I OCTAVE LEVENSPIEL and KENNETH B.

BISCHOFF
Illinois institute of Technology, Chicago 16, 111.

Backmixing in the Design of

Chemical Reactors
This article provides a useful perspective on the impor-
tance of backmixing in several types of reactors. Design
charts are given which set limits for the practicing engi-
neer in the field of reactor kinetics and design
--- A

'4'
ADEQ~ATE design of chemical flow
reactors rests on knowledge of two fac-
tors-the over-all rate a t which the
reaction proceeds and the extent of
backmixing occurring in the reactor.
T h e first factor involves the determina-
tion of the rate equation, which for
homogeneous reactions may be found
most conveniently by using small scale
laboratory batch equipment. As back-
mixing does not occur in such setups,
the effect of reaction rate may be isolated
and examined alone. Backmixing may
also be studied independently, as in
hydrodynamic studies in which the
progress of the fluid is examined as it
passes through the reactor vessel. h-u-
merous techniques have been employed
and the results of the studies reported in
many different ways.
I n flow reactors backmixing, reaction
rate and degree of conversion are tied
together in a complex fashion. Because
the role of backmixing is difficult to
evaiuate in such situations, much present
design is based on the assumptions that
no backmixing occurs or that the con-
tents of the system are well mixed and
uniform in composition.
Complete Backmix and
Nonbackmix Reactors
T h e idealized situation in which there
is no backmixing is called plug flow,
piston flow, slug flow, tubular flow, or
nonbackmix flow, and is characterized
by the fact that flow through the reactor
is orderly and the residence times of all
fluid elements are alike. For this situa-
tion, the volume of reactor required to
effect a fractional conversion, x , of
reactant A is given by
where r is the reaction rate of A and F
is the feed rate of A into the reactor.
T h e other extreme in flow conditions is
characterized by so great a n extent of
backmixing that any fluid in the reactor
has an equal chance of being found a t the
reactor outlet. This idealized situation
is called complete or total backmix

VOL. 51, NO. 12 DECEMBER 1959 1431

 Backmixing and Local single dimensionless group--" the locaI
Longitudinal Dispersion Number longitudinal dispersion number" defined
by Dlud, where D is the longitudinal
Flow patterns in reactors may vary dispersion coefficient.
greatly; however, the resulting back- In the local longitudinal dispersion
mixing may often be characterized by a number, u is the fluid velocity and d is
-
VrO
V. IO2 I

STREAMLINE FLOW
IN PIPES

D
Udt
0 01 01 IO

f = I- X = FRACTION OF REACTANT REMAINING

Figure 1. The volume of a backmix

reactor i s greater than that of the plug
flow reactor required for the same
I I I
duty. The size ratio rises rapidly with IOZ lo3 lo4 10'
increase in both fractional conversion
R e . Sc d t UP . =- udt
and reaction order E
?
PDV Dv
IO
flow, or stirred tank flow; the volume of
reactor required for conversion x of TURBULENT FLOW
reactant A is given by IN PIPES
V , = Fx/r (2 1
T h e progress of many reactions may be -
D
I
udt
approximated by the simple rate law
r = kC" (3)
where n is the order of the reaction. For
such cases, when fluid density remains
unchanged, the comparison of sizes of 0.I I

reactors for a given feed rate is found from

these equations to be
lo3 lo4 IO ' IO6

I
I
1

(4)
I PACKED BEDS

This result (Figure 1) shows that

except for a zero-order reaction the
complete backmix reactor always re-
quires a larger volume than a plug flow
reactor for a given feed rate? and that GASES I
the effect of backmixing becomes in- , I
creasingly important for higher reaction 011
order and for approach to complete 10- lo-' I IO IO2 lo3
conversion of reactant. Hence approxi-
mations to ideality which may be per- -dP G o
P
missible a t low fractional conversions IO, I I I
would lead to large errors a t high frac-
tional conversions. PACKED BEDS
-D
UdpE
LIQUIDS
Partial Backmix Reactors
Because a real reactor exhibits some
degree of backmixing, the requiredvolume I1 I I I I
for a given duty should lie somewhere be- 0.I I IO IO2 lo3
tween the two extremes given by Equa-
tions l and 2. T h e problem then is
twofold: to determine the extent of
backmixing by a quantitative measure Figure 2. Experimental findings on backmixing b y numerous investigators
and then to use this measure with rate local dispersion coefficients vary greatly with Reynolds numbers for flaw in pipes ( A and
data to determine the necessary reactor B) but are rather insensitive over wide ranges of Reynolds numbers far flow through packed
size. beds ( C and D)

