A REVISION OF MULTIPLICITY AND

PARAMETRIC SENSITIVITY CONCEPTS IN NONISOTHERMAL

NONADIABATIC PACKED BED CHEMICAL
REACTORS

J. PUSZY&KI,t D. ANITA. V. HLAVitEK$ and H. HOFMANNO

Department of Chemical Engineering, Institute of Chemical Technology. 166 28 Prague 6, SucbMtarova 5,
Czechoslovakia

(Received 1 September MO; accepted ZS February 1981)

Ab&nct--One-phase models including axial as well as axial and radial dispersion effects were analyzed. Reliable
numerical methods based on the finitedifference approach and the Newton-Raphson method were proposed to
solve these models. It was shown that two stable steady states may exist in very long nonadiabatical packed bed
reactors. A relation between parametric sensitivity and multiplicity was discovered. Practical conchsions im-
portant for design of industrial nonadiabatical packed bed reactors are drawn.

1. INTItODUCTION effects are negligible provided the length of the bed is

General mathematical models of packed bed reactors long enough. Hlav&Eek et al. calculated the regions of
may be very complex if all important transport processes multiple steady states for an adiabatic[lO] and a
have to be considered. It is known that in certain cir- nonadiabatic 1l] reactor. They pointed out that multi-
cumstances a tubular packed bed reactor, where a highly plicity may exist only for lower values of Peclet number.
exothermic reaction takes place, becomes extremely Although there are many papers dealing with both
sensitive to small changes in operating conditions. A parametric sensitivity and multiplicity in chemical reac-
slight perturbation of inlet and/or system tors, the relation between parametric sensitivity and
parameters may result in excessive hot spot tem- multiplicity has not been obvious so far. In this paper we
peratures. Bilous and Amundson[l] referred to this wish to reexamine the question of the importance of
effect as to the parametric sensitivity. Barkelew [2] and axial dispersion in nonadiabatic reactors. The goal of this
Dente et n1.[3] performed extensive numerical cal- paper is (i) to devise a method for calculation of long
culation of parametric sensitivity for a one-dimensional packed bed reactors if axial dispersion is considered, (ii)
one-phase model. Agnew and Potter[4] extended the to find relations between multiplicity and parametric
Barkelews analysis to a twodimensional network model sensitivity, (iii) to indicate that multiple steady states
making use of the Deans-Lapidus[S] two-dimensional may exist also in very long packed bed reactors and (iv)
network of perfectly mixed stages. A review of methods to draw conclusions important for design of industrial
which can be used for determination of parametric sen- packed beds.
sitivity was published 161. For description of packed beds
dispersion may be important. Frequently, because the 2 RI!XCTfJRMODEL EQUATlONS
genera1 model including axial as well as radial dispersion Consider the equations governing a packed bed
has been considered too formidable for numerical catalytic reactor under the assumptions of constant phy-
solution[7], the dispersion effects have been analyzed sical properties. After adopting the idea of smooth vari-
independently. The model considering only axial dis- ations in bed the dimensionless differential equations are:
persion effects was analyzed by Carberry and Wendel[E],
and others[9-111. The model including only radial dis-
persion phenomena was investigated by Finlayson[lZ].
Valstar[l3] and Hla&ek[6]. The general model con-
sidering both the axial and radial dispersion processes (1)
was analyzed only by Young and Finlayson[l4].
For adiabatic reactors Carberry and Wendelf and ae
AT-=- 1 a% a% 1 ae
HlavLEek and MarekI91 indicated that axial dispersion &- Pe;? +& (z+xG >
e
-E+DaBf(y)exp - . (2)
ton leave from Dept. of Chemical Engineering, Technical az L+d
University of Wroclaw, Wroclaw, Poland. When there is no radial dispersion in the inlet and outlet
*Author to whom correspondence should be addressed.
Present address: Dept. of Chemical Engineering, Catholic Uni- regions the boundary conditions for &=0 and f = 1
versity of L.euven, de Croylaan 2. B-3030 Leuven (Heverlee), reduce to Danckwerts boundary conditions [ 141:
Belgium.
8Iustitut fiir technische Chemic I, Universitjit Erlangen-Mm-
berg, 8520 Erlangen, Egerlandstrasse 3, FRG.

1605
1606 J. Pusz~tb~r et 01.

z=*.aY=O ?!!=-3i(~-(+)c ANALYSIS OF MULTlPLIClTY AND PARAMETRIC SENSITMTY

* a2 a2 (i) Parametric sensitivity and multiplicity
The parametric sensitivity concept has been used
extensively in the titerature for a