DESIGN OF PACKED CATALYTIC REACTORS

John Beek

Shell Development Compony, Emeryville, California

I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11. Reduction of Chemical and Rate Equations to an Independent Set .
A. Definition of Stoichiometric Matrix and Submatrices . . . . . . . . . . . . . . . 205
B. Relations between Conversions and Concentrations . . . . . . . . . . . .
C. Calculation of Additive Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
D. Introduction of Virtual Conversions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
E. Illustration . . . . . . . .
F. Exceptional Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
111. Equations Describing Simultaneous Reaction
A. Statement of Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
B. Derivation of Conservation Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
C. Adiabatic Reactor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
D. The One-Dimensional Approximation . . . . . . . . . ................ 222
E. Calculation of the Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IV. Estimation of Transport Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A. Introductory Discussion . . . . . . . . . . . . . . . . . . . . . . . .
B. Velocity Profile . . . . . . . . . . . . .
C.Eddy Diffusivity . . . . . . . . . . . . . .................
D. Other Contributions to Effective

V. Numerical Solution of Equations .....................

A. Automatic Computer Assume

E. Introduction of F as Radial Variable . . . . . . . . . . ............... 240

F. Explicit Partial DiBlerence Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
G. Implicit Partial Difference Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
H. Adiabatic Reactor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
I. Illustrations . . . . . . . . . . . . . . . . . . . . . .
VI. Stability of Packed Tubular Reactors . . . . . . . . . . . . . . . . . .
A. Statement of Problem . . . . . . . . . . . . . . . . . . . .
B. Description of Barkelew's Criterion . . . . . . . ............... 268
VII. Scale Models of Packed Tubular Reactors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
A. General Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . ............... 259
B. Derivation of Requirements for the General Case . . . . . . . . . . . . . 201
C. Selection of Important Quantities ......................... . 1 262
203
204 JOHN BEEK

D. Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
E. The One-Dimensional Approximation .............................. 265
F. Example ............. ............ .,. 267
Nomenclature . . . . . ................................... 268
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269

1. Introduction
It is understood by practicing chemical engineers that the design of
complicated pieces of equipment has not been reduced to a straightfor-
ward operation: there is no formula into which we can substitute the
required conditions to give explicitly the corresponding design parameters.
The design process involves, rather, the calculation of the conditions that
would be realized in the equipment with the design parameters fixed a t
various sets of values, and then selecting a suitable set of parameters for
the design.
It is the purpose of this chapter to discuss presently known methods
for predicting the performance of nonisothermal continuous catalytic re-
actors, and to point out some of the problems that remain to be solved
before a complete description of such reactors can be worked out. Most
attention will be given to packed catalytic reactors of the heat-exchanger
type, in which a major requirement is that enough heat be transferred to
control the temperature within permissible limits. This choice is justified
by the observation that adiabatic catalytic reactors can be treated almost
as speciaI cases of packed tubular reactors. There will be no discussion
of reactors in which velocities are high enough to make kinetic energy
important, or in which the flow pattern is determined critically by accel-
eration effects.
In order to provide a logical framework for the discussion, an example
will be worked out in enough detail to illustrate the methods used. There
is no question of providing a recipe for designing nonisothermal reactors;
methods of working that are useful will be presented, and their applica-
tion to a more or less typical problem will be described.
It will be supposed that the kinetics of all the reactions that are going
on and the thermodynamical and molecular transport properties of all
the substances present are known, and that it is desired to find out how
the composition of the efluent from a reactor depends on the conditions
that are imposed. The conditions that must be fixed are the composition,
pressure, temperature, and flow rate of the reactant mixture, the dimen-
sions of the reactor and of the catalyst pellets, and enough properties of
the heat-transfer medium to determine a relation between the tempera-
ture of the tube wall and the heat flux through it.
Only systems in a quasi-steady state will be considered; that is, the
 DESIGN OF PACKED CATALYTIC REACTORS 205

discussion will not include transient states in which the change is fast
enough to give the heat capacity of the system any importance, but will
apply to those systems in which a progressive deterioration of the catalyst
makes changes that become significant only after several hours. Some
attention will be given to a cooperative property of the system that
Amundson has called parametric sensitivity, but this has reference to the
magnitude of differences between two steady states corresponding to some
difference in the conditions.
The discussion will also be restricted to reactors in which the range
of temperature is too wide to permit the use of an average temperature
to characterize the whole catalyst bed. No sharp line can be drawn fol-
lowing this restriction, but it clearly includes any case in which the rate
of reaction cannot be described satisfactorily as a linear function of tem-
perature over the whole range covered by the reactor. The only way to
find out in doubtful cases whether the variation of temperature is sig-
nificant is to make the calculation taking the variation into account.
With the above restriction, an explicit representation of the output of
a reactor as a tabulated function of the input conditions is out of the
question. In most cases, the output must be calculated by numerical
methods. This means that the conditions within the reactor must be cal-
culated stepwise, starting with the known conditions at the inlet. This
process may be thought of as finding a numerical solution for a set of
differential equations, or, following the suggestion of Deans and Lapidus
(D3), as solving a set of difference equations, moving along the reactor
by steps of a particle diameter. The important feature that all such proc-
esses have in common is that, in order to find out what comes out of the
reactor, it is necessary to find out what the conditions are everywhere in
the reactor. The first problem is to formulate the equations that are used
to get this information. Toward this end, the equations that describe con-
ditions within the reactor are formulated, suggestions are made for eval-
uating the transport properties that appear in the equations, and then
methods for solving the equations are given. Finally, approximate meth-
ods are presented for evaluating the parametric sensitivity of reactors,
and for changing the scale of a reactor without changing the course of
the reaction.

II. Reduction of Chemical and Rate Equations to an independent Set

A. DEFINITION MATRIXAND SUBMATRICES

OF THE STOICHIOMETRIC

1. The Complete Matrix

It is not unusual to encounter a reacting system in which a given set
of products may be obtained by more than one reaction path. In a simple
206 JOHN BEEK

case there is no difficulty in writing down the stoichiometric relations, but

in complicated situations, such as the catalytic reforming of naphthas, i t
is useful to have a systematic scheme for relating the concentrations of
the various substances to the concentrations of a minimum number of key
substances. The scheme employed here will take advantage of the com-
pact language of matrix mathematics ( A l , H4). An algebraic treatment
of multiple-reaction problems has been given by Brota (B8a). The reader
should note that a knowledge of matrix methods is not essential for un-
derstanding the later sections of this article.
The quantities involved in such a scheme are derived from the sto-
ichiometric coefficients that appear in the chemical equations, the equa-
tions being written in such a form that the coefficients tell the number of
moles of the substance in question produced by one inole of the reaction.
A familiar example of multiple reaction paths is the catalytic oxidation of ethyl-
ene to ethylene oxide, where the by-products water and carbon dioxide are pro-
duced both by direct oxidation of ethyIene and by further oxidation of ethylene
oxide. For this example, the equations may be written in the form
0 = -Q& - 02 + GH40
0 = -CzHa - 3 02 + 2 COz - 2 HZO
O = - $02 - CzH,O + 2 CO2 + 2 HZ0
The array of coefficients on the right-hand side of the equations is manipulated as
a matrix, which may be denoted as S, the matrix of stoichiometric coefficients. In
the example,

s=
(
-1
-1
0
-6
-3
-$
: : ;)
-1 2 2
(2-1)

If any inert substances are present, they are represented in the matrix by columns
of zeros. In this and the other matrices used in the calculation, the substances and
the quantities that pertain to them are arranged by columns, while the reactions
and the quantities that pertain to them are arranged by rows. For example, the
set of heat capacities, represented as C,, is a row matrix, while the set of reaction
rates R is represented as a column matrix.
2. The Reduced Matrix
The number of independent reactions is just the rank of the matrix S,
that is, the rank of the largest nonsingular square submatrix of S. In the
example above, the second row is the sum of the other two, so that the
rank cannot be 3. On the other hand, the four elements in the upper left-
hand corner obviously form a nonsingular submatrix, so that the rank
is 2. A nonsingular square submatrix s having the rank’ of S, is selected
* Throughout this review the first digit of a numbered equation refers to the
section numbered with the corresponding Roman numeral.
 DESIGN OF PACKED CATALYTIC REACTORS 207

as a basis for the calculation. The choice amounts to selecting a set of

key substances and a corresponding set of key reactions] in terms of
which the composition of the system is to be represented. There is some
advantage in choosing this matrix from columns referring to substances
that have easily determined concentrations. In the example, a good choice
would be the first two rows of the columns for ethylene oxide and carbon
dioxide; here there is the additional advantage that the submatrix is
diagonal] having the form

It is desirable a t this point to rearrange the matrix S so that the

submatrix s appears in the upper left-hand corner. Then S is writ,ten in
the form

S=(
1 0 -1
0 2 -1
-1 2
as though the chemical equations were
0
-$
-3
-3
0
2
2 1 (2-3)

0 = CzHpO CzH, - 4 0 2
O = 2 CO, - CzH4 - 3 02 + 2 HpO
0 = -CzHdO + 2 CO, - $ 0 , + 2 Hz0
3. Other Submatrices
Two other submatrices of S are needed in the calculation: they are
SR,made up of the rows of S that appear in s, and S,, made up of the
columns of S that appear in s. In the example, they are

SR = (01 0
2
-1
-1
-$
-3
0
2 (2-4)
and

s, = ( i :)
-1 2
The fact that
s = s&'sR
will be made use of later.

B. RELATIONS
BETWEEN CONVERSIONS
AND CONCENTRATIONS

1. Fundamental Relation
It is the relation between the concentrations and the conversions that
gives meaning to the stoichiometric coefficients. The relation is
208 JOHN BEEK

Y = YO + XTS (2-7)
or, for a single reaction,
Y = YO+XS
We note that XT is the transpose of X,in which rows and columns are
interchanged. Equation (2-7), when put into words, says that the number
of moles of each substance present in a certain mass of reacting mixture
is the number of moles there initially plus whatever amount has been
formed by all the reactions that have occurred. The conversion X is de-
fined as the number of moles of the corresponding reaction, as written,
that have occurred in a mass of reactant equal t o the average molecular
weight of the initiaI mixture, and the concentration Y is the number of
moles of the substance in question in this same mass; Yo,the initial value
of Y,is just the initial mole fraction.
2. Composition in Terms of Concentrations of K e y Components
Now the key components, those that correspond to the columns of s,
are singled out for special attention. The set of the concentrations of these
components is denoted as y, and a part of Eq. (2-7) is written in the
form
+
y = so XTSc (2-8)
The formula for calculating Y from y is obtained by writing Eq. (2-7)
with S,s-lSB substituted for S. The equation becomes
+
Y = Yo XTScs-'SR
+
= Yo (y - Yo)s-'SR
= (Yo- Y S - ' s R ) +
Y(s-'SR) (2-9)
The last form of this equation is in the best form for computing, since Y
is found from a constant plus a multiple of y . I n following the course of
a set of reactions, the two matrices in parentheses would be calculated
once for all, and then used as constants. Equation (2-9) gives a syste-
matic procedure for finding the concentrations of all components from the
concentrations of the key components, which are followed through the
reactor by using Eq. (2-8) either directly or indirectly.
3. Calculation of Mole Fractions
I n order to calculate the mole fraction of a component, the total num-
ber of moles in the reference mass must be known. This total number of
moles is the sum of all the elements of Y.Sometimes it is more con-
venient to calculate this sum with a subsidiary equation than to sum the
concentrations directly. The column matrix formed by summing each row
of SR is denoted by v ; i t gives the net number of moles formed by the
 DESIGN OF PACKED CATALYTIC REACTORS 209

key reactions. Then the desired relation is obtained by summing rows in

the postfactors of Eq. (2-9) , giving the equation
ZY (1 - yos-'v)'+ y(s-'v)
= (2-10)
C. CALCULATION OF AN ADDITIVE PEOPERTY
Any additive property of the mixture can be obtained directly from
the formula for Y.Let P be a row matrix, the elements of which are the
molar values of the property for the several components. Then F, the
additive property of the mixture, is found from the equation
P=YPT
= (Yo- ps-'SR)PT + P(S-'SRPT) ( 2 1 1)
It must be remembered that F calculated in this way refers to the mass
of mixture equal to the initial average molecular weight, that is, to EY
moles.
From the point of view of the computer, it is worth noting that in
each of the working equations, just as many multiplications as there are
independent reactions are used in calculating on element of the desired
matrix.

D. INTRODUCTION CONVERSIONS
OF VIRTUAL

1. Definition of Virtual Conversions

It is sometimes useful to construct a set of virtual conversions, related
to virtual reaction rates, that can be used instead of the concentrations
for describing the composition of a reacting mixture. The column matrix
x can be defined by putting xTs for the term XTS,that appears in Eq.
(2-8), that is, for the amounts of the key components that have been
formed. From the equation
XTS = XTS, (2-12)
it follows immediately that
XT = XTS,s-' (2-13)
and that

(2-14)
The last result shows that the set of virtual conversions can be used to
calculate all the concentrations. By considering the variation of 2, the
relation
(R = RTS,S-' (2-15)
may be seen to hold, in which is the column matrix of virtual rates.
210 J O H N BEEK

2. FORMULAS
FOR OTHER QUANTITIES

The formulas of Eqs. (2-9), (2-lo), and (2-11) can be transformed

to expressions in x. The results follow.
Y = Yo + X T S R (2-16)
ZY = 1 XTY + (2-17)
P = YoPT X T S R P T + (2-18)
The concentrations of key substances and the virtual conversions are
practically equivalent as variables for determining the composition of a
reacting mixture.
E. I~LUSTRATION
The uae of these relations may be illustrated by continuing with the example
of the oxidation of ethylene. The factors needed are calculated, and then used in
the expressions for the desired quantities.

s-1 = (; ;)

Y = (-;) (2-21)

s-'Y = (-;) (2-22)

ZY = 1 + ay10 + tY1 (2-23)

The concentration of oxygen is
Y4 = YrO + + f ~ i " &ao - t ~ -i &i (2-24)
and its mole fraction is this concentration divided by EY. In terms of the virtual
conversions, similar expressions are obtained.

S,s-' = (2-25)
-1 1
EY = 1 - 321 (2-26)
R1 = R1 - Ra (2-27)
RZ= Rz 4- Ra (2-28)
F. EXCEPTIONAL CASES
For lack of information about the kinetics of the reactions, it is some-
times necessary to approximate the selectivity for a desired product in
the form of some function of the conversion with respect to some one
 DESIGN O F PACKED CATALYTIC REACTORS 211
reaction, giving no attention to the detailed effects of conditions on the
rates of side reactions. In such a situation, with the rates of side reac-
tions nominally determined by the conversion for one reaction, the side
reactions are not independent, and only one concentration variable needs
to be followed, no matter what the rank of the stoichiometric matrix is. I n
fact, there is formally only one reaction, with nonintegral stoichiometric
coefficients that may vary with conversion.
Another situation in which the number of independent reactions is
less than the rank of the formal stoichiometric matrix arises when the
rates of two reactions are always in the same ratio. In this situation, the
two reactions should be combined into one, and instead of the two cor-
responding rows of the matrix, the appropriate linear combination of
them should be used.

111. Equations Describing Simultaneous Reaction and Transport Processes

A. STATEMENT
OF ASSUMPTIONS

1. Smooth Variation of Properties

It is in order to state first the problem in rather genera1 form, and in
as complete detail as is warranted by our knowledge of the processes that
are involved. In the first place, a bed of catalyst pellets is treated as
though it were statistically homogeneous, its average density depending
on distances from walls or other disturbances to the structure of the
packing, but no account being taken of the actual detailed location of
individual pellets, or of the actual shape of passages among the pellets.
Accordingly, it is assumed that such quantities as the superficial mass
velocity and the friction factor vary smoothly through the bed, and, as
a further consequence, that the temperature, pressure and composition
vary smoothly.
Since it is obvious that the smooth variation of these quantities does
not extend to the surface of catalyst pellets, much less through them, it
must be explained that the smoothness has significance only for a succes-
sion of points that are in some sense in equivalent positions relative to
the particles of packing. We might, for example, think of a network of
points that are roughly in the centers of the holes among the pellets:
taking account only of these points, the temperature can be described as
a fairly smooth function of position. An alternative network would be
the points a t the centers of catalyst pellets.
As the statistical description of the packed bed must depend on some
kind of averaging to represent the properties of the bed as functions of
position, the property is averaged over a representative region having the
212 JOHN BEER

same order of magnitude as a particle of packing. For the solid, this

region is just the volume of a particle. For the fluid, the region can be
thought of as bounded by constrictions in the passages among the parti-
cles, or, somewhat vaguely, as being a hole in the packing. I t % an essen-
tial weakness of this statistical description of random packed beds that
these regions vary in size and shape, so that the average of a property is
only approximately a smooth function of position.
In the traditional view, the desired smoothness is induced by con-
sidering only averages over regions that include enough particles to give
statistical stability to the averages, but that are still small compared to
the size of the bed. This concept is hardly satisfactory when the diameter
of the tube is only 10 times the diameter of the particles, so that the whole
range of radial variation is covered in 5 particle diameters. The average
over a small region is subject to larger fluctuations, but at least it gives
more meaning to the concept of smooth variation of properties in the
packed bed. Baron (B3) has discussed the effect of spatial fluctuations
on the behavior of packed reactors; it may be hoped that our knowledge
of the structure of packed beds will develop enough so that this effect can
be taken into account, but a t present it can only be characterized as
small.
2. EDDYDIFFUSION DOMINANT
Another feature of packed beds that must be discussed before mate-
rial and enthalpy balances can be formulated is the apparent eddy dif-
fusion that is observed, both transverse to the flow and parallel to the
flow. It is generally agreed that the transverse diffusion results from a
random walk of elements of fluid through the passages in the bed (B3),
or equivalently, from mixing of converging streams (Bl, R1) and that
the diffusion paralIeI to the flow arises primarily from the stagewise mix-
ing in the interstices of the bed (A4, K2),with some contribution from
the variation of mass velocity with radial position (Cl,C5). I n the usual
formulation of the equations, terms for the accumulation of material and
of heat by both transverse and parallel diffusion have been included (see,
for example, 54). The radial transport of heat in a heat-exchanger type
of tubular reactor cannot, of course, be neglected, and the description
of the mixing in a packed bed by a diffusion coefficient is so good that
the term for the transverse diffusion can be included without further dis-
cussion.
3. AXIALDIFFUSIVE
PROCESSES
NEGLECTED
There are two reasons for adopting a different attitude toward the
apparent diffusion parallel to the flow. One is that in the reactors that
 DESIGN OF PACKED CATALYTIC REACTOBS 213
one expects to encounter in commercial practice, the effect is small. The
Peclet number for axial diffusion is based on the particle diameter, and
is between 1 and 2; then the Peclet number based on the length of the
bed, which gives some idea of the importance of diffusion in modifying
a process going on in the bed, is between 1 and 2 times the number of
particle diameters in some characteristic length, such as the length of a
region in which the temperature or concentration is changing rapidly. In
any case, this length has the same order of magnitude as the length of the
bed, so that the significant Peclet number is a t least 100. Another obser-
vation indicating the negligible magnitude of the diffusive effect is that
with a Peclet number of 2, the concentration 3 particle diameters up-
stream from a steady source is about 1/400 what it is just downstream
from the source.
The second reason goes back to the nature of the processes giving rise
to the apparent diffusion. What is observed is that an impulsive signal
introduced into a fluid flowing through a packed bed is dispersed as it
goes downstream, or that a harmonic signal is attenuated. From the ex-
tent of the dispersion or attenuation an apparent diffusivity is calculated,
and this diffusivity is useful in predicting to the first order the small
effects of the dispersive process. It must be kept in mind, however, that
if the dispersion arises only from the processes mentioned above, that is,
from stagewise mixing and the variation of velocity, it does not lead to
one of the important effects of diffusion, which is the propagation of a
signal upstream. Only if there is a significant amount of backmixing, so
that a signal is sent upstream, is it reasonable to represent the dispersive
process as diffusion. The importance of this distinction was emphasized
by Beek and Miller (B4)who pointed out that if there is backmixing,
boundary conditions a t both ends of the reactor must be used to define
the conditions within the reactor; this is just the same as the situation
that was shown by Wehner and Wilhelm (W2) to hold when a term for
axial diffusion is introduced into the differential equations describing the
system.
From the point of view of practical computation, the equations with-
out axial diffusion are solved by starting at the inlet to the reactor and
computing the conditions stepwise downstream, making no adjustment
for what happens further downstream. When there is axial diffusion, the
conditions a t all points in the reactor have to be adjusted to meet bound-
ary conditions a t both ends, and the difficulty of the computation is mul-
tiplied by a large factor.
Deans and Lapidus (D3), assuming that backmixing is negligible,
have recommended that, instead of trying to use diffusion to represent
an essentially different dispersive effect, one should formulate a set of
214 JOHN BEEK

difference equations that describe the processes that are actually going
on in the packed bed. There is no question but that this scheme gives
the best representation of the behavior of a packed bed that is a t present
available, but this good representation is very expensive in computing
time, both because i t requires axial steps about equal to a particle diam-
eter and because the simultaneous difference equations that must be
solved are not linear in the unknown quantities. As will be explained
below, this method is the one to use when the reactor is so short-meas-
ured in particle diameters-that the dispersive effects must be taken into
account.
In summary, the argument for writing the equations describing the
reactor without accounting for axial dispersion is that this effect usually
has very little importance, while the extra effort required to account for
it is large. The equations derived in this section will be based on the
assumptions that the axial dispersion is negligible, and that the condi-
tions within the bed are sufficiently smooth functions of position to be
related by differential equations. These assumptions involve the reserva-
tion that the bed is not extremely short.
4. Heat-Transfer Coefficient at the Wall
Another assumption that will be made hardly needs stating, as i t has
come to be generally accepted. It will be assumed that there is a local
resistance to heat transfer from the fluid to the wall of the reactor, giving
rise to what is practically a discontinuity in the radial temperature profile
a t the inside surface of the wall. It has not been possible to determine
the exact character of the wall effect that is observed, so that the descrip-
tion used here is not certainly the best one. On the basis of the experi-
ments that have been made, however, this description is indistinguish-
able from an alternative one that assigns a reduced thermal conductivity
to a layer adjoining the wall (see Y1 and Y3).

B. DERIVATION
OF CONSERVATION
EQUATIONS

1. Nature of Transverse E d d y Diffusion

a. Concentration Based on Mass
There is a sound experimental basis (B6, L1) for the conclusion that
in packed tubes when the Reynolds number is above 100, the transport
of both matter and heat transverse to the flow is dominated by eddy dif-
fusive processes. To the extent that these processes dominate, all con-
served entities in the fluid diffuse alike, provided the gradients of the
concentrations of these entities in amount per unit mass of fluid are used
as driving forces for the diffusion. I n all that follows, it will be assumed
 DESIGN OF PACKED CATALYTIC REACTORS 215
that molecular diffusion is negligible in comparison with eddy diffusion.
The system has the property that the flux of an entity transverse to the
flow can be related to a Peclet number by the equation

J , = - (Gdp/Np,)dc/ax (3-1)
the same Peclet number serving for all conserved quantities. I n this
equation, J , is the component in the x-direction of the superficial diffu-
sive flux, G is the magnitude of the superficial mass velocity, c is the
concentration of a conserved entity, and x is a coordinate perpendicular
to the direction of flow. All flows and rates are based on unit area or
volume of the reactor, without regard to what fraction of that area or
volume is occupied by packing.
It may be noticed that the quantity Gd,,/Np, does not have the di-
mensions of a diffusivity, but of a diffusivity multiplied by a density.
This situation arises from the fact that the amount per unit mass, rather
than the amount per unit volume, is used for a concentration. It is neces-
sary to use the mass of fluid as a basis for concentration, since the mass
but not the volume is conserved in the flow.

b. Definition of Eddy-Diffusion Operator

The equations describing the variation of concentration and tempera-
ture are derived from material and enthalpy balances. The enthalpy bal-
ance is used instead of a heat balance because the enthalpy is a conserved
quantity in the flow. The balances are based on the diffusive flux shown
in Eq. (3-1), which will now be used in the form

J = -E gradc (3-2)
This form for the equation is chosen to take advantage of the vector no-
tation; and to make it possible to indicate a purely transverse flux as
arising from a gradient of concentration that may have a nonzero com-
ponent parallel to the flow. In this equation, J is the superficial diffusive
flux, and the eddy diffusion coefficient E may be thought of either as a
tensor of second rank, or, in Gibbs’ nomenclature, as a dyadic. -E oper-
ates on a gradient of concentration to give a flux in a direction opposite
to the projection of the gradient of concentration on a plane perpendicular
to the flow, with a magnitude equal to the product of Gdp/NPeand the
magnitude of the projection.
In a steady state the balance involves accumulation by convection,
by diffusion, or conduction, and by reaction. The statement that the sum
of these must be zero takes the form

-div(Gc-Egradc-kk,gradc)+A = O (3-3)
216 JOHN BEER

in which A is the rate of production of the entity in question by reaction,

and the term div k, grad c is added to account for diffusive transport by
processes in addition to eddy diffusion. k , is taken as zero for the diffu-
sion of material.
2. Material Balance
All the required material balances can be written in one equation by
putting for c the set of concentrations of key components, designated by
the row matrix y. Corresponding to Eq. (2-8), the reduced matrix of
local production rates may be seen to be RTS,.With these substitutions,
the equation giving the material balances for the key substances is
-div (Gy - E grad y) RTSc = 0 + (3-4)
Only eddy diffusion is involved in the material balance. In the steady
state, the condition
div G = 0 (3-5)
permits some simplification of Eq. (3-4) to the form
-G grady + div ( E grady) + RTS, = 0 (3-6)
3. Enthalpy Balance
a. General Statement
For the enthalpy balance, c becomes H,the enthalpy of unit mass of
fluid, which is naturally expressed in terms of the temperature and com-
position of the fluid. I n this case, the source term becomes zero, corre-
sponding with the conservation of enthalpy in the flow. The heat of reac-
tion enters when the enthalpy balance is transformed into an equation
involving the temperature. The enthalpy balance is then
- + +
- G grad H div ( E grad H ) div (k, grad T) = 0 (3-7)
The term div (k, grad T) accounb for the conduction of heat through the
packing. Since the conduction operates partly in series and partly in
parallel with the eddy transport, the exact relationship with the mass
velocity is complicated, and has not been successfully analyzed. Argo and
Smith (A3) have given the most complete discussion of the problem.

b. Relation to Temperature and Composition

The notation H is now introduced for the matrix of the partial molar
enthalpies of the components of the fluid. Then the enthalpy is expressed
in the form [see Eq. (2-S)]
H = (Yo- y0s-'SB)HT + y+S&HT (3-8)
 DESIGN OF PACKED CATALYTIC REACTORS 217
and the gradient is
+ +
grad H = (Yo- P S - ~ S R JTS-ISR)grad HT (grad ~)s--ISRH~(3-9)
When the partial enthalpies vary significantly with pressure and com-
position as well as with temperature, three terms appear in the working
expression for grad H. In the derivation that follows, the partial enthal-
pies will be assumed to depend only on the temperature, which is equiva-
lent to saying that they are just the molar enthalpies of the separate
components. If there is an important effect of composition or pressure, it
is easy to carry the extra term through the derivation.
With the elements of H depending on the temperature only, their
variation can be expressed in the form
grad H = C, grad T (3-10)
with C, standing for the matrix of molar heat capacities of the compon-
ents. Then it follows that
grad H = (Yo- yOs-'SR + JS-'SR)C,~grad T + (grad ~ ) S - ' S R H (3-11)
~
+
The product (Yo- y0s-lSR Y S - ~ S R )CPTmay be recognized as C,, the
heat capacity of the unit mass of mixture.
Each of the terms in grad H contributes to div ( E grad H ) . On ex-
panding, it is found that
div (E grad H ) = div (ECpgrad 2') +
div [E(grad y)s-lS~H~](3-12)
The last term in this equation may be expanded to give an expression
involving more directly measured quantities.
The expression is
div [$?(grad Y)s-ISRH~]= [div (E grad y)]s-lS~C,~
+
(grad T) * (E grad Y)s-'SRC,* (3-13)
Making use of these relations, the enthalpy balance may be written as
- - +
- C,E grad T - G (grad ~)s-'SRH~ div (EC, grad T )
+
[div ( E grad ~ ) ] S - ' S R H4- -
~ (grad T ) ( E grad Y)s-'SRC,~
+ div (k, grad T) = 0 (3-14)
c. Use of Material Balance to Introduce Rates
This equation may be considerably simplified and made easier to use
by combining it with the material balance. Equation (3-6) is multiplied
by s-lSRHT as a postfactor, and then subtracted from Eq. (3-14). The
result is
-C,G grad T + div (EC, grad T ) + div (k, grad T )
+ (grad T) (E grad Y)S-~SRCJ+ RTQ = 0
* (3-15)
218 JOHN BEER

In writing this equation, Q, the matrix of heats of reaction, has been

substituted for the product -SHT, that is, for the negative of the iso-
thermal enthalpy change for the reaction. It will be noticed that the
third term does not appear in a so-called heat balance, which is formu-
lated to account for the heat accumulated and produced in a region,
using EC, for the eddy thermal conductivity. Such a formulation fails
because heat is not conserved in the flow.
It may be worth while to review the different kinds of multiplicity
involved in the symbols appearing in Eqs. (3-6) and (3-15). Equation
(3-6) is merely a shorthand way of writing the material balance for each
of the key components, each term being a row matrix having as many
elements as there are independent reactions. The equation asserts that
when these matrices are combined as indicated, each element in the result-
ing matrix will be zero. The elements in the first two terms are obtained
by vector differential operation, but the elements are scalars. Equa-
tion (3-15), on the other hand, is a scalar equation, from the point
of view of both vector analysis and matrix algebra, although some of its
terms involve vector operations and matrix products. No account need
be taken of the interrelation of the vectors and matrices in these equa-
tions, but the order of vector differential operators and their operands as
well as of all matrix products must be observed.

d. Equations in Terms of Virtual Rates and Conversions

An alternative formulation of these equations is in terms of the vir-
tual conversions of selected key reactions. From Eqs. (2-8) and (2-12) it
may be seen that
grad J = grad xTs (3-16)
and from Eq. (2-15) that
RTS, = !HTs (3-17)
With these substitutions, Eq. (3-6) becomes

-G . grad x T s + div (E grad xT)s + ?XTs= 0 (3-18)

On dropping the factor s and transposing, the equation in x is obtained.

-G . gradx + div (Egradx) + % = 0 (3-19)

In order to make the corresponding transformation of Eq. (3-15), the
term RTQ must be expressed in terms of 3. The reduced matrix 2is in-
troduced, representing the heats of reaction of the key reactions, being
equal to -&HT. The sequence of substitutions is
 DESIGN OF PACKED CATALYTIC REACTORS 219
RTQ = -RTSHT
= -RTS,s-'S~HT
= %Tss-12
= %T2 (3-20)
After making this substitution and putting xT for ys-l in Eq. (3-15),
the equation in x is
-C,G grad T + div (EC, grad T ) + C,SRT(grad T ) . (E grad x)
+ div (k, grad T ) + QT%= 0 (3-21)
e. Expansions in Terms of Measured Quantities
In using either of Eqs. (3-15) or (3-21), the term div (C$ grad T )
must be evalueted. An expansion of the term gives the expression
+ -
div (C$3 grad T ) = C, div ( E grad T ) (grad C,) ( E grad T) (3-22)
The quantity that requires attention is grad C,. C, can be expressed
-as it was in Eq. (3-ll)-as a linear combination of the heat capacities
of the components. When the gradient is expanded to show its dependence
on composition and temperature, the result is
grad C , = (grad Y)s-~SRC,~
ac T grad T
+ (YO- y0s-1SR + ys-lS~)--& (3-23)

The evaluation of this expression is time consuming, a situation that ex-

plains the fact that the variation of the heat capacity is neglected if it
is not excessively large. It may be remarked that the elements of SRCPT
are the values of AC, for the key reactions, and so that s ~ ( d C , ~ / d T )
involves the variation of AC, with temperature. In many cases both of
these quantities are small enough to be negligible.
In the numerical solution of these equations, the heat capacity of the
fluid must be calculated a t every mesh point if i t is not constant, so that
it may be more convenient to use a finite-difference approximation for
grad C,. The factor ( E grad y)s-lSnCPTappearing in Eq. (3-15)must
be evaluated in any case.
4. Equations for a Tubular Reactor
The flow is practically parallel to the axis of a cylindrical reactor and
the whole system is symmetrical about the axis of the cylinder. Then the
fluxes E grad T and E grad y have only radial components, and the
-
product (grad C,) ( E grad T ) is just the product of the radial derivative
of C, with the magnitude of E grad T. It takes the form

(3-24)
220 JOHN BEER

in which T is the dimensionless radial coordinate, and dC,/dr is evalu-

ated by putting the operator d / d r for grad in Eq. (3-23). The divergence
of E grad T in this case is
aG aT
div (E grad T) = Np,raa[ g
r ar ( r $)+ ar
-g-]
(3-25)

When the relations given above are incorporated into Eq. (3-15), and
the resulting equation is divided by C,G, the working equation is ob-
tained.