1432 INDUSTRIAL AND ENGINEERING CHEMISTRY

 B A C K M I X I N G I N REACTORS

some size characteristic-i.e., tube diam- fluidized bed or of vessel in an all-fluid

eter for flow in pipes or particle size in
packed beds. D/ud varies from 0 for
plug flow reactors to for complete
backmix reactors; a n increase in D/ud
corresponds to a rise in the degree of
backmixing. Dispersion groups in
general are reciprocals of the correspond-
ing mass transfer Peclet groups.
For pipes and packed beds, D / u d
correlates simply with the dimensionless
groups characterizing the flow condi-
tions. Figure 2 , A , is the correlation
for streamline flow in pipes; both
Reynolds and Schmidt numbers affect
the local dispersion number. This rela-
tionship was obtained theoretically and
verified by three experimental runs (70).
Figure 2, B, the recommended correla-
tion for turbulent flow in pipes, is based
on numerous laboratory and field studies
with both liquids and gases. Prepara-
tion and detailed discussion of Figure 2,
A and B, may be found elsewhere (7).
Figure 2, C, shows packed bed local
dispersion numbers found by various
investigators (7-4, 6, 8). T h e results
suggest a larger local dispersion number
for liquid systems as compared to gas sys-
tems. Figurc, 2, D. represents a compre-
hensive study of liquid dispersion in or-
dered and randomly packed columns and
shows two distinct regimes of flow (5).
Reactor longitudinal
Dispersion Number
Though D/udcharacterizes the intensity
of backmixing a t any point in a reactor,
the “reactor longitudinal dispersion num-
ber,” D/uL, must be used in conjunction
with the reaction rate equations to
determine the reactor size. L is some
measure of the length of fluid path
through the reactor. I n cylindrical
reactors, this is measured most con-
veniently by the length of packed or
reaction chamber. For vessels in which
the intensity of fluid mixing varies from
position to position, such as the spherical
reaction chamber or the vessel with large
end effects, the effective length has yet to
be determined.
If the local dispersion number may be
considered uniform throughout the re-
actor, D,’uL may easily be found. For a
packed bed reactor
D l u L = (D/ud,)(d,/L) (5)
and for a tubular reactor
D/uL (D/udi)(dt/L) (6)
Chemical Reaction and Backmixing
T h e differential equation governing
the fractional conversion of reactant as a
function of axial distance 1 in a reactor
of length L is given by
D -d2f - u df
- - kCon-’f” = 0
dlz dl (7)
in which f = 1 - x is the fraction of
reactant remaining (9).
First-Order Reaction. For a first-
order reaction, or rl = 1, Equation 7 has
been solved ( 7 7 ) under the appropriate
boundary conditions corresponding to
any homogeneous reactor vessel or any
catalytic reactor bed in which the
intensity of backmixing is uniform (con-
stant D,ud throughout). In dimension-
less form the solution relates the frac-
tional conversion with both the reactor
dispersion number, D / u L , and the rate
group for the first-order reaction, kL,’u.
Thus
Figure 3, a graphical presentation of the
above result in useful form, was prepared
by combining Equations 1 and 8 and
compares sizes of reactors required for a
given feed rate for nonbackinix and
partial backmixing conditions.
Second-Order Reactions. T h e efFect
of backmixing on reactor size require-
ment when a second-order reaction is
taking place can be found by solving
Equation 7 for n = 2. An analytical
solution for cases of reaction order other
than 1 is not kno1z.n because of the non-
linearity of the differential equation.
Therefore, the differential equation was
solved numerically on an IBM 650
digital computer. A fourth-order
Runge-Kutta method was used, and i t
was found that 25 increments were
sufficient for the desired accuracy. Be-
cause the problem was of a boundary
value nature rather than initial value-
i.e., both the slope, df,‘dl, and the ordi-
nate, f, were not known a t either bound-
ary-the method of solution was neces-
sarily trial and error. The process was
started a t the reactor outlet where the
slope was known to be zero ( 7 7 ) . A
value of the ordinate a t this position was
estimated and the Runge-Kutta method
was used to calculate back to the reactor
inlet boundary, a t which point the
boundary condition
where f ( O + ) is the fraction of reactant
remaining just within the reactor en-
trance, had to be satisfied to within a
given error of 0.01-Le.. absolute values
1 - f(0-t) + D/uL d f ( O + ) ’ d ( l ’L) ab-

200

100
60
40
V
v 20
-
- V O

V.
10

6
4

I
0 001 001 01 10 0.0I 0.I 1.0
f = I - X = FRACTION OF REACTANT REMAINING
f = I-X=FRACTION OF REACTANT REMAINING
Figure 3. The volume o f an actual reactor with a given Figure 4. The volume of an actual reactor with a given
D/uL compared t o that of the ideal reactor required for the D/uL compared to that of the ideal reactor required for the
same duty same duty
For a first-order reaction, analytical solution of the differential equation is For a second-order reaction, a numerical solution of differential equation
possible (J J ) was obtained on an IBM 650 computer

VOL. 51, NO. 12 DECEMBER 1959 1433

solute values L 0.01, If it was satisfied,
the answers were punched out and a
new set of values were calculated for
a different value of D/uL. If not
satisfied, the initial assumed ordi-
nate was corrected by the computer, and
a new curve was calculated. This proc-
ess was repeated until the boundary
condition at the reactor inlet was satis-
fied. The last correctedvalue of the reac-
tant concentration a t the outlet was used
as the total conversion for plotting.
Figure 4 compares ideal and actual
reactor sizes for a given feed rate. I t was
prepared by combining Equation 1 with
the result of the trial and error numerical
analysis described above.
Fractional-Order Reaction. Inter-
polation between Figures 3 and 4 allows
estimation of reactor size to effect a
given conversion for fractional order
reactions involving a single reactant.
Backmixing does not affect a zero-order
reaction.
Example
A gaseous material undergoes a com-
plex series of changes when in contact
with a solid catalyst. To investigate the
kinetics of this reaction a small labora-
tory reactor is constructed consisting of a
1-inch schedule 40 pipe packed to a
depth of 1.5 inches with solid ‘/l-inch
diameter catalyst pellets. For a feed
rate of 2.7 cubic feet per hour or GO =
49 Ib./hr.-sq. ft. the reactant is 99%
decomposed. T h e reaction is of the
first order with respect to the reactant.
What depth of V,-inch diameter cat-
alyst is required to yield a 9970 decom-
position in a larger reactor to be con-
structed of 12-inch schedule 30 pipe if the
feed rate is 4000 cubic feet per hour or
Go = 547 Ib.,’hr.-sq. ft.? Additional
information and assumptions: p =
0.01 cp. ; isothermal conditions through-
out and no net change in the number of
moles of material passing through the
reactor; neglect any nonuniform flow
patterns due to channeling a t the pipe
wall.
Solution. FOR I-INCHREACTOR.The
particle Reynolds number
From Figure 2, C, the local dispersion
number
D/udp = 0.5
Thus the reactor dispersion number
D / u L = ( D / u d p ) ( d p / L )=
(0.5)(0.25/1.5) = 0.0833
iYow from Figure 3
--
SVo/SV E VIVO E 1.35
but for the experimental run
- = vol. feed/hr. -
sv - A
vol. reactor
[2.7/(7.5 X = 3600 hr.-l
1434 INDUSTRIAL A N D ENGINEERING CHEMISTRY
Therefore, for the ideal case of plug flow
- A , moles A/cu. ft.
57’0 = 1.35 (3600) = 4860 hr.-1
FOR12-INCHREACTOR.
T h e particle
Reynolds number
and from Figure 2, C, the local dispersion
number D/ud, = 0.55. At this point
the unknown length, L, enters in both
variables S V and DjuL, so the solution
involves successive trials.
Assume to start that no backmixing
occurs in this reactor. For this case
using the value of S V o found above
- vel. feed/hr. -
SVO = 4860 h r . 7 =
vol. reactor
4000
(0.797)L
Therefore the reactor depth
L = 1.032 feet = 12.4 inches
Check the assumption of plug flow
made above.
D/uL = (Djudp)(d,/L) =
(C.55)(0.25/12.4) = 0.0111
From Figure 3
Therefore the assumption is justified,
and L = 12.4 inches.
From this example, we see that back-
mixing usually plays only a minor role in
large packed bed reactors, although it
may be significant in small laboratory
reactors.
Precautions
A number of precautions must be
observed in applying this procedure in
the scale-up of process equipment.
The first involves the isothermal require-
ment. As heat effects depend on the
surface-volume ratio of the reactor,
scale-up will result in larger heat effects
and probably nonisothermal temperature
distributions which must be accounted
for by a complex analysis or a mean
reaction rate constant. Another pre-
caution involves the unjudicious extra-
polation of backmixing data from small
to large equipment. Gross flow pat-
terns could vary considerably; this
probably is the case between laboratory
and industrial-sized fluidized units with
their different degrees of bypassing of the
fluid in the form of bubbles. T h e
degree of backmixing may not be uni-
form throughout a reactor because of
entrance effects, nonuniform cross sec-
tion, etc. This may be dealt with by
using a n average D/uL or going directly
to the distribution functions from which
the D values are obtained.
Nomenclature
a = defined by Equation 9, dimen-
sionless
= reactant
C = concentration of A , moles A /
cu. ft.
CO = initial concentration of reactant
d, = particle diameter in a packed
bed, ft.
d, = diameter of pipe or tubular
reactor, ft.
D = longitudinal dispersion coeffi-
cient or the effective axial
diffusion coefficient, sq. ft./
hr.
D, = molecular diffusion coefficient,
sq. ft./hr.
Dlud = local longitudinal dispersion
number, a reciprocal Peclet
group for mass transfer, di-
mensionless
D/uL = reactor longitudinal dispersion
number, a reciwocal Peclet
group for mass transfer,
dimensionless
f = 1 - X, fraction of reactant A
remaining, dimensionless
F = feed rate of A into reactor,
moles A/hr.
= superficial mass velocity, 1b.l
Go
(hr.) (sq. ft.)
k = reaction rate constant,
moles A l--rr
(2T) hr,-l
1 = axial distance from entrance of
reactor, ft.
L = length of pipe or reactor, ft.
n = order of reaction as defined in
Equation 3
1’ = reaction rate, rate of disappear-
ance of reactant A , moles A /
(hr.) (cu. ft.)
m = F/VCo space velocity, hr.-I
U = average velocity of flow, ft./hr.
V = volume of reactor, cu. ft.
X = fraction of reactant A converted
into product, dimensionless
e = porosity, dimensionless
J
! = viscosity, lb./(ft.) (hr.)
= density, lb./cu. ft.
P
SUBSCRIPTS
0 = nonbackmix, slug or plug flow
situation
= complete backmix or stirred
tank situation
literature Cited
(1) Carberry, J. J., Bretton, R. H.,
A.I.CI1.E. Journal 4, 367 (1958).
(2) Danckwerts, P. V., Chem. Eng. Sci. 2, 1
(1953).
(3) Deisler, P. F., Jr., Wilhelm, R. H.,
IND.END.CHEM.45, 1219 (1958).
(4) Ebach, E. A,, Ph.D. dissertation,
University of Michigan, Ann Arbor,
1957.
(5) Jacques, G. L., Vermeulen, T., U. of
Calif., Berkeley, Rept. UCRL-8029 (No-
vember 1957).
(6) Kramers, H., Alberda, G., Chem. Eng.
Sci. 2,173 (1953).
(7) Levenspiel, O., IND.ENG.CHEM.50,
343 (1958).
(8) McHenry, K. W., Jr., Wilhelm, R . H.,
A.1.Ch.E. Journal 3, e3 (1957).
(9) Smith, J . M., “Chemical Engineering
Kinetics.” Chap. 11, McGraw-Hill,
New York, 1956.
(10) Taylor, G. I., Proc. Roy. Soc. 219A,
186 (1953).
(11) Wehner, J. F., Wilhelm, R. H.,
Chem. Eng. Sci. 6,89 (1956).
RECEIVED
for review May 1, 1959
ACCEPTED August 28, 1959