Blood Vessel Branching: Beyond the Standard Calculus Problem

Author(s): John A. Adam

Source: Mathematics Magazine, Vol. 84, No. 3 (June 2011), pp. 196-207
Published by: Taylor & Francis, Ltd. on behalf of the Mathematical Association of America
Stable URL: [Link]
Accessed: 20-04-2018 18:04 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@[Link].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
[Link]

Mathematical Association of America, Taylor & Francis, Ltd. are collaborating with
JSTOR to digitize, preserve and extend access to Mathematics Magazine

This content downloaded from [Link] on Fri, 20 Apr 2018 18:04:47 UTC
All use subject to [Link]
196 MATHEMATICS MAGAZINE

Blood Vessel Branching: Beyond the

Standard Calculus Problem
J O H N A. A D A M
Department of Mathematics & Statistics
Old Dominion University
jadam@[Link]

As one who is very thankful to have survived a heart attack in 1996, I have more than
a passing interest in the vascular system. Perhaps this is due to the following statement
by my cardiologist: “Your left anterior descending coronary artery is 99.9% blocked.”
While that might have been an exaggeration, the sight of my artery narrowing to a
thread on the angiogram was certainly a cause for concern.
The vascular system is the collection of all arteries, veins, and capillaries that permit
the flow of blood from the heart, around the body, and back to the heart again. A
combination of fluid flow and diffusion brings every cell in a healthy body within
reach of the nutrients and gases it needs. The combined length of all this plumbing is
rather large: Indeed, in 1967 the mathematical biologist Robert Rosen [5] made the
claim that the total length of the vascular system in large mammals is of the order of
50,000 miles!
The purpose of this article is to describe a significant application of mathematics
to the understanding of the vascular system. Over a century ago, in 1878, Wilhelm
Roux [6] described the relationships among the angles and radii of bifurcating blood
vessels, showing how they arise from some simple principles of optimality. Our goals
here are to rederive Roux’s results for the benefit of a wider audience, to examine their
implications, and to establish how relevant they may be to the human vascular system.
To accomplish this, we first study the equations for the flow of fluid in a pipe and es-
tablish a fundamental result, which is standard fare (or once was) in most engineering-
style calculus books. We then present a sequence of increasingly faithful models of
blood-vessel branching. In these models, we judge optimality using a succession of
so-called cost functionals.
Our treatment is based on some rather obscure material, referred to in passing by
Thompson [9], and in somewhat more detail by Rosen [5], but expanded considerably
here.
At the outset we acknowledge some simplifying assumptions.
The first concerns the nature of fluids. Newton modeled the simplest kind of fluids
by imagining them to consist of thin layers that slide past one another, developing a
resistance to the flow. An important factor of proportionality—the ratio of “drag” to
the velocity gradient perpendicular to the direction of flow—is called viscosity. A fluid
with the property that the viscosity is independent of the forces applied to it, is called
a Newtonian fluid. Naturally enough, if the viscosity does depend on the applied force,
then the fluid is called non-Newtonian.
In this paper we model blood as a Newtonian fluid. This proves to be a reasonable
approximation except when blood cells are forced through the tiniest capillaries, which
are barely as large as the blood cells themselves. In these capillaries, it is not reasonable
to think of layers of fluid sliding past each other, so our models do not apply.
Furthermore, the pressure driving the whole system is far from constant; there are
short time lags between the high pressure and the peak blood flow because of the
Math. Mag. 84 (2011) 196–207. doi:10.4169/[Link].84.3.196. c Mathematical Association of America

This content downloaded from [Link] on Fri, 20 Apr 2018 18:04:47 UTC
All use subject to [Link]
VOL. 84, NO. 3, JUNE 2011 197
inertia of the blood. In addition, the blood vessels are flexible, not rigid (though of
course, hardening of the arteries occurs with age). Cumulatively, this can modify the
blood dynamics, as do the surface waves along the vessel walls, induced by the changes
in pressure and flow generated by the heart. Such waves can be reflected at various
points in the arterial tree: at branching junctions, or where the vessel diameter changes
significantly. By contrast, we always assume time-independence and a constant shape
for the blood vessels.
Having acknowledged these realities, we proceed to build our first model.