(3-26)

*=
The corresponding equation derived from Eq. (3-6) is
A [ i a ( r g ) + ~ - - a1r aGaJ
]+RTSo
(3-27)
az ~ ~ T ar~ r ~ 2

The calculation of the factor (ay/ar)s-'S,CPT may be illustrated by the ex-

ample used in Section 11, the catalytic oxidation of ethylene. The two modes of
oxidation of ethylene are used MI the key reactions, and ethylene oxide and carbon
dioxide are selected as the key components. The reactions and the corresponding
stoichiometric matrices are shown in Section 11. In the range of temperature be-
tween 20O0 and 350°C., the heat capacities of the reactants are nearly enough
linear functions of the temperature, and accordingly the product SRC~',the matrix
of changes of heat capacity for the key reactions, will be a linear function of the
temperature. The matrix is found to be

SRCpT =
+
-3.57 5.44 x 10-ST
3.56 - 5.71 X
with T expressed in degrees K. Then [see Eq.(2-191
+
s-'SRCpT = (i ;)( -3.57 5.44 x 10-ST
3.56 - 5.71 X 10-'T
+
-3.67 5.44 X 1 O - T
1.78 - 2.86 X 10-'T
Finally, this matrix is multiplied by the row matrix ay/ar as a prefactor, giving the
result

9 s-lsBcp~=
ar
*
ar
(-3.67 + 5.44 x 10-3~)
+z
a'a (1.78 - 2.86 X 10-aT) cal./mole deg.
 DESIGN OF PACKED CATALYTIC REACTORS 22 1

If the bed of catalyst has a uniform cross section but is not a cir-
cular cylinder, another dimension enters into consideration. In cases
where a fairly high symmetry is retained, the numerical solution of Eqs.
(3-6) and (3-15) is certainly possible, but just as certainly not easy or
fast. If the symmetry is low, or if the boundaries are complicated, as,
for example, in a large bed through which cooling tubes are passed, the
problem becomes very difficult. The best approach in such a case is prob-
ably to estimate the radius of an equivalent circular cylinder. A rough
guide to such an estimate can be found in two-dimensional potential
theory, but the fact that the rate is not a linear function of composition
and temperature makes it impossibIe to form an accurate estimate in any
general way.
5. General Form of Eddy-Diffusion Operator
If the reactor has a converging or diverging flow pattern, a more gen-
eral expression for E is needed. I n tensor notation, the expression is

(3-28)

This expression is derived from the conditions given above for the ap-
plication of the operator E .
There is some question whether the conditions in a nonadiabatic reac-
tor that does not have high symmetry can be calculated with a reasonable
effort. Even when the pressure is practically constant, the flow pattern
and the reaction interact through the pressure drop in a very complicated
way. The distribution of flow in an unsymmetrical packed bed is hard to
calculate in the simplest case, when the temperature and molecular weight
are constant.
6. Boundary Conditions for the Case of a Tubular Reactor
The conditions a t the entrance to a reactor tube must be specified to
fix one boundary condition. In general, the temperature and composition
would be specified as functions of radial position, but the performance of
the reactor depends very little on anything but the average conditions in
the reactant fluid.
The boundary conditions a t the wall, on the other hand, influence the
performance of the reactor critically, and should be determined as accu-
rately as possible. For the equations of concentration (or conversion) the
condition a t the wall is that the flux of material normal to the wall is
zero, which requires that the directional derivative of concentration nor-
mal to the wall be zero. For the tubular reactor, with cylindrical sym-
metry, the condition is expressed by the equation
222 JOHN BEEK

(3-29)

For the equation giving the temperature, the condition is that the
flux of enthalpy normal to the wall as calculated from the eddy diffu-
sivity and the thermal conductivity match the flux as calculated for
transfer to the wall. The corresponding equation is

I n this equation h, is the coefficient for transfer of heat to the wall

and T, is the temperature of the wall.
REACTOR
C. ADIABATIC
An adiabatic reactor with a regular flow pattern is much simpler to
work with than one with exchange of heat. Except for resistance to ex-
change of material and heat between fluid and catalyst, the space velocity
is the only significant space variable. The natural independent variable
to use is U, the reciprocal of the molar space velocity, or the volume
traversed divided by the total molar flow. The relation between the con-
centration and the reciprocal space velocity is given by the equation

(3-31)

The temperature is fixed by the condition that the enthalpy is constant.

Equation (3-8) provides the required relation.

D. THEONE-DIMENSIONAL
APPROXIMATION
1 . Over-All Heat-Transfer Coefficient
When most of the resistance to heat transfer between the axis of the
tube and the control medium is locaked a t the wall of the tube, -the
variation of conditions in a cross section of the tube becomes relatively
unimportant, and a useful approximation is obtained by using average
conditions in a cross section to represent the whole cross section. The
central problem in using such as approximation is to estimate the rate
of heat transfer through the wall of the tube, given the average tempera-
ture. The heat transfer is determined by the temperature of the fluid next
to the wall, which is between the temperature of the control medium and
the average. The heat-transfer coefficient must be modified by the ratio
of the differences of the wall and the average temperatures from the tem-
perature of the control medium.
A simple approximation for this ratio, based on the assumption that
 DESIGN OF PACKED CATALYTIC REACTORS 223
the radial temperature profile is parabolic, was given by Beek and Singer
(B4).Their result may be expressed in the form

(3-32)
where Nsi = rzht/k,.
The temperature is averaged.over the mass flow, and k, is evaluated
for the conditions close to the wall. Using the ratio given above, the over-
all heat-transfer coefficient U can be calculated. It is
u = h J ( 1 -k 4NBj) (3-33)
leading to the following equation for the average temperature.

(3-34)
The corresponding relation for the concentration is

(3-35)
The simplest way to estimate the average rate is to use the rate at
the average temperature in the cross section. The average rate calculated
in this way is usually too low, because the rate usually varies more
strongly than linearly with the temperature. It is possible to calculate
the average rate in a way that is consistent with the assumption of a
parabolic radial temperature profile, but the improvement obtained in
this way is usually small compared to the error involved in the assump-
tion itself. The best procedure is to use the rate corresponding to the
average temperature, and to restrict the use of the one-dimensional ap-
proximation to cases in which the profile is sufficiently flat. The profile
is increasingly flat when the Biot number is decreased.
2. Limitations
No general criterion can be given, with assurance that the error will
be less than some specified value if the criterion is met. The reaction used
below to illustrate the solution of the equations affords a comparison for
one case. Table I shows that the error in this case is significant when
Nsi = 4.3.
E. CALCULATION
OF THE PRESSURE

I n order to complete the description of the system, the pressure within

the reactor must be known. In the present state of knowledge, it is a
sufficient approximation to consider the pressure in a cross section of the
reactor as constant, and to calculate i t as a function of axial position
224 JOHN BEEH

TABLE I
EFFECT OF BIOTNUMBERON TEE ACCURACY
OF THE ONE-DIMENSIONAL
APPROXIMATION

,
T - To ("C.)
~

NB~ One-dimensional approximation Correct value

4.3 42.4 52.2

3 39.8 46.5
2 37.2 41.3
1 33.5 35.2

only, using an empirical friction factor to estimate the axial pressure

gradient. The equation is

(3-36)

IV. Estimation of Transport Properties

A. I~TTRODUCTORY DISCUSSION
After the differential equations and boundary conditions have been
formulated, there remain the problems of evaluating the various coef-
ficients that appear, and carrying out the numerical solution of the
equations. The following quantities are required.
1. The velocity distribution.
2. The Peclet number for eddy diffusion.
3. The contribution of other modes of transport to the effective
thermal conductivity.
4. The heat-transfer coefficient a t the wall.
5. The friction factor.
It is immediately apparent that all these quantities vary to some
extent with position in the reactor. The Peclet number presumably
changes with the structure of the packing as the wall is approached,
and changes in temperature and composition of the reacting fluid affect
the intensive properties of the fluid that determine rates of transport.
The only justification for ignoring the actual variation is that the un-
certainty about the value of any one of the listed quantities is as large
as the correction that would be made for a change in the properties of
the fluid. A possible exception to this rule is encountered in the case
of a liquid flowing with a Reynolds number in the range below 100. In
this situation, a large change in temperature might, through its effect
 DESIGN OF PACKED CATALYTIC REACTORS 225

on the viscosity and in turn on the Reynolds number, have a significant

effect on the Peclet number. I n such a situation, some improvement in
precision could be obtained by treating the Peclet number as a point
function of conditions in the reactor.
B. VELOCITY PROFILE
Smith and his students (M4, S2, D4) have shown that the velocity
in a packed tube varies strongly with radial position, and have given a
qualitative explanation of the effect, relating it to the larger void fraction
in the part of the bed near the wall. The experimental work that has
been done does not, however, lead to an accurate prediction of the
veIocity profile in a given situation. It appears that the inherent lack
of reproducibility in packing a tube has obscured the results of the most
careful measurements. The kind of difficulty that is encountered is shown
in Table 11,which gives differences between measurements with cylindri-
TABLE I1
DIFFERENCES
IN VELOCITY
PROFILES
IN BEDS
OF CYLINDERS
AND SPHERES,
AT 4 RADIALPOSITIONS

Diameter of particle (in.)

Average velocity
(ft./sec.) 1/4 3/8 1/2

1.01 0.12 -0.03 0.06

0.19 -0.24 0.03
0.11 -0.10 -0.09
-0.24 0.16 -0.17
1.62 0.12 -0.06 0.09
0.20 -0.12 0.08
0.07 -0.12 -0.12
-0.20 -0.06 -0.15
2.14 0.14 0.00 0.13
0.14 -0.22 0.06
0.03 -0.09 0.09
-0.24 0.06 0.15
2.64 0.15 0.02 0.10
0.16 -0.12 0.06 ’

0.02 -0.13 , -0.13

-0.11 -0.03 -0.15

cal and spherical packings at a sequence of radial’positions, taken ,from

Table 3 of the paper by Schwartz and Smith (S2). The quantities given
in the original table are ratios of velocities a t a given radial position
226 JOHN BEEK

to average velocities, all measured a t a distance of 2 in. above the top

of the bed, Table I1 gives the entries for spheres minus the entries for
cylinders having the same diameter, with 3 sizes of particles and 4 flow
rates. The measurements were made in a 4-in. tube, a t distances from
the axis of 0.32, 0.55, 0.70, and 0.88 times the radius of the tube: each
figure given represents the average of 3 runs.
With the measurements subject to fluctuations of 20 or 30%, no ac-
curate description of the profile is possible. All that can be said is that
with moderate ratios of tube to particle diameter, the maximum velocity
is about twice the minimum, and that when the particles are relatively
small, the profile is relatively flat near the axis. It is fairly well es-
tablished that the ratio of the velocity a t a given radial position ta the
average velocity is independent of the average velocity over a wide
range. Another observation that is not so easy to understand is that
the velocity reaches a maximum one or two particle diameters from the
wall. Since the wall does not contribute any more than the packing to
the surface per unit volume in the region within one-half particle di-
ameter from the wall, there is no obvious reason for the velocity t o drop
off farther than some small fraction of a particle diameter from the
wall. I n any case, all the variations that affect heat transfer close to
the wall can be lumped together and accounted for by an effective heat-
transfer coefficient. Material transport close to the wall is not very im-
portant, because the diffusion barrier a t the wall makes the radial vari-
ation of concentration small.
It is desirable to have some systematic representation of the velocity
profile that has the right features, and preferably is easy to use. It is
clear that no polynomial in T of reasonable degree will give the flat
region near the axis that is required for small particles. A convenient
empirical formula for the profile across a slab that would have the right
properties has the form of a constant, the velocity that would be ob-
tained with a very large thickness, plus a hyperbolic cosine function of
the distance from the middle. The analogous formula for a cylinder has
a constant plus the modified Bessel function of the first kind and order
zero, written lo.After imposing the condition that the velocity have
the right average, an expression involving two adjustable parameters
may be written. It is
G- +
1 ~ 1 / 2 d r ~ / [exp
d , (- e ~ ~ / d ~ ) ] i ~ ( e ~ ~ ~ / d ~ )
+
G - 1 2 ~ 1 / 5 Z J i G [exp ( - 6 T 2 / d p ) ] ~ ~ ( & 2 / d p ) (4-1)
in which Il denotes the first-order modified Bessel function and I, and 6
are adjustable parameters. x determines how much higher the velocity
is near the wall than a t the axis, and 6 determines how flat the profile is
 DESIGN OF PACKED CATALYTIC REACTORS 227
in the middle region. This expression is derived from the assumptions
that the limiting value of the ratio of velocity near the wall to the
velocity a t the axis as the radius increases is a fixed constant, equal to
+
1 A; and that the limiting shape of the profile near the wall is
invariant to particle diameter if the distance from the wall is measured
in particle diameters. Figure 1 shows curves calculated from this empiri-
cal expression with A = 1.5, 8 = 1, and two values of r z / d p : 12.8 and 4,
together with corresponding sets of experimental results from the meas-
urements of Schwartz and Smith (52). The comparison shows that the
expression of Eq. (4-1) does not represent these results a t all well. At
the same time, it is true that no expression giving a reasonable variation

0 0.2 .0.4 0.6 0.8 1.o

r2
FIQ.1. Representation of velocity profile.

of the profile with conditions could give a good representation of the

figures shown in Table 11. What is needed is an experimental study
comprehensive enough to fix the average behavior of the system on a
statistical basis, with all significant systematic errors eliminated or cor-
rected for. A qualitative representation of the trends is all that can
be attained on the basis of the experiments that have been made.
DIFFUSIVITY
C. EDDY
Bernard and Wilhelm (B6) showed that the radial diffusivity in a
packed tube is proportional to mass velocity and particle diameter when
the Reynolds number is above 200, and that the diffusivity in this range
is so large that molecular diffusion could make no significant contribu-
228 JOHN BEEH

tion to it. They inferred from these results that the diffusion in this
range of Reynolds number could be related usefully to a Peclet number
equal to about 10, with the diameter of the particles serving for the
characteristic length. Their results for the diffusion of GOz in air at
low Reynolds number showed the expected dependence of $he Peclet
number on Reynolds number, the Peclet number decreasing in a region
where the effect of molecular diffusion is expected to appear. The dif-
fusion of dye in water, however, showed an unexplained increase a t
very low flow rates. The experiments with water were repeated by
Latinen (Ll), who found that if precautions were taken to insure that
no gas accumulated in the bed, the Peclet number continued to increase
as the flow rate was decreased, reaching a value of about 80 a t a Reynolds
number of 7. Several other workers have made measurementa of radial

z
d

diffusivity in packed beds, the results all being in agreement that the
radial average of the Peclet number is close to 10 when the Reynolds
number is above 100, and that i t increases as the Reynolds number is
decreased below this value. Latinen’s results may be taken as representa-
tive; Fig. 4-2 is a graph of his results, as smoothed by him, but adjusted
by a factor to make the Peclet number approach 10 a t high Reynolds
number.
It must be kept in mind that these measurements were made with
water, in which the molecular diffusivity is negligibly small. When the
fluid is a gas, with a Schmidt number having the order of magnitude of
unity, the Peclet number for molecular diffusion, being the product of
Schmidt number and Reynolds number, has the order of magnitude of
the Reynolds number. Even if the effect of molecular diffusion is multi-
plied by the fraction void and by another factor of 5, to take care of
 DESIGN OF PACKED CATALPTIC REACTORS 229
the tortuosity of the diffusion path, the contribution becomes con-
siderable when the Reynolds number is less than 20. The descriptions
of heat and mass transfer assembled in this article should not be used
for a Reynolds number less than 40.

D. OTHERCONTRIBUTIONS
TO THE EFFECTIVE
THERMAL
CONDUCTIVITY
1. Discussion by Argo and Smith
A detailed discussion of the transport of heat in packed beds was
given by Arga and Smith (A3). These authors recognized that no com-
plete theory can be constructed on the basis of what is known now, but
they attempted to take account of all the processes that would be ex-
pected to take part in the transport. The expression they derived was
necessarily based on a much simplified description of the packed bed,
but presumably has the right kind of dependence on the various quanti-
ties that appear. In the notation of this chapter, i t is

The terms in order represent the effects of molecular conduction, of

eddy conduction, of radiation through the void space, and of conduction
through the solid. The quantity h’ that appears includes all paths for
the transport from one particle to the next, being composed of t e d s
for radiation, point contacts, and convective transfer to and from the
fluid. It is remarkable that this expression contains no adjustable
parameters, being based completely on the theoretical analysis and on
correlations established for special conditions.
Most of the measurements in this field have been made under con-
ditions such that only the terms for eddy transport and for conduction
through the solid need be considered. If it is assumed that, by analogy
with mass transport, the modified Peclet number increases when the
Reynolds number is decreased below 100, all the evidence (Ala, B9, C2,
M1, Q1) indicates that the conduction through the solid is a t least as
important as the eddy transport when the Reynolds number is less than
50, even with such poorly conducting solids as glass. This feature of
the experimental results is well reproduced by the Argo-Smith theory.
It is in the region of very high conductivity that the theory fails.
The inclusive quantity h’ is a t least as large as the convective heat-
transfer coefficient, so that a lower bound for the conduction term is
given by calculating it using h instead of h’. A good illustration is af-
forded by the fact that the term calculated in this way for a bed of
steel spheres with air flowing through it a t a Reynolds number of 500
230 JOHN BEER

is 0.31 times the term for eddy transport, assuming a Peclet number
of 10 for the eddy transport. This is to be compared with a calculated
ratio of 0.10 for alumina, with a conductivity of 5 x cal./cm. sec.
deg. (S3). The experiments of Singer and Wilhelm (54) and of Kwong
and Smith (K3) show that the effective conductivity is not significantly
changed by the substitution of steel for ceramic or glass packing in this
range of Reynolds number, although there is a substantial increase in
the range below 100. Chu and Storrow (C3) observed a significant in-
crease in the over-all heat-transfer coefficient a t high Reynolds numbers
but this is not certainly an effect of larger effective thermal con-
ductivity. Thus the term accounting for conduction through the solid,
which is needed to explain the effective conductivity a t low Reynolds
numbers, leads to too great a sensitivity to high values of the solid con-
ductivity at high Reynolds numbers.
2. Suggested Simplification
Even if it is supposed that the expression as given is a good approxi-
mation, and that the experiments somehow failed to indicate the large
effect, it must be concluded that the experimental basis for the calculation
of the effective conductivity is uncertain. Until some theory is developed
that represents the experimental results better, or until the Argo-Smith
theory is confirmed by experiment, i t appears that it would be useful to
have a simplified expression, containing parameters that can be adjusted
to fit the experimental results. A suitable expression is
0.6hdpk,
k, =
2k* + 0.7hdp (43)
which is exactly analogous to the last term of Eq. (4-21, but with the
smaller h substituted for h’, with the factor 0.6 instead of 1 - B, and
with the factor 0.7 introduced in a term in the denominator to reduce
the effect of a large increase in the conductivity of the solid when the
Reynolds number is high. Compared with the last term of Eq. (4-2),
this empirical expression gives a somewhat poorer correspondence with
most of the experiments a t low Reynolds numbers but 8 decidedly im-
proved correspondence for Reynolds numbers of 200 or higher. The effect
of void fraction is ignored, first, because its effect has not been measured,
and second, because the fraction of the volume occupied by the solid
has opposing effects, by way of the area for conduction and by way
of the path of conduction per particle.
3. Radiative Transport
There is no experimental basis for including the factor c in the term
for radiative conduction. Polack (Pl)concluded from his measurements
 DESIGN OF PACKED CATALYTIC REACTOR6 231
that Diimkohler’s estimate of 0.5 for the coefficient (D2)is acceptable.
Although the effect of radiation must ultimately increase as the fraction
void is increased, the variation in the usual range encountered is pre-
sumably small and may be neglected.
4. Prqosed Expression for Efective Conductivity
The effect of molecular conduction is negligible when the Reynolds
number is greater than 40. These considerations lead to the following
proposed working expression for the effective thermal conductivity in a
packed bed.

This expression is compared with the authors’ correlations of their ex-

perimental results in Fig. 3, which shows the reciprocal of the apparent
Peclet number.
The decision to maintain an upper limit of 10 on the apparent Peclet
number for heat conduction a t high Reynolds numbers is justified by

FIQ.3. Radial thermal conductivity. Broken lines are calculated for k. = 2 x lo-“
and 3 x lo-’ cal./cm. sec. deg. Curves: (1) Aerov and Umnik (All; (2) Bunnell
and others (B9) ; (3) Campbell and Huntington (C2) ; (4) Coberly and Marshall
(C4); (5) Maeda ( M l ) ; (6) Quinton and Storrow (Ql).
232 JOHN BEEK

the results of the measurements on material diffusion. The effective dif-

fusivity can be measured much more easily than the thermal conduc-
tivity, and the evidence for the eddy character of the diffusion is over-
whelming. The conclusion that a Peclet number of 10 describes material
diffusion a t high Reynolds numbers leads directly to a limiting value of
10 for thermal conduction, since enthalpy must be transported a t least
as effectively as matter.
There is room for substantial improvement in our knowledge of
thermal conduction in packed tubes. At present, the effect of conduction
through the solid is quite uncertain, and this effect must have some im-
portance a t low Reynolds numbers. I n view of the very large experi-
mental effort that has gone into this field, it does not seem to be an in-
viting field to enter, but the serious questions that remain are worthy
of attention,
One remark should be added to this discussion. The eddy-conductive
term in Ice, containing the heat capacity of the fluid as a factor, is ade-
quate as long as there is no reaction. When a reaction is going on, the
very concept of thermal conduction loses its precision, and for an ac-
curate description of the system the equations of Section 111must be used.
E. HEAT-TRANSFER COEFFICIENT AT THE WALL
There is even more uncertainty in estimating the heat-transfer coef-
ficient a t the wall of the tube than in estimating the effective thermal
conductivity in the bed of catalyst. The measurement is essentially a
difficult one, depending either on an extrapolation of a temperature pro-
file t o the wall or on determining the resistance at the wall as the dif-
ference between a measured over-all resistance and a calculated re-
sistance within the packed bed. The proper exponent to use on the flow
rate to get the variation of the coefficient has been reported as 0.33 (C4),
0.47 (C2), 0.5 and 0.77 (Hl), 0.75 (A2) , and 1.00 (Ql).
In the face of these discrepancies, any choice of a formula for calcu-
lating h, must be to some extent arbitrary. It seems reasonable to depend
on the close similarity between the processes involved in transfer to the
wall and to the packing. Thoenes and Kremers (Tl) have presented a
thorough discussion of mass transfer from a particle of packing to the
fluid, which can be translated into a treatment of heat transfer by sub-
stituting Nusselt and Prandtl numbers for Sherwood and Schmidt num-
bers. These authors suggest using three terms to represent the contribu-
tions associated with conduction in laminar flow, with eddy conduction,
and with stagnant conduction. The heat-transfer form of the expression
they recommend for flow through a body-centered cubic lattice of
spheres normal to the (100) plane is
 bESIGN OF PACKED CATALYTIC REACTORS 233

+ +
NN,, = 2.42Nh5$Nfi55 0.129Nfi,0.8Np,0.4 1 . 4 N ~ o O . ~ (4-5)
This expression gives a rather good representation of the broken line
recommended by Hougen and Watson [see (H6), p, 9871 for random
packing.
While it is desirable to take account of all effects in a description of
the transport of heat, a complete description must await a deeper analysis

Fro. 4. Heat-transfer coefficient to wall. Broken line is N Naccording

~ to Thoenes
and Kramera (TI). Curves: (1) Aerov and Umnik (A2); (2) Coberly and Marshall
(C4); (3) Hanratty (cylinders) (111); (4) Hanratty (spheres) (Hl); ( 6 ) Quinton
and Storrow (Ql); (6) Yagi and Wakao (Y2).

or a larger experimental effort. It is not entirely satisfactory to show

the laminar, turbulent, and stagnant conductivities as being simply ad-
ditive. Nevertheless, the success of the expression given by Thoenes and
Kramers in representing a smooth relation between the Nusselt and
Reynolds numbers shows that the form of the expression is useful, a t
least as an empirical formula.
Figure 4 shows some experimental measurements of h, in comparison
234 JOHN BEEK

with the Thoenes-Kramers formula. As might be expected from the dif-

ferent conditions, the transfer coefficient for the packing is larger than
for the wall. A reasonable approximation can be obtained by taking the
coefficient for the wall as a constant multiple of the coefficient for the
packing. A further approximation, involving dropping the term for
stagnant conduction, is admissible if the Reynolds number is restricted to
be greater than 40. The simplified expression for the Nusselt number is
+
NN,,= 3.22N~,$$Np,4~0.117N~o.8Npr0.4 (4-6)
and, with a factor of 0.8 to account for the difference between transfer
to a particle and to the wall, the heat-transfer coefficient a t the wall be-
comes
+
h, = (/If/&) (2.58N&3$Npr4' 0.094N~~~*~Np,O~~) (4-7)
Hanratty (Hl) has pointed out that there is some evidence that
heat transfer to the wall is different with spherical and cylindrical
particles. The experiments indicated that the less obstructed flow near
the wall in a bed of spheres made the transport to the wall more like
what it is in an empty tube, leading to an exponent of 0.77 on the
Reynolds number. Hanratty recommends for air flowing among spheres
the correlation
h, = (?~f/d,)(0.25N&'.~) (4-8)
This correlation can be well represented in the range of Reynolds
numbers covered by the experiments by an expression of the Thoenes-
Kramers form. It is
h, = (/If/&) (0.203N~36Npr3* -k O.~~ONR~.*NP,O.~)(4-9)
Until further experimental work has given some reason to change, it
would be desirable to use Eq. (4-7) for particles that pack next to
the wall like cylinders, and Eq. (4-9) for particles that pack next t o
the wall like spheres.
F. FRICTION
FACTOR
The review published by Ergun (E2) provides a definitive descrip-
tion of pressure drop in packed tubes when the ratio of particle diameter
to tube diameter is sufficiently low. In addition, although the complicated
relationship between the diameter ratio, the fraction void and the friction
factor can not be accurately represented without some explicit dependence
of the friction factor on the diameter ratio, Ergun showed that his
correlation does work for a wide variety of experimental conditions. The
friction factor is calculated from the expression

f =7
1-e
(1.75 + 150
 DESIGN OF PACKED CATALYTIC REACTORS 235
with the Reynolds number based on the diameter of spherical packing
that gives the same superficial area per unit volume of bed as does the
actual packing.
In using this friction factor for a tubular reactor, the Reynolds num-
ber is evaluated at some estimated average condition, and then the
corresponding friction factor is used for the whole bed. In calculating
the axial pressure profile, the average composition and temperature in
a cross section are used to estimate the density of the fluid, and this
density is used with the average superficial mass velocity to estimate
the axial derivative of pressure.

V. Numerical Solution of Equations

A. AUTOMATIC COMPUTER ASSUMED
It will be assumed in discussing the solution of the differential
equations relating conditions within a tubular reactor that the calcula-
tions are to be carried out with some sort of automatic digital computer.
The problem is just manageable with a desk computing machine in
the one-dimensional approximation when there is only one independent
reaction, but even in this simplest case, this is an expensive way to
work, particularly if a fairly large number of solutions is required.

B. ONE-DIMENSIONAL APPROXIMATION
If the one-dimensional approximation is adequate, the problem is
reduced to a routine integration of a set of ordinary differential equations.
Procedures for integrating such sets of equations are given in standard
works on numerical analysis (see, for example, H5, M2, and M3). The
working equations for simple forward-diff erence equations are given in
the next paragraph.
Equations (3-34), (3-35), and (3-36) may be writen in the form

% = Fi(+l, &, . . . (5-1)

dz
in which the are the conversions, the temperature, and the pressure,
and j is the number of independent reactions.
Then the Fa are

R = fGZ/PdP (5-2)

F - %Tq--2u (7' - TJ (5-3)

- C,G r2CPG
Fi = R/?G (2 < i < j + 3) (5-4)
236 JOHN BEER

It must be remembered that heat capacity, heat of reaction, density,

and the rates are, in general, variables in these equations, being known
functions of the dependent variables.
When the cooling medium is a boiling liquid, the dependence of its
temperature on the length of the tube must be calculated from the con-
ditions in the pool of liquid. When it is a circulating liquid, flowing
concurrently with the reacting fluid, its temperature can be calculated
from a heat balance. This heat balance is expressed by the equation

W,C, 2 = U(2w3)(T - T,) (5-5)

in which W , is the mass rate of flow of the cooling medium per reactor
tube and C, is its specific heat capacity. With the definition
w = 21772u/wccc (5-6)
the derivative of To can be written as
Fj+s =IW(T - Tc) (5-7)
The working equation, based on the form of Eq. (5-1), is
h,n+l = &,n f kFi(+ln, hn, .- t h+8,n) (5-8)
where k is written for the increment in x. As the truncation error in
this expression is of the order of k2,so that the error in the solution is
of the order of k , i t is advantageous to calculate the solution with two
different values of k , and to extrapolate linearly to a zero value for k.
Methods having smaller truncation errors are given in the references
cited above.

