Chapter 7.

Approximate Shortcut Methods for Multicomponent

Distillation

The previous chapters served as an introduction to multicomponent distillation. Matrix methods are
efficient, but they still require a fair amount of time even on a fast computer. In addition, they are
simulation methods and require a known number of stages and a specified feed plate location. Fairly
rapid approximate methods are required for preliminary economic estimates, for recycle calculations
where the distillation is only a small portion of the entire system, for calculations for control systems, and
as a first estimate for more detailed simulation calculations.
In this chapter we first develop the Fenske equation, which allows calculation of multicomponent
separation at total reflux. Then we switch to the Underwood equations, which allow us to calculate the
minimum reflux ratio. To predict the approximate number of equilibrium stages we then use the empirical
Gilliland correlation that relates the actual number of stages to the number of stages at total reflux, the
minimum reflux ratio, and the actual reflux ratio. The feed location can also be approximated from the
empirical correlation.
7.1 Total Reflux: Fenske Equation
Fenske (1932) derived a rigorous solution for binary and multicomponent distillation at total reflux. The
derivation assumes that the stages are equilibrium stages.
Consider the multicomponent distillation column operating at total reflux shown in Figure 7-1, which has
a total condenser and a partial reboiler. For an equilibrium partial reboiler for any two components A and
B,

(7-1)
Figure 7-1. Total reflux column

Equation (7-1) is just the definition of the relative volatility applied to the reboiler. Material balances for
these components around the reboiler are
 (7-2a)
and

(7-2b)
However, at total reflux, B = 0 and LN = VR. Thus the mass balances become

(7-3)
For a binary system this naturally means that the operating line is the y = x line. Combining Eqs. (7-1) and
(7-3),

(7-4)
If we now move up the column to stage N, the equilibrium equation is

The mass balances around stage N simplify to

Combining these equations, we have

(7-5)
Then Eqs. (7-4) and (7-5) can be combined to give

(7-6)
which relates the ratio of liquid mole fractions leaving stage N−1 to the ratio in the reboiler.
Repeating this procedure for stage N−1, we obtain

(7-7)
We can alternate between the operating and equilibrium equations until we reach the top stage. The result
is
 (7-8)
If we define αAB as the geometric average relative volatility,

(7-9)
Eq. (7-8) becomes

(7-10)
Solving Eq. (7-10) for Nmin, we obtain

(7-11)
which is one form of the Fenske equation. Nmin is the number of equilibrium contacts including the partial
reboiler required at total reflux. If the relative volatility is constant, Eq. (7-11) is exact.
An alternative form of the Fenske equation that is very convenient for multicomponent calculations is
easily derived. Equation (7-11) can also be written as

(7-12)
(DxA)dist is equal to the fractional recovery of A in the distillate multiplied by the amount of A in the feed.

(7-13)
where (FRA)dist is the fractional recovery of A in the distillate. From the definition of fractional recovery,

(7-14)
Substituting Eqs. (7-13) and (7-14) and the corresponding equations for component B into Eq. (7-12)
gives
 (7-15)
Note that in this form of the Fenske equation, (FRA)dist is the fractional recovery of A in the distillate,
while (FRB)bot is the fractional recovery of B in the bottoms. Equation (7-15) is in a convenient form for
multicomponent systems.
The derivation up to this point has been for any number of components. If we now restrict ourselves to a
binary system where xB = 1 − xA, Eq. (7-11) becomes

(7-16)
where x = xA is the mole fraction of the more volatile component. The use of the Fenske equation for
binary systems is quite straightforward. With distillate and bottoms mole fractions of the more volatile
component specified, Nmin is easily calculated if αAB is known. If the relative volatility is not constant,
αAB can be estimated from a geometric average as shown in Eq. (7-9). This can be estimated for a first
trial as
αavg = (α1αR)1/2
where αR is determined from the bottoms composition and α1 from the distillate composition.
For multicomponent systems calculation with the Fenske equation is straightforward if fractional
recoveries of the two keys, A and B, are specified. Equation (7-15) can now be used directly to find Nmin.
The relative volatility can be approximated by a geometric average. Once Nmin is known, the fractional
recoveries of the non-keys (NK) can be found by writing Eq. (7-15) for an NK component, C, and either
key component. Then solve for (FRC)dist or (FRC)bot. When this is done, Eq. (7-15) becomes