Viscous fluid flow

We begin by describing the flow of a fluid, such as blood, through a cylindrical pipe,
such as a large blood vessel. We will derive equations for the velocity of the fluid
(which varies with location in the pipe) and for the amount of fluid that passes through
the pipe per unit time.
Our model is ultimately based on the Navier-Stokes equations. These equations
arise from applying Newton’s second law of motion to viscous fluids, which we imag-
ine to consist of thin layers, each small element of which must obey conservation
of momentum. In simplest form, the force on each element of fluid is the sum of a
pressure-gradient term and a viscous term. For an incompressible Newtonian fluid, the
viscous term, mathematically, is proportional to the Laplacian of the velocity field.
This is important because it implies that momentum is diffused in the same way heat
diffuses; indeed in the simplest example of unidirectional time-dependent flow, the
Navier-Stokes equations reduce to the heat (or diffusion) equation.
Now consider the flow of a fluid in a uniform circular cylindrical pipe of radius a.
The fluid is assumed to be incompressible and independent of time, and subject to
a fluid pressure that decreases along the pipe in the positive direction. The flow is
described by a velocity field u, which is a vector field, giving the velocity of the fluid
at each point inside the pipe. Under these circumstances the velocity field u is in the
axial direction only, which means that, in cylindrical coordinates, u = h0, 0, u(r )i,
0  r  a. In the presence of a uniform pressure gradient Px along the pipe, the fluid
velocity u(r ) satisfies the differential equation
✓ ◆
1 d du Px
r = , (1)
r dr dr µ

where µ is the coefficient of viscosity. (You can find a derivation of this equation in
any standard text on fluid dynamics [3].) Note that the left-hand side of equation (1) is
just the radial part of the Laplacian in cylindrical coordinates. Integrating the equation
twice, we find that

Px r 2
u = u(r ) = + B ln r + C,
4µ

where B and C are constants to be determined. If u(0) is to be well defined then B ⌘ 0

of course, and because of friction, the velocity is zero at the inner surface of the pipe,
so u(a) = 0, and this implies that C = Px a 2 /4µ. Therefore the velocity profile is a
parabolic one, moving the direction opposite the pressure gradient:

Px 2
u(r ) = (a r 2 ). (2)
4µ

This content downloaded from [Link] on Fri, 20 Apr 2018 18:04:47 UTC
All use subject to [Link]
198 MATHEMATICS MAGAZINE
We can use (2) to compute the rate at which volume flows through the vessel. The
technical term for volume flow rate is the volume flux per unit time through any cross
section (as in F IGURE 1). It is given by the integral
Z a
V (a) = 2⇡r u(r ) dr,
0

where V is expressed in units of volume/time. The mass flux is determined by multi-

plying this expression by the (constant) fluid density ⇢.

a r

Figure 1 Fluid flow in a cylindrical pipe

Substituting the above expression for u,

Z
⇡ Px a ⇡ Px a 4
V (a) = r a2 r 2 dr = . (3)
2µ 0 8µ
This result is known as Poiseuille’s law. Since Px is the constant negative pressure
gradient, we may write it in terms of the pressure difference, 1P, over the length L
of the pipe, that is, Px = 1P/L, so the volume flux (or equivalently, the how much
blood flows per unit time) is
⇡ 1Pa 4
V (a) = / a4 L 1.
8µL
The fourth power makes volume flux very sensitive to the radius of the pipe, which
has significant implications for vascular disease in particular. Halving the radius of
the tube results in a volume flux one sixteenth of the original, and even a reduction in
radius of about 16% will halve the volume flow rate!
The reciprocal of V has units of time/volume and can be taken as a measure of the
frictional resistance. Given a fixed pressure drop, the frictional resistance is directly
proportional to the length of the pipe. The dependence of the resistance on a 4 is
perhaps harder to visualize, but a standard analogy may prove helpful. Poiseuille’s
law is effectively a hydraulic metaphor for Ohm’s law in electrical circuits. This law
states that the voltage E in the circuit is equal to the product of the current I and
the resistance R: that is, E = IR. In fluid flow, the pressure drop 1P is the analog
of the voltage, and the volume flux V is equivalent to the current. Thus the hydraulic
resistance can be expressed as
1P 8µL 4
= /a L. (4)
V ⇡a 4
The idea, then, is that the resistance to the flow of a fluid through a pipe, or a system
of pipes, measures the amount of work required to pump the fluid through the system.
The lower the resistance, the more efficient the system will be. Our first method to find
the best angle of branching (and hence the location) of a smaller vessel is to minimize
resistance along the branching path. Thus, our first cost functional is based on the total
hydraulic resistance of the system to the flow.

This content downloaded from [Link] on Fri, 20 Apr 2018 18:04:47 UTC
All use subject to [Link]
VOL. 84, NO. 3, JUNE 2011 199
The standard calculus model
We consider the case of a smaller vessel branching off from a larger one. For the
time being, we consider the larger vessel to be straight. F IGURE 2 shows the basic
configuration: The main blood vessel of radius r1 has a narrower vessel of radius r2
(< r1 ) branching at an angle ✓ to the axis of the former.
We consider the points A, B, and C to be fixed, with BC perpendicular to AC and
with distances denoted by c = |AC| and b = |CB|. The choice of ✓ determines the
location of the point O along with the distances L 1 = |AO| = c b cot ✓ and L 2 =
|OB| = b csc ✓ .

r2

L2
B

r1 A O C

L1
c

Figure 2 Blood vessel branching from an artery

The first cost functional we examine is one found in many elementary calculus
books [8]. We use (4) to calculate the minimum total hydraulic resistance along the
path AOB.
Since we have established that the hydraulic resistance, according to Poiseuille’s
law, is proportional to length and inversely proportional to the fourth power of the
radius for each component of the system, we may express the total resistance L1 as

L1 L2
L1 = k 4 + 4
r1 r2

c b cot ✓ b csc ✓
=k 4
+ .
r1 r24
The constant k will be significant later. To minimize resistance as a function of ✓, set
"✓ ◆ #
4
d L1 kb r 2
= 4 csc2 ✓ cos ✓ = 0.
d✓ r2 r1

Then the only critical value is

✓ ◆4
1 r2
✓m = cos ,
r1
and a quick verification shows that L1 assumes an absolute minimum on the interval
(0, ⇡/2) at ✓m . We substitute some relative values for the radii to test our model: If
r2 = 0.5r1 , then ✓ ⇡ 86 ; and if r2 = 0.8r1 , then ✓ ⇡ 66 .

This content downloaded from [Link] on Fri, 20 Apr 2018 18:04:47 UTC
All use subject to [Link]
200 MATHEMATICS MAGAZINE
The first angle is almost a right angle. Is our model sound enough to justify such
an extreme prediction? One concern is that the cost functional relates only to the path
AOB; the implied assumption is that the effect of ✓ on the path AOC can be neglected.
We will address that issue soon; but first, we consider variations of the cost functional.

Varying the cost functional

In addition to minimizing the total resistance to flow, Roux and Rosen consider min-
imizing the cost of maintaining the structure. It seems reasonable to suppose that the
maintenance cost should be proportional to the volume of the system. The volume
along the path AOB is given by
L2 = K L 1r12 + L 2r22
where K is another constant. This is minimized when
✓ ◆2
r1
✓ = cos 1 ,
r2
which is not the same as the value ✓m that minimizes the previous functional L1 .
We can also combine these functionals. Our next cost functional includes a term
proportional to resistance and a term proportional to volume:
L = L1 + L2
✓ ◆
L1 L2
=k + 4 + K L 1r12 + L 2r22
r14 r2
✓ ◆ ✓ ◆
k k
= L 1 4 + K r12 + L 2 4 + K r22
r1 r2
⌘ L 1 A1 + L 2 A2 , (5)
where k and K are constants of proportionality. Proceeding in the same manner as
before, we find that

dL 2 A1
= A2 b csc ✓ cos ✓
d✓ A2
which is zero when
 "✓ ◆ ✓ ◆#
1 A1 1 r2 4 k + K r16
✓ = cos = cos .
A2 r1 k + K r26

This result depends on the relative values of k and K . For any positive values of the
constants, it agrees with the original minimizer ✓m only in the two limiting situations:
(i) r2 ! r1 and (ii) r1 ! 0.

Observations
How can we choose a cost functional? More generally, how can we choose a good
model? One test is how well its predictions match observations.
Roux gave a set of empirical observations about about vascular branching (cited
separately by Thompson [9] and Rosen [5]). In paraphrased form, they are:

This content downloaded from [Link] on Fri, 20 Apr 2018 18:04:47 UTC
All use subject to [Link]
VOL. 84, NO. 3, JUNE 2011 201
(1) If an artery bifurcates into two equal branches, they deviate at equal angles from
the main stem.
(2) If the branches do not have the same radius, then the larger branch (i.e., the con-
tinuation of the main artery) makes a smaller angle with the original direction
than does the smaller branch.
(3) Branches that are so narrow that they do not significantly diminish the flow in the
main stem branch off at large angles (typically between 70 and 90 ).

Considering units
One way to understand the combined functional (5) is as a measure of the power
dissipated by the flow. Power is defined as work done per unit time, in this case by the
flow, and work has the dimensions of force times distance. To make this clear, we need
to understand the units of the constants k and K .
In dimensional analysis, use is often made of the units of mass, length, and time
(denoted by [M], [L], and [T ], respectively). These define the dimensions of particu-
lar physical quantities, and are called fundamental units. They are analogous to basis
vectors for a vector space, such as the unit vectors {i, j, k} for the space R3 . Thus,
the dimension of force being mass ⇥ length / time2 is denoted [M][L][T ] 2 . (Other
choices of fundamental units are possible; for example, mass, length, and force could
have been used just as well. The author has written elsewhere about this and the valu-
able Buckingham Pi Theorem [1].)
Considering our combined functional (5), we see that the first parenthetical expres-
sion has units [L] 3 . But from the definition above, power has units
2 1
[M][L][T ] [L][T ] = [M][L]2 [T ] 3 .
Therefore, for the term itself to be in units of power, the constant k must have units of
[M][L]5 [T ] 3 .
By the same token, the second parenthetical expression has units [L]+3 , so the constant
K must have units
[M][L] 1 [T ] 3 .
These considerations would be the starting point for any serious inquiry into what
determines the values of the constants k and K .

Optimizing a single vessel

Next, we consider the contribution to L from a single unbranched vessel of radius r
and length L:
4
L = k Lr + K Lr 2 . (6)
Biologically, this means that the vessel’s contribution to the total cost is determined
by two competing factors: the power required to maintain the volume of blood needed
to fill the vessel, and the power required to pump blood through that same vessel. The
first is proportional to the inverse fourth power of the radius, and for optimality this
requires that the radius be as large as possible; the second is proportional to the square
of the radius, and so the radius should be as small as possible [11].

This content downloaded from [Link] on Fri, 20 Apr 2018 18:04:47 UTC
All use subject to [Link]
202 MATHEMATICS MAGAZINE
If we expect that the branching configuration minimizes the functional, then we
should expect that the radius of each single vessel should minimize it as well. From
equation (6),
dL 5
= ( 4kr + 2K r )L
dr
which is zero when k = K r 6 /2. Since L00 (r ) > 0, this is a minimum. For the optimal
value of r we have
L = Lr 2 , (7)
where = 3K /2 is another constant.

Branching versus bifurcation

Thus far the cost functionals used have referred to a smaller vessel branching off a
larger one, not bifurcation. This will now be amended.
From this point forward we consider all three branches in F IGURE 3. We consider
the points A, B, and C fixed. The choice of the point O (in the same plane) determines
the branching angles and ✓.
The radii and lengths of the vessels AO, OB, and OC are denoted respectively by
r0 , L 0 , r1 , L 1 , and r2 , L 2 .

A r1, L 1 B
r0, L 0
O

r2, L 2

C
Figure 3 Bifurcation

Optimizing the configuration

The important relation (7) will be used in connection with the three diagrams in F IG -
URE 4. In each diagram we consider a small change in the location of the bifurcation
point O to a new point O 0 . In each case, the distance from O to O 0 is , which is small
compared with the distances AO, OB, and OC.
Now from the theorem from the calculus of variations known as the Principle of
Minimum Work [7], we know that a small first-order change in the minimal configu-
ration results in a second-order change in the cost functional, so to the first order, the
value of the functional is unchanged. In each part of F IGURE 4, this means that the
change in L, summed over the three vessels in the configuration, must be zero (to first
order in ). We will see what this implies in each diagram.
Using the law of cosines for F IGURE 4(a), we have
✓ ◆
2
(O 0 B)2 = L 21 + 2 2 L 1 cos ✓ ⇡ L 21 1 cos ✓ .
L1

This content downloaded from [Link] on Fri, 20 Apr 2018 18:04:47 UTC
All use subject to [Link]
VOL. 84, NO. 3, JUNE 2011 203
A r1, L 1 B
r0, L 0
O
O0

r2, L 2
(a)
C

A B

O0

(b)
C

A B
O0

(c)
C
Figure 4 First-order changes to the configuration

Therefore
✓ ◆
0
O B ⇡ L1 1 cos ✓ = L1 cos ✓,
L1
so the path OB has been shortened by approximately cos ✓. Similarly, the path OC is
shortened by cos , and of course, the path AO has been increased by the amount .
Now we apply the result L = Lr 2 for each branch. The total increment in cost is
zero, and this is equal to the sum of these cost increments. After canceling the common
factors and , we find that

r02 = r12 cos ✓ + r22 cos .

In F IGURE 4(b), the length of OB is increased by . It is readily shown using the
law of cosines that AO is reduced by approximately cos ✓ , OC is increased by ap-
proximately cos(✓ + ), and once again, the cost increments sum to zero, so that
r12 = r02 cos ✓ r22 cos(✓ + ). (8)
The result corresponding to F IGURE 4(c) is most easily accomplished by interchanging
r1 and r2 and ✓ and in the result (8):
r22 = r02 cos r12 cos(✓ + ).
From these three equations we can solve for the three angles to obtain
r04 + r14 r24 r04 + r24 r14 r04 r14 r24
cos ✓ = ; cos = ; cos (✓ + ) = . (9)
2r02r12 2r02r22 2r12r22

This content downloaded from [Link] on Fri, 20 Apr 2018 18:04:47 UTC
All use subject to [Link]
204 MATHEMATICS MAGAZINE
These results can be taken further. For optimal flow in a single vessel, the flow rate
f (for a given value of L) turns out to be proportional to the cube of the radius. When
vessels branch the flow must split into two parts, which means that f 0 = f 1 + f 2 and

r03 = r13 + r23 . (10)

(This equation might appear to contradict Poiseuille’s law (3), which might suggest
that r04 = r14 + r24 . But applying Poiseuille’s law in this case would require identical
pressure gradients Px in all three branches, which is not the case in general.)
Eliminating the superfluous radius in each case we obtain
4/3 4/3
r04 + r14 r03 r13 r04 + r24 r03 r23
cos ✓ = ; cos = , (11)
2r02r12 2r02r22
and a similar equation for cos(✓ + ).
To examine some consequences of this newest cost functional, let r1 = r2 , corre-
sponding to equal branchings from the primary vessel. According to (11), this implies
that ✓ = : Vessels with equal radii branch off at equal angles to the main stem. This
is exactly the first empirical law of Roux.
Next let r2 = ↵r1 with 0 < ↵ < 1. The formulas in (9) imply that

r04 + r14 (1 ↵ 4 ) r04 r14 (1 ↵ 4 )

cos ✓ = ; cos = .
2r02r12 2↵ 2r02r12
Is it possible to determine from these equations which of the two angles ✓ or is
the larger? For angles in (0, ⇡/2), the equation cos(✓ ) > cos( ) is equivalent to
2
1 + ↵ 2 r14 > r04 .

But r03 = r13 + r23 leads to r04 = (1 + ↵ 3 )4/3r14 . This and the easily verified inequality
2 4/3
1 + ↵2 > 1 + ↵3

show that for ✓ , 2 (0, ⇡/2), ✓ < . Thus the larger vessel branches off at a smaller
angle than the narrower vessel. This is the second empirical law of Roux.
Now suppose that r2 ⌧ r1 (so that ↵ ⌧ 1). This also implies that r1 . r0 . Then
using the results

r04 r14 (1 ↵ 4 ) 4/3

cos = and r04 = 1 + ↵ 3 r14 ,
2↵ 2r02r12
we may write
4/3
1 + ↵3 1 + ↵4
cos = 2/3
.
2↵ 2 1 + ↵ 3
For a branching angle = 70 , this gives ↵ ⇡ 0.34; and if = 80 , then ↵ ⇡ 0.17,
clearly indicating that the branching angle steepens as the vessel radius narrows. Re-
taining only the leading terms in ↵ we obtain

↵ ! 0 as ↵ ! 0,

so ! ⇡/2 as ↵ ! 0, that is, the branching angle approaches 90 as the vessel radius
tends to zero. These statements are the third empirical law of Roux. Thus with the cost

This content downloaded from [Link] on Fri, 20 Apr 2018 18:04:47 UTC
All use subject to [Link]
VOL. 84, NO. 3, JUNE 2011 205
functional (5) all three empirical laws have been verified, leading one to conclude that
the model may well be fairly realistic for many animal vasculatures, particularly for
the smaller vessels.

Calculating total length

We now apply these ideas to examine the statement of Rosen, quoted at the begin-
ning of the article, regarding the total length of the (large) mammalian vascular sys-
tem. Suppose for simplicity that the primary branch (aorta) bifurcates into two vessels
with equal radii: if r1 = r2 the statement r03 = r13 + r23 is equivalent to r1 = 2 1/3r0 ⇡
0.794r0 . Suppose further that every vessel in a generic vascular system arises from an
equal bifurcation of the immediately preceding larger vessel. In mammals, a typical
capillary radius is about 5 microns, and it is of interest to calculate how many bifur-
cations from a given primary vessel are required to reach this radius. Therefore the
number n of bifurcations from an aorta of radius r0 that will result in such a capillary
must satisfy the equation
6
(0.794)n r0 = 5 ⇥ 10 m.
3
For an animal such as a dog, r0 ⇡ 0.5 cm = 5 ⇥ 10 m, so
3
(0.794)n = 10 3 , and n = ⇡ 30. (12)
log10 (0.794)
It follows from this that the number of vessels in the system resulting from the final
bifurcation is 230 ⇡ 1.07 ⇥ 109 , which is close to the estimate of 1.2 ⇥ 109 given by
Rashevsky [4] and Zamir [11]. As pointed out by Rosen [5], the agreement becomes
even closer when one realizes that observationally, it is probably difficult to distinguish
vessels in the final bifurcation from those arising from several preceding ones (the 29th
or even 28th bifurcation).
Now we try to estimate of the total length of the vascular system. In the absence
of much physiological information (but see below), and in the spirit of a back-of-
the-envelope calculation, we suppose that the primary vessel, the aorta, has length
L 0 , and that each bifurcation k produces twice as many vessels, each of length L k =
⌘L k 1 , where it is assumed that 0 < ⌘ < 1. Then the total length of the system after n
bifurcations is
L n = L 0 + 2⌘L 0 + 22 ⌘L 1 + 23 ⌘L 2 + · · · + 2n ⌘L n 1
2 3 n
= L 0 1 + 2⌘ + (2⌘) + (2⌘) + · · · + (2⌘)
⇥ ⇤
(2⌘)n+1 1 1
= L0 , ⌘ 6= . (13)
2⌘ 1 2
For discussion purposes, based on the table below, we suppose that L 0 = 40 cm,
and take various values for ⌘ > 1/2. For ⌘ = 2/3,
⇥ ⇤
(4/3)31 1
L 30 = 0.4 m ⇡ 9 ⇥ 103 m ⇡ 10 km;
1/3
whereas if ⌘ = 7/8, then
⇥ ⇤
(7/431 1
L 30 = 0.4 m ⇡ 2 ⇥ 107 m = 2 ⇥ 104 km.
3/4

This content downloaded from [Link] on Fri, 20 Apr 2018 18:04:47 UTC
All use subject to [Link]
206 MATHEMATICS MAGAZINE
If ⌘ > 1 (a given bifurcation is longer than the immediately preceding one), the
length explosion is even more rapid: ⌘ = 6/5 results in L 30 ⇡ 6 ⇥ 107 km! If Rosen’s
estimate of 50,000 km for the total length of the vascular system in large mammals
is accurate, then we would on average expect ⌘ to be slightly less than 15/16 on the
basis of this approach. It would appear that since real vascular systems are unlikely to
have this degree of predictability, our version of Rosen’s estimate is at best an educated
guess!
The table below, adapted from that in Zamir [11], is based on the so-called “classical
model,” wherein the arterial tree was thought of as being highly ordered on a global
scale. It remains essentially unmodified today apart from some numerical changes,
but the main difficulty, according to Zamir, is not its numerical accuracy, but its cen-
tral premise, namely that all these vessels have a well defined beginning, end, length,
and diameter. Such conceptual considerations, while interesting, would take us too far
afield in this paper. Nevertheless, the approximate length of the arterial tree (of the
dog) is obtained from level nine of the classical model, yielding about 108 cm or 1000
km, again, considerably lower than Rosen’s estimate.

TABLE 1: Hierarchical structure of the arterial tree

Level Vessel Description No. of Vessels Length (cm) Diameter (mm)

1 aorta 1 40 10
2 large arteries 40 20 3
3 main branches 600 10 1
4 secondary branches 1.8 ⇥ 103 4 0.6
5 tertiary branches 7.6 ⇥ 104 1.4 0.14
6 terminal arteries 106 0.1 0.05
7 terminal branches 1.3 ⇥ 107 0.15 0.03
8 arterioles 4⇥ 107 0.2 0.02
9 capillaries 1.2 ⇥ 109 0.1 0.008

Finally, we summarize the appropriateness of the fundamental result (10), which is

really the basis for both the estimate for the number of bifurcations in the vascular sys-
tem of large mammals, and the derivation of Roux’s empirical laws. According to Za-
mir, in his discussion of the branching structure of arterial trees, “A consideral amount
of data from the cardiovascular systems of man and animals has been shown to support
these results, though with a considerable scatter in some cases.” [11] Such a cube law is
also consistent with there being a constant shear force throughout the arterial network,
a conclusion reached by several authors. (Zamir’s papers give details.[10, 11])
Regarding the number of bifurcations in the arterial tree, Zamir also writes “If the
global structure of the tree were purely ad hoc, the number of branching sites along
certain paths could, simply by chance, be very large, perhaps of the order of thousands.
Yet it is usually found to be surprisingly small, of the order of 30 or so.” Furthermore,
the average increase in cross-sectional area at an arterial bifurcation, that is, the ratio

r12 + r22
A= ,
r02

This content downloaded from [Link] on Fri, 20 Apr 2018 18:04:47 UTC
All use subject to [Link]
VOL. 84, NO. 3, JUNE 2011 207
is found to be about 1.26 for symmetrical bifurcations, and somewhat lower for non-
symmetrical ones. As is easily shown from equation (10), used in conjunction with the
additional statement (used above) that r2 = ↵r1 , 0 < ↵ < 1,
1 + ↵2
A= .
(1 + ↵ 3 )2/3
For ↵ = 1, A = 21/3 ⇡ 1.26; for ↵ = 0.7, A ⇡ 1.22, and for ↵ = 0.5, A ⇡ 1.16.
Indeed, the function A(↵) is readily seen to possess a single maximum at (1, 21/3 ) on
any interval [0, b], where b > 1. Recall that the reciprocal of this maximum value,
2 1/3 , arises naturally in the corresponding bifurcation problem discussed above. It
is also noteworthy that the accepted value for the increase in cross-section area from
the aorta to the capillary beds is about 1000 [2]. Since we have identified the likely
bifurcation number in this system as n = 30, the definition of A requires that

A30 = 1000, so that A = 100.1 ⇡ 1.26.

Acknowledgment I am grateful to Dr. Thomas Klevan and Professor M. Zamir for their kind advice.

REFERENCES
1. J. A. Adam, Mathematics in Nature: Modeling Patterns in the Natural World, Princeton University Press,
Princeton, 2006.
2. A. C. Burton, Physiology and Biophysics of the Circulation, Year Book Medical Publishers, Inc., 1965.
3. S. N. Curle and H. J. Davies, Modern Fluid Dynamics, Vol. 1, Van Nostrand Reinhold, Wokingham, England,
1968.
4. N. Rashevsky, Mathematical Biophysics, Vol. 2, Chapter XXVII, Dover, New York, 1960.
5. R. Rosen, Optimality Principles in Biology, Butterworth, London, 1967.
6. W. Roux, doctoral thesis on the bifurcation of blood vessels, cited by Thompson [9].
7. H. Sagan, Introduction to the Calculus of Variations, Dover, New York, 1992.
8. J. Stewart, Calculus, Early Transcendentals, 5th ed., Brooks/Cole, Belmont, CA, 2003.
9. D’Arcy W. Thompson, On Growth and Form, Dover, New York, 1992.
10. M. Zamir and H. Chee, Segment Analysis of Human Coronary Arteries, Blood Vessels 24 (1987) 76–84.
11. M. Zamir, The Branching Structure of Arterial Trees, Comments on Theoretical Biology 1 (1988) 15–37.

Summary Calculating the optimal angle for blood vessel branching is a standard calculus problem. However,
optimality in that setting is judged by a cost functional that turns out not to give realistic results. We study
a sequence of improvements to the cost functional, finally arriving at one that passes an important modeling
test: From this last functional, we derive three empirical laws of blood vessel branching, originally proposed by
German zoologist Wilhelm Roux.

JOHN A. ADAM is Professor of Mathematics at Old Dominion University in Norfolk, VA. His research interests
are currently associated with theoretical problems in meteorological optics. He enjoys nature photography (of the
point-and-shoot kind), and is a frequent contributor to the Earth Science Picture of the Day website (http:
//[Link]/). He is author of Mathematics in Nature: Modeling Patterns in the Natural World and
coauthor of Guesstimation: Solving the World’s Problems on the Back of a Cocktail Napkin (both Princeton
University Press). His most recent book, A Mathematical Nature Walk (also Princeton) was published in June
2009. He enjoys being a grandparent of four small children: the geometric mean of their ages is 1.19 years (at the
time of writing).

This content downloaded from [Link] on Fri, 20 Apr 2018 18:04:47 UTC
All use subject to [Link]