OF ERROR
C. SOURCES
1. Introduction
When the radial variation of temperature must be taken into ac-
count, the problem assumes an entirely different character. Each of the
equations is now a partial differential equation, and both radial and
axial profiles must be calculated; a mesh or network of radial and axial
lines is set up, and the temperature and composition are calculated for
each intersection. A great deal of work has been done on the formulation
of difference equations for solving the related diffusion or heat-conduction
equations; most of this has been directed towards the case in which
there is only one dependent variable and in which the source is a linear
function of that variable. Although the results obtained for one dependent
variable are only partially applicable to the multiple-variable problem,
 DESIGN OF PACKED CATALYTIC REACTORS 237

some discussion of the way the formulation of the difference equations

affects the accuracy and stability of solutions is in order here.
The analysis of a difference equation is greatly simplified if it is
thought of as an approximation to the corresponding differential equation
a t a specified point, which will usually be a mesh point. From this point
of view, the approximation to each term arises naturally from con-
siderations of symmetry or convenience, and just as naturally the error
in the approximation can be related to the Taylor expansion of the
term in question about the reference point.
2. Truncation Error
The accuracy of a solution is affected by two properties of the dif-
ference equations and by the rounding error involved in the calculation.
The first property of the difference equations is the truncation error,
that is, the discrepancy between the differential equation written for
some point and the difference equation that is supposed to correspond
to it. The second property is the order of the difference equation with
respect to the axial step. These effects can be illustrated by the way
they appear in the solution of a first order ordinary differential equation.
Suppose the equation
y' + by = 0 (5-9)
with b positive, is to be solved. Let k be the step length. Then a forward-
difference approximation to the derivative a t z = nk is
yn' (Y.+I - yn)/k (5-10)
which leads to the difference equation
- Yn + bkyn = 0 (5-1 1)
This has the solution
yn = y0(1 - bk)" (5-12)
which may be expanded in the form
yn = yoe-bkn(l - ?jb2k2n + (5-13)
As the solution of the differential equation is
y = yoe-bE (5-14)
and kn is the same as z , the term -Sb2k2n gives a measure of the
fractional error in the approximate solution, when there is no rounding
error. This error is proportional to k when z is fixed. At the same time,
it may be seen that there is no tendency for a systematic growth of the
rounding error.
238 JOHN BEEK

3. Satellite Solutions
An alternative approximation to the derivative is the central-dif-
ference form
Y2 e!(Ye1 - Yn-l)/2k (5-15)
which gives the second-order difference equation
yn+l+ 2bkyn - yn-1 = 0 (5-16)
Assuming that an estimate of y1 is somehow available, the solution of
this difference equation is [see (H4), p. 2381.
gn = B(-bk + C F E F ) n + C(-bk - dl + bek2)" (5-17)
where
(5-18)

and
(5-19)

The first thing to be observed about Eq. (5-17) is that the second
quantity in parentheses is both negative and greater in magnitude than
unity. As a result of this situation, the successive powers of this quantity
will oscillate increasingly and without limit as n is increased. It is clear,
then, that the approximate solution will not even resemble the solution
of the differential equation unless the coefficient C is made sdiciently
small by a proper choice of yl. It does not do any good to make the
step small, because the quantity multiplied by C is (-l)"ebs in the first
approximation, and so nearly independent of Ic.
The spurious or satellite term in the solution is introduced by using
a second-order difference equation to approximate a first-order dif -
ferential equation. An extra condition is needed to fix the solution of
the second-order equation, and this condition must be that the coef-
ficient of the spurious part of the solution is zero. I n the general case of
a nonlinear difference equation, no method is available for meeting this
condition exactly.
If the coefficient C is exactly zero, B becomes yo. With this value of
B, the solution of Eq. (5-17) may be expanded to the form
+ +balcsn+
yn = yoe-akn(l * *) (5-20)
I n this result, the fractional error a t a given value of z, that is, of kn,
is proportional in the first approximation to ka, instead of to k, as in
Eq. (5-13). This means that the approximation to the derivative given
 DESIGN OF PACKED CATALYTIC REACTORS 239
by Eq. (5-15) is in this sense better than that of Eq. (5-10). Before
discussing the difference between these forms in terms of the truncation
error, the effect of rounding errors on the numerical solution of Eq.
(5-16) must be considered.
It was seen that if y1 is not chosen to make C practically zero, the
solution of Eq. (5-16) oscillates widely. The same requirements apply
to the subsequent steps: thus, if a significant error is introduced a t any
step by rounding, that error will be magnified as the solution proceeds.
For this reason, a high accuracy in the individual numerical operations
is required when a difference equation that suffers from this sort of
instability is used. I n using this central-difference approximation with
nonlinear equations, however, the problem of getting started with a
proper value of y1 is usually more serious than the problem of controlling
roundoff errors.
4. Estimation of Order of Error
The approximations of both Eqs. (5-10) and (5-15) are for the
derivative ,of y a t z = nlc. The error involved in the approximations may
be estimated by comparing the value of y n f l calculated from the cor-
responding difference equation with the exact value. The forward-dif-
ference formula gives
+
yn+l S Yn kyI: (5-21)
which is to be compared with the Taylor expansion
yn+1 = yn + kyn' + Wy," + O(kS) (5-22)
In this case, then, the truncation error in the expression 'for yn+l is
dominated by a term proportional to k2. The central-difference formula
gives
yn+t= y*1+ 2kyI: (5-23)
instead of the complete expression
yn+t = yn-1 +2kyd + +
Qk3ynl'l O(k6) (5-24)
from which it may be seen that the truncation error is of the order of k3.
These truncation errors in the stepwise calculation lead plausibly to
the magnitude of the errors a t the nth step that are indicated in Eqs.
(5-13) and (5-20). The accumulation a t each step of an error of a cer-
tain magnitude gives after n steps an error of n times the magnitude.
Since the number of steps to reach a given length is inversely pro-
portional to the size of the step, i t may be expected that the order of
the truncation error in the solution will be one less than the order of
the truncation error for one step. Accordingly, the second-order trun-
240 JOHN BEEK

cation error in the forward-difference expression gives an error of the

order of Ic in the solution, while the third-order truncation error in the
central-difference formula gives an error of the order of k2 in the solution.
D. STABILITY IN PARTIAL DIFFERENCE EQUATIONS
Each of the kinds of error exhibited in the numerical solution of
the ordinary differential equation can appear in the solution of the partial
difference equations used to solve a partial differential equation. In ad-
dition, a different kind of instability can appear, arising from a different
kind of spurious or satellite solution. A good discussion of this kind of
instability is given in a paper by O’Brien et uE. (01); a more rigorous
treatment is given by Lowan (L2).The essential point is that the step
from one profile to the next is made by calculating each value on the
new profile as a linear combination of the values on old profiles, together
with any source term involved. Equation (5-35) given below is an ex-
ample of the nature of the step. An equivalent representation of a step,
then, is to say that the set of values on the old profile is multiplied by
a square matrix to get the set of values on the new profile. If the set
of values on a profile is thought of as a linear combination of charac-
teriatic vectors of the matrix, it may be seen that one contribution to
the solution after a large number of steps is the same linear combination
of characteristic vectors, each now multiplied by a large power of the
corresponding characteristic value. From this observation it may be
inferred that if any characteristic value of the matrix of coefficients in
the partial difference equation is greater in magnitude than unity, the
solution cannot be stable.
The classical example of the instability that results is the difference
formulation given by Richardson (R2)for the heat-conduction equation.
He proposed using a central-difference formula [as in Eq. (5-15)] for
the derivative with respect to time, together with the usual central-
difference formula for the space derivative. The resulting equation has
a truncation error of the third order in both time and space steps, but
the solution is unstable for any length of step. Thus, this natural and
accurate formulation is not available for use if more than a few steps are
to be taken.
E. INTRODUCTION OF 13 AS RADIAL VARIABLE
In working with a tubular reactor, or with some other form that can
be well enough represented by an equivalent tubular reactor, a significant
advantage in accuracy is gained by taking account of the symmetry
about the axis. Since all the properties in the system must be even
functions of radial position, the square of the radial coordinate is
 DESIGN OF PACKED CATALYTIC REACTORS 241
preferable to the first power as one of the independent variables. If the
substitution u = 1.2 is made, the operators ( l / r ) ( d / d r ) r and (a/&)
change to 2 ( d / d u ) u % and 2u34 ( d / d u ) . This means that in Eqs. (3-26)
and (3-27), quantities like ( l / r )(a/&) [r(dT/dr)] become 4(d/du)
[ u ( d T / d u ) ] ,and quantities like (dG/dr) ( d T / d r ) become 4u(dG/du)
(dT/du).

F. EXPLICIT
PARTIAL
DIFFERENCE
EQUATIONS

1. General Discussion
There are two practical approaches in formulating the working dif-
ference equations for the packed tubular reactor. The simple one is
to use a forward-difference equation for the axial derivative and central-
difference formulas for the radial derivatives. The leading terms in the
truncation error are then proportional to k2 and to kh2, where h and k
are written for the radial and axial steps. This means that, in order to
take advantage of the accuracy of the approximations for the radial
derivatives, Ic must have the same order of magnitude as h2, so that
k2 and kh2 will be comparable. This is a serious limitation on the length
of the axial step that can be used.
2. Condition for Stability
There is another limitation that has nearly the same effect. If the
equation is written
%
! = wu -
a24 + (other terms) (5-25)
az au2

so that the corresponding difference equation has the form

- +m.n - wu 4m-l.n - 24m.n + 4 m + l . n + (other terms) (5-26)
Ic h2
the solution may be unstable if
k > h2/2w (5-27)
In Eq. (5-26), the first index is used for the radial position, the second
for the axial position. The limitation of the inequality (5-27) does not
restrict the order of magnitude of k, but it does put a quantitative limit
on its value.
When more than one dependent variable appears, with the variables
linked through source terms, that is, through the rates, the limit on
the axial step may be greater or less. The conditions imposed a t the
boundary can also have some effect on the stability. All that can be
242 JOHN BEEK

said in the more general case is that the inequality (5-27) provides a
rough rule, but that somewhat shorter steps may be required.
Two considerations regarding truncation error that enter into the
derivation of the partial difference equations should be pointed out. I n
some published formulations of these equations, the first radial derivative
has been approximated by a forward-difference expression (Kl, S5, W1) .
This unsymmetrical formula has no advantage over a symmetrical or
central-difference expression, but has a greater-lower order-truncation
error. The central-difference approximation

is used in the equations given below.

When equal increments in r are used, the symmetry about the axis
can be depended on to derive an accurate expression for the required
radial derivative a t the axis. When, as in the present formulation, equal
increments in r2 are used, the first increment in r is so large that the
analogous formula gives too great a truncation error. If r2 or u is con-
sidered as the radial variable, no useful additional property of symmetry
about the axis is available. In order to keep the truncation error small,
a three-point forward-diff erence approximation is used for the radial
derivative a t the axis. This accounts for the appearance of +2,,, in the
formula for +O,nS1.
3. Boundary Condition a t the W a l l
A convenient way to take care of the boundary condition a t the
wall is to introduce a so-called image point, a virtual axial profile one
increment outside the wall. The value of the variable on this profile
is found from the boundary condition, and then used just as though it
were a t a regular mesh point in calculating the value at the wall for the
next step.
4. Formulation of Equations
For a reason that will become apparent when the other type of
difference equation is given, the terms that are not linear in the de-
pendent variables are separated out for special treatment. With this dis-
tinction in mind, Eqs. (3-26) and (3-27) are written schematically in
the form
(5-29)

with the definitions

w = 4d,/Np,r~~ (5-30)
 DESIGN OF PACKED CATALYTIC REACTORS 243
and
4dp aG
' = Nper22G
(5-31)

o and 9 as thus defined are independent of x . When the variable is con-

version, the term S is just the virtual rate divided by the molar mass
velocity. When the variable is temperature, S includes the terms as-
sociated with the variation of heat capacity. The relations are
+ =x: S = %/G (5-32)

With these definitions, the difference equations are the following:

+ 2h-Ic [ w m ( 2 m + 1 ) + m h ~ m ] + m + l , n + k S m , n (0 < m <M) (5-35)

+M,n+l = 2 k M 2 W ~ ~ ~ - 1 , n

+ (1 - 2 k M 2 w ~- kNsi [ w M ( 2 M + 1) -k 'MI) @M,n

+ 2[ w M ( 2 M + 1 ) f v ~ ] T c , n
k N Bi
kSM,n (5-36)
I n these equations, the index M is used for the points a t the wall. Since
u, the square of the dimensionless radius, varies from zero to unity, M
is just the reciprocal of h.
In practice, this set of equations is written with $C representing each
conversion in turn, and finally the temperature. When 4 represents a
conversion, the value of N B i is taken to be zero; when it is temperature,
N B i takes the value appropriate to the conditions next to the wall.
In conjunction with these equations, the temperature of the cooling
medium and the pressure may be followed with the simple forward-dif-
ference equations, neglecting any radial variation of pressure. The equa-
tions are
TC,n-+l = Tc,n +
IcW(TM,n - Tc.n) (5-37)
and
+
I .

Pn+l = P n kjGn2Xn2/pndp (5-38)

244 J O H N BEER

I n Eq. (5-37), W is defined in terms of the heat-transfer coefficient

across the wall instead of an over-all coefficient to the average tempera-
ture in the cross section of the tube. The definition is
w = 2?rrzhJWcCc (5-39)
The derivatives appearing in S can be approximated by finite dif-
ferences on the nth profile. The quantity k, can be treated as a constant,
since the uncertainty in its determination is so great that any refinement
would be wasted.

G. IMPLICIT DIFFERENCE
PARTIAL EQUATIONS
1. General Discussion
An alternative approach to the formulation of the difference equa-
tions is to sacrifice some convenience in solving the equations for the
advantage of being able to take longer steps. It has been found that a
central-difference approximation to the axial derivative can be used if
the radial derivatives are approximated by suitable averages of dif-
ferences calculated on the new profile and on one or more old profiles.
+
To fix the thoughts, suppose that the ( n 1)th profile is to be calculated.
Then a suitable average to use for the radial derivative would have equal
+
weights on the (n 1) th and the (n - 1)th profiles, and the rest on
the nth profile. This arrangement is symmetrical about the nth profile,
where the differential equation is being approximated, so that the trun-
+
cation error is small. If the weights on the ( n 1)th and ( n - 1)th
profiles are larger than 4, that is, if the weight on the nth profile is
less than Q, the solution of the difference equation is stable for any
length of step (01).
A convenient choice of weight is 9 for the new profile, giving a weight
of zero on the nth profile. I n the general case, the approximation to the
second radiaI derivative a t the point m,n is

+ (1 - 24(4m-l,n - 24m,n + 4ny-1,n)

+ ~(4m-1,ni-l - 24m,n+1 + &n+1,n-t1)] (5-40)

and the approximation to the first derivative takes a similar form. With
expressions like these for the radial derivatives, and with a central-dif-
ference approximation to the axial derivative, the difference equation
corresponding to Eq. (5-29) when w = 0.5 is
 DESIGN OF PACKED CATALYTIC REACTORS 245

+m,n+l - +m,n-l
2k
= mum (+m-l,n-I - 2+m,n-1 + +m+l,n-I ++m-~,n+l - 2h,n+1 + +m+l,n+l)

+ + mhltrn) 1
(urn (+rn+l,n-l - 6m-1.n-1 + +m+l,n+l - +m-~,n+J + Sm,n

(5-41)
At this point it may be seen why some terms in addition to those
involving reaction rates are grouped in S. The coefficients o and 7 may
depend on the radial position, but they can be treated as independent
of temperature and composition, and accordingly they can be assigned
values a t each mesh point. The terms appearing in S, on the other hand,
have coefficients that are affected by the dependent variables; since the
unknown values of the dependent variables at three mesh points in the
new profile appear in the difference equation, it is important for them to
have known coefficients, so that the set of simultaneous equations can
be solved in a straightforward way.
2. Boundary Conditions
The boundary conditions are used in setting up equations for the
values a t the axis and the wall much as for the forward-difference
equations. The principal change is that, in order to control the truncation
error, the first difference equation represents the differential equation
localized on the nth profile a distance (a) h from the axis, instead of on
the axis. The difference between the temperature of the reacting fluid
and of the cooling medium is taken as the average at the ( n- 1)th and
+
(n 1)th profile, and similarly the density and molecular weight used
to calculate the change in pressure are obtained as averages for the
profiles.
3. Formulation of Equations
The working equations are
246 J O H N BEEK

4. Pressure
Equation (5-48) is written for the case of a gaseous reactant, in
which P / p is rather insensitive t o P. This form of the equation for prw
sure drop is used, rather than one analogous to Eq. (5-38), because the
long steps that can be taken with this set of difference equations demand
an equation with a small truncation error. At the same time, the re-
quirement for stability makes it impossible to use the conditions on the
nth profile to fix the difference in pressure, so that an average on the
+
(n- 1) th and ( n 1)th profiles must be used. Equation (5-46) is an
explicit equation for Pn+12 if Pn+l/pn+lcan be estimated without using
 DESIGN OF PACKED CATALYTIC REACTORS 247
an accurate value of Pn+l.If the reactant is a liquid, the appropriate
equation is