(7-17)
If two mole fractions are specified, say xLK,bot and xHK,dist, the multicomponent calculation is more
difficult. We can’t use the Fenske equation directly, but several alternatives are possible. If we can
assume that all NKs are nondistributing, we have

(7-18a)

(7-18b)
As shown in Chapter 5, Eqs. (7-18) can be solved along with the light key (LK) and heavy key (HK) mass
balances and the equations
 (7-19)
Once all distillate and bottoms compositions or values for Dxi,dist and Bxi,bot have been found, Eqs. (7-11)
or (7-12) can be used to find Nmin. Use the key components for this calculation. The assumption of
nondistribution of the NKs can be checked with Eq. (7-10) or (7-17). If the original assumption is invalid,
the calculated value of Nmin obtained for key compositions can be used to calculate the LNK and HNK
compositions in distillate and bottoms. Then Eq. (7-11) or (7-12) is used again.
If NKs do distribute, a reasonable first guess for the distribution is required. This guess can be obtained
by assuming that the distribution of NKs is the same at total reflux as it is at minimum reflux. The
distribution at minimum reflux can be obtained from the Underwood equation and is covered later.
Accurate use of the Fenske equation obviously requires an accurate value for the relative volatility. Smith
(1963) covers in detail a method of calculating α by estimating temperatures and calculating the geometric
average relative volatility. Winn (1958) developed a modification of the Fenske equation that allows the
relative volatility to vary. Wankat and Hubert (1979) modified both the Fenske and Winn equations for
nonequilibrium stages by including a vaporization efficiency.

Example 7-1. Fenske equation

A distillation column with a partial reboiler and a total condenser is being used to separate a mixture
of benzene, toluene, and cumene. The feed is 40 mol% benzene, 30 mol% toluene, and 30 mol%
cumene and is input as a saturated vapor. We desire 95% recovery of the toluene in the distillate and
95% recovery of the cumene in the bottoms. The reflux is returned as a saturated liquid, and constant
molal overflow (CMO) can be assumed. Pressure is 1 atm.
Equilibrium can be represented as constant relative volatilities. Choosing toluene as the reference
component, αbenz−tol = 2.25 and αcumene−tol = 0.21. Find the number of equilibrium stages required at
total reflux and the recovery fraction of benzene in the distillate.
Solution
A. Define. The problem is sketched below. For A = toluene (LK), B = cumene (HK), C = benzene
(LNK), we have αCA = 2.25, αAA = 1.0, αBA = 0.21, zA = 0.3, zB = 0.3, zC = 0.4, FRA,dist = 0.95,
and FRB,bot = 0.95.
 a. Find N at total reflux.
b. Find FRC,dist at total reflux.
B. Explore. Since operation is at total reflux and relative volatilities are constant, we can use the
Fenske equation.
C. Plan. Calculate Nmin from Eq. (7-15), and then calculate FRC,dist from Eq. (7-17).
D. Do It. Equation (7-15) gives

Note that αAB = αtol−cumene = 1/αBA = 1/αcumene−tol. Equation (7-17) gives

which is the desired benzene recovery in the distillate. Note that

E. Check. The results can be checked by calculating FRC,dist using component A instead of B. The
same answer is obtained.
F. Generalize. We could continue this problem by calculating Dxi,dist and Bxi,bot for each component
from Eqs. (7-13) and (7-14). Then distillate and bottoms flow rates can be found from Eqs. (7-
19), and the distillate and bottoms compositions can be calculated.

7.2 Minimum Reflux: Underwood Equations

For binary systems, the pinch point usually occurs at the feed plate. When this occurs, an analytical
solution for the limiting flows can be derived (King, 1980) that is also valid for multicomponent systems
as long as the pinch point occurs at the feed stage. Unfortunately, for multicomponent systems there will
be separate pinch points in both the stripping and enriching sections if there are nondistributing
components. In this case an alternative analysis procedure developed by Underwood (1948) is used to
find the minimum reflux ratio.
The development of the Underwood equations is quite complex and is presented in detail by Underwood
(1948), Smith (1963), and King (1980). Since for most practicing engineers the details of the
development are not as important as the use of the Underwood equations, we will follow the approximate
derivation of Thompson (1980). Thus we will outline the important points but wave our hands about the
mathematical details of the derivation.
If there are nondistributing HNKs present, a “pinch point” of constant composition will occur at minimum
reflux in the enriching section above where the HNKs are fractionated out. With nondistributing LNKs
present, a pinch point will occur in the stripping section. For the enriching section in Figure 7-2, the mass
balance for component i is

(7-20)
 Figure 7-2. Distillation column

At the pinch point, where compositions are constant,

(7-21)
The equilibrium expression can be written in terms of K values as

(7-22)
Combining Eqs. (7-20) to (7-22) we obtain a simplified balance valid in the region of constant
compositions.

(7-23)
Defining the relative volatility αi = Ki/Kref and combining terms in Eq. (7-23),

(7-24)
Solving for the component vapor flow rate, VMin yi,j+1, and rearranging

(7-25)
Equation (7-25) can be summed over all components to give the total vapor flow rate in the enriching
section at minimum reflux.

(7-26)
In the stripping section a similar analysis can be used to derive,

(7-27)
Since the conditions in the stripping section are different than in the rectifying section, in general
and .
Underwood (1948) described generalized forms of Eqs. (7-26) and (7-27) which are equivalent to
defining

(7-28)
Equations (7-26) and (7-27) then become polynomials in φ and and have C roots. The equations are
now

(7-29)
and

(7-30)

If we assume CMO and constant relative volatilities , Underwood showed there are common
values of φ and which satisfy both equations. Equations (7-29) and (7-30) can now be added. Thus, at
minimum reflux

(7-31)
where α is now an average volatility.
Eq. (7-31) is easily simplified with the overall column mass balance

(7-32)
to

(7-33)
ΔVfeed is the change in vapor flow rate at the feed stage. If q is known

(7-34)
If the feed temperature is specified a flash calculation on the feed can be used to determine ΔVfeed.
Equation (7-33) is known as the first Underwood equation. It can be used to calculate appropriate values
of φ. Equation (7-29) is known as the second Underwood equation and is used to calculate Vmin. Once
Vmin is known, Lmin is calculated from the mass balance

(7-35)
The exact method for using the Underwood equation depends on what can be assumed. Three cases will
be considered.
Case A. Assume all NKs do not distribute. In this case the amounts of NKs in the distillate are:

while the amounts of the keys are:

(7-36)

(7-37)
Equation (7-33) can now be solved for the one value of φ between the relative volatilities of the two
keys, αHK−ref < φ < αLK−ref. This value of φ can be substituted into Eq. (7-29) to immediately calculate
Vmin. Then

(7-38)
And Lmin is found from mass balance Eq. (7-35).
This assumption of nondistributing NKs will probably not be valid for sloppy separations or when a
sandwich component is present. In addition, with a sandwich component there are two φ values between
αHK−ref and αLK−ref. Thus use Case C (discussed later) for sandwich components. The method of Shiras et
al. (1950) can be used to check for distribution of NKs.
Case B. Assume that the distributions of NKs determined from the Fenske equation at total reflux are also
valid at minimum reflux. In this case the DxNK,dist values are obtained from the Fenske equation as
described earlier. Again solve Eq. (7-33) for the φ value between the relative volatilities of the two keys.
This φ, the Fenske values of DxNK,dist, and the DxLK,dist and DxHK,dist values obtained from Eqs. (7-36) and
(7-37) are used in Eq. (7-29) to find Vmin. Then Eqs. (7-38) and (7-35) are used to calculate D and Lmin.
This procedure is illustrated in Example 7-2.
Case C. Exact solution without further assumptions. Equation (7-33) is a polynomial with C roots. Solve
this equation for all values of φ lying between the relative volatilities of all components,
αLNK,1−ref < φ1 < αLNK,2−ref < φ2 < αLK−ref < φ3 < αHK−ref < φ4 < αHNK,1−ref
This gives C-1 valid roots. Now write Eq. (7-29) C-1 times; once for each value of φ. We now have C-1
equations and C-1 unknowns (Vmin and Dxi,dist for all LNK and HNK). Solve these simultaneous equations
and then obtain D from Eq. (7-38) and Lmin from Eq. (7-35). A sandwich component problem that must
use this approach is given in Problem 7.D15.
In general, Eq. (7-33) will be of order C in φ where C is the number of components. Saturated liquid and
saturated vapor feeds are special cases and, after simplification, are of order C-1. If the resulting
equation is quadratic, the quadratic formula can be used to find the roots. Otherwise, a root-finding
method should be employed. If only one root, αLK−ref > φ > αHK−ref, is desired, a good first guess is to
assume φ = (αLK−ref + αHK−ref)/2.
The results of the Underwood equations will only be accurate if the basic assumption of constant relative
volatility and CMO are valid. For small variations in α a geometric average calculated as

(7-39)
can be used as an approximation. Application of the Underwood equations to systems with multiple feeds
was studied by Barnes et al. (1972).

Example 7-2. Underwood equations

For the distillation problem given in Example 7-1 find the minimum reflux ratio. Use a basis of 100
kmol/h of feed.
Solution
A. Define. The problem was sketched in Example 7-1. We now wish to find (L/D)min.
B. Explore. Since the relative volatilities are approximately constant, the Underwood equations can
easily be used to estimate the minimum reflux ratio.
C. Plan. This problem fits into Case A or Case B. We can calculate Dxi,dist values as described in
Cases A or B, Eqs. (7-36) and (7-37), and solve Eq. (7-33) for φ where φ lies between the
relative volatilities of the two keys 0.21 < φ < 1.00. Then Vmin can be found from Eq. (7-29), D
from Eq. (7-38) and Lmin from Eq. (7-35).
D. Do It. Follow Case B analysis. Since the feed is a saturated vapor, q = 0 and ΔVfeed = F (1 − q) =
 F = 100 and Eq. (7-33) becomes

Solving for φ between 0.21 and 1.00, we obtain φ = 0.5454. Equation (7-29) is

where
Dxi,dist = F zi(FR)i,dist
For benzene this is
Dxben,dist = 100(0.4)(0.998) = 39.92
where the fractional recovery of benzene is the value calculated in Example 7-1 at total reflux. The
other distillate values are
Dxtol,dist = 100(0.3)(0.95) = 28.5 and Dxcum,dist = 100(0.3)(0.05) = 1.5
Summing the three distillate flows, D = 69.92. Equation (7-29) becomes

From a mass balance, Lmin = Vmin − D = 44.48, and (L/D)min = 0.636.

E. Check. The Case A calculation gives essentially the same result.
F. Generalize. The addition of more components does not make the calculation more difficult as long
as the fractional recoveries can be accurately estimated. The value of φ must be accurately
determined since it can have a major effect on the calculation. Since the separation is easy,
(L/D)min is quite small in this case. (L/D)min will not be as dependent on the exact values of φ as it
is when (L/D)min is large.

7.3 Gilliland Correlation for Number of Stages at Finite Reflux Ratio

A general shortcut method for determining the number of stages required for a multicomponent distillation
at finite reflux ratios would be extremely useful. Unfortunately, such a method has not been developed.
However, Gilliland (1940) noted that he could empirically relate the number of stages N at finite reflux
ratio L/D to the minimum number of stages Nmin and the minimum reflux ratio (L/D)min. Gilliland did a
series of accurate stage-by-stage calculations and found that he could correlate the function (N−Nmin)/(N
+ 1) with the function [L/D − (L/D)min]/(L/D + 1). This correlation as modified by Liddle (1968) is
shown in Figure 7-3. The data points are the results of Gilliland’s stage-by-stage calculations and show
the scatter inherent in this correlation.
Figure 7-3. Gilliland correlation as modified by Liddle (1968); reprinted with permission from
Chemical Engineering, 75(23), 137 (1968), copyright 1968, McGraw-Hill.
To use the Gilliland correlation we proceed as follows:
1. Calculate Nmin from the Fenske equation.
2. Calculate (L/D)min from the Underwood equations or analytically for a binary system.
3. Choose actual (L/D). This is usually done as some multiplier (1.05 to 1.5) times (L/D)min.
4. Calculate the abscissa.
5. Determine the ordinate value.
6. Calculate the actual number of stages, N.
The Gilliland correlation should only be used for rough estimates. The calculated number of stages can be
off by ± 30% although they are usually within ± 7%. Since L/D is usually a multiple of (L/D)min, L/D = M
(L/D)min, the abscissa can be written as

The abscissa is not very sensitive to the (L/D)min value, but does depend on the multiplier M.
The optimum feed plate location can also be estimated. First, use the Fenske equation to estimate where
the feed stage would be at total reflux. This can be done by determining the number of stages required to
go from the feed concentrations to the distillate concentrations for the keys.

(7-40a)
Now assume that the relative feed location is constant as we change the reflux ratio from total reflux to a
finite value. Thus

(7-40b)
The actual feed stage can now be estimated from Eq. (7-40b).
An alternate procedure that is probably a more accurate estimate of the feed stage location is Kirkbride’s
method (Humphrey and Keller, 1997). The ratio of the number of trays above the feed, Nf − 1, to the
number below the feed stage, N − Nf, can be estimated as,

(7-41)
Since neither procedure is likely to be very accurate, they should only be used as first guesses of the feed
location for simulations.
The Gilliland correlation can also be fit to equations. Liddle (1968) fit the Gilliland correlation to three
equations. Let x = [L/D − (L/D)min]/(L/D + 1). Then

(7-42a)
while for 0.01 < x < 0.90

(7-42b)
and for 0.90 ≤ x ≤ 1.0

(7-42c)
For most situations Eq. (7-42b) is appropriate. The fit to the data is shown in Figure 7-3. Naturally, the
equations are useful for computer calculations. Erbar and Maddox (1961) (see King, 1980, or Hines and
Maddox, 1985) developed a somewhat more accurate correlation that uses more than one curve.
As a rough rule of thumb we can estimate N = 2.5 Nmin. This estimate then requires only a calculation of
Nmin and will be useful for very preliminary estimates.

Example 7-3. Gilliland correlation

Estimate the total number of equilibrium stages and the optimum feed plate location required for the
distillation problem presented in Examples 7-1 and 7-2 if the actual reflux ratio is set at L/D = 2.
Solution
A. Define. The problem was sketched in Examples 7-1 and 7-2. F = 100, L/D = 2, and we wish to
estimate N and NF.
B. Explore. An estimate can be obtained from the Gilliland correlation, while a more exact
calculation could be done with a process simulator. We will use the Gilliland correlation.
C. Plan. Calculate the abscissa

determine the ordinate

 from the Gilliland correlation, and then find N. (L/D)min was found in Example 7-2, and Nmin in
Example 7-1. The feed plate location is estimated from Eqs. (7-41) and (7-40).
D. Do It.

The corresponding ordinate (N − Nmin)/(N + 1) = 0.27 using Liddle’s curve. Since Nmin = 3.77, N =
5.53. From Eq. (7-40a), NF,min is calculated as

Where xLK,dist was found from Example 7-2 as

and
xHK,dist = xcum,dist = 0.021
Then, from Eq. (7-40b),

E. Check. A check of the Gilliland correlation can be obtained from Eq. (7-42b). With x = 0.455 this
is

or (1 − 0.283) N = Nmin + 0.283, which gives N = 5.65. The 2% difference between these two results
gives an idea of the accuracy of Eq. (7-42) in fitting the curve.
A check on the value of Nf can be obtained with Kirkbride’s Eq. (7-41). To use this equation we need
to know the terms on the RHS. From Example 7-2, F = 100 and D = 69.92. Thus, B = 100 − 69.92 =
30.08. The HK = cumene and the LK = toluene. The feed mole fractions of both are 0.30. From
example 7-2:
Dxcum,dist = 1.5. Then xHK,dist = (Dxcum,dist)/D = 1.5/69.92 = 0.02145.
Dxtol,dist = 28.5. Then Bxtol,bot = Fz − Dxtol,dist = 30.0 − 28.5 = 1.5, and
xLK,bot = Bxtol,bot/B = 1.5/30.08 = 0.04987.
Then, Eq. (7-41) becomes,

and (Nf − 1)/(N − Nf) = 1.1898

which gives Nf = 3.46 if we use N = 5.53 or Nf = 3.7 if we use N = 6. Thus, the best estimate is to use
either the 3rd or 4th stage for the feed. This agrees rather well with the previous estimate.
 A complete check would require solution with a process simulator.
F. Generalize. The Gilliland correlation is a rapid method for estimating the number of equilibrium
stages in a distillation column. It should not be used for final designs because of its inherent
inaccuracy.

7.4 Summary—Objectives
In this chapter we developed approximate shortcut methods for binary and multicomponent distillation.
You should be able to satisfy the following objectives:
1. Derive the Fenske equation and use it to determine the number of stages required at total reflux and the
splits of NK components
2. Use the Underwood equations to determine the minimum reflux ratio for multicomponent distillation
3. Use the Gilliland correlation to estimate the actual number of stages in a column and the optimum feed
stage location
References
Barnes, F. J., D. N. Hansen, and C. J. King, “Calculation of Minimum Reflux for Distillation Columns
with Multiple Feeds,” Ind. Eng. Chem. Process Des. Develop., 11, 136 (1972).
Erbar, J. H., and R. N. Maddox, Petrol. Refin., 40(5), 183 (1961).
Fenske, M. R., “Fractionation of Straight-Run Pennsylvania Gasoline,” Ind. Eng. Chem., 24, 482
(1932).
Gilliland, E. R., “Multicomponent Rectification,” Ind. Eng. Chem., 32, 1220 (1940).
Hines A. L., and R. N. Maddox, Mass Transfer. Fundamentals and Applications, Prentice Hall,
Englewood Cliffs, New Jersey, 1985.
Humphrey, J. L., and G. E. Keller II, Separation Process Technology, McGraw-Hill, New York,
1997.
King, C. J., Separation Processes, 2nd ed., McGraw-Hill, New York, 1980.
Liddle, C. J., “Improved Shortcut Method for Distillation Calculations,” Chem. Eng., 75(23), 137
(Oct. 21, 1968).
Shiras, R. N., D. N. Hansen and C. H. Gibson, “Calculation of Minimum Reflux in Distillation
Columns,” Ind. Eng. Chem., 42, 871 (1950).
Smith, B. D., Design of Equilibrium Stage Processes, McGraw-Hill, New York, 1963.
Thompson, R. E., “Shortcut Design Method-Minimum Reflux,” AIChE Modular Instructions, Series
B, Vol. 2, 5 (1981).
Underwood, A. J. V., “Fractional Distillation of Multicomponent Mixtures,” Chem. Eng. Prog., 44,
603 (1948).
Wankat, P. C., and J. Hubert, “Use of the Vaporization Efficiency in Closed Form Solutions for
Separation Columns,” Ind. Eng. Chem. Process Des. Develop., 18, 394 (1979).
Winn, F. W., Pet. Refiner, 37, 216 (1958).
Homework
A. Discussion Problems
A1. The Fenske equation: