideas … and that if
we try to get the best of both worlds, we generally make things worse.

Plus and minus

We have at least one charged polymer (e.g. a polyacrylic, a
polysulfonate, a polyquat) and an oppositely charged
molecule that is either a polymer or something a bit
complex like a surfactant. And we may have some added
salt.

Under the right conditions, the “complex coacervate”

appears most commonly as some sort of scattering or turbidity, maybe along with some visible phase
separation, depending on the size and density of the drops. The scattering droplets aren’t so much a neutral
pairing crashing out of solution, but pairings that contain a lot less water, but still plenty of it, in water that
contains a lot less of the starting pair. The droplet detection limit depends on the wavelength of
interrogating radiation – x-rays see the onset of coacervation before optical turbidity meters or the human
eye. If (e.g. for surface cleaners, Surface Cleaning) you want the coacervate at the surface, clever FTIR
experiments (Analytical Techniques) can extract insightful signals.

It is strange to say “right conditions” because this means “failure” to remain as an homogenous solution. In
the classic case of 2-in-1 shampoos we want to convert a homogenous solution containing lots of
surfactant, some charged polymer and some oil into a phase-separated state which deposits charged
polymer and oil (tangled in the polymer) onto the surface. The transformation must happen when the whole
system is diluted. In the ideal case, the deposit is at least kinetically stable but the small surfactant
molecules wash out, leaving the charged polymer and oil to do what the formulator requires.

This phase separation is best shown via ternary phase diagrams, as discussed in the general overview
below. Before then we can look at a binary phase diagram often shown in the context of coacervates:
 The classic coacervate plot shows the ratio of charged
species, Z, against concentration, C. When this is 1
you have the two phases. When it is small or large
you have a single phase, in between, the higher the
concentration of each (at the same ratio) the more
likely you are to have the coacervate. You can get
similar plots with change of the salt concentration (or,
more likely, ionic strength). With any of these
changes you get more or less turbidity or a different
FTIR signal. For example, a gross imbalance of the
charges (wrong ratio) means that there’s little reason
for a new species to be formed. And at high salt
concentrations, charge effects are so shielded that
little of interest can happen. If you change not the polymers but the % charged species in the polymer, then
effects are unpredictable. And if you’ve had great success with polymer A and opposite charged B and
want to change A to C or B to D, you are probably starting from ground zero.

In 1957 Voorn/Overbeek theory was developed to describe the phenomena. You can readily generate V/O
parameters to fit your data (I could produce a V/O app!), but that’s no help because just about every
assumption behind V/O theory is wrong. Clever work by smart theoreticians has produced (according to
reviews) at least 3 powerful theories that could/should explain coacervates but, as the experts acknowledge
they are each wrong/limited in at least one major way and none of them is usable in any real-world context.

What can you do about it?

A few chapters reach the same sad conclusion that there are no useful toolkit tools. Sometimes this is
because researchers have been vainly using the wrong approach and things can be fixed with a switch to a
more productive one. For coacervates a lot of very smart people have worked hard to come up with better
predictive tools but it seems a genuinely tough problem that awaits some genius breakthrough to give us
something for our toolkit. Sure, we understand the general phenomena, but we have no workable, appable
tools to make development less painful.

Although what follows still doesn’t help directly with coacervates, the logical approach that I
learned from a lecture by Dr Tobias Halthur provides some structured thinking to the relevant
issues. It also shows that coacervates are one example of a range of inter-related phenomena.
Generally we want the best of both worlds as these systems are often intended to give us controlled
viscosity in our formulation. If A gives viscosity, but not enough, and the same with B, maybe
A+B will give us what we need. Sadly, the opposite is often true. As Dr Halthur acknowledges, a
lot of the ideas below are explored in the book Surface Chemistry of Surfactants and Polymers by
Kronberg, Holmberg and Lindman. Of course, all blame for oversimplifications of this complex
topic goes to me.

The right viscosity behaviour

It is a common formulation requirement to obtain a relatively high low-shear viscosity along with a low
high-shear viscosity. The chapters on Thickeners_Associative Thickeners, Thickeners_Polymeric
Thickeners and Thickeners_Wormlike Micelles cover some key themes.

432 FST
 Surfactancy Coacervation

The main theme is that entanglement provides viscosity. With conventional high MW polymers,
entanglement is easy to obtain, but although shear thinning occurs (Flow_Shear dependent) it is not the
rapid decline that is often required in paints, and personal care products. HEURs and Wormlikes are
entangled, but now we get the desired properties because the hydrophobic forces that create the tangles are
easily broken by shear … then recover rapidly.

If you can get the desired properties in your specific formulation with one of those three approaches, then
life is easy. When they each fail, for different reasons, it’s tempting to think that a combination of two of
them might solve the problem. The point of this addition to the Coacervates story is that many such
attempts are guaranteed to fail, for reasons that are not at all obvious.

Ternary phase diagrams

Most of us are uncomfortable with ternary diagrams. The
Surfactancy_Phase Diagrams chapter attempts to reduce our
discomfort and the screen shot from [Link]
practical-surfactants/[Link] shows how to read the values at the
point shown by the circle. The trick is to read from a line that is
parallel to the face of the triangle opposite the component of interest
and from the scale that reaches 100 at the component of interest. The
line relevant for Water is horizontal, so it intersects the blue and the
green scales. But the blue scale reaches 100 at W so you read that it’s
46%. The Surfactant line is parallel with the right edge of the triangle
and because it is the red scale that reaches 100 at S, you read off 27%.

The next thing we need to be comfortable with is drawings of “tie

lines”. Before we do, note that in this phase diagram (from
[Link] the
order of W, S, X has been shifted from the previous plot and the
helpful grid showing the axes has been removed. It is frustrating that
this view is different from the previous one, but you have to accept
that everyone plots their phase diagrams the way they happen to like
them. You have to get used to reading whatever is put in front of you.

This time, the circle is at [W67, X17, S16] and is in the middle of a
group of lines within a boundary. These are tie lines. The entire region inside the boundary is unstable – the
system spontaneously breaks into 2 phases. We often say that “the surfactant (or polymer) crashes out of
solution” which implies a region with pure solution and a region of pure surfactant/polymer. That is almost
never the case. Instead you have a region with a high water content and a low level of surfactant/polymer,
then a region of high surfactant/polymer and lower water content. The tie lines tell you the contents of the
two phases. In this case at one end you have 77% water and 3% X (maybe our polymer) and at the other
end it is 45% water, 50%X and 5% S.

One final detail about tie lines. In the example above they are
more-or-less horizontal. These are “segregation” tie lines. If
they are more-or-less vertical then they are “association” tie
lines. We see the difference in this diagram. The segregation
type simply gives region of higher and lower concentrations
of the respective ingredients. The association type has regions

FST 433
of higher and lower concentrations of both ingredients. If you want a single phase, you may not be
concerned whether phase separation is via segregation or association. But if you find two phases, although
it’s obvious from the diagrams what is going on, to optimize a real-world formulation, understanding which
type of failure you are looking at, gives you some root cause and cure ideas.

Clues from the CAC

We are all familiar with the critical micelle
concentration, CMC (Surfactancy_CMC and
Langmuir) which we can obtain by plotting just about
any parameter such as surface tension or conductivity
versus surfactant concentration. Here we plot surface
tension in blue. At the start, the value is that of the
water. When the plot reaches a plateau, extra surfactant
is going into micelles rather than reducing surface
tension further.

In the orange curve we start with a solution of our

desired polymer, and the polymer itself has reduced the starting surface tension. As you add surfactant, at
the Critical Association Concentration, CAC, extra surfactant goes into associating with the polymer, so the
measured parameter remains constant. The association is in the form of local micelles formed next to the
polymer chain. The CAC is not dependent on the polymer concentration. As you continue to increase the
surfactant concentration, at the saturation point (which does depend on polymer concentration) the extra
surfactant is free, so you carry on down the curve to the CMC.

Why do you care about this? If you attempt this and find an unchanged CMC, you know there is no
significant surfactant/polymer interaction. If you find a reasonable CAC you get a good idea about
surfactant polymer interaction and can use data such as the saturation concentration to work out surfactant/
polymer interactions. If CAC << CMC then you probably have an undesirable phase separation and it may
not be worth exploring the system much further.

How do different types of surfactant interact with different types of polymers? We can answer this in a
simple table:

Surfactant Polymer CAC? Comment

Ionic Opposite charge CAC << CMC Coacervate
Ionic Polar nonionic CAC < CMC Usable
Nonionic Polar or charged No effect No effect
Zwitterionic Polar or charged No effect No effect
All Hydrophobic CAC < CMC Usable

434 FST
 Surfactancy Coacervation

Phase diagrams

The sorts of phase diagrams of interest to us are really small

corners of the full triangle. Here we go from 100% to 90% water
and 0% to 10% of surfactant S and polymer X. The (associative) tie
lines are in the range of 98% to 94% water and just a few % each of
S and X. So when in the following we talk about single phases we
mean “single phase in the region of interest”. Outside this region,
where we might have 20% surfactant, we are likely to have a single
phase because the effects of interest here are swamped, so there is
no need to explore the larger diagram.

It is a general rule that polymers of different types tend to phase separate. There is usually some enthalpic
penalty (“unlike isn’t compatible with unlike”) and entropic gains from mixing are too small to
compensate. For the purposes of this discussion, we can call a micelle a polymer, in that it requires some
extra entropic boost to promote miscibility with a real polymer.

As an example, an ionic surfactant’s counterions are relatively constrained when ordinary micelles are
formed, and are more free when the surfactant associates with a polar nonionic polymer. This extra freedom
allows a single phase to be formed. If you add salt which reduces the constraints of the counterions around
the micelle, then the entropic gain is smaller and you get segregative phase separation. If, for whatever
reasons, you have to have a surfactant and a polar nonionic polymer, then avoid extra salt and, if possible,
choose a surfactant with a relatively small tendency to produce large micelles.

You can get some entropic gains from oppositely charged surfactant and polymer when the charge densities
aren’t too high, so CAC < CMC. But stronger interactions give associative phase separation. This takes us
back to coacervates, and we’ll discuss some more once we’ve finished this overview.

When surfactant and polymer are the same charge, there’s no chance for entropic gain, so there’s phase
separation.

Similarly, nonionic surfactants and polar polymers have no reason for entropic gain, so there’s phase
separation.

This leaves us with one useful category: surfactants of any type with
hydrophobic, or hydrophobically modified, polymers. Because the system is
driven by hydrophobic interactions, there is plenty of room for compatibility.
Indeed, if you take a typical HEUR-like associative thickener (see
Thickeners_Associative Thickeners) and start adding surfactant, the viscosity
can increase. The micelles can start linking up the HEUR in new ways. But too
much of a good thing is always bad. Although viscosity increases for a while, the
micelles start interfering with the HEUR mechanism, so the viscosity starts to
decrease.

The way out of this problem is to have micelles that themselves can tangle. A
combination of wormlike micelles and HEUR-like systems can give higher viscosities than either system
on its own.

FST 435
Back to coacervates

Here we show three possible coacervate zones. Assuming that we

want a phase separation (for example to deposit a hair conditioner)
how do we organise the system to deposit in one of our chosen
zones?

The fancy answer is that we do systematic phase diagram

explorations checking for turbidity over a range of different WXS
ratios, shown with the blue dots. We do these while varying ratios of
key parameters such as:

• Surfactant tail length

• Charge concentration on the polymer
• Salt concentrations

Even with that sparse blue dot matrix, that’s 45 experiments per variant, so with “hi” and “lo” values for
each variant we’re up to 270 experiments. In practice, most of us aren’t going to do this.

The preceding sections provide some heuristics for some minimum experiments to reach a satisfactory
outcome.

Increasing the surfactant tail length makes both the CMC and the CAC lower. Because this is a coacervate
CAC, maybe 2 orders of magnitude smaller, that means a shift in the direction of 2.

A higher charge concentration leads to stronger attraction of the surfactant. This will shift one edge in the
direction of 2 but will also extend the recovery in the direction of 3. There will also be an extension of the
phase separated zone towards the X direction.

Although salts can encourage phase separation for ionics + neutrals by reducing entropic gains, they also
reduce the attractions of the oppositely charged surfactant and polymer for coacervates. This will lead to a
smaller 2-phase zone, somewhere between 1, 2 and 3.

These intuitions will not be precise, but they are still helpful. We
know that we will be diluting our formulation from somewhere
outside the small subsection of the ternary diagram and that it will
reach the 100% water corner. If we happened to have formulated
with a coacervate in regions 2 or 3, then our formulation will fail to
deliver whatever package we intended. Instead of a the careful
analysis of the phase diagram via 270 experiments, the idea would
be to identify potential dilution curves and test for the appearance of
coacervates. This has the advantage that the experiments themselves
are much simpler – start at some interesting concentrations of salt,
potential surfactants, potential polymers and measure turbidity during titration with water. With one or two
formulations identified as interesting, add a few well-chosen blue dot WXS experiments to be confident
that there is sufficient margin of error in the dilution curve. Region 1 in the diagram is robust against
modest starting-point errors. If you had a success from just touching Region3, then the extra experiments
would confirm that this would not be a reliable formulation strategy.

436 FST
 Surfactancy Coacervation

It's not easy

The first draft of this chapter emphasised that there are too few usable tools to help us formulate these
complex systems. The extra insights in the extended version still don’t provide the sorts of tools that could
make your life much easier. Instead, the hope is that with some extra logical background, it will be
somewhat easier to think through why things go wrong or go right and what adjustments might be needed
to accommodate changes required by other parts of the formulation.

If any reader has some better ideas they would like to share with the community, I’d be delighted to update
the chapter.

FST 437
Surfactancy CPP and phases
Links
Dishwashing Liquids, Hair Shampoos

Critical Packing Parameter, CPP, is really only relevant for concentrated surfactants in water and
salt solutions. In principle it can help explain the various liquid crystal phases we observe in such
solutions, though careful academic work shows that the predictive power is not very impressive.

It can be argued that it is useful for understanding salt and cosurfactant effects that can change a
surfactant solution from low to (super) high viscosity.

A, V and l
A surfactant molecule has a head area of A, a tail volume of V and a tail length of l. We can then calculate
the CPP via:

V
CPP = Al

If we draw this as a cartoon and happen to have V ~ Al, so CPP~1 we find:

[Link]

When the shape is “balanced” then the molecules tend to pack in a cubic lattice. A range of examples that
you can find in the app (in this case by changing only V) are:
 Surfactancy CPP and phases

If you keep the same V and l but change A then you can go from spherical micelles to an hexagonal
arrangements (highly viscous) then onto lamellae, cubic (again viscous) the inverse phases. This is often
done with simple surfactants like SLS or SLES which have relatively large effective head areas because of
charge-charge repulsion. As you add NaCl, the charge-charge repulsion decreases, the head area decreases,
so CPP increases.

The graphics showing the structures were kindly generated by Seth Lindberg at P&G.

My assertion that CPP is a worthless idea has been regularly challenged. Eventually I had to admit
that it had some use in thinking through wormlike micelles. More recently I had to double the
number of applicable domains from 1 to 2. That’s a small number of plausible uses for an idea that
features so heavily in surfactant science courses.

The hexagonal structures can elongate into Thickeners_Wormlike MicellesThe highly viscous wormlike
phases give the impression of luxury formulations, appreciated by those making Soaps and Washing and
Hair Shampoos.

Packing parameters
CPP is all about concentrated surfactants in water. The idea is essentially worthless for understanding
emulsion behaviour because we have oil which interacts in unknown ways with the tail regions of different
surfactants and, as we know from Surfactancy_HLD, systems have complex behaviours and any surfactant
can, in the right circumstances, be o/w or w/o. An alternative approach to HLD uses the idea of Packing
Parameters which reflect the current geometry of the system. There is unfortunate confusion between the
valid PP approach and the invalid CPP approach to emulsions.

If you want to describe HLD effects by saying that the oil, salt and temperature change the PP, that’s fine,
but because there is no viable way to predict PP values in real formulations the idea is noted simply for
completeness.

FST 439
Surfactancy Critical Capillary Number
Links

There are at least two cases where a Critical Capillary Number, the ratio of intertial versus surface
tension forces, has a big impact on surfactant-related formulations.

Capillary numbers
Across the whole of fluid flow phenomena the same “dimensionless numbers” keep cropping up. One of
these is the Reynolds number, familiar in terms of the transition from laminar to turbulent flow. Another is
the capillary number, Ca, which is the ratio of inertial to surface tension forces. For a fluid with velocity U,
viscosity η and surface tension γ:

Uη
Ca = γ

The dimensions of the units (m/s, Pa.s, N/m) cancel out so Ca is dimensionless.

Ca and Emulsification
Many of us have tried, and failed, to make an o/w
emulsion when the oil happens to be viscous. [The same
logic applies to w/o emulsions but we will focus on o/
w]. It’s not that it’s difficult, it’s impossible (at least for
normal dispersion methods). The problem is shown via a
graph:

The Y-axis is the capillary number required to create an

emulsion. The definition of Ca is based on the
interfacial tension (IFT), γ, the radius of the desired
drops, R, the viscosity of the continuous phase, ηc and a
velocity gradient G which depends on your disperser
speed.

ηcGR
Ca = γ

It is no surprise that you need a certain value of G, and therefore of Ca to be able to break up the oil drops.
If the viscosity of the oil, ηd for dispersed phase, is approximately the same as ηc for the water continuous
phase then the emulsification is efficient. If your aqueous phase contains some additives that raise its
viscosity and the oil stays the same viscosity, you have to work harder to get an emulsion. At a ratio of 0.1
you need relatively almost 10x the capillary number.

The surprise is when you try to disperse a viscous oil. Once it reaches 4x the viscosity of the aqueous
phase, the required Ca is infinite – you can’t disperse it.
 Surfactancy Critical Capillary Number

For those who need to create an emulsion of a viscous oil in water a standard method is to use
Surfactancy_Emulsion Inversion.

As we shall see in the next section, an alternative to increasing G by increasing the energy of the disperser,
is to decrease the IFT.

Removing oil
Although oil removal is a general problem, a specific example is removing remnants of oil from an oil well,
a process called EOR, Extended Oil Recovery. A similar process is removing chemicals from contaminated
soil.

If you pump a surfactant solution through the narrow pores which are coated with oil or contaminant, it
follows the general rule that low viscosity solutions simply punch a hole through higher viscosity liquids
and the results of your efforts are some slightly oily surfactant solution. But if the capillary number is
greater than a critical value, the removal becomes efficient.

[Link]

Here, φ is the amount of oil remaining and CB is a combination of capillary and Bond numbers which, for a
horizontal process, is simply the capillary numbers – the Bond number captures extra buoyancy forces for
vertical extraction.

Because increasing Ca via velocity or viscosity requires greatly increased pumping energy, it is more
effective to increase Ca by reducing the IFT. The best way to reduce the Surfactancy_Interfacial Tension
and Rigidity is via smart use of Surfactancy_HLD. It is no coincidence that much of the scientific work on
HLD came from the oil industry.

FST 441
Surfactancy Dynamic Surface Tension
Links
Dishwashing Liquids, Hair Shampoos, Soaps and Washing, Laundry Liquids, SkinCare, Surface Cleaning,
Water-based Adhesive, Emulsion Polymers

For most formulation purposes, it is not surface tension but dynamic surface tension (DST) which
matters. Classic surface tension is measured over minutes, formulations require responses in
milliseconds or seconds. Being able to measure and understand DST is, therefore, rather important.

Classic DST
Countless academic papers on DST consider sigmoidal plots like these:

[Link]

Such plots can be analysed via complex theories, yet the results of the analyses seem not to offer many
usable insights. This frustration was sensed by Rosen who (as per the app) provided a set of parameters that
could at least identify power laws (how sharply the surface tension decreased) and timescales (central point
of the rapid decline).

For the formulators this isn’t much help. If you use a linear, instead of a log plot with the same data, it
doesn’t look too interesting:
 Surfactancy Dynamic Surface Tension

Changing the X-axis scale to 1 second (not shown) we can see that the surface tension halves in about 0.3s.

If you continue to read the Rosen analysis included with the app, there is a long chain of reasoning which is
hard to master. It would be nice to have an alternative view.

DST Choice

[Link]

Instead of Rosen’s long chain of logic based on plausible parameters, we have a direct calculation based on
parameters described in the Surfactancy_CMC and Langmuir chapter. We have Γm, and a (which is 1/K
from the CMC app) and the CMC. That is the left-hand graph. An additional factor is the diffusion
coefficient D which varies surprisingly little over a broad range of surfactants.

From those inputs we can calculate the DST behaviour. The value of τD is the characteristic diffusion time
which defines when the surface tension starts to fall sharply. This depends on D (of course) and also on h,
which is the depth of liquid needed to provide enough surfactant molecules to reach the surface during that
time. The larger h, the larger (worse) the τD.

h2
τD = D

FST 443
For good DST behaviour we need a small h which is given by the core surfactant parameters and C which
is the concentration of free surfactant:

Γm
h=
(
a 1+ a
C
)
For a fast DST you need a large C. But if you have a “good” surfactant with a small CMC, C hits a
maximum at that point – after that, any extra surfactant goes into creating micelles. So for good DST
performance you need a “bad” surfactant with a high CMC.

This explains why coating formulators regularly add “acetylenic” surfactants which, at concentrations in
the range of 0.1% (large by normal surfactant standards) produce excellent DST behaviour. And, yes, these
have high CMC values.

The calculation of DST with time is done by a “standard” equation from Ward and Tordai. None of
the standard references tell you how to implement this complicated differential equation.
Fortunately, Prof Paul Stevenson needed to use this equation and his bright graduate student,
Xueliang Li, found a relatively easy way to implement it. They then kindly published their method
which is, indeed, easy to implement.

It is frustrating that so much potentially usable science is unused because smart academics don’t
take the trouble to tell the rest of us how to implement their smart theories. And wonderful when
you find those who do.

Measuring DST
The standard modern technique is a bubble pressure tensiometer. A bubble of known radius R is created at
2γ
the end of a capillary. There is a maximum pressure of P= R . The bubble can be created over timescales
from 10ms upwards, with the maximum pressure, and therefore γ, decreasing as the speed of creation
decreases. It is, as always, somewhat more complicated than that, but the principle is simple and direct.

444 FST
 Surfactancy Emulsification

Surfactancy Emulsification
Links
Sun Screens, SkinCare, Water-based Adhesive, Emulsion Polymers, Pharma Formulation

The language of “macroemulsions”, “nanoemulsions” and “microemulsions” is confusing. The

first are in the µm size range, the second in the nm range and both are only kinetically stable. The
third are also in the nm range but are thermodynamically stable, you can create them by simple
mixing.

Macroemulsions
Put a disperser into a continuous phase (say water) containing a dispersed phase (say oil) and the oil drops
are broken into smaller ones (for an exception when the oil is viscous see Surfactancy_Critical Capillary
Number). These drops readily coalesce Surfactancy_Emulsion Coalescence.

Now put a suitable surface active agent into the mix so the interfacial tension (IFT) is reduced and multiple
things can happen:

• The driving force for droplets to coalesce (their surface tension) is reduced.
• There can be a charge or steric barrier around the oil drop, reducing the tendency to coalesce.
• There are Marangoni effects that move surfactant molecules to fill any random holes in surfactant
coverage, helping to protect the drop.

It remains unclear which of these is the most important. What is clear is that the thermodynamic energy
required to create the drops is very small, yet the kinetic energy used to create them is very large – the
process is massively inefficient. A simple app captures much of this:

[Link]

If the emulsion is 100 ml of 1µm drops with 1g of surfactant of MW 400 and a head area of 45 Ų, then
you’ve created 2.1014 drops, with a total surface area of 600m², equivalent (this is a popular unit for
emulsions) to 1/10 of a football field, and have covered the drops to 109%. Perfect. The energy needed to
create that surface area is 3 J, but you would have used kJ of disperser energy.

To give a feeling of what it takes to create small emulsion drops, the Emulsion Drop Size Model brings
together some complex physics to provide some estimates:
[Link]

You must read the text of the app to know what all the inputs and outputs mean, but the take-home message
is that “it’s complicated”. So it’s not surprising that macro-emulsification remains more of an art than a
science. With a key exception.

For those who have an efficient dispersing system and can spend the necessary time to reach equilibrium,
the previous model relating surfactant to surface area does a reasonable job of telling you what you can
hope for. If, in the example, the surfactant had been present as 0.5 g, giving 50% theoretical coverage, then,
by simple arithmetic, the final drop size will be ~ 2µm instead of 1µm.

Choice of surfactant

There are a few unhelpful historical “rules” to help choose the surfactant. Bancroft’s rule says: “The phase
in which an emulsifier is more soluble constitutes the continuous phase.” Those who love HLB say that you
get o/w with high HLB and w/o with low HLB. As is well known, both rules have so many exceptions as to
be worthless.

Using Surfactancy_HLD it is simple. Knowing the temperature, salinity, oiliness of the oil (EACN), you
need a surfactant with a characteristic value, Cc that gives an HLD in the range -0.5 to -1 for an o/w or 0.5
to 1 for a w/o. It’s as simple as that.

Because we know that HLD values closer to 0 give low IFT, and because generation of drops is much easier
at low IFT, an efficient way to create an o/w emulsion is to have HLD ~ -0.1. Unfortunately, the low IFT
also means that there is strong Surfactancy_Emulsion Coalescence. Hence the trick common to ethoxylates
to raise the temperature to a point where HLD ~ -0.1, obtain a fine emulsion relatively easily, then rapidly
cool the emulsion to HLD ~ -1, where the emulsion is stable.

The reason that HLD lets us formulate emulsion rationally is that it is a curvature-based theory. There is
complete agreement that -ve curvature gives o/w and +ve curvature gives w/o; after all the Winsor R value
is all about this. What is not clear is why the curvature has this effect.

The great Langmuir thought he’d spotted the obvious reason – the surfactants at the interface curve in the
same way as the emulsion drop. This is obvious but wrong. The curvature is over the nm scale while a 1µm
drop is flat when viewed at that scale. An ingenious way to relate nm-scale curvature to µm-scale drops is
to focus on the moments when, during their creation, drops have a chance to break up o/w or w/o. That
breakup will take place starting with a nm-scale hole. If that curvature is the right way round (leading, say,
to an o/w net result) then the hole will open and extend. If it’s the wrong way round, it will not open:

446 FST
 Surfactancy Emulsification

In the top pair, the curvature naturally favours the

opening from (for simplicity of drawing) a large
tail and small head, giving a w/o emulsion (HLD
>0). With an HLD < 0, it would be hard to open
the water drop, so oil drop openings will be
favoured so you get o/w.

This “Kabalnov wedge” theory is controversial,

and others might reach similar conclusions via
other routes. What’s not in doubt is that curvature
rules emulsification and the HLD approach to
curvature is the most practical tool available to use.

Nanoemulsions
Nanoemulsions have a key advantage that their rate of Surfactancy_Emulsion Creaming and Flocculation
is very low, so if they are also well stabilised against Surfactancy_Emulsion Coalescence they can be great
emulsions. The three classic problems for nanoemulsions are that:

• They take a lot of energy in the disperser system to create the small drops and, as the Emulsion
Surface Area app emphasises;
• They require more surfactant to cover the extra surface area;
• They are prone to Ostwald ripening.

The first reason is why it is popular to make nanoemulsions via clever Surfactancy_Emulsion Inversion
schemes where “inversion” can be used loosely.

A classic example is the ethoxylate trick of getting to HLD = 0 via temperature (so there is no necessity to
invert by going to HLD > 0) so that very small droplets at super low IFT are easy to make. With a rapid
cooling quench to restore HLD to ~ -1, the IFT increases so the barrier to coalescence is raised.

There are similar tricks that can be played with intelligent use of the HLD equation and different EACN
values for the oil (start with a low EACN to create HLD ~ 0 then “quench” with a high EACN oil to reach
HLD ~ -1), or do a similar trick with a surfactant, starting with a higher Cc then quenching with a low Cc
molecule.

The alternative route is to use catastrophic phase

inversion where you start with a small amount of
water so the system is forced to be w/o then add
enough water to create the inversion to the correct
o/w state. The reason this works is that any flipping
of curvature, via HLD or catastrophic phase
inversion, goes via a point of low IFT. The smart
formulator works out how to do the flipping
through knowledge of the Salager diagram that
shows both inversion points with the fine
emulsions that they can create. This diagram from
the [Link]
surfactants/[Link] app makes this point.

FST 447
There’s nothing you can do about the need for more surfactant – the surface area has to be covered.

The issues around Ostwald ripening are covered in that chapter: Surfactancy_Emulsion Ostwald.

Microemulsions
Creating a microemulsion is easy. Set up the ingredients so that when combined HLD = 0, have enough
surfactant to allow all the dispersed phase to be solubilized, and carry out a normal low-energy mixing to
allow the microemulsion to form.

A frequent criticism of this strategy is that it requires “too much surfactant”. Yes, a naïve formulation with
no attention to the NAC part of HLD-NAC might require too much surfactant. By attending to those
aspects of head area and tail length that relate to efficiency and by either adding “linkers” to extend the
head and tail lengths or using “extended surfactants” that have a neutral (propylene oxide) middle section,
efficient systems can be created. You should also acknowledge that it is easier to create efficient
microemulsions with smaller rather than larger EACN values.

A problem in practice is that a new batch of the “same” surfactant or oil might have a different Cc or
EACN so your carefully-tuned formulation might not work. That is why a culture of “scans” to measure Cc
and EACN values is important for success – you can readily adjust your surfactant or oil blend to
accommodate batch-to-batch variations.

What spoils many such schemes is that surfactants can form strange phases with other surfactants. This
means that an otherwise perfect blend based on Cc values might fail because the surfactants tie each other
up in a nasty phase. It has also been found that although microemulsions are thermodynamically stable, the
order of addition can create kinetically stable alternative states that block the formation of the
microemulsion.

So this chapter comes with a negative guarantee. I cannot guarantee that you can get a perfect emulsion by
following a few basic rules – our state of knowledge of surfactant interactions is too limited. But I can
guarantee that if you don’t follow the rules, your chances of success are a lot smaller.

448 FST
 Surfactancy Emulsion Coalescence

Surfactancy Emulsion Coalescence

Links
Water-based Adhesive, Pharma Formulation

Coalescence, the joining of smaller emulsion drops to make bigger ones has two impacts. The first
is obvious, when we see a fine emulsion fall apart into bigger and bigger drops that cream easily.
The second is hidden, when the two drops we’ve just created with our disperser coalesce quickly,
requiring another pass of the disperser to get them to re-form.

We can’t model the processes exactly, but we can learn a lot from the theory and the app.

Diffusion, Collisions, Coalescence

To coalesce, the drops must collide and to collide they must diffuse through the continuous phase. So we
start with the diffusion coefficient, D, for a drop of radius r in a medium of viscosity η and where kT is
Boltzmann constant times temperature:

kT
D= 6πηr

The coalescence rate, CR, when you have n particles per unit volume is given by a two-part equation. The
first is the collision rate and the second represents the chances of a collision resulting in coalescence,
involving the activation energy, E, the barrier over which the drops have to go in order to merge:

E
CR = 8πDrn2.e− kT
Substituting for D we end up with:

E
4kT 2 − kT
CR = 3η n e

V 1
Because the radius cancelled out, it looks as though it is not important, but because n = Vdrop there is a r6
dependence, one of the many reasons it is harder to make small emulsion drops. From the coalescence rate
we can calculate the number of particles at time t, nt, starting from an original number n0via:

1
nt =
1
n0 +t ( 4kT
3η )exp(− kT )
E

nt
We can then plot n over time:
0
[Link]

In this screen shot, the number of drops decreases to 20% of the original over 1 hour. If you change the
radius, the % oil or the viscosity, timescales change somewhat. But change E from 20 to 18 kT and
coalescence is almost complete in 20 min.

That is the strength and weakness of the model. It is good that it makes us focus on the size of the
activation barrier, but it is hard to know how to engineer the system to ensure a large E.

We can imagine some terms that make up E. To help us we need to look at two drops joining:

There are 3 sorts of barriers:

1. Charge or steric repulsion we can lump under Dispersions_DLVO, which depends on the net
charge or the steric repulsion;
2. Interfacial rigidity which depends on the surfactant packing in the interface;
3. An effect around the need to create a fresh curved interface which depends on the IFT – with a
low IFT leading to low stability.

Effects 2 and 3 are discussed in Surfactancy_Interfacial Tension and Rigidity

Although we can say that E = EDLVO + ERigidity + ECurvature and although we can appify many aspects of
the above ([Link] it is more important to
understand the principles.

For DLVO stabilisation, a large charge on the surface and a low salt concentration helps for o/w but fails
for w/o because (as discussed in the app) charge effects don’t work in environments with low dielectric
constants. For steric barriers, a large ethoxylate chain can provide the stability for o/w and a large

450 FST
 Surfactancy Emulsion Coalescence

hydrophobic tail provides the stability for w/o.

Interfacial rigidity is relatively little studied. One common observation is that “bad” surfactants, such as
polymerics, which have problems of slow kinetics when forming emulsions are “good” once at the
interface as they are hard to remove.

Although we tend to think of coalescence as applying only to emulsion drops once formed, very rapid
coalescence during emulsification means that you can’t make the emulsion. So it is worth stating here that
the IFT effect is also relevant during Surfactancy_Emulsification because, as the “Kabalnov Wedge”
shows, the curvature at the narrow bridge when drops are prospectively coalescing might be the one way
round to encourage it or the other way to discourage it. If you are trying to create an emulsion and, at some
point, had 50:50 o/w and w/o drops, if the curvature encouraged w/o drops to coalesce then the resulting
emulsion would be o/w.

FST 451
Surfactancy Emulsion Creaming and
Flocculation
Links
Water-based Adhesive, Pharma Formulation

The rise of oil emulsion drops to the surface of an o/w emulsion is “creaming”. If you are
interested in the settling of water drops in a w/o emulsion, just invert the direction.

Although we start with simple creaming, we swiftly move on to flocculation.

Stokes Equation

[Link]

Creaming is governed by Stokes equation. The velocity v depends on the radius r, the viscosity η and the
density difference between oil and water, Δρ. In addition, the Richardson & Zaki dependence on volume
fraction φ is included:

2gr2Δρ 5.65
v= 9η (1 − φ)

To avoid creaming, increase the viscosity, increase the volume fraction of oil and decrease the drop radius.
Because the drop radius is likely to increase because of coalescence, this simple equation is only of use to
get an idea of what to expect at the start of creaming. If you can reduce v at the start, the side effects of
coalescence are also reduced.

Flocculation
As the drops rise, they become more crowded and can behave either like a “floc”, i.e. a strong assembly of
individual drops, or they can coalesce into larger drops. The practical difference in behaviour isn’t
significant at this level of simulation – in each case you get drops that want to rise faster, blocked by the
steadily-increasing concentration of similar flocs/drops.

As explained in the app text, the complex process can be modelled via a sophisticated technique from U
Sofia. It requires two views:

1. The volume fraction of oil from the top (0%) to the bottom (100%) at a chosen time
 Surfactancy Emulsion Creaming and Flocculation

2. How the height of the clear zone increases over time

[Link]

On the left we see that after 60min an originally 30% tube has cleared below 87% of the tube, is unchanged
up to 30% and rises to the close-packed maximum of 67% at the surface.

On the right, the clear zone has risen from 0% up from the bottom to 50%, where it is plateauing.

It takes a while to get used to interpreting the plots, but after a while it makes sense. Flocculation is
complex, so it is better to carry out lots of virtual experiments before trying to interpret real ones.

The algorithm in the paper by Prof Gurkov’s team at U Sofia was hard to implement. Prof Gurkov,
when asked for help, found the text of their code written in old-fashioned, simple, C for a computer
that no longer existed. Fortunately, old-fashioned C is easily translated into modern JavaScript and
the app was quickly written.

Readers will have noticed a number of references to the surfactant science group at Sofia. They
have contributed massively to our understanding of surfactant formulation and have always been
generous with their help when I wanted to appify their work.

FST 453
Surfactancy Emulsion Inversion
Links
Emulsion Paint, Emulsion Polymers

Emulsion inversion is a powerful technique for producing products that might otherwise be
impossible to create. The downside of this power is that it can wreck the formulations of those who
don’t understand and embrace its power.

From one core diagram it is possible to develop a deep understanding of a phenomenon that causes
so much confusion.

Two Types of Inversion

The Y-axis of this diagram from
Salager shows that you can
invert an emulsion by changing
the Surfactancy_HLD around 0.
If you start with 50:50 oil to
water with a surfactant system
with HLD < 0 you will have an
o/w emulsion. If you increase
the HLD (e.g. for an ethoxylate
system, heating it, for an ionic
system by adding salt, then you
go via a Type III microemulsion
into a w/o emulsion when HLD
> 0.

But even if HLD < 0, if you

have only a small fraction of water, there isn’t enough continuous phase of water to make o/w, so you
switch to w/o. Theoretically this should happen around 70% oil (and the inverse, making o/w when HLD >
0, at 70% water) because you are beyond the close-pack limit of oil (or water) droplets.

This effect when there is insufficient continuous phase can be called “catastrophic inversion”.

In real life, the point of catastrophic inversion is uncertain as it depends on how far you are from HLD = 0,
on the direction of the transition and on how good or bad your disperser is.

Because those who, historically, specialised in catastrophic inversion seemed generally unaware of HLD
theory, the literature is very confusing. They would change some parameter and see an unexpected
inversion – unexpected to those who were unaware that that parameter (a different oil, change of salinity,
change of temperature, change of surfactant blend) might have changed the HLD.
 Surfactancy Emulsion Inversion

Low IFT
It is well-known that at HLD = 0 it is very easy to
create a fine emulsion, because the interfacial
tension, IFT, is very low. Such an emulsion is also
unstable, because of the low IFT, so you start your
emulsification at this HLD = 0 point then do
something (e.g. cool for o/w emulsions via
ethoxylates) to take HLD away from 0, allowing a
kinetically stable emulsion.

Because IFT is controlled by curvature, it is also

low at the catastrophic inversion point, where the
curvature suddenly flips. So emulsification is very
fast, and then requires you to keep going beyond the inversion point to enter a stable domain with a higher
IFT.

Accidental inversion in production

In the lab we can add all our ingredients in whichever order we prefer and start our disperser, so we tend to
get the results we expect. But in production the stirrer might be in the bottom 1/3 of the vessel and the order
of addition of ingredients may be fixed by different constraints. So it is possible to start up the stirrer with
the wrong oil-to-water ratio. If that happens to be in the right zone to create an emulsion that is the inverse
of what is required, and if the stirrer starts up rapidly, you might create an inverted emulsion. It can be very
hard to re-invert an emulsion, especially if the mixing/stirring is not very efficient.

This gives a rule for the starting of mixing in production. Know the oil-to-water ratio, know the HLD,
know which way round is right and which is wrong. If at startup you are already the right way round, you
can start your stirrer at full speed while adding the other ingredients. If you are the wrong way round, then
start the stirrer very slowly and keep it slow till you are safely in the correct zone.

After a long day of teaching surfactant science theory there was an “open questions” session. A
question appeared about rules of starting production with fast or slow mixing each giving different
problems for different products. Total panic – how to answer such an obscure question. A wise co-
trainer, seeing my panic, whispered “The inversion Z-diagram”. That’s why this “accidental
inversion” section appears in the FST. You can’t know everything, you have to be “out there”
learning from the real world and, if you are lucky and have a wise colleague, you, and others, learn
new things.

Emulsions of viscous oils

In the Surfactancy_Critical Capillary Number chapter you find that it is impossible (at least for
conventional dispersers) to create a good emulsion if the viscosity of the dispersed phase is more than 4x
that of the continuous phase.

If you need an o/w emulsion with a viscous oil, start with the easy problem of creating a w/o emulsion
using your desired emulsifier which is good for o/w, but with a small fraction of your total water. You will

FST 455
be able to make a w/o inverted emulsion. As you now increase the amount of water you reach the
catastrophic inversion point where IFT = 0, and you immediately get a fine o/w emulsion. Flood that with
more water to reach your desired ratio with oil and the emulsion is stable.

The moment of inversion is obvious. Because the oil is highly viscous, the w/o emulsion is also viscous.
But at the inversion point, water becomes the continuous phase so viscosity reduces drastically.

In some industries this way of making emulsions is routine. To those who have never heard of it, it is an
awesome way to make a seemingly impossible product.

Theory is fine, but it’s scary if you’ve never needed to use it. A company asked me to help them
achieve an “impossible” emulsion via inversion, in a system I knew nothing about. They also knew
that I’d never worked on a real inversion project. The Salager theory is so beautiful that I was able
to make some predictions that seemed reasonable.

Failure of predictions in complex systems is normal – real life doesn’t always follow simple
models. You learn from the failure. But when, as happened in this case, the predictions work out
OK, it’s most satisfying.

456 FST
 Surfactancy Emulsion Ostwald

Surfactancy Emulsion Ostwald

Links
Water-based Adhesive, Pharma Formulation

If you make a fine emulsion you often find that the few larger drops (see Size Distribution below)
get larger while the small get smaller. This effect, which regularly spoils otherwise great emulsions
is called Ostwald Ripening. A few tricks can help reduce the damage that Ostwald can cause.

The big get bigger

[Link]

The app shows that if the “average” size (see Size Distribution below) starts at a radius, r0, of 250nm then
after 24 hours this particular emulsion will have droplets over 800nm.

2γ
The “pressure” inside a drop is r , where γ is the interfacial tension, IFT. If some of the oil inside the drop
happens to migrate to a drop with a larger r, the overall energy of the system decreases. The original drop
now has an even smaller r, to the relative driving force increases. It’s a one-way journey, the big get bigger
and the small get smaller … till they disappear under the super-high pressure once they get into the low nm
range.

The rate at which this can happen depends on how easy it is for a molecule to leave one drop and migrate to
the bigger one. If the solubility of the oil, c, is high, then this process is easy. So one rule for avoiding
Ostwald ripening is to use insoluble oils. Hexadecane emulsions (we will return soon to hexadecane) do not
change size over time. Obviously the time taken also depends on the diffusion coefficient, D – the higher it
is, the faster the ripening.

The equation captures these factors:

 8γDcVt
rt3 − r03 = 9RT

V is the molar volume of the oil and we have RT, gas constant time temperature on the bottom. In the app,
T is assumed to be 298°K.

As expected from the driving force, decreasing γ, having a better surfactant, decreases the speed of
ripening. The other reason for having a good surfactant, which spends its time efficiently at the interface, is
that there is little risk of surfactant micelles carrying the oil between droplets.

The Hexadecane trick

As mentioned above, hexadecane has a such a low solubility in water that there is no drop-to-drop
transmission, so there is no ripening. The trick is that if you add a small amount of hexadecane to your real
oil, which might have a higher solubility and unacceptable ripening speeds, the ripening starts … then
stops. What happens is that the hexadecane concentration builds up inside the smaller drops as the oil
leaves, and this concentration difference between smaller and larger drops gives a chemical potential
difference (some call it an osmotic pressure difference) that balances the surface energy difference, at
which point the ripening process stops.

If you have water-in-oil emulsions then a little bit of salt in the water has the same effect in stopping
Ostwald.

Size distribution

[Link]

Discussion about Ostwald requires analysis of the emulsion drop size. The “average” size, mentioned
above, is ill-defined. Looking at the app, the size distribution has at least 4 numbers, each of which is valid
and, in the right circumstances, meaningful. D[1,0] is the “number average” diameter, based on counting
the total number of drops. D[3,2] is the “area average” and D[4,3] is the “volume average”. Although there
4 3
are a small number of larger drops, because volume is 3 πr , each drop contains much more oil than the

458 FST
 Surfactancy Emulsion Ostwald

smaller drops, so the average in terms of volume or weight is, in this case, 2x larger than the number
average. D50 is the size where half the drops have a mass less than it and half more.

This digression into particle size is important for discussions of Ostwald. You can’t understand the
phenomenon unless you:

• Have a good measure of particle size (most likely an analytical centrifuge);

• Understand how to interpret the particle size data.

The ripening equation describes the “average” radius, but which one. The r³ term in the equation tells us
that it must be the number average radius because the effect depends on its cube.

FST 459
Surfactancy Foam Drainage
Links
Dishwashing Liquids, Hair Shampoos, Soaps and Washing

Foams can easily fail by drainage – all the water runs out of the foam which then collapses. How
this happens, and how to avoid it, is non-intuitive but makes sense once explained.

Where is the water that drains?

This image from
[Link]
practical-surfactants/[Link]
gives an exaggerated view of the
Plateau borders at the edge of each
foam wall and of the nodes where
the borders meet.
 Surfactancy Foam Drainage

We can see this in more detail from this close-up. The key
fact is that although the volume of the foam is made up by
the faces/walls, the amount of water in those walls is <1%
of the total. So when we worry about drainage, we need to
worry about water running down drainpipes (the borders)
that intersect at the nodes.

Yes, once a lot of water has drained, then the walls start to
drain and collapse, but our focus should be on the main
drainage. Although the precise choice of surfactant might
affect the “disjoining pressure” (see
[Link]
[Link]), that keeps the walls from collapsing, because the
drainpipe drainage really controls the foam stability, it is a
surprise to find that the surfactant has relatively little to do with stability.

What controls drainage?

[Link]

Unlike most apps, the lines in this graph never change (except when you switch modes, explained below).
When you change the key parameters, the change is in the times recorded. Also, because the theory behind
the calculations is yricky, we delay discussing it until we find out what is and is not important for good
drainage behaviour.

First, we need to know what the graph means. Let’s look at the 60s line. It’s saying that although the foam
started with 0.1 fraction of liquid at the top of the tube, z=0, that wet front has sunk down to ~25mm from
the top, so that the foam from 25mm upwards has steadily decreasing amounts of water. After 2 min, the
wet front has fallen to 60mm from the top, and so forth.

How can we increase the time for the front to move downwards, i.e. to reduce drainage? As hinted above,
we don’t control the time via the surfactant. Instead we control it via the liquid fraction, ε, the viscosity, µ,
and the bubble diameter, D:

FST 461
 • Decreasing ε by a factor of 2 increases the time by √2 (for channel mode, below, this is 2x)
• Increasing the viscosity µ by a factor of 2, increases the time 2x
• Decreasing the bubble diameter, D, by a factor of 2, increases the time 4x

Generally you make a foam with the wetness (ε) required for your purpose and your viscosity is similarly
dictated by other factors in your formulation. The D² dependence tells us that for a long-lasting foam, you
should focus on creating small bubbles. See Surfactancy_Foaming where you might discover that we are
often bad at focussing on this key parameter for foams.

Before discussing how drainage happens, we need to check one more factor.

Surfactant effects
The nature of the surfactant doesn’t feature in the above list. Certainly the surface tension of the foam
makes no difference except to the uninteresting final curve which is the equilibrium one.

But there is one option in the app which is dependent on the surfactant. The plot shown is in “Node” mode
where the walls of the channels are so mobile that they are essentially liquid, so water descends in “plug”
flow and overall flow gets limited by the flow capacity of the nodes. If you switch to “Channel” mode, the
walls behave like classic pipes with viscous drag. This means that the drainage is generally slower.

A rigid surfactant or a floppy surfactant with a classic rigidifying additive such as myristic acid is therefore
preferred if you want to reduce drainage and have a tougher foam wall that will resist the effects of
drainage.

In the app, the differences can be hard to spot because they have different dependencies on ε. This means
that at large ε values, channel flow can be a bit faster than node mode.

The theory
The theory itself turns out to be simple in concept, but had to be developed over many years by different
groups to model the whole process as everything depends on everything as the drainage continues.

The drainage literature is confusing, partly because the problem is difficult, partly because
different academic groups have their favourite ways of addressing the issues, using different
notations and theoretical approaches. But sometimes you get lucky. An email to Prof Saint-Jalmes,
who had a theory somewhat different from other approaches, resulted in a generous explanation of
how the theories were complementary and how his approach was especially convenient for
creating an app that related nicely to his team’s extensive experimental data.

Here we just look at the core theory. The diameter of the pipes is directly related to the size of the foam
bubbles. If you halve the bubble size, you halve the diameter of the pipes. Although flow down a pipe goes
1 1
as 3, halving the bubble size also decreases the pipe’s length by a factor of 2, so the net effect is a
r r2
effect. The common theories concentrate on the pipe length L which conveniently is given by L = D/2.7.
We want a small L because over a given height of foam, the smooth flow through the pipe is interrupted
more often.

462 FST
 Surfactancy Foam Drainage

So our core equations for the velocity of the drainage front, vf, are, depending on density ρ and gravity g:

0.007ρgL2ε
Channel : vf = µ

And

0.002ρgL2√ε
Node : vf = µ

Again, the details don’t much matter and there’s not much you can do with µ nor with ε (which changes as
the foam drains, making the calculations more complex), so for stable foams you probably want a stiffer
surfactant system with channel-dominated flow, but mostly you should focus on getting smaller bubbles
which give the smaller length L, with more interruptions to smooth flow. It’s as simple as that.

FST 463
Surfactancy Foam Ostwald
Links
Dishwashing Liquids, Hair Shampoos, Soaps and Washing

We often see foams decaying via a decrease in the number of small bubbles as the large bubbles
get larger. This is Ostwald ripening, and the ways to tackle it are limited but clear.

Changing bubble size distribution

[Link]

The chapter on Surfactancy_Emulsion Ostwald uses a simple Ostwald formulation. Here we can take
advantage of the sophisticated work of the team at U Sofia who provided a complex algorithm and much
help to look at a more realistic case where there is a bubble size distribution.

2γ
The basics of Ostwald are simple. The pressure inside a bubble is given by r , so smaller bubbles have a
larger pressure. If some of the gas in the small bubble moves to a larger bubble, the larger bubble gets
larger and its pressure decreases. The next effect is that gas continues to move from smaller to larger
bubbles.

In a foam the process is complicated by the fact that as the bubbles change size, they pack differently so
there are kinetic barriers to the process happening.

The app gives you, in the left-hand graph how the bubble size distribution changes over your specified time
period while the one on the right shows the rise of the “average” bubble size. The word average is in quotes
because there are various definitions of average depending on number, area or volume averages, as
described below.
 Surfactancy Foam Ostwald

You define the starting distribution with a peak value and a starting width which produces a skewed
Gaussian.

Slowing Ostwald
The driving force is internal pressure. So a better surfactant, with a lower surface tension, will give a slower
growth. As you can’t change these values by more than a factor of 2, the effect on Ostwald is relatively
small.

The ability for a gas molecule to move from a smaller to a larger bubble depends on the solubility of the
gas in the water. CO2 has a much higher solubility than air, so CO2 foams (beer, espresso crema) ripen very
quickly. We can simulate this via the Permeability, k which is 50 in the screen shot given that we are using
air with a Henry constant ~ 0.015. If we add CO2 so that Henry becomes 10x larger, then k becomes 500
then we find that we get the same degree or ripening in 170s rather than 1200s.

But if you can’t change your gas from, say, air, how can you slow things down? The key is to reduce the
permeability of the monolayer itself, kml. A factor of 2 decreases k by a factor of 2 and it now takes 3000 s
to reach the same increase in size:

FST 465
How do we reduce kml? This depends on the stiffness of the surfactant layer. The standard trick used across
the personal care industry is to add some myristic acid (or similar) to the formulation.

The black myristic acid is useless at forming the foam,

that’s the job of the red surfactant which might be SLS.
As soon as the foam is formed, the highly insoluble
myristic acid prefers the interface and, because it’s
neutral, packs more tightly than the anionic surfactant.
The foam elasticity ([Link]
practical-surfactants/[Link]) increases and
the ability of gas molecules to permeate through the
foam wall decreases.

Bubble size
Using a method described in the text of the app, (it needs a 90° prism, an LED light and your smartphone
camera) it’s not so hard to get a good picture of your bubbles. Load your image into the app and you get the
bubble size distribution.

466 FST
 Surfactancy Foam Ostwald

[Link]

It is worth noting that when I needed to know how to measure the bubble size, it was, once again, the U
Sofia group who told me how to do it. They’d not invented the technique, but they’d worked out how to do
it effectively. Via the app, the technique has spread widely and is much appreciated. Occasionally someone
tries a 60° prism, but that fails. And although I say “use a smartphone camera” it’s not so easy to get a
good, even, in-focus image. Because the quality of image analysis is 90% the quality of the original image,
invest some time and experimentation into getting a high-quality image; it’s worth it.

FST 467
Surfactancy Foam Rheology
Links
Dishwashing Liquids, Hair Shampoos, Soaps and Washing

Foam rheology is about Modulus, Yield Stress and shear-dependent Viscosity. Four simple inputs
give us the behaviour, which in turn tells us that for a strong foam, small bubbles is the smart way
to go.

[Link]

Controlling foam behaviour

The basic “feel” of a foam is captured by the 3 components of its rheology:

1. Modulus – how springy it is if we try to compress it;

2. Yield Stress – its resistance to flowing;
3. Viscosity – how viscous it is when you get it flowing.

When the air fraction is φ, the surface tension is γ and the bubble radius is R then the modulus, G, is given
by:

1.4φ(φ − 0.64)γ
G= R

The yield stress, Y, is given by:

0.5(φ − 0.64)γ
Y= R

The equations for two types of shear-rate-dependent viscosity are provided in the app. The two types are
 Surfactancy Foam Rheology

1. Sliding friction – bubbles just moving past each other smoothly;

2. Surface dissipation – bubbles expand & contract during motion and therefore dissipate energy.

You can click between them to see their differences. It is easy to take a sliding foam from simple
surfactants and change it to a surface dissipation foam by making the foam walls more rigid via a small %
addition of myristic acid or similar poor surfactant-like molecules that prefer to pack at the interface once
the basic surfactant has created the foam.

Yet again, these core equations come from the team at U Sofia who have provided so much great, usable
foam science.

Make small bubbles

Although surface tension makes a difference, the practical range of control is limited. Viscosity is usually
close to water unless you add glycerol, so again there’s not much room for change. Going to small bubbles
is, therefore, the only way you can readily produce a “stronger” foam. The trick for making small bubbles
is to focus less on the surfactant and more on your foaming mechanism. As shown in
Surfactancy_Foaming, a good foaming technique can boost the foam more easily than playing with the
formulation.

FST 469
Surfactancy Foaming
Links
Dishwashing Liquids, Hair Shampoos, Soaps and Washing, Laundry Liquids, Toothpastes

There is no mystery to foaming – any bubble in any liquid can create a “foam”. Once there is
sufficient density of bubbles (a foam fraction above ~74%) a real foam is in existence. Now the
question is how stable the foam is. Most “reasonable” surfactants and plenty of nano-scale
particles or lumps of protein can provide the stability. So now we’re on to real foam volume and
quality. This is less dependent on the surfactants than is imagined.

Create small bubbles, quickly

This image from [Link]
[Link] shows what we are trying to achieve – lots of bubbles, maybe
with volume fraction over 90% packed into a stable structure. As the
description above sets out, just about anything can be made to foam and with
the right amount of air being chopped down to small-enough foam bubbles, a
credible foam can be created via any reasonable surface active agent that can
create a reduction in air-water interfacial tension or provide a semi-solid
array (“Pickering foam”) of small particles.

This description is very different from the standard descriptions of lists of

surfactants that are said to be strong or weak foamers, of special synergies of co-foamers and the general
confusion about foam quantity and stability.

It is also different from analyses of the stability of isolated bubbles where analyses of disjoining pressure
dependence on DLVO (see [Link] elasticity
(see [Link] are of interest. These effects
are real, but seem to play a minor role in real foam stability.

We can state this because a ground-breaking paper from the Denkov group in U Sofia, referenced in the
foam-making app [Link] used high-
powered theoretical tools to analyse extensive foam creation data from a wide range of “normal”
surfactants and concluded that there was no real difference between the quantity and quality of foam
produced, so the theoretical tools provided no significant insights.

Instead, they pointed out, the key to making a good foam is to ensure a rapid reduction in bubble size
through proper application of shear forces. As soon as you have small bubbles then Surfactancy_Foam
Drainage decreases (proportional to radius squared) and the Surfactancy_Foam Rheology shows higher
modulus and yield stress (proportional to 1/radius).

For those who want plenty of stable foam, concentrating on the foaming mechanics is more important than
worrying about surfactant formulations. For example, a weak and useless fire-fighting foam created via
inefficient aspirator nozzles, can be replaced with stunning stable foams using the CAF (Compressed Air
 Surfactancy Foaming

Foam) technique. A bubbly hand soap wash can be turned into a rich shaving foam using a Japanese
foaming net (rapid rubbing with a fine mesh that efficiently breaks up bubbles) or a simple latte whisk.

Bad foaming needs better surfactants

If you create a foam by forcing bubbles through a sintered disk, you are automatically guaranteeing a poor
foam because there is insufficient shear to make the relatively large bubbles smaller. So carrying out
detailed analyses in this “foam analyzer” is an analysis of an unnecessarily failed foam. Just because a
piece of equipment is flashy and complicated doesn’t mean that it produces data relevant to your
application.

If you use a kitchen blender, or repeatedly invert cylinders of foaming solution in other “standard tests” you
again are working with a foam that is less good than it should be.

So in these cases, the surfactant system matters. Getting good Surfactancy_Dynamic Surface Tension
effects, and using ill-defined foaming synergies become necessary and tedious tasks. The problem is ill-
defined and despite decades of work we seem to be no nearer to finding helpful scientific rules.

If your organisation is demanding that you make another me-too foaming product, but this time with a more
luxurious foam, try to innovate by not doing what everyone else is doing. Use your science and ingenuity to
create a novel system that makes a great foam from a me-too formulation. Astonish them with the quality
of a foam from a Japanese shaving net, or bring in a cheap latte whisk. Once colleagues see that it’s the
foamer, not the foam, they might be persuaded to embark on a fresh approach.

Take inspiration from the amateur YouTubers (some of whom I’ve enjoyed helping) who are keen
to make great foams for cement blocks, children’s parties, protecting crops from frost and other
amazing applications. Invariably they evolve methods that naturally create very small bubbles, or
break up bubbles efficiently from large to small.

FST 471
Surfactancy HLD
Links
Dishwashing Liquids, Hair Shampoos, Soaps and Washing, Laundry Liquids, SkinCare, Water-based
Adhesive, Emulsion Polymers, Pharma Formulation

If we are formulating with surfactants and oils, using different salt concentrations and
temperatures, it is a good idea to use a formula that captures all 4 factors, allowing us to move
around surfactant space as we adjust our formulations for different end uses, for changes of raw
materials and for going greener. Unlike all other systems, HLD lets you do exactly that.

Hydrophilic Lipophilic Difference

Let us have a temperature T, a salinity S in equivalents of g/100cc of NaCl, and oil with an oiliness EACN
and a surfactant with a characteristic value, Cc. We can calculate the HLD via the simple formula:

HLD = F(S) − [Link] − α(T − 25) + Cc

While k ~ 0.16 for most systems, the other parameters are specific to surfactant types:

• For ethoxylates F(S) = 0.13S, α = -0.06

• For ionics F(S) = ln(S), α = 0.01
• For APGs, Spans … F(S) = 0, α = 0 and for polyglycerols α ~ -0.01

Even if we don’t know what HLD means, these bullet points immediately alert us that formulating with
ethoxylates, ionics and sugar-based requires different tricks for each class, especially the sugar-based which
have no dependence on salinity or temperature.

Battling HLB

When HLB was first invented, it did a good job of describing some attributes of the then new
ethoxylate surfactants. But it didn’t take into account their large T-dependence and didn’t have a
coherent way to deal with oils. Somehow it got extended to all surfactants, despite the fact that
there was no objective way to measure their values. In the 1970’s the first steps towards HLD were
taken, but the HLB orthodoxy blocked adoption of the obviously better idea. If HLB had died back
then we would have had 30+ years of data gathering on HLD and we would all be in a better place.
Killing off HLB has been extraordinarily difficult, but its demise is accelerating, and it will,
hopefully, be soon forgotten.

The meaning of HLD

When HLD = 0 we have a system balanced between oil and water. The Surfactancy_Interfacial Tension and
Rigidity is super low making it easy to create microemulsions and to break up oil drops for emulsification
 Surfactancy HLD

and for cleaning oils from surface.

• When HLD < 0 we have oil-in-water curvature of the interface. If you want to make o/w
emulsions, it’s a good idea to know that your HLD < 0.
• When HLD > 0 we have water-in-oil curvature of the interface. If you want to make w/o
emulsions, it’s a good idea to know that your HLD > 0.
• Because for ethoxylates we can increase HLD with T we can take an HLD < 0 o/w system at room
temperature to the HLD = 0 inversion point at a higher temperature, and, if we choose, into HLD
> 0 to make a w/o system at even higher temperature. This simple fact means that to call any
surfactant (especially an ethoxylate) ready for o/w or w/o is misleading.

Using a graphic kindly

provided by VLCI in
Amsterdam, who are
experts in formulating
with HLD, we can see
the whole of HLD space
in one go, including the
types of emulsions
(called Winsor Types I,
II, III or IV), their visual
appearance and
solubilization
capabilities. It is
impressive both that a
single theory
encapsulates so much
and that VLCI’s graphic
is able to convey that information so elegantly.

Although the calculation of HLD is trivial, it is easier to get an app to do it, taking into account the different
coefficients for the different types of surfactant:

[Link]

Sometimes you find a theory out of desperation. I needed a microemulsion for a cosmetics project
and from the literature I could create one using decane as the oil. It had the right properties in tests
but I needed to change to IPM, isopropyl myristate. All it needed was a change in the surfactant
package as IPM is a different oil from decane. A search of the literature showed that HLB, CPP,
Winsor R etc. were useless. Oh, there’s another theory called HLD. A couple of weeks exploring

FST 473
 HLD in the literature and in the lab resulted in a perfect IPM microemulsion. Obviously everyone
would want to use such a simple and usable theory, so I put the first spreadsheets, programs and
apps into the public domain. That was in 2010 and people (though far fewer) are still saying that
they prefer HLB or that HLD is “too hard”.

What is the Cc of my surfactant?

Sometime in the future, all surfactant suppliers will provide the Cc values not just of each surfactant but of
each batch of surfactant. The last point is important. Commercial surfactants are mixtures and even small
changes of the chemicals in the mix can change the Cc by a significant amount. This explains the often-
found failure of a new batch of the “same” surfactant, or how the “same” surfactant from a different
supplier can behave differently even though standard analyses say they are the same. One of the reasons for
the existence of HLD is that we can rationally formulate both large changes (as in the diagram) and small
tweaks for different batches of a surfactant.

There is a list of currently known Cc values on [Link]

but for your surfactant you either need a good value from the supplier or measure it yourself. Here’s how to
do that:

[Link]

You have a set of tubes containing 50:50 oil and water. You systematically vary one of the variables
(salinity, EACN, Cc) across the tubes and you look for the 3-phase tube where HLD = 0, the swap between
o/w and w/o. Because you know all the other variables, you know the Cc of your surfactant.

In the early days of trying to popularise HLD, a small amount of funding gave the chance for
myself and VLCI to attempt to measure the Cc values of ~20 surfactants, kindly provided by some
companies who were willing to explore this relatively unfamiliar approach to surfactant science.

474 FST
 Surfactancy HLD

This was a mad thing to try to do. We had no apps for guidance, we’d only done a few such scans
before, and surfactants are often uncooperative We made lots of errors out of ignorance (we now
know never to mix SLS and AOT – they hate each other) and we had a few failures. But over a few
weeks we got reasonable values (or upper/lower limits) on ~ 15 surfactants.

The secret of success was hard work, self-criticism, iterative learning from failure … and having
the wise advice of Prof Edgar Acosta available by email and (back then) Skype.

Although scans are rarely as perfect as the screen shot, they are done routinely. If, for some reason, your
preferred scan gives strange surfactant phases then you have to try an alternative. One variant is to make a
scan with a known good surfactant that gives clean scan tubes, then to add a small amount of your new
surfactant. After shaking and equilibration, the 3-phase tube might have shifted by 1. This gives you a new
Cc for your mix and from a simple calculation based on mole fraction of the two surfactants, you get the
unknown Cc value.

[Link]

In this simple example, the Cc of a 50:50 wt% mix of surfactants with values of -1 and 1 does not give Cc
= 0 because #1 is a lower MW so there is a larger molar fraction of #1, resulting in the negative Cc.

This mixture rule is very powerful. If you need a precise Cc for a great formulation but don’t have (or can’t
afford, or don’t like the green credentials of) a single surfactant at that value, you create a mix from 2
available, affordable and greener surfactants.

Other scans are becoming increasingly popular. The Acosta Fast method (see a description at
[Link] is one, the PIT-slope method from U
Lille is another as is their clever salinity scan that uses smart mixing to increase salinity while keeping
other ratios constant. Some find that conductivity measurements are faster for identifying phase changes.
Choose whichever technique works for your specific circumstances. There’s a big difference between a
scan for a new, unknown surfactant, and a scan to detect subtle batch-to-batch variations of a known one,
ionics are different from non-ionics, ethoxylates are different from everything else. There’s no single right
answer, but the worst answer is to keep going with a slow and painful technique that discourages you from
getting the information you need.

What is the EACN of my oil?

The “oiliness” of octane is 8 and of dodecane is 12. Suppose we have an oil which is not a pure alkane but,
in a scan, behaves like decane? We give it an Equivalent Alkane Carbon Number of 10. If you do plenty of
scans with plenty of oils you can find values from 20+ down to -5. What is an alkane with -5 carbon atoms?
Of course it’s meaningless in terms of alkanes, but this linear scale lets us formulate a wide range of oils.

FST 475
How do we know the EACN? Again, suppliers should tell you their values. Or you could look them up on
the list on the site below, or you can measure them by scans, or you can create them from a rational blend
(this time in terms of vol%) of two known oils:

[Link]

As with Cc values, EACN values of the “same” oil can vary – for example typical plant-based oils. So
doing a scan to measure each incoming batch of oil tells you whether you need to tweak your formulation
for the new batch, e.g. making a slight adjustment to the surfactant ratio.

Polar oils
Although HLD can handle salts, temperature, surfactants and oils, it has a problem with “polar oils”. These
are molecules like octanol that can be considered an oil or a surfactant but don’t behave as either. A small
amount (e.g. a fragrance molecule in a formulation) can make a large change to HLD. There happens to be
a good theory for tackling polar oils issues. It is Tchakalova’s CIT (Constant Interface Thickness) approach,
captured in [Link] In practice, the approach is
far too hard for standard formulators, so we have to check their effects on our individual formulations.
Fortunately, the culture of scans that we use to measure EACNs, Ccs and batch variations can also be used
for polar oils. One example of doing this is the “Pit Shift Method” which is the 4th app inside
[Link]

Take a scan and find that the transition tube is, say, #4. Now add a small amount (e.g. 0.5%) of the polar oil
to each tube. After re-equilibration we might find a shift towards #5. Repeat with another 0.5% and now we
are clearly in #5. With luck (and it’s usually the case) you will find a linear relationship. This means that
you can rationally modify your HLD to take into account the amount of polar oil required in your
formulation. If you need to increase the amount for some other reason, you can rationally tweak, say, the Cc
blend to accommodate it. A shift to a different fragrance mix? Re-do the 0.5% scans and get a new
correlation.

Yes, it would be nicer if we could predict polar oil effects. But until we can, some quick scans can get us
quickly to the right formulation.

Cosurfactants, linkers, extended surfactants

HLD takes you to the right part of formulation space, but how much surfactant should you add to get the
required effect? HLD cannot tell you, you need the more powerful HLD-NAC theory,
[Link] A key factor in HLD-NAC is the ξ
value which tells you the efficiency of that surfactant with that oil (sadly there is no general-purpose ξ
theory so far). Here we don’t worry about ξ but take on board the fact that for a given oil at the same HLD
different surfactant blends at the correct Cc can have smaller or larger ξ values and, therefore, different
efficiencies.

If the efficiency isn’t as high as you would like then you can try adding some cosurfactants or linkers (the
terminology if vague). Many formulators get into a mess trying this because any extra molecule can change

476 FST
 Surfactancy HLD

both the HLD and the efficiency. Let’s show that with a diagram:

The additive might change the HLD only. If you weren’t aware of this change, you might think the additive
makes things worse. It might change the HLD and the phase volumes. Again, if you weren’t aware of the
HLD change you would think that the additive made things worse when in fact it is just what you required
– but at a shifted HLD (which you can shift back via the Cc blend). Yes, if you are lucky the phase volumes
get larger so your experiment would be a success. But formulators can’t rely on luck.

So each time you try an additive to improve the system, test the additive on some scan tubes and if you
have an improvement in efficiency (good) with a shift in HLD (bad), you just shift the HLD back to where
you wanted it.

How do cosurfactants/linkers work? By effectively extending the tail region and/or the head region to
improve the ability of the surfactant to control larger phase volumes. Why not just use a surfactant with a
longer tail? If you can, that’s the best option. But often a C14 surfactant will have the right solubility and
phase behaviour while a C16 version is insoluble or makes undesirable phases with the other surfactant in
the blend.

“Extended surfactants” are a different story. They have a similar head and tail region but add a “neutral”
polypropylene oxide middle region that doesn’t mess up the surfactant behaviour while increasing the
influence over the oil and water.

The disadvantage of the higher efficiencies of extended surfactants is that they are often slower to act.

Silicone surfactants and oils

The Cc and EACN numbers are linked via the traditional carbon chemistry of alkyl chains, aromatics and
modest functionalities such as esters. Silicones inhabit a different world. Although they are “hydrophobic”
that doesn’t mean that they are “oleophilic”; indeed, they are often insoluble in common oils.

This means that the simplest advice is “the best surfactant for a silicone is a silicone surfactant”. Or, in
other words, “Don’t try to use HLD for silicones”.

This general separation of standard and silicone surfactants provides insights into why, as discussed in
Surfactancy_Anti-foaming, silicone anti-foamers are generally more successful or, rather, less unsuccessful.

FST 477
Surfactancy Interfacial Tension and Rigidity
Links
Laundry Liquids, SkinCare

Interfacial Tension, IFT, is the “surface tension” between oil and water. It dominates how the oil
and water interact across the interface. When you are at an HLD = 0 balance (Surfactancy_HLD)
the IFT isn’t just low but super-low, enabling many desirable actions in our surfactant systems.

Interfacial rigidity
The aligned surfactants at an interface will add some rigidity to the interface. This rigidity, Er, has an
influence on what happens when emulsion drops collide and along with the Marangoni stabilization effect
(where surfactant molecules rush in to fix any temporary reduction in surfactant concentration at the
interface) exerts control over emulsions and emulsification. It is obviously linked to curvature values, as it
is harder to bend a more rigid interface. Yet it is little talked about and data on values is sparse. Instinct
suggests that those surfactants with a high rigidity will give emulsions that are more stable against
Surfactancy_Emulsion Coalescence, yet the relevant theory there uses an activation barrier, of unspecified
origin, rather than interfacial rigidity.

The only rule of thumb is that for ionics it’s around 1kT and for ethoxylates around 4kT. We’ll use those
values in a moment.

Interfacial tension
The energy to open up 1m² of fresh interface (J/m²) or the force exerted by the interface along 1m of, say, a
Wilhelmy plate, (N/m) is the interfacial tension. For a stable emulsion we want this value to be high. To
create fresh interface (e.g. cleaning oil from a surface, Surfactancy_Roll-up and Eötvös Number, or
sweeping oil from rocks Surfactancy_Critical Capillary Number) we want a low value.

It is directly related to the curvature of the interface though we might prefer to use radius, R, as being 1/
curvature.

Er
IFT =
4πR2

When we invert an emulsion, going from o/w to w/o, we invert the curvature so it becomes 0. This means
that the radius passes through infinity at that point and the IFT is 0. This point is defined via
Surfactancy_HLD theory. If we apply the more sophisticated HLD-NAC theory we find that R doesn’t go
fully to infinity, it is merely a large value. We see all this in the app:
 Surfactancy Interfacial Tension and Rigidity

[Link]

In the screenshot, the system reaches HLD = 0 with a surfactant of Cc = 0. At that point the IFT is 0.006
mN/m. A typical o/w emulsion created at HLD < -3 will have an IFT > 1 so there is a reduction of a factor
of 150 by tuning the formulation to HLD = 0.

Much less effort is required to create a fine emulsion when the IFT is super low, a fact used (but often ill-
understood) in PIT (Phase Inversion Temperature) emulsification, in catastrophic phase inversion
(Surfactancy_Emulsion Inversion a sudden flip in curvature unrelated to HLD, but still a high radius
interface), and by those who use HLD to create emulsions via other rational variations of S, T, EACN and
Cc.

Measuring IFT
If you can measure IFT with anything other than a spinning drop tensiometer (SDT), the values you are
measuring are not too interesting. Because the real interest is with super-low IFTs you need a device that
measures these low values.

[Link]

We have a narrow tube containing our aqueous phase and we inject a small sample of oil, which will form a
spherical drop. You spin the tube and the centrifugal forces force the drop into a cylindrical shape:

FST 479
 Via a microscope you measure the radius R and from it you calculate
the IFT.

IFT γ is calculated from R, the density difference Δρ and the angular

velocity ω via:

Δρω2R3
γ= 4

It sounds easy, but those who have worked with classical SDTs find them to be very hard to use.
Fortunately, the newer generation of SDT devices are easier so such measurements can become more
routine.

480 FST
 Surfactancy Phase Diagrams

Surfactancy Phase Diagrams

Links
Hair Conditioners, Laundry Liquids, Pharma Formulation

Most of us avoid surfactant phase diagrams.

We find their ternary shapes confusing and examples like this look like
too much hard work (which is probably true).

But we can do a lot with relatively simple diagrams and with the help
of a few apps they turn out to be easy to create and easy to analyse.

What are we trying to achieve?

We seldom have just a single surfactant and water – generally we have at least 3 components, such as
surfactant, water and oil. To navigate around that space a ternary diagram is compact, informative and, to
most of us, confusing.

So first of all, get used to reading a diagram. The Phase Diagram Explorer lets you view a variety of
different formats and lets you set your own level of difficulty in exploring it. Here we have chosen to have
the most helpful output, telling us where our mouse pointer is … and why those are the values.

[Link]
 I was having lunch at a P&G facility. We were discussing phase diagrams and I commented that I
found them so difficult that I was writing an app so I could understand them myself. The stranger
next to me was delighted. “I teach phase diagrams here, and everyone is confused by them. Can I
help you to write the app”. The stranger was Seth Lindberg who is expert in both in phase
diagrams and graphics. We agreed on a joint project. The phase diagram apps are correct because
of Seth’s critique of iterative versions and are informative because of Seth’s graphics skills.

You can quickly graduate from this training app to one that shows the tie lines and shaded zones from
microemulsion systems:

[Link]

You not only get the Water, Oil, Surfactant readout (42:52:6) but you also see the 3 phases that appear in
this region. Things that are mysterious to most of us, such as “tie lines” and the triangular areas without
lines, become straightforward once you explore them with your mouse.

If 3 dimensions aren’t enough for you, you can even view

data in 4D. This is a fancy “prism plot” and is about the
limit of what can be achieved on a 2D screen:
[Link]
[Link] . It is a powerful example of how theory
and visualization can come together. The theoretical
microemulsion ternary diagrams can be seen over a range
of HLD values so you can grasp what happens if, for
example, the temperature changes for an ethoxylate
system.

482 FST
 Surfactancy Phase Diagrams

Additionally, what about the scary diagram in the

introduction to this chapter which was extracted
from [Link]
surfactants/[Link]? Just move your mouse over it
and you can see what the phase is, what it looks like
in a polarizing microscope and how the surfactant
molecules are packed. Although it’s great if you
have a polarizing microscope to explore complex
phase diagrams, they aren’t necessary for most of
us who only need to explore a limited part of
surfactant space.

Your own phase diagram

So how do you create your own phase diagram?
If you happen to have a polarizing microscope,
one neat trick is to place a sample of your
surfactant onto a microscope slide, place a drop
of water on one part and place a cover slip as
shown. Looking down themicroscope you can
quickly see if the surfactant likes to get into interesting phases as you go from low to high concentrations.

For more complex setups, you need to look at a few well-chosen tubes and work out what phases they
contain. For this you don’t need a polarizing microscope. The techniques described in the QPD (Quick
Phase Diagrams) app, developed by Seth Lindberg at P&G have proved very popular because they are both
simple and powerful. Here, using Seth’s graphic, is how to know what phase is in your tube:

FST 483
[Link]

A few other tips and tricks (e.g. a light box with some sheets of polarizer) can be found on the Quick Phase
Diagram web page.

484 FST
 Surfactancy Roll-up and Eötvös Number

Surfactancy Roll-up and Eötvös Number

Links
Dishwashing Liquids, Hair Shampoos, Laundry Liquids

Oil is most efficiently removed by surfactant from a

surface via “roll up”, where the blob of oil just
detaches or, snap-off, where a big chunk comes off.
This can only happen when the Eötvös number, a ratio
of gravitational and interfacial forces is above a critical
number (the yellow line) which, in turn, depends on the
contact angle of the oil with the surface.

The best way to get a high Eötvös number? Have a low

interfacial tension. The best way to get a low interfacial
tension? Formulate to HLD = 0.

Why do we want roll-up?

We know that surfactants can wrap around small blobs of oil, creating an emulsion. So if you can get the oil
from your clothes or dishes into the water, the surfactant can wrap around it and keep it from the surface.
But it takes a lot to overcome the problem of Cleaning_Boundary removal. If you can get a large blob of oil
to simply rise into the water, that would be ideal.

The force lifting the oil depends on its radius, R, gravity g and the density difference Δρ between the oil
and water. The retaining force is the interfacial tension, γ. The balance between them is the Eötvös number
(named after a famous Hungarian scientist) E given by:

ΔρgR2
E= 2γ

When E > Ecrit, the yellow line in the diagram, you get roll-up or, at least, necking & snap-off which are
better than nothing.

What is Ecrit?

It is rather easy to clean oil from cotton. Cotton is so hydrophilic that the oil is hardly attracted, so the
contact angle is high, and a small E will remove it. The same oil on a polyester fabric is much harder to
remove – the fabric and oil are more compatible, the contact angle is low and a high E is required. For each
contact angle there is a critical value of E, Ecrit, above which the oil will be removed. The calculation of
Ecrit is described in the app.
[Link]

In this example, the contact angle is 67°, Ecrit is 5 and the calculated E value from the IFT of 0.27 mN/m is
5.6, so the oil is “unstable”, i.e. will be removed.

Getting a low γ
The only way to get a high E is from a low γ. We know from Surfactancy_HLD theory that when salinity,
temperature, surfactant and oil are matched the Hydrophilic Lipophilic Difference is 0. And because we can
go directly from HLD to Surfactancy_Interfacial Tension and Rigidity we can find that γ becomes super
low (orders of magnitude) at this point.

So for good cleaning it’s easy. Find a surfactant which, at the required washing temperature, and for the
intended oil (typically a triglyceride fat with an EACN ~ 12) gives HLD ~ 0.

It’s never so simple

Of course this relatively simple theory cannot capture the full complexity of cleaning oil from surfaces. At
the very least, the mechanical motion with a dish washing brush or inside a washing machine will give a
boost similar to a larger Ecrit. But what is not disputed is that tuning your surfactant formulation to be
closer to the ideal of HLD ~ 0 will greatly aid cleaning.

As discussed in Dishwashing Liquids, the formulations have been optimised for foaming rather than
cleaning, so they are surprisingly ineffective. Formulations with HLD closer to zero do exist so you have a
choice – nice foam or efficient cleaning.

486 FST
 Thickeners Associative Thickeners

Thickeners Associative Thickeners

Links
Sun Screens, Emulsion Paint

If we need real, permanent thickening of a system, then choose Thickeners_Polymeric Thickeners.

When you want something to be able to go quickly from thick to thin and the reverse, then you
need associative thickeners or, in special cases, Thickeners_Wormlike Micelles. Although ASE
thickener aren’t associative, their HASE variants are, so they are included here because their
behaviour is so similar.

A fragile network
We start with the classic HEURs, Hydrophobically Modified Ethoxylated Urethanes, because they are
relatively simple molecules with a rich, complex behaviour for good (the desired thickening behaviour) and
bad (side-effects discussed later).

The orange parts are hydrophobic alkyl chains, say C16. They are attached by urethane links
(green dots) to the blue ethoxylate chain. This hydrophobic-hydrophilic combination can form
micelles, orange blobs in the next image, but with the ethoxylate loops sticking out. These
molecules in the 10-35K MW range are called “telechelic” with “chelic” meaning “claw”.

Some of the hydrophobic ends can go from micelle to micelle, allowing a network to build up.
Here it’s linear but it could equally be branched. At low shears, and short timescales these long
pseudo-polymers can tangle and create high viscosities and a desirable yield stress Flow_Yield
Stress so that a paint, coating or cosmetics is suitably “thick” for application, yet easy to apply
under shear from an applicator or finger.

If it’s a paint or cosmetic emulsion, why not simply thicken by increasing the volume fraction of
emulsion particles? Because as shown in Flow_Particle Viscosity, viscosity rises alarmingly
around the φ > 0.55 region so small changes can have a catastrophic effect on the performance of
the formulation.

Thickener rheology
Most of us don’t know what rheological behaviour we would like, but we would like to be able to have
tests that can reveal internal details of good, OK and bad formulations. Standard low strain rheology does
an especially good job for associative thickeners because we know in advance what an ideal system looks
like – as shown in the app:
[Link]

The left-hand graph shows what we might find if we measured low-shear viscosity at different
concentrations of the thickener. There’s a linear, uninteresting, increase up to some specific point, then the
viscosity takes off. If you do such tests for a range of good/OK/bad thickeners in your specific system, you
get a good idea of how they compare in this practical aspect.

The right-hand graph shows a typical “frequency sweep” where G' & G"
(Flow_G' and G''). The G' is the elastic portion that gives you the gel-like
behaviour you need. The higher G', the “stronger” the system. The log plot in
the screenshot is deceptive (but is what is commonly shown in the literature)
– this little image is the same data in linear format, showing more clearly the
massive increase in viscosity around the 1/s timescale.

The timescale, ω, at which the response changes from fluid to gel-like is controlled by a relaxation time τR.
You will want to control τR as much as you want to control the maximum G'

The reason we do these plots is because τR and G' are super important for formulation properties. For
example, in paints you are interested in the performance on the brush or roller and in the subsequent
levelling and sag: Coating_Levelling Theory. For cosmetics you will have other criteria such as the more
subjective “feel”. Although you might have a sense that one formulation is better than another, it’s better to
have numbers that can tell you whether the better formulation has a different τR, a different G' max or some
sweet-spot combination.

Influencing the thickening

You can change 3 things in your choices of HEURs

1. Hydrophilic chain length – biggest impact on G' and τR. Going from C14 to C16 then C16 to C18
can give successive increases of 10x in each property.
2. Ethoxylate chain length – longer chains give somewhat greater thickening, but because the
increase is apparent at the same molar concentration (same number of molecules), the effect is
smaller when we measure it in terms of wt%.

488 FST
 Thickeners Associative Thickeners

3. Nature of the urethane bridge - the bridges affect hydrophobicity and rigidity so there are no
obvious rules of how to choose

You can also choose, say, HASEs, Hydrophobic Alkali-Swellable Emulsion Polymers. These tend to be less
well-defined than the HEURs so their rheological response is not so “sharp”. In the app, reduce the G' α
and G" β values from their HEUR value of 1 to see how the rheological response broadens.

The other influence, for good or bad, on the thickening is surfactants. In a clean lab formulation you can
add increasing amounts of your favourite surfactants to see the maximum G' or low-shear viscosity increase
then, usually, decrease. For typical surfactants the effects are modest.

At first the tails of the surfactants displace one end of a HEUR, freeing it to link to another micelle. So the
viscosity goes up. Eventually this becomes self-defeating – so many HEUR molecules are displaced that
there’s no chance of a good network.

If you add “pre-wormlike” surfactants which, thanks to a Critical Packing Parameter ~ 1 (Surfactancy_CPP
and phases) are keen to form strong micellar phases, then you can get larger increases. Indeed, if you add
real wormlike micelles (Thickeners_Wormlike Micelles, with an app remarkably similar to the one in this
chapter) then you are potentially doubling up on the thickening effect – with the warning that these
complex systems can be unpredictable.

Real-world emulsion thickening

It’s a necessary first step to get used to HEURs, HASEs etc. in simple, clean formulations. Their behaviour
is complex enough, and you have so many parameters to choose from.

As soon as they are applied to, say, real-world emulsion polymer formulations, complications appear.

The first is one you’ve prepared for. Those formulations may have their own surfactants which might be
free enough to start interacting. Depending on their levels and their interactions with the HEUR (and you
may have no idea of either) they might give you a thickening boost or (hopefully unlikely) will take you
over the boost limit.

If you make your own emulsion polymers you can handle this rationally. If you buy them in, and the
supplier is reluctant to reveal any useful information, you have to rely on intelligent guesswork.

Despite the uncertainties, at least this effect is only changing the basic rheology and with your background
knowledge of good/OK/bad you can tweak your HEUR type or quantity to hit your sweet spot.

The other effect is very different – and is another good reason for using rheology. As described in
Dispersions_Rheology (High shear) (which covers high volume fractions), the viscosity of an emulsion can
increase dramatically if some sort of fractal particle association takes place. Those familiar with the terms
“depletion flocculation” or “bridging flocculation” can imagine how the pseudo-polymeric associations of
thickeners can act to attract the particles causing at best an increase in viscosity and at worst a complete
collapse of the formulation.

The theories behind these effects are poorly described by any tool other than Dispersions_Scheutjens-Fleer
theory. Read that chapter if you are keen to understand the effects and have some clue as to how to fix the
problems.

FST 489
Even if you don’t want to know about the theory, why should you worry about an increase in viscosity –
after all we’re using thickeners, so some extra thickness sounds like a free lunch.

Put two formulations into a rotational viscometer and do a sweep of shear rate. The first formulation has no
such particle/thickener interactions and you see the viscosity reduce sharply at low shear rates. The second
formulation will start at a somewhat higher viscosity (not a big problem) but the fall-off is very slow. You
now have a very different type of thickener. If you wanted this type you could probably have achieved it by
increasing the volume fraction of emulsion.

Smart mapping
There is no realistic chance of you solving your thickener optimization challenges using pure theory – you
have too many real-world uncertainties.

The alternative, trial & error plus “experience” has not been proven to be a great alternative.

What the theory tells us is that regularly carrying out the same experiments on a range of good/OK/bad
formulations builds up a picture of:

• Viscosity versus concentration

• G' & G" over a range of frequencies
• Shear rate behaviour

Because each is related to a different aspect of formulation space you will be able to map how formulation
ingredients interact across this complex space. With some sensible data analytics, you will be able to
extract predictive tools that work in your formulation space.

It’s a minimax – minimum work for maximum payback.

490 FST
 Thickeners Polymeric Thickeners

Thickeners Polymeric Thickeners

Links
Sun Screens, Emulsion Paint

Thickening with a polymer is either a nuisance (when you want a high concentration of polymer in
a formulation, but it makes it too viscous to handle) or a necessity (when you need a controlled,
higher viscosity for some purpose). The theory is the same and covered in Flow_Polymer
Viscosity. In this short chapter we focus on the general need to achieve a desired viscosity with the
least added polymer and the fewest downsides.

Lots of tangles
Viscosity comes from entanglement of polymer chains. You may or may not be surprised that entanglement
that causes viscosity is the same as causes adhesion, Adhesion_Entanglement, but that means we can
borrow a diagram that shows that entanglement is 3 crossings:

The dotted line is a real or virtual interface. In the left-hand image, the polymer chain of interest crosses it 3
times. This means that it is entangled. If you pull up on the big arrow at the top left, the lower loop gets
tangled with the other loops.

In the right-hand image, the chain has been cut so it now only crosses 2 times. When you pull up, the loop
is merely intermingled and can be pulled out. In terms of viscosity, this is a much smaller effect, closer to
the φ* (volume fraction) and C* (concentration) situations discussed shortly.

Obviously if you have a larger MW you will get more tangles. But why do some polymers appear regularly
as thickeners while others don’t? Because good thickeners have a low Mc, Critical Entanglement MW. If
you play with:
[Link]

you will find that in this very simplified view of blobs of polymers, the θmin “bend angle” has a big effect
on how much the polymer sticks out into the solvent, as does the monomer size b. Some polymers stick out
a lot for their MW and so have a bigger chance to encounter other chains and, therefore, entangle. Although
φ* is merely overlapping, rather than entangling, there’s a good chance that a small φ* will lead to a small
MC

If you go to

[Link]

you see that the viscosity starts to take off somewhat above a C* and massively at Ce which is when
entanglement kicks in.

492 FST
 Thickeners Polymeric Thickeners

In that app you need to enter Mc so how do you know what it is? Well, your suppliers should give you these
values routinely, but they don’t. If they did you would know that a 100,000 MW polymer with Mc = 10,000
will have 10 tangles while one with Mc = 25,000 will only have 4. So the same MW of the first gives much
higher viscosities, more tangles, than the other.

On that page you can find a list of all the Mc values I happen to know. Here it is in plain text:

Polymer Mc values

Polyethylene: 3300; Poly(vinyl fluoride): 3300; Poly(ethylene terephthalate): 3400; Poly(ether

ether ketone): 3400; Poly(ether ketone): 3500; Nylon 6,12: 4000; Polyoxyethylene: 4200;
Poly(1,4-butylene succinate): 4400; Poly(epsilon-caprolactam): 4800; Poly(epsilon-caprolactone):
4800; Poly(vinylidene fluoride): 5200; Polycarbonate (Bisphenol-A): 5200; Poly(propylene oxide):
5600; Polyoxymethylene: 5800; Poly(3-hydroxybutyrate): 5800; Polypropylene: 6400;
Poly(glycolic acid): 6800; Poly(1,4-butadiene): 6800; Polylactic acid: 8600; Poly(vinyl alcohol):
10500; Polytetrafluoroethylene: 10800; Poly(vinyl chloride): 11500; Poly(acrylic acid): 13800;
Poly(methacrylic acid): 15100; Polyisobutylene: 15200; Poly(vinyl acetate): 15400;
Polyacrylonitrile: 15800; PDMS: 16600; Poly(methyl methacrylate): 20400; Poly(ethyl
methacrylate): 27000; Poly(vinyl butyrate): 27600; Polyacrylamide: 28000; Poly(vinylidene
chloride): 29200; Polystyrene: 31000; Poly(ethyl 2-cyanoacrylate): 59000;

It is, incidentally, no coincidence that all the great engineering polymers, PE, PET, PEK, Nylon, PC have
relatively low Mc values – they are strongly internally tangled to be tough. PMMA and PS, both known to
be brittle, have a high Mc. Superglue, Poly(ethyl 2-cyanoacrylate), is an especially weak polymer with a
massive Mc!

Solubility
The other aspect of the previous app is the Flory-Huggins χ parameter, Dissolution_Flory-Huggins. When
it’s much less than 0.5 you can readily dissolve your polymer as long as it is largely amorphous –
crystalline polymers, like PE, are insoluble because the solvent can’t disrupt the crystalline structure.

What is critical is that at low χ values, as the first app shows, the polymer opens up (at χ = 0.5 it’s a tight,
Gaussian ball), giving you greater practical entanglement and viscosity. The entanglement from a low Mc is
only a potential – the polymer has to be opened up before it can tangle.

How do you match a polymer to a solvent to get a low χ value? Via Dissolution_Hansen Solubility
Parameters, where you match the HSP values of your solvent (blend) and polymer. If it’s an aqueous
system, you simply choose your favourite high MW, low Mc, water-soluble polymer.

Flow behaviour
So far we’ve been concerned with low shear viscosity. As we (usually) need the formulation to flow at
some point then we need a good flow curve – plot of viscosity versus shear rate and, as it’s a good idea to
think in this way, how the viscosity depends on the shear stress:

FST 493
[Link]

Here the viscosity plummets at 10/s which may or may not be what you desire. How do we predict this
behaviour from the polymer-solvent properties? At the time of writing, I don’t know. So you need good
access to a good rotational rheometer to get these data.

In practice
This means what you already know: there aren’t too many useful thickeners which have the dual ability of
providing large viscosity at a small concentration. Either you can’t get a high enough MW or the Mc is too
large, or the polymer you like, even if sufficiently soluble, isn’t happy enough in the solvent to expand and
tangle.

If your problem is the polymer/solvent compatibility, then via HSP you might be able to tweak your solvent
blend to give you the lower χ value for greater entanglement.

And even if you get the right low-shear viscosity, there’s no guarantee that you’ll get the lower viscosity
you require at a higher shear rate. There seems to be no substitute for a good rheometer and, perhaps, some
building up of flow curves for different polymer/solvent/χ-value/Mc combinations – another potential
example of “smart mapping” where you gather the same meaningful data over a range of relevant samples
to help provide some useful rules in an area where theory, as in this case, lets you down in terms of pure
prediction.

494 FST
 Thickeners Wormlike Micelles

Thickeners Wormlike Micelles

Links
Hair Conditioners, Hair Shampoos, Soaps and Washing

When you need a convenient thickener for a personal care product, adding some NaCl to a simple
surfactant mix can do the job via the creation of wormlike micelles.

At the right combination of surfactant, concentration and

salinity, some surfactants that were in typical spherical
micellar solutions of low viscosity can transition to long
“wormlike” micelles, shown in this image kindly provided by
Seth Lindberg.

Although they are relatively delicate, if one micelle crosses

another there is an entanglement force, so the system can be
highly viscous. In soaps and shampoos this allows the
formulator to provide a “luxury” viscous feel using only the
surfactant and some sodium chloride.

Unlike Thickeners_Polymeric Thickeners, which can be too

sticky, the relative delicacy of the wormlike micelles means
that they can break up easily and, of course, viscosity decreases rapidly as water is added, reducing both the
surfactant and salt concentrations. The reduction in salinity and surfactant concentrations changes the form
of the micelles (usually back to spherical micelles) so the entanglement viscosity is lost.

Like other forms of entanglement (see Adhesion_Entanglement) the effect on properties such as viscosity
depends on concentration to a high power (~3.5) above a critical entanglement concentration C*. We can
capture such behaviour in an app:
[Link]

C* is defined as a user input and the dramatic change in slope of viscosity is clear from this log plot.

Of equal interest is the rheology of these systems. Because the micelles are only loose clusters, if you probe
them over long timescales, they are easily broken. Using the standard rheological measures G' (elastic) and
G" (viscous) (Flow_G' and G''), we see from the right-hand graph that at low speeds (ω is the frequency of
oscillation), viscosity is low, building to a high value at a specific speed which corresponds to the
relaxation rate of the system. At higher speeds, the system becomes effectively an elastic solid, with a high
G' and low G".

Flows like honey, wobbles like jelly

As explained by Neil Cunningham of the Centre for Industrial Rheology, the implications of the
right-hand graph are that at low shear rates, gently tipping a bottle, the wormlike formulation
should flow like honey (or, as in his excellent TA Webinar, corn syrup), while at high speeds it is
elastic and wobbles like jelly.

The rheology curves are classic Maxwell equations. Given a reference G value, G0, and a relaxation
timescale τ, the values are given by:

ω2τ2
G ' = G0
1 + ω2τ2

ωτ
G ' ' = G0
1 + ω2τ2

The low shear viscosity, η, shown in the left-hand graph comes directly from the Maxwell equation:

η = G0τ

496 FST
 Thickeners Wormlike Micelles

The viscous and rheological effects are very temperature dependent via τ’s dependence on an activation Ea
in an Arrhenius equation:

Ea
− RT
τ = Ae
Although the equations are useful, where do you get the values? The answer is that you use simple
viscosity tests to get C* and standard rheometer measures to get τ. Then you have to reverse fit to match the
graphs in the apps. Although this is unsatisfactory, it is still a relatively fast way to get core data on your
system and how it changes with surfactants and salinity.

Upsetting the worms

So far we’ve thought of nice wormlike micelles in pure water with added salt. Our real formulation
challenges involve plenty of other ingredients intended for other purposes. As has been noted by many, a
standard Hair Shampoos can be nicely thickened, yet a simple marketing request to change the fragrance
totally destroys the effect. One or more of the fragrance molecules gets into the hydrophobic micellar core
and flips the subtle balance that had previously given the desired viscosity. Whether you can recover the
viscosity by changing the salt concentration is unknowable in advance.

As there is no usable theory to predict the effects of additives to a wormlike formulation we resort to the
standard approach in the FST: have a good rheometer with a well-defined method that lets you quickly scan
formulations with systematic variations of salinity and ingredients. As you accumulate enough data, who
knows, your AI might be able to provide you with the desired predictive power. This can’t happen if you
gather data randomly, with different methodologies, on unreliable rheometers. Not even an AI can
disentangle such data.

FST 497
- Extras -
As the book progressed I needed to add one “extra” then a second. If readers find the need for more extras,
I’ll be happy to add them.
 Analytical Techniques

Analytical Techniques
Here we look at some ideas that are slightly away from the standard uses of standard analytical
techniques, providing some deep formulation insights. The inspiration for this chapter came from a
lecture by Dr Tobias Halthur of CR Competence AB. He kindly provided some key images, below,
so you get his expert view of the topics.

NMR Relaxation for particle/solvent interactions

We add our particles to a solvent in whatever manner we choose, and put the
results into a normal NMR tube. The tube is then inserted into a small, relatively
simple benchtop NMR machine (this image is from Mageleka) that does not
measure an NMR spectrum. Instead, it issues a chain of pulses and follows the
relaxation of the pulses. A run with the solvent on its own establishes the standard,
relatively slow relaxation time. If the solvent has little interaction with the
particles, the measured relaxation time will be little changed from the reference.
But strong solvent-particle interactions provide a large change in relaxation time.

Via a systematic comparison of the relaxation behaviour of the particles with different solvents you get a
good idea of how the particle interacts with its surrounds.

If the particle is covered by some sort of dispersant, the relaxation effects are more subtle. If the dispersant
is nicely extended into the solvent, then the solvent will see a relatively fluid boundary and the relaxation
time will change much less. Like all techniques, you have to have good hypotheses and use good sense to
interpret the results.

You don’t have to use the benchtop machine. A classic NMR machine can also work in relaxation mode.
It’s your choice of tying up your fancy NMR machine or having a benchtop machine for regular particle-
solvent analyses.

FTIR with smart subtraction

A typical FTIR spectrum is relatively complex and if you need
to do a before/after comparison of some process or reaction, or
if you need to compare your formulation with, say, a similar
formulation from a competitor, it can be difficult to extract
meaning from a side-by-side comparison. Even in this
simplified spectrum it’s hard to know what is going on between
sample A and B. Real, more complex spectra are even harder to
compare.
 The obvious alternative is to subtract one spectrum from another
and look for differences. However, small differences in
concentration or measurement setup can lead to too many
differences, positive and negative.

Smart subtraction requires you to be confident that at least one

peak in the spectrum should be an unchanging reference between
the samples. So you adjust the subtraction so that this peak (or
maybe 2 or 3 peaks) are effectively removed. In this example, simple B-A subtraction gives differences
everywhere. If we know that the -OH peaks around 3400 cm-1 should be unchanged, we tune the
subtraction so this area is flat. Now the resulting negative and positive peaks should represent real
differences between the samples. With knowledge of what the peaks showing differences might be, it’s
possible to formulate a hypothesis of what’s going on.

FTIR of aqueous complexes

Systems such as Surfactancy_Coacervation, Thickeners_Associative Thickeners and
Thickeners_Wormlike Micelles involve complex interactions between water, surfactants, polymers, salts
and additives. For some users, the bulk properties are important, for others (such as the complexes left on
the surface of Surface Cleaning) it is the surface effects that are important.

In both cases it might seem difficult to get meaningful FTIR spectra, but with the right setup, you can get a
lot of useful information from these complex systems. The review by David Scheuing, Fourier transform
infrared spectroscopy in surfactant science: A personal view, J Surfact Deterg. 2022;1–25, is intended for
the practical formulator in this complex space. Although the review is for “surfactant science” the
principles apply to similarly complex systems.

Information from bulk solutions

For bulk solutions the trick is to have a precision spaced (~25µm), temperature-controlled aqueous cell,
with windows of water-insoluble CaF2 or BaF2. Temperature control is vital because the IR absorbance of
water changes significantly even over 1°C. You take a reference spectrum of whatever “water” is relevant
(typically you will have specified salts and pH), then use that, via smart subtraction, to remove as much as
possible of the water spectrum in order to amplify the relatively weak and complex signals from your
sample. The main peak ~3400 can be relatively easily nulled out. The peak that might cause more trouble
with functionalities in your system is the one at 1640 which is itself influenced by interactions with solutes.
Switching to D2O may be necessary if you really need data in this region.

Although you can imagine that, say, carboxylate absorption for a soap should shift at the CMC (which it
does), it is not obvious why a spectrum might provide any insights into the surfactant tail region; it is just
some boring CH2 groups with a terminal CH3. However, as the review points out, the various wagging,
bending, stretching modes change in relative intensity depending on the degree of ordering in the tail. The
ordering of tails in a spherical micelle is different from tails in wormlike micelles, and there are diffierences
between tails of systems with different curvatures at oil/water interfaces. The review has many specific
examples of spectra from real-world systems showing the sorts of effects of interest: wormlike micelles,
phase changes, ethoxylate behaviour at the phase inversion temperature, microemulsions, mixed surfactant
systems.

Interfacial FTIR

500 FST
 Analytical Techniques

Those familiar with ATR (Attenuated Total

Reflectance) know that a crystal can be placed
next to a surface and a spectrum easily gathered
in the region where the light partially penetrates across the interface. But the spectrum is the average over
the 1-2µm penetration depth, so is useless for picking up nm-scale effects such as coacervate deposition.
However, using a multi-reflection crystal, you can get 10-25 reflections so the effective pathlength is
10-25x longer. This means that a signal from a nm-scale layer can be observed after subtraction of the bulk
signal, as in the previous section.

This is fine if you want to see interactions of surfactants and polymers with the surface of your, say, Ge
crystal. But generally we need a specific surface. There’s no problem in principle of adding a layer thinner
than the penetration depth. The limitation is your confidence in applying a layer to your expensive crystal
and being able to remove it later. As the review notes, it’s not so hard to add alumina (sputtering) or titania
(from solution) to the surface.

If you take the trouble to work out how to do this, you can be rewarded with insights that tend to over-turn
the simplistic diagrams we like to make of polymers and surfactants at surfaces. For example, a monolayer
of SDS resolutely refuses to provide the ordering we assume is the norm – the various CH2 modes don’t
change into those of packed layers. And when you add polymers and electrolytes, it is clear that what’s
obtained on the surface is a loose fluffy agglomerate, with the final form depending not just on the
stoichiometry, but also on the order of addition. These structures have a large element of kinetic control,
despite what the thermodynamics would suggest.

The ellipsometry and QCM experiments shown below very much confirm these sorts of conclusions.

A fascinating set of studies included in the review starts with a “thick” (~1 µm) layer of an oil and follows
its removal via detergents in real time. Using oils of different melting points and ethoxylates with different
HLD behaviour it was possible to gain deep insights into the processes involved.

Ellipsometry
This image, kindly provided by Dr Halthur sums up ellipsometry:

FST 501
If a beam of circularly polarized light is reflected from a smooth surface, the amplitude and the phase of the
reflected polarized depend on the complex refractive index of whatever is on the surface, and on its
thickness. “Complex refractive index” means, in normal language, its refractive index and its absorption
coefficient for light at the given wavelength.

The classic technique is done with dry coatings in air and is sensitive down to low nm of thickness. With
care and experience, the experiments can be done under liquid, which makes it more interesting to
formulators. What you get from the measurements are:

• Thickness
• Refractive index
• Adsorbed mass – dry weight (even though that mass might be swollen)

A specific example is shown below, comparing/contrasting with the QCM-D.

QCM-D (Quartz Crystal Microbalance with Dispersion)

Another image from Dr Halthur helps us to grasp QCM-D:

502 FST
 Analytical Techniques

Put a quartz crystal into your solution and get it to oscillate at its resonant frequency. Because of the
viscosity of the liquid, this frequency will be different from that in air. If you have a QCM-D machine (D
for Dispersion) you can see how the amplitude of the oscillations decay when the stimulating signal is
switched off.

Now gently flow a solution of interest over the crystal. If something in the solution wants to accumulate on
the surface, the mass of the crystal increases, so the resonant frequency decreases. The effect is exquisitely
sensitive so you can pick up nm deposits. . What you get from the measurements are:

• Total adsorbed mass (including the liquid)

• Visco-elastic properties

A further image from Dr Halthur lets us see a specific example (the reference is included in the image) of
how the two techniques reveal complementary information:

FST 503
Some mucin, a polymer with thiol groups at each end is deposited from solution and rinsed. It may or may
not be attached by one of the thiol groups and we can imagine many ways that it sits on the surface – flat
and compact or sticking up as a loose bundle. The ellipsometry measures its dry mass, which is relatively
low – no surprise because it’s a small deposit. The thickness is reasonably large, and the refractive index is
very close to that of water. The QCM-D seems to tell a different story. The mass increase is large. But then
we recall that the device records the total mass, including water absorbed along with the mucin. The low
refractive index seen in the ellipsometry now makes sense because it is seeing a highly swollen mucin
layer.

When lactoperoxidse is added, the mass increases, substantially, according to the ellipsometry, but the
QCM-D shows only a small mass increase. The ellipsometric thickness shows a sharp decrease and the
refractive index increases – all indicating that the complex of the two polymers is highly compact. Finally
the viscosity and shear modulus from the QCM-D both show a significant increase, which is consistent
with the data from the ellipsometer indicating a more substantial complex layer.

With beautiful results and excellent diagrams, it all seems rather straightforward. Like all sophisticated
techniques, things aren’t so hard if you have high-quality equipment and years of training. The point is not
for you to rush out and become instant experts but to know that these capabilities exist and, as with all
modern technology they tend to get more automated and more powerful so, maybe, more routine for when
you need to better understand what’s happening at an interface.

Beware Superduperometers
The best analytical technique is the one that gives you key information quickly. You ask the right question
and the equipment gives you helpful answers.

However, it is frequently assumed that the best technique is some superduperometer, a machine so new,
powerful, complex, expensive that it must be good. If you put your sample into the machine it will generate

504 FST
 Analytical Techniques

lots of amazing numbers you couldn’t have obtained any other way. But if you are not asking it the right
question and it’s not giving you helpful answers, it’s a waste of time.

Yes, there are times when a critical issue can only be resolved using a fancy machine. Go ahead, use it. The
techniques highlighted in this chapter are, in the hands of those I know who use them, great examples of
how to learn stuff you otherwise could not know. But academia and expensive corporate central labs are
full of fancy machines that provide very little problem-solving insight for a lot of money.

And this is the 21st century. It is amazing what you can do with some ingenuity, an Arduino, some sensors,
a 3D printer and a good hypothesis. Add some LED lighting, decent optics and a laptop or phone to record
some images or video and you can get a feel for what is going on at some key interface. I measured the
coefficient of friction of various cheeses on various packaging films on my kitchen table. An order to
Amazon and a wait of a few days and I could measure some important thermal conductivities that a
megacorp admitted would have taken months to organise. When a really nice pump system, required for a
key experiment, got damaged in transit, we got the data we needed using a syringe and a steady hand. You
can have controlled RH chambers with a fish tank, some water and some well-chosen crystals. Someone I
know replaced a complicated robotic paint-brush cleaning system by using disposable 3D printed brushes
and saved $1000’s by using a standard oven rather than a fancy automated oven … by using the robot to
press the On button on the cheap oven.

Two workhorse rheometers will probably be better for you than one fiddly, expensive one locked in an
expert’s lab. A routine FTIR measurement when I was visiting a megacorp was a nightmare because instead
of a couple of lab machines they had a super-fancy one, run by an expert in a “Do not enter” lab which you
had to book days in advance.

Another problem with superduperometers is that you often have to submit your sample and wait till the data
comes back. Despite your careful explanation, what they measured turns out to be what they thought you’d
asked, not what you really asked. A good rule is “Never get anything measured on a fancy machine without
you being present”. What is obvious to you isn’t obvious to the operators, so they will miss pointing their
ion beam gun at the one thing that could solve your problem and, instead, gather data to prove something
you already know. This is a slow and expensive way to learn nothing.

Find the simplest, quickest way to answer a clear question to help solve a real problem. Sit at the set-up and
start asking the questions … and discover that there’s a glitch in the idea. Modify, try again, find another
glitch … and quickly you get the answers you need. Lean, agile, fail fast aren’t the answer to everything,
but for formulations they are much more likely to be helpful than any superduperometer.

FST 505
Open Analyzer
Many readers will be formulating for greener products. It turns out that good intentions are not the
same as good outcomes, so it’s important to check sooner, rather than later, that your precious
resources aren’t going to be spent on something that is not only not better for the planet but
actually worse than the current product or process.

A full LCA (Life Cycle Analysis) is the gold standard for this. But they are cumbersome and
bureaucratic. Sometimes a less formal “what if” tool can give you the answers you need. Open
Analyzer is a handy tool for early exploration of whichever part of the system is important to you.

Interactive What Ifs

[Link]

Suppose you wanted to know how to reduce the CO2 footprint of a cup of coffee. You might want to reduce
cost of transportation, roasting, packaging, fertilizers … things that are someone else’s responsibility. But if
you assemble all the relevant parameters and play with the effects of, say, halving them (by clicking on any
of the pink ovals) you find that most make little difference … compared to the efficiency with which you
heat the water to make your cup of coffee.

I could tell you as a fact, or I could provide a full LCA, but having a model that you can interact with and
try out “what ifs” is an efficient way to understand what’s going on.

But suppose you found this model was lacking some key parameters. No problem, you copy or download
 Open Analyzer

the model, alter it in Excel or Google Docs (the data are just simple .csv) then load your new version.
Because it’s just an app, you can have two tabs open with a different model in each to be able to compare
and contrast.

If you look at other examples you will find attempts to replace some conventionally produced chemical
with a bio-based one, and learn that after maybe 10 years of hard work you might save 1 kT CO2/year, the
equivalent of 1 second of current human emissions. Is it a good idea to put that much effort into such a
small saving? Would it have been better to put your precious resources into something that could save 1
MT?

Another example is the famous “plant bottle” for fizzy drinks, using bio-based ethylene glycol. The Open
Analyzer analysis shows that it is worse for the planet than the conventional one, a truth backed up by a full
LCA. Happily, the originators of the plant bottle decided that using recycled PET was a much better
strategy, which it is.

How to use Open Analyzer

The inputs

You need an Excel .csv file containing the Nodes. If click Copy Model on any of the datasets that interest
you then paste into Excel, or if you click the Download Model, you will see that the format is relatively
simple, though getting everything right is, I admit, tricky. Each node is identified by a unique id, then has a
set of other properties. One key problem is that in the world of Javascript, everything is Case Sensitive. So
when you specify an id you can't specify an Id or an ID. And if you have a node called MyNode you cannot
have formulae based on myNode or Mynode (unless you confused yourself by creating different nodes with
those id values!). If you have an optional file with the same name but with a .html extension then this
assumed to be the HTML you wish to include on the page to help explain what is going on. If you write this
in an ordinary text editor at the least provide an HTML headline using a few HTML tags

This is your title

and start a paragraph with the

tag.

Loading a Network

From the combobox you just select a model and the job is done. To load your own then (for obscure
Javascript security reasons) you need to load the Nodes file first then, optionally, the .HTML file.

Sensitivity Analysis

If you select one of the inputs and provide a value, you automatically get a sensitivity analysis! Each time
you change something, the values of every node are automatically put on the Clipboard using the minimum
and maximum of that slider along with your current value. Just paste into Excel and you can analyse it any
way you want. Why not add a Sensitivity Analysis function? Because it complicates the interface and doing
it automatically gives you the maximum benefit for the minimum work

Your own input files

Here are what each of the inputs mean for Nodes:

FST 507
 • id: The name of the node.
◦ If this id = xxx or XXX then everything in that line will be ignored. This lets you add
full-line comments to your .csv file
◦ If this id = hierarchy then the label field should be UD, DU, LR, RL to describe up/down,
down/up, left/right or right/left hierarchical layout. For this you need an extra column
called "level" where you give numbers starting at 1 for the different hierarchical levels.
The quality of the view is not guaranteed as hierarchies are complex to get right. Clone
the HomeCompost-H example to check how it works
• label: What appears in the box
• units: Units (optional) that precede the value on the lower line of the label
• title: Tooltip pop-up text. The column and individual tooltips are optional
• color: Choose between pink, blue, yellow, salmon, green, red, cyan, magenta and white. There are
complicated reasons why you can't just enter your own colours. My convention is pink for inputs,
blue for simple intermediates, salmon for aggregated intermediates and yellow for stand-out
conclusions. You can choose your own scheme, but this one seems to work well over a range of
models
• val: The value of a known input. This must always be the actual number. So if something is 1
million, it must be 1000000. If this node produces calculated values, put a ?
• conv: To display the number in the box, multiply by this value. So to see 1000000 as 1.00, set
conv to 0.000001 or 1e-6. Use m in the label to indicate to the user that the value is in millions
• from:, to:, step: Most inputs make sense only over a range which you specify in converted units as
from and to, with step showing the precision on the slider. So if you want to go from 100 thousand
to 10 million in steps of 100 thousand and your units are in millions you would have from:0.1,
to:10, step:0.1
• equn: This is the trickiest, but not so hard. Suppose you have two inputs called Factor1 and
Factor1 and you wanted the node to be their sum. You just put {Factor1}+{Factor2}, i.e. you type
the sum, but put the names in curly brackets so the software can work out what is what. You can
use any Javascript code you like1. With even the smallest amount of effort (and attention to
brackets) you can easily tell the meaning of
3*([Link]({Factor1})+[Link](4.2+{Factor2},2))
• comment: This comment is for your own use as a comment about that row. But if a comment is
present it appears as a new line within the Tooltip, under the "title" text

Errors in your network

It is entirely normal (i.e. it has happened to me many times) to have some errors in your .csv file. For
example if you have a node called MyNode you might include {Mynode} in a formula, or you might have
typed (MyNode} with one curved and one curly bracket. When you load a model, the code attempts to alert
you with messages that should make it easy to spot and fix the problem. You may get more messages than
you might like (an error may have knock-on effects), but it is better to have too many than too few. When
you correct the .csv, reload it and keep going till you get none of those sorts of errors. Of course, you will
probably have other sorts of errors in your logic so you will need a few further cycles of refinement.

Capturing a different version of a model

You may have a model with one set of inputs then decide you want to keep a version with inputs you have
just adjusted on screen. Click the Model → Clipboard button and the model is ready to paste into Excel and
save as a new .csv. Or click Download Model → CSV if you prefer to have the file delivered directly to
your download directory.

508 FST
 Open Analyzer

A database of input values?

Those in the green biotech world who want to use Open Analyzer to model their processes would love to
have a set of standard values for many of the common processes. I could easily create a Standard Values
database if I had the data. But I'm not a biotech expert. If anyone wishes to help create a database of values
I will be happy to do the coding and, of course, their contribution would be fully acknowledged.

FST 509