5. Stability
Although the solution of these equations is stable in the technical
sense that the fluctuations do not grow without limit, it must be observed
that there is a serious practical limitation on their use when the source
term is small, and that they are useless when the source term is ex-
tremely small. The limitation is associated with the appearance of
fluctuations of limited magnitude, which may be large enough to obscure
the results. The common situation in a heat-exchanger type of reactor
is that the rate of heat production and of heat transfer are fairly well
balanced, in which case the solutions are smooth. In the extreme case of
the heat-conduction equation without source, however, the fluctuations
quickly develop to the same magnitude as the desired solution. Jim
Douglas (D5)has devised a cure for this difficulty. His scheme is to
calculate the radial derivatives with weights Q, -Q, $, and 4 on the
+
( n - 2) th, ( n - 1) th, nth, and (n 1) th profiles, respectively. I n spite of
the fact that more profiles must be stored and used in the calculation,
this formulation is better than the forward-diff erence or explicit scheme
when the storage space is available.
6. Technique for Solving the Simultaneous Equations
Equations (5-42) to (5-44) constitute a set of simultaneous linear
equations in the unknown values of the quantity on the new profile, each
of the sets for temperature and the conversions being independent. The
matrix of coefficients is the same for each conversion, and differs for
the temperature only in one element, the one containing the heat-transfer
coefficient. An important property of each matrix of coefficients is that
it is independent of the axial position, so that for the purpose of the
calculation it is a constant matrix. Another important property is that
only three diagonals of the matrix contain elements that are not zero.
These properties make it possible to throw the calculation of the un-
known quantities into a very simple form. The essential feature of the
calculation is that coefficients in two two-term recursion formulas are
constructed from the matrix elements, and then these coefficients are
used to calculate, first, a set of ancillary quantities, and then the desired
quantities. The procedure is exactly what would be used in eliminating
the unknown quantities successively from the equations and then sub-
 248 JOHN BEEK

stituting the quantities that have been calculated back into the equations
in order to calculate the others.
Equations (5-42)to (5-44)are transformed by dividing each one by
the magnitude of the first coefficient appearing on the left-hand side.
After this transformation, they can be written schematically in the form
(Po - bo41 = co (5-48)
-dm-1+ %(Pm - bm(Pm+l = cm (5-49)
-(PM-l+ UM4M = C M (550)
+
In these equations, the second subscript, n 1, has been dropped; all the
terms on the right-hand side have been collected into the Cc, and the u4
and b4 are constants for a given mesh size, defined by the relation of
these equations to the set (5-42)to (5-44).The next operation is to con-
struct a set of coefficients from the ad and bt, by the following procedure.
po = bo (5-51)
pm = bm/(um -~~-1) (m = 1,2,3,. . . ,M - 1) (5-52)
qm = l/(um - pm-1) (m = 1,2,3,. . ,M )
* (5-53)
These coefficients are computed a t the start of the calculation and used
in each succeeding step that is based on the same mesh size.
The unknown quantities are calculated by using two recursion for-
mulas in succession. They are
Do = Co (5-54)
D m = ~~(Dtn-1Cm)+ (m = 1,2,3,. . M ). (5-55)
(PM = DM (5-56)
(Pm = D m +
pm(Pm+l (m = 0, 1,2, . . . ,M - 1) (5-57)
With this method of calculation, the unknown quantities are found from
the C, with only two multiplications per mesh point, and the rounding
errors are kept small.
7. Technique for Starting
There is still one problem in connection with using Eqs. (5-42) to
(5-46)that has not been touched upon, namely, the selection of a pro-
'
cedure for getting started. The difference equations show how t o get the
solution for a profile when the values on the two preceding profiles are
known, but do not tell how to take a step when only the initial conditions
are given. Equations (5-34)to (5-38)cannot be used, because they give
too large a truncation error with a large axial step. One simple way to
proceed is to start with short steps, using the explicit formulas, and then
 DESIGN OF PACKED CATALYTIC REACTORS 249
to double the length of the step repeatedly until the desired length of
step is reached. The fastest and most convenient procedure would be to
double the length of the step a t every step after duplicating the first short
step. Unfortunately, this procedure leads to irregular behavior of the
solution, apparently because the initial profile represents the (n - 1) th
profile for all the steps taken until the final step length is reached. I n
order to get a good approximation to the solution of the differential equa-
tion, it is necessary to take two or three steps of each size during thc
starting period.
8. Special Treatment of Radiative Heat Transfer
A troublesome situation arises when radiation makes a large contri-
bution to the transport of heat. The corresponding term in the difference
equation is one of those that were grouped with the source term, to be
evaluated on the nth or intermediate profile, because i t has a variable
coefficient. If this term becomes as large as the term accounting for eddy
transport, the solution of the difference equation becomes unstable. The
difficulty can be surmounted by representing the coefficient of (dT/du) as
the sum of a constant value somewhere near the middle of its range, and
the difference from this constant value. Then the part of the term having
the constant coefficient can be calculated as the average of the approxi-
+
mations on the ( n - 1) th and ( n 1) th profiles, leaving the remainder
to be evaluated on the nth profile. The variable factor in the coefficient
is then broken into two parts, as follows:

(5-58)

By choosing TI, the reference temperature, so that the value of T3/C,

at TI is somewhat higher than one-half of its maximum value, the vari-
able part is made smaller than the constant part. The variation of heat
capacity with conversion is not likely to be large enough to make trouble,
but in any case a value of heat capacity corresponding to an intermediate
conversion can be used.

H. ADIABATIC
REACTOR
1. General Case
By using the reciprocal space velocity as an independent variable,
the adiabatic reactor can be treated as a one-dimensional reactor. Equa-
tions (3-31) and (3-8) provide the basis for the calculation. P u t in terms
of the virtual conversions, these equations are
ax
-=(R (5-59)
da
250 J O H N BEEK

and
+
H = YoHT x T S ~ H T (5-60)
= constant
Equation (5-60) can be represented adequately in nearly all cases with
terms of not more than the second degree in temperature, so that the
temperature required to determine the rate can be calculated. If the heat
capacity does not vary too much, Eq. (5-60) can be used as a linear rela-
tion in temperature.
2. Onlg One Independent Reaction
When there is only one independent reaction, Eq. (5-60) establishes
a unique relation between conversion and temperature for a given initial
condition. I n this case, it may be advantageous to calculate the recipro-
cal space velocity from the conversion by a quadrature, instead of inte-
grating (Eq. 5-59) as written. The relation is

u(z1) = [g (5-61)

Some quadrature formula, such as Simpson’s rule, gives the soIution faster
than does numerical integration of the differential equation by a general
method.
I. ILLUSTRATIONS
As an illustration of the use of these methods, a simple example of a
schematic catalytic reaction may be treated. Two reactions are supposed
to occur in a constant ratio. The reactions are
A+ B C +D= (AH = -84.7 kcal.)
and
A + 2B = 2E (AH = -79.3 kcal.)
The combined reaction is
A + +
1.068B = 0.932C 0.9320 0.136E + (AH= -84.3 kcal.)
The rate of the reaction has been found to be

with the parameters

k~ = 6.69 X lo-’ mole./cm.* sec. atm.
p = 0.69 atm.-’
A = 1.158 X deg.
 DESIGN O F PACKED CATALYTIC REACTORS 251

Although the rate of the side reaction is specified only approximately as

being proportional to the rate of the main reaction, the maximum tem-
perature is limited to 380°C.
The feed to the reactor is a mixture of A and B with an inert gas,
with the composition
N A = 0.095
Ne = 0.887
The following properties are estimated for this mixture a t 600°K.
n = 41.4 g./mole
C, = 28.7 cal./mole deg.
p = 1.54 X lo-* g./cm. sec.
kj = 1.24 X cal./om. sec. deg.
The following dimensions and conditions for the reactor are to be studied.
The catalyst pellets are in the form of short cylinders.
r2 = 1.87 cm.
2 = lo3 cm.
d p = 0.318 cm.
E = 0.39

Fc, = 1.6 X cal./cm. sec. deg.

T,= 320°C. = 593°K.
P = 14.62 atm.
G = 1.942 X low3(initial mol. wt.)/cm.2 sec.
The tubes are cooled by a boiling liquid, with an effective heat-transfer
coefficient of 2.85 X cal./cm.2 sec. deg.
From these conditions and properties, the transport properties are
calculated.

N C
- -&
Pr -nk,= 0.86
NRe= n G d = 166
P
Np, = 10.1 (Fig. 2)
N N , ,= 23.4 [Eq. ( 4 4 1
h = 9.1 X lW3 calJcrn.2 sec. deg.
h, = (0.8)h = 7.3 X 10-8 cal./cm.2 sec. deg. [Eq. (4-711
252 JOHN BEEK

1- - -
1 1
Or
hr - h, + 2.85 X
hr = 5.8 X lo-* cal./cm.2 sec. deg.
= 5.3 x cal./cm. sec. deg. [Eq. (4-3)]
kc = 2k, 4- ;7hd,
The conduction by radiation is based on an average temperature of 630°K
and an emissivity of unity for broken-in catalyst. It is
k, = (2)(1.37 X 10--8)(630)*(0.318) [Eq. (4-411
2.2 X lo4 cal./cm. aec. deg.
=
Then the effective thermaI conductivity is

k, =
C Gd
+ + k*
kc

= 1.75 X lo4 +
0.53 X lo-* f 0.22 X
= 2.50 X 10-8 cal./cm. sec. deg.

Nsi = = 4.32

The first questions to be answered are whether the pressure drop and
the difference in temperature between the fluid and the catalyst are large
enough to be important. Again using 630'K. as a representative tem-
perature, the density of the mixture is found to be 1.17 X g./cm?
The friction factor is

f+:= (1.75 + 150 -

NR3
= 23.8
Then the pressure drop is
41.4)(1.942 X lO+)Y
(23'8) r((1.17 X 10-2)(0.318)
=
= 2.46 X 104 g./cm. sec?
= 2.4 X 10" atm.
which is a small enough fraction of the pressure to be negligible.
A representative rate of reaction is for the conditions N A =0.06,
~ or, referred to
T = 630. The rate is then about 7 x loM7m o l e / ~ m .sec.,
the outside surface area of the catalyst particles, 7 x 10-7/13, or 6 X
lo-* mole/cm.2 sec. The corresponding flux of heat is 5 x cal./cm.2
sec., which, with a heat-transfer coefficient of 9 x cal./cm.2 sec. deg.,
gives a temperature difference of about 0.6'C.
 DESIGN OF PACKED CATALYTIC REACTORS 253
The example will be carried through without taking account of the
variation of heat capacity with conditions or of the variation of mass
velocity with radial position, because the computing program that was
available for this calculation does not accommodate these effects. As
pointed out above, the extra terms arising from these effects are treated
as contributing to the source term in the equation for the temperature.
The essential details of the calculation with two radial increments are
shown. First the difference equations are written in literal form [see Eqs.
(5-42) to 5-44) 1.
+ +
(0.75 kw/h)+o.n+i (0.25 - kw/h)4i,wi
+
= (0.75 - ko/h)+o,+~ (0.25 + +
k~/h)+l,+i 1.5lcs0,n 0.5k&,n +
-O.5(k@/h)+oo,n+1 (1 2kw/h)h,n+l - 1*5(kw/h)&,n+l
= 0.5(kw/h)40,+1 + +
(1 - 2k~/h)h,n-1 1.5(ku/h)&,n-i 2kSl.n +
+ + +
-2kM’~&,n+l+ [I 2kWw hNBi(M 0.5)]&.n+i = 2kM2~&,+1
+ + +
[I - 2kM20 - kwNBi(M 0.5)]&,n-1+ 2kS2.n 2kwN~i(M 0.5)Tc +
The table below shows what the symbols represent in the separate equa-
tions for conversion and temperature.

The numerical values for h and M were not used in these equations
because the literal form indicates better how they enter. The values 0.5
and 2 are inserted for h and M, and k is taken t o be 5 cm. Each of the
resulting equations is divided through by the absolute value of, the co-
efficient of the first,term on the left, giving the set of working equations.
The working equations for the temperature are written in terms of the
variable t = T - T,, a transformation that has the effect of making T,
the origin for the scale of temperature.
Xo.n+l - 0.09951~1.n+1
= 0.35074~0,,i + + +
0.54975~1,,-1 3.4777 X I03Ro,, 1.1592 X 108R1,,
-XO,n+l +9.547821,n+1- 3X~.n+l
= X0,n-1 + + +
1.547821,n-1 3X2,n-l 2.8567 X 104R1,,
-Xl,n+l + 1.6935~,+1
- +
- Z1.n-1 - 0.30653~2,+1 3.5709 X 108Rz.n
254 J O H N BEEK

to.n+l - O.2094&,n+1
== O.18585t0,n-l + O.6O472t1,n11 + 8.9742 X 10BRo,n+ 2.9914 X 1O6R1,n
-to,n+l + 7.8841S1,n+1- 3&,n+l
= tO,n-1- 0.11589t1,n-1 + f 5.8788 X IO7&,n
3t2,n-l

-ti,n+l + 2*8350L2,n+1= - 1.8639t~,n-1+ 7.3485 X 106Rz,n

t1,n-1

From this point on, the calculation is best shown in tabular form. The
profiles a t z = 90 cm. and z = 95 cm. are used to calculate the profiles
a t z = 100 cm.

z = 90 cm. z = 95 om.

m loax t (C) 102, t (0 10VZ mole/cm.S Bec.

0 4.7618 42.607 4.8869 39.615 8.1925

1 4.5518 25.158 4.7060 23.489 5.3394
2 4.5038 11.469 4.6645 10.720 3.6302

These values are used to calculate the terms on the right-hand side of the
difference equations. The quantities a,b , and C appear in Eqs. (5-48) to
(5-50), p and q in (5-51)to (5-53), and D in (5-54) to (5-57).
Calculation of conversion profile a t z = 100 cm.:

~~ ~

0 - 0.099510 4.5193 0.099510 - 4.6193 5.0016

1 9.5478 3.0000 26.844 0.31752 9.4482 3.3194 4.8472
2 1.6935 - 3.3008 - 1.3760 4.8114 4.8114
~ ~~ ~ ~ ~ ~~

Calculation of temperature profile a t z = 100 cm.:'

~~

m a , bm Cln P, 1/am 0, tm

0 - 0.20944 32.081 0.20944 - 32.081 36.652

1 7.8841 3.0000 105.484 0.39090 7.6747 17.925 21.823
2 2.8350 - 6.4505 - 2.4441 9.9732 9.9732

The following table shows how the temperature at the axis of the
tube and the average conversion vary along the reactor.
 DESIGN OF PACKED CATALYTIC REACTORS 255
~~~~~ ~ ~~

Axial Average
z temperature conversion
(em.) ("C.1 ( X 102)
~~

0 320 0.00
10 333 0.44
20 345 0.97
30 356 1.57
40 365 2.18
50 370 2.75
60 372 3.27
70 371 3.72
80 367 4.09
90 362 4.41
100 356 4.65
150 337. 5.53
200 330 6.14
300 326 6.97
400 324 7.56
500 323 7.99
600 322 8.31
700 322 8.56
800 321 8.76
900 321 8.91
1000 321 9.02

The conversion a t 1000 cm., 9.02 X is to be compared with the

initial mole fraction of substance A , 9.5 x The fractional conver-
sion of 0.95 is in a suitable range. The maximum temperature, 372OC., is
below the limit imposed, but not enough below to leave very much room
for increasing the production rate.
Some effects of changing the size of the steps used in the calculation
on the accuracy of the results are shown below, using the axial tempera-
ture a t a length of 60 cm. as illustration.

k (cm.)
h ( = 1/M) 10 5 2.5 1.25 0.625

0.5 376.42 373.73 373.05 - -

0.25 - 373.07 372.47 372.30 -
0.125 - - 372.38 372.24 372.20

An obvious method for increasing the production rate is to raise the

temperature of the cooling medium, and at the same time to improve the
256 JOHN BEEK

heat transfer by decreasing the radius of the reactor tubes and increasing
the flow rate. The performance of a reactor with the radius decreased to
1.27 cm. and with the mass velocity doubled has been calculated. With
the larger mass velocity the pressure drop is multiplied by 4, but it is
still not excessive.
The following new parameters are obtained.
rz = 1.27 cm.
G = 3.884 X 10” (initial mol. wt.)/cm.2 sec.
N R =~ 332
N p e = 10.0

N N =~ 32.7
ht = 1.02 X calJcrn.2 sec. deg.
k, = 4.41 X 10” cal./cm. sec. deg.
Nsi = 2.94
With these parameters and with the cooling medium a t 35OoC., the max-
imum temperature is 385OC., above the limit. With Toa t 345”C., the
maximum temperature is 37loC., which is safe enough, and the conver-
sion a t a length of 1000 om. is 0.0902. This choice of conditions then
seems to be acceptable, and has the advantage that the production rate
is twice as high as in the previous set of conditions, while the number of
tubes required is only about 10 percent larger.
By exploring the effects of changing mass velocity, radius of the tube,
and temperature of the control medium, an optimum set of conditions can
be determined. In addition to these variables, the diameter of catalyst
pellets and the pressure are sometimes disposable parameters, and can
be adjusted to improve the performance of the reactor.
There is a simple method for speeding the calculation of conversion
that should be pointed out. After the conditions in the upstream end of a
reactor have been calculated, including a region well beyond the point
where the temperature is a maximum, the conditions in the rest of the
reactor can be calculated accurately enough with only 2 or 3 radial in-
crements, or, if the Biot number is not too large, with the one-dimensional
approximation. In this region , the temperature has no intrinsic interest,
and it is only necessary to estimate the average conversion.
In the case of the first example given above, the accuracy of this kind
of continuation was tested, starting in each case with substantially cor-
rect profiles. The intervals over which the calculation was carried and
the results of the calculation of the average conversion are s h o w in the
following table.
 DESIGN OF PACKED CATALYTIC REACTORS 257

Interval (cm.) M=2 One-dimensional Correct

1-80 0.04206 0.03986 0.04179

80-300 0.07094 0.07037 0.07087
100300 0.07090 0.07061 0.07087
300-1000 - 0.09024 0.09023

The temperature reached a maximum in this case a t a length close to

60 cm.

Vl. Stability of Packed Tubular Reactors

A. STATEMENT OF THE PROBLEM

No complete picture of the stability of a reactor can be worked out

without specifying the kind of control that is operating, and then, using
this specification together with the properties of the reactor, studying the
transient behavior of the system, a t least in the locally linearized form.
The method of Deans and Lapidus (D3) for finding the transient condi-
tions in a packed tubular reactor can be adapted to this kind of study,
but a t great expense in computing time. It must be recognized, however,
that there is no simple solution to the problem, and that the best pro-
cedure for checking a combination of reactor and control system may be
to use a single tube of a multitube reactor as an element in an analog
computer. Even this scheme is not always adequate, as may be seen by
considering as an example the dynamic response of a multitube reactor
that is cooled by a boiling liquid, coupled with the condenser that re-
moves the heat and returns the liquid. I n this case, the resistance to two-
phase flow and mixing offered by the bank of tubes cannot be simulated
accurately with a single tube.
I n critical cases it may well be worthwhile to make a complete analy-
sis of stability. In many cases, however, enough can be learned by study-
ing what Bilous and Amundson (B7) called parametric sensitivity. These
authors derived formulas for calculating the amplification or attenuation
of disturbances imposed on an unpacked tubular reactor originally in a
steady state, with the idea that if the disturbances grow unduly the per-
formance of the reactor is too sensitive to the conditions imposed on it,
that is, to the parameters of the system. The effect of feedback from a
control system was not considered. As pointed out by the authors, it
would be a much more complicated task to apply their procedure to a
packed reactor, but it still would entail f a r less computation than a study
of the transient response.
258 JOHN BEER

A simple criterion for easy controllability was proposed by Wilson

(W4). H e pointed out that if
Ea(Trn - Tc)/RoTrn2< 1 (6-1)
where T, is the maximum temperature in the reactor, there can be no
problem about control. This is an excessively conservative requirement,
as was shown by Barkelew. It can be thought of as a sufficient but not a
necessary condition for controllability.

OF BARKELEW'S
B. DESCRIPTION CRITERION
1. Assumptions Involved
Wilson's criterion is based on the one-dimensional reactor. A more
useful criterion of parametric sensitivity based on this model was given
by Barkelew (B2). Barkelew proceeded by assuming that only one reac-
tion is important, that its rate may be expressed by the form

and that the temperature of the cooling medium is constant. He then

defined a new set of variables and parameters:
Xf =1 - Y/YO (6-3)
7 r ( T - T,)
= 6-41
N = 2U/r2CpkeyTa (6-5)
= Q~or/Cp (6-6)
With these variables, the relation between dimensionless temperature T
and the fractional conversion X , is

2. Nature of the Criterion

The dependence of T ~ the , maximum dimensionless temperature, on
the parameters N and S was studied by carrying out numerical integra-
+ +
tions of Eq. (6-7), using either l (YXor ( 1 p X ) - l for g ( X ) .The key
to the criterion developed by Barkelew is in the observation that the
family of curves showing T ~ / asS a function of N / S for constant values
of 5 is bounded below by an envelope, and that there is for each curve
a rather sharp change from a small slope to a large one in the neighbor-
hood of the contact with the envelope. A large slope for this curve means,
of course, a maximum temperature that is sensitive to the parameters,
 b E S I G N O F PACKED CATALYTIC REACTORS 259
This sharp curvature is particularly pronounced when S is large, which
is to say when the product of activation energy and heat of reaction is
large. Barkelew proposed as a condition for sensitivity that the combina-
tion of N and S be such that the value of N / S is above the point of con-
tact with the envelope. Given this criterion, other families of curves were
constructed, showing the critical relationship between N / S and S for
various combinations of the parameters a, p, and TO, the initial dimen-
sionless temperature. Several such families of curves are shown in the
original paper.
The introduction of the factor g(Xlt) adds a considerable generality
to the treatment. Examples given by Barkelew of the application of these
forms of g ( X , ) are:
Second-order reaction, -1 5 a < 0,p = 0.
Product-inhibited reaction, a = 0, p > 0.
Reactant-inhibited reaction, cr = 0, -1 < p < 0.
Autocatalyzed reaction, a > 0, p = 0.

VII. Scale Models of Packed Tubular Reactors

A. GENERALDISCUSSION
1. Requirements for a Useful Scale Model
The concept of a scale model of a reactor is useful because in some
situations of practical importance it permits us to predict rather accu-
rately the effects of changing certain design parameters of a reactor even
though the kinetics of the reactions are not known, provided other param-
eters are changed in a corresponding way. A familiar problem is that
of increasing the diameter of a reactor without changing the course of
the reactions going on, and without any detailed information about how
the rates depend on local conditions. In spite of the fact that no exact
scale model of a heat-exchanger type of reactor can be made, the require-
ments for a good approximation can be met in many cases when only a
moderate change in scale is required.
The rules given by Damkohler (Dl) for changing the scale of cata-
lytic reactors without changing the course of the reaction were derived
primarily by dimensional analysis. A better idea of the requirements for
scaling up can be obtained by a detailed examination of the coefficients
in the differential equations and boundary conditions describing the
reactor, with the independent variables in the equations transformed to
a modified reciprocal space velocity and a dimensionless radial variable.
In an exact scale model, these coefficients are all the same as they are in
260 JOHN BEEK

the prototype, so that the temperature, pressure, and composition depend

in just the same way on the new variables, and no essential feature of
the reacting system is changed. The construction of a useful scale model
depends on the possibility of choosing the design parameters in such a
way that the more important coefficients are approximated rather well,
without changing any coefficient enough to change the desired relation-
ship seriously. It will be seen that no useful model can be made when
either the pressure drop through the reactor or the temperature difference
between the catalyst and the fluid has a significant effect on the reactions,
and that the construction of the model is much simplified when the radial
variation of conditions can be ignored.

2. Role of the Activity of the Catalgst

In the treatment of the scaling problem given by Johnstone and
Thring (Jl), the activity of the catalyst is considered as a disposable
parameter. Since the rate of reaction appears in the dimensionless quan-
tities chosen by Damkiihler (Dl) to characterize a reactor, the activity
of the catalyst enters implicitly. The situations in which the activity of
a catalyst can be changed without affecting the selectivity or apparent
activation energy are practically restricted to those in which the catalyst
particles are impermeable, and the activity can be adjusted by changing
the size of the particles. This means, in effect, that the activity is a t our
disposal only to the extent that it can be changed by making the recip-
rocal change in the particle diameter. In principle, the catalyst could be
diluted with an inert packing, but this scheme is not useful in designing
new equipment. In order to show how it enters the relations, an activity
factor a, which would normally be unity, is carried through the calcula-
tions.

3. Parameters to be Chosen
The parameters nominally at our disposal are the diameter of the
reactor, the mass velocity, and the particle diameter. The problem of
changing the scale will be discussed from the point of view of an engineer
who wants to match the performance of two reactors of different diam-
eters, and who must then choose a corresponding mass velocity and an
accessible-and acceptable-combination of particle diameter and activ-
ity of catalyst.
In all that follows, it will be assumed that neither radiation nor mo-
lecular conduction will affect the performance of a scale model. If either
is so important that its variation cannot be ignored, no useful scale model
can be designed.
 DESIGN OF PACKED CATALYTIC REACTORS 261

B. DERIVATION
OF REQUIREMENTS FOR THE GENERALCASE
1. Transformation of Axial Variable
The axial variable in Eqs. (3-6) , (3-15), (3-28), and (3-30) is trans-
formed by the substitution
2= (Q/a>C (7-1)
This substitution is made, and the terms involving the rates are multi-
plied by the activity factor a. After some rearranging, the equations take
the form

2. Necessary Approximations
As the model is to be constructed so that the intensive properties of
the reacting fluid are to be invariant to the change of scale, such quan-
tities as the heat capacity and the rate and heat of reaction are also
invariant. In his treatment of packed catalytic reactors, Bosworth [see
(B8), p. 3181 assumes that the diffusivities and the thermal conductivity
remain constant when the scale is changed. Since these quantities are
approximately proportional to the mass velocity and the particle diam-
eter, the resulting rules for scaling can not be correct. The presence of
6
such factors as (dG/dp)/G and Tv in the first two equations requires
that the velocity profile must retain its shape when the scale is changed,
that is, that G must change in the same ratio everywhere.
There are two quantities appearing in Eq. (7-3) that cannot be kept
invariant. These are k,/Gd, and (dk,/d,)/Gd,. Owing to the fact that
k, and G vary with r in different ways, there is no way to keep the ratios
fixed everywhere when G is changed. The best that can be done is to
keep their average values unchanged.
The difference in temperature between the surface of a catalyst pellet
262 J O H N BEEK

and the fluid surrounding the pellet can have an important effect on the
rate, and can be changed significantly by changing the particle diameter
or the flow. The only quantity that is in question here is the heat-transfer
coefficient which, for a given composition and temperature of fluid, can
be written in the form
h = g(NRe)/dp (7-6)
where the function of Reynolds number written as g ( N R e ) can be ex-
pressed as proportional to some power of the Reynolds number, with the
exponent between 0.5 and 0.6 in the range of general interest.
3. List of Invariant Quantities
The following quantities are then to remain unchanged in any exact
scale model:
ki = GdP/a~2'Npe (7-7)
(7-8)
(7-9)
(7-10)
(7-11)
(7-12)
h = B(N&)/dp (7-13)
It is obvious that no scale model can be constructed when all these
quantities are important. The question that must be answered is under
what conditions useful models can be designed. I n the discussion, it will
be convenient to replace the second quantity by the product of the first
and second, that is, by
k2' = ht/ar2 (7-14)
If it is assumed that ht/h is independent of Reynolds number, k,' may
be replaced by another ratio,
ki' = ~ ( N R ~ ) / ~ W $ (7-15)
C. SELECTIONOF IMPORTANT
QUANTITIES
When the resistance to heat transfer is appreciable both within the
bed and at the wall, all the ratios except ks and k7 are important, and
some approximation must be made. Some progress may be made by ob-
serving that the performance of such a reactor is much more critically
determined by heat transfer than by mass transfer, and that eddy trans-
port, which provides mass transfer, makes the major contribution to heat
transfer. The problem can be attacked, then, by giving attention to the
 DESIGN OF PACKED CATALYTIC REACTORS 263
principal terms arising from the effective thermal conductivity, and using
an average value of mass velocity as the governing quantity. The quan-
tity that replaces kl, ka, k4, and k5 in this approximation is

(7-16)
This quantity is obtained by dividing the eff ective thermal conductivity
by urZ2,and then by Cp.As the problem is only approximately solved
in this way, kl’ may be calculated with average values of C p and G.
There are just two cases in which anything simple can be said about
the way the activity of a catalyst changes when the particle diameter is
changed. One is when the particles are porous and small enough to make
the activity independent of size, and the other is when the particles are
impermeable, in which case the product of activity and diameter is con-
stant. These will be considered in turn.
1. The activity is constant. The k’s are fixed by the prototype reac-
tor. The new G and dp are chosen to give the fixed values of kl’ and kz”
with the new r2. I n order to illustrate how the required changes in G and
d, affect the requirements on k6 and k7, an approximate calculation can
be carried through, neglecting conduction through the solid, and assuming
that g(NRe)= (const) NRen, where n is between 0.5 and 0.6. The equa-
tions for G and d,, are then
(7-17)
and

($>”($>”’ ($-) = (7-18)

These yield the relations

(7-19)
and

($) ($>’”’
= (7-20)

It may be seen that G varies nearly as the square of r2, but that d p does
not change very much. This is favorable for the assumption that the
activity is constant. It is immediately obvious that, on the other hand,
neither ko nor k7 will be very nearly constant, k6 changing like a t least
the cube of r2 and k7 being proportional to r2. I n this case, then, neither
pressure drop nor heat transfer to or from the catalyst can have any in-
fluence on the reaction on either scale. If these effects are negligible, a
good scale model can be designed.
264 JOHN BEEK

2. The activity of the catalyst is inversely proportional to the diam-

eter of the particles. Using the same assumptions as for case 1, the rela-
tions become
(7-21)
and
,(7-22)
These give the results

(7-23)
and

(7-24)

Here again, G varies as something like the square of r2, and d, varies but
little, leading to substantially the same conclusions for this case as for
the other.
The problem of establishing the proper axial temperature profile and
heat transfer coefficient in the heating or cooling medium has not been
touched upon. If the temperature and transfer coefficient are determined
by flow conditions, the total flow rate of the medium is kept in a con-
stant ratio to the total flow rate of reactants, and the size of the channels
in which the medium flows is then chosen to match the original value of
the ratio kz’. If heat is transferred to a boiling liquid, there is
either the obvious solution of changing nothing or there is no solution,
depending on whether the boiling point at the top and bottom of the
pool of liquid is or is not practically the same. I n the second case, the
vertical temperature profile might be matched approximately by using
a liquid with a different normal boiling point, and a t a different pressure.
When conduction through the solid is taken into account, the varia-
tions of both G and dp are somewhat increased, but the variation of dp
is still gratifyingly small. I n both cases considered, it is advantageous
to have the required increase in particle diameter small, in the first case
because i t makes the assumption of constant activity more secure, and
the second because the accompanying decrease in activity is small. In
some situations a multitubular reactor with tubes of larger diameter may
demand less metal in the tubes even though it contains a somewhat larger
volume of catalyst.
 DESIGN OF PACKED CATALYTIC REACTORS 265

D. LIMITATIONS
The results reached for the packed reactor in which both conduction
through the bed and heat transfer a t the wall are important may now
be summarized. When the diameter of the reactor is changed, the mass
velocity and partide diameter are adjusted to keep k~' and kz", defined
by Eqs. (7-16) and (7-15), a t their values in the prototype reactor, any
variation of the activity with particle diameter being taken into account.
The effectiveness of this procedure depends on the assumption that sev-
eral conditions that actually change in going from one scale to the other
change little enough to have a negligible effect. These conditions are: the
pressure drop in the reactor, the shape of the velocity profile, the ratio
of the diffusivity for heat to the diffusivity for material, and the tem-
perature difference between the reacting fluid and the catalyst. The first
and fourth of these must be negligible on both scales. The condition on
the pressure drop will usually make it impossible to change the scale by
any large factor, because of the strong dependence of pressure drop on
mass velocity,
E. THEONE-DIMENSIONAL APPROXIMATfON

The requirements for approximating the radial heat transfer in a

packed reactor with a constant over-all heat-transfer coefficient are dis-
cussed in Section 111, above. When these requirements are met, Eqs.
(3-32), (3-33) , and (3-34) are used to describe conditions in the reactor.
With the substitution of Eq. (7-1), these equations take the form

(7-25)

(7-26)

(7-27)

The only quantities that must be kept constant in this caae are
ka = U/ar, (7-28)
and the k6 of Eq. (7-12) to take care of the pressure drop. I n addition,
there is the requirement that the catalyst and the fluid surrounding it
be a t practically the same temperature, or that, alternatively, the k, of
Eq. (7-13) remain unchanged.
With only one principal quantity k8 to be kept invariant, something
may be done about one or the other of the quantities and Ic,. Be-
cause of the complicated dependence of U on G and d, in the range of
practical concern, a simple approximate relation cannot be constructed.
Some idea of the possibilities may be indicated with two illustrations.
First, consider the case when Ic6 is important but k , is not. Nothing
can be done if the product ad, is constant and the Reynolds number is
high, because then the friction factor is insensitive to Reynolds number,
and ?7 must be nearly constant. If the activity is constant and the
Reynolds number is high, d, varies about as the cube of %. With the
same assumptions about the effective thermal conductivity and the heat
transfer coefficient that were used for the three-dimensional reactor, with
the additional assumption that in the prototype reactor htcr2/lce= 2, the
relation between and r2 obtained by eliminating d, is
+
(G/Go)-o~a/[l 0 . 5 ( r ~ / r z ~ ) ( ~ / ~=) -O.67(r2/rz0)
~~~] (7-29)
From this relation it is found that if r2 is to be multiplied by the factor
1.1, is multiplied by 1.3 and d, by 2.2. This result shows that even in
the case of the one-dimensional reactor, no useful increase in scale is pos-
sible if pressure drop in the reactor has a significant effect on the reaction.
If the extra degree of freedom is needed to keep the coefficient for
heat transfer to the catalyst constant, the situation is not quite so bad.
In order to fix k,, d, must vary as something between the first and the
three-halves power of G. Continuing with the same assumptions as in
the case above, short tables have been calculated to show the relations
when the activity is constant and when it is inversely proportional to the
diameter of the particle.

TABLE I11
PARAMETERS IN SCALE MODELSOF A ONE-DIMENSIONAL REACTOR

a = aa a = aaddpO/dp

rzlrza GIGO dP/dPQ rdna GIGO dP/$

1.1 1.16 1.25 1.18 1.1 1.15

1.2 1.42 1.69 1.37 1.2 1.31
1.3 1.78 2.39 1.58 1.3 1.48
- - - 1.79 1.4 1.66

It may be seen that there are disadvantages in making any consid-

erable increase of scale in either case. If the activity is constant, the
particle size increases so much that great stress is put on the approxima-
tions used in calculating the parameters. If the activity is inversely pro-
portional to the diameter of the particles, the production rate per unit
volume of reactor is seriously decreased.
 DESIGN OF PACKED CATALYTIC REACTORS 267
Kjaer ( K l , p. 79) has given the appropriate modification for over-all
heat-transfer coefficients to the average temperature and to the tempera-
ture of the thermowell when a reactor tube contains a coaxial thermowell.
F. EXAMPLE
The use of the foregoing rules for constructing scale models may be
illustrated by changing the scale of the reactor used as the principal
example in Section V. Consider the problem of making a scale model in
which the mass velocity is 0.6 times what it is in the prototype, with the
same activity of catalyst, so that the length of the reactor is changed
from 10 meters to 6 meters. The corresponding radius of tube and diam-
eter of catalyst particle are to be calculated.
The quantities to be kept constant are k ~ ’ and
, k2‘, defined as

Because of the fact that the Nusselt number and the Peclet number are
not simply expressed in terms of the particle diameter it is convenient t o
use them as intermediate variables in the calculation. Since the mass
velocity is fixed, the Reynolds number and the Nusselt number are deter-
mined by the particle diameter. The calculation proceeds by finding the
values of r2 from selected values of d,, as determined from k~‘, and kz’
separately. By inverse interpolation, the value of dp that makes the two
values of r2 equal is found.
The quantities that are calculated in turn from a tentative value of
dp are NRe,NPe,NNu,k,, r 2 ( k [ ) ;and for k2‘, ht and then r 2 ( k d ) .The
calculation gives
d, = 0.326 cm.
r2 = 1.53 cm.
N p e = 11.0
G = 1.165 X (initial mol. wt.)/cm.2 sec.
k, = 1.680 x cal./cm. sec. deg.
The calculated profiles given below show how the scale model com-
pares with the prototype in this case. The distance from the entrance
is 60 cm. for the prototype and 36 cm. for the model.
This illustration does not pretend to represent the correspondence to
be expected between a real scale-model reactor and its prototype, but to
show that if the transport properties can be accurately estimated, a use-
ful scale model can be constructed in simple cases. In the real situation,
there may be large uncertainties in the best obtainable estimates of the
transport properties, and there may be in addition significant effects of
268 JOHN BEEE

Conversion T - To ("C.)
Dimensionless
radius Model Prototype Model Prototype

0.00 0.03593 0.03556 62.19 62.47

0.60 0.03366 0.03357 40.80 40.94
0.71 0.03222 0.03231 30.81 30.88
0.87 0.03144 0.03164 21.93 21.97
1 .oo 0.03121 0.03144 13.99 14.01

the variation of heat capacity with temperature and composition, and of

the radial variation of mass velocity.

Nomenclature
Note: definitions which are not followed by parenthetical dimensions refer to dimen-
sionless entities.
activity factor h heat-transfer coefficient, Section
coe5cienta in implicit difference IV; radial increment, Section V
equations (Mt-8 T-1: none)
rate of production of an entity hr heat-transfer coefficient between
[(amount) L - W ] reacting fluid and control medium
coefficients in implicit difference (Mt"T-1)
equations hw heat-transfer coefficient between
concentration of an entity reacting fluid and wall (Mt-aT-1)
[(&mount)M-'1 H enthalpy of reacting fluid (UP)
heat capacity of control medium H matrix of molar enthalpies of com-
(LY"T-1) ponents (MLat? mole-')
known t e r m in implicit difference J diffusive flux of an entity
equations (Tor none) [(amount) L-t-11
heat capacity of reacting fluid k axial step, Section V; rate con-
(Lgt-T-1) stant, Section M (L: t-1)
matrix of molar heat capacities of k, effective radial t h e m 1 conduc-
components (MLgt-IT-' mole-') tivity in packed tube (MLt-T-1)
awdiary quantity in solution kf thermal conductivity of reacting
of difference equations (T or fluid (MU-aT-1)
none) ki invariants in Section VII (vari-
diameter of catalyst pellet (L) OW)
modified eddy-diffusion coefficient Ir, radial thermal conductivity in ex-
(ML-'t-1) cess of eddy conductivity
energy (enthalpy) of activation (MLt-8T-1)
(MDt-3 mole-') m index of radial step
friction factor M number of radial increments
general expression for a derivative JT average molecular weight of mix-
(various) ture (M mole-1)
functions defined in text, Sections n index of axial step, Section V;
VI and VII (various) numerical constant, Section VII
superficial MME velocity N parameter used in Barkelew's cri-
(ML-*t-') terion
 DESIGN OF PACKED CATALYTIC REACTORS 269

mole fraction of the ith compo- W coefficient in axial derivative of Tc

nent (L-1)
Biot number, htrz/k. W c mass rate of flow of control me-
Nusselt number, hd,/kf dium (Mt-l)
modified Peclet number, Gd,/E X conversion, defined in Section I1
Reynolds number, G d p / p (mole M-1)
pressure (ML-W) X j fractional conversion
coefficients in recursion formulas x matrix of virtual conversions for
heat of reaction (-AH)(MUt-a key reactions (mole M-1)
mole-') K matrix of conversions (mole M-1)
matrix of heats of reaction Y concentration (mole M-1)
(MLet-2 mole-') J matrix of key concentrations
matrix of heats of key reactions (mole M-1)
(MLet-2 mole') Y matrix of concentrations (mole
dimensionless radial coordinate M-1)
radius of tubular reactor ( L ) z coordinate in direction of flow
rate of a reaction (mole L-3t-1) (L)
molar gas constant (MUt-2T-l 2 length of reactor (L)
mole-') a parameter in rate expression
matrix of rates of reactions (mole -?j parameter in rate expression
L-St-1) y parameter in rate expression (T-1)
matrix of virtual rates of key re- r modified reciprocal space velocity
actions (mole L-st-') (M-1L3t)
generalized source term, Section 7 coefficient in schematic differen-
V; parameter in Barkelew's crite- tial equation (L-1)
rion, Section VI (Tor none: none) ff constant in expression for velocity
reduced stoichiometric matrix profile
matrix of stoichiometric coeffi- A constant in expressionfor velocity
cients profile
matrix of stoichiometric coeffi- p viscosity (ML-It-1)
cients for key components Y change in number of moles in a
matrix of stoichiometric coeffi- reaction
cients for key reactions Y matrix of changes in number of
absolute temperature (T) moles for key reactions
temperature of control medium pi density of fluid (ML-8)
(TI Q reciprocal space velocity (M-lDt)
temperature of wall of a tube ( T ) 7 dimensionless temperature
over-all heat-transfer coefficient + general dependent variable (van-
(Mt-ST-') ous)
weighting coefficient on radial o coefficient in schematic difleren-
profile tial equation (L-1)

REFERENCES
Al. Acrivos, A., Znd. Eng. Chem. 47, 1553 (1955).
Ala. Aerov, M. E., and Umnik, N. N., J. Tech. Phys. (US.S.R.) 21, 1351 (1951).
A2. Aerov, M. E., and Umnik, N. N., J. Tech. Phys. (UBSB.) 21, 1364 (1951).
A3. Argo, W. B., and Smith, J. M., Chem. Eng. Progr. 49, 443 (1953).
A4. Aris, R., and Arnundson, N. R., AJ.ChB. Journal S, 280 (1957).
B1. Bakhurov, V. G., and Boreskov, G. K., 3. App2. Chem. (USBB.) 20, 721
(1947).
270 J O H N BEEK

B2. Barkelew, C. H., Chem. Eng. Progr. Symposium Ser. 65, No. 25 (Reaction
Kinetics and Unit Operations), 37 (1959).
B3. Baron, T., Chem. Eng. Progr. 48, 118 (1952).
B4. Beek, J., Jr., and Miller, R. S., Chem. Eng. Progr. Symposium Ser. 55, No. 25
(Reaction Kinetics and Unit Operations), 23 (1959).
B5. Beek, J., Jr., and Singer, E., Chem. Eng. Progr. 47, 534 (1951).
B6. Bernard, R. A., and Wilhelm, R. H., Chem. Eng. Progr. 46, 233 (1950).
B7. Bilous, O., and Amundson, N. R., A.I.ChJZ. Journal 2, 117 (1956).
B8. Bosworth, R. C. L., “Transport Processes in Applied Chemistry.” Wiley, New
York, 1966.
B8a. Brotz, W., “Grundriss der chemischen Reaktionstechnik,” pp. 35-44. Verlag
Chemie, Weinheim, Germany, 1958.
B9. Bunnell, D. G., Irvin, H. G., Olson, R. W., and Smith, J. M., Ind. Eng. Chem.
41, 1977 (1948).
C1. Cairns, E. J., and Prausnitz, J. M., Chern. Eng. Sci. 12, 20 (1960).
C2. Campbell, J. M., and Huntington, R. L., Petroleum Refiner 31(2), (1952).
C3. Chu, Y. C., and Storrow, J. A,, Chem. Eng. Sci. 1, 230 (1952).
C4. Coberly, C. A,, and Marshall, W. R., Jr., Chem. Eng. Progr. 47, 141 (1951).
C5. Converse, A. O., AJ.Ch.E. Journal 6, 344 (1960).
D1. Damkohler, G., 2.Elektrochem. 42, 846 (1936).
D2. Damkohler, G., in “Der Chemie-Ingenieur,” Vol. 3, p. 421. Akademische Ver-
lagsgesellschaft, Leiprig, 1937.
D3. Deans, H. A., and Lapidus, L., AJ.ChE. Journal 6, 656, 663 (1960).
D4. Dorweiler, V. P., and Fahien, R . W., A.I.Ch.E. Journal 5, 139 (1959).
D5. Douglas, J., personal communication.
El. Ebach, E. A., and White, R. R., A.I.Ch.E. Journcll 4, 161 (1958).
E2. Ergun, S., Chem. Eng. Progr. 48, 89 (1952).
F1. Fahien, R. W., and Smith, J. M., A.I.Ch.E. Jaurnat 1, 28 (1955).
G1. Gamson, B. W., Thodos, G., and Hougen, 0. A., Trans. A1.Ch.E. 39, 1 (1943).
H1. Hanratty, T. J., Chem. Eng. Sci. 3, 209 (1954).
H2. Hartman, M. E., Wevers, C. J. H., and Kramers, H., Chem. Eng. Sci. 9, 80
(1958).
H3. Hatta, S., and Maeda, S., Chem. Eng. (Tokyo) 13, 79 (1949).
H4. Hildebrand, F. B., “Methods of Applied Mathematics.” Prentice-Hall, New
York, 1952.
H5. Hildebrand, F. B., “Introduction to Numerical Analysis.” McGraw-Hill, New
York, 1956.
H6. Hougen, 0. A., and Watson, K. M., “Chemical Process Principles.” Wiley, New
York, 1947.
J1. Johnstone, R. E., and Thring, M. W., “Pilot Plants, Models, and Scale-up
Methods in Chemical Engineering.” McGraw-Hill, New York, 1957.
K1. Kjaer, J., “Measurement and Calculation of Temperature and Conversion in
Fixed-bed Catalytic Reactors.” Gjellerups, Copenhagen, 1958.
K2. Kramers, H., and Alberda, G., Chem. Eng. Sci. 2, 173 (1953).
K3. Kwong, S. S., and Smith, J. M., Ind. Eng. Chem. 49, 894 (1957).
L1. Latinen, G. A., Dissertation, Princeton University, Princeton, New Jersey, De-
cember, 1951.
L2. Lowan, A. N., “The Operator Approach to Problems of Stability and Con-
vergence,” Scripta Mathematica, New York, 1957.
M1. Maeda, S., Tech. Repts. Tohoku Univ. 16, 1 (1952).
 DESIGN OF PACKED CATALYTIC REACTORS 271
M2. Mickley, H. S., Sherwood, T. K., and Reed, C. E., “Applied Mathematics in
Chemical Engineering.” McGraw-Hill, New York, 1957.
M3. Milne, W. E., “Numerical Calculus.” Princeton Univ. Press, Princeton, New
Jersey, 1949.
M4. Morales, M., Spinn, C. W., and Smith, J. M., Znd. Eng. Chem. 43, 225 (1951).
01. O’Brien, G. G., Hyman, M. A,, and Kaplan, S., J . Math. Phys. 29, 223 (1951).
P1. Polack, J. A,, Dissertation, Massachusetts Inst. Technol., Cambridge, 1948.
Q1. Quinton, J. H., and Storrow, J. A., Chem. Eng. Sci. 5, 245 (1956).
R1. Ranz, W. E., Chem. Eng. Progr. 48, 247 (1952).
R2. Richardson, L. F., Phil. Trans. Roy. Soc. London, Ser. AblO, 307 (1910).
S1. Schuler, R. W., Stallings, V. P., and Smith, J. M., Chem. Eng. Progr. SympoGum
Ser. 48, No. 4 (Reaction Kinetics and Transfer Processes), 19 (1952).
S2. Schwartz, C. E., and Smith, J. M., Znd. Eng. Chem. 45, 1209 (1953).
S3. Sehr, R. A., Chem. Eng. Sci. 9, 145 (1958).
54. Singer, E., and Wilhelm, R. H., Chem. Eng. Progr. 46, 343 (1950).
S5. Smith, J. M., “Chemical Engineering Kinetics.” McGraw-Hill, New York, 1956.
S6. Strang, D. A., and Geankoplis, C. J., Znd. Eng. Chem. 50, 1305 (1958).
T1. Thoenes, D., Jr., and Kramers, H., Chem. Eng. Sci. 8, 271 (1958).
W1. Walas, S. M., “Reaction Kinetics for Chemical Engineers.” McGraw-Hill, New
York, 1959.
W2. Wehner, J. F., and Wilhelm, R. H., Chem. Eng. Sci. 6, 89 (1956).
W3. Wilke, C. R., and Hougen, 0. A., Trans. A.Z.Ch.E. 41, 445 (1945).
W4. Wilson, K. B., Trans. Znst. Chem. Engrs. 24, 77 (1946).
Y1.Yagi, S., and Kunii, D., A J . C h b . Journal 6, 97 (1960).
Y2. Yagi, S., and Wakao, A.Z.Ch3. Journal 5, 79 (1960).
Y3. Yoon, C.Y., Ph.D. Thesis, Massachusetts Inst. Technol., Cambridge, Septem-
ber, 1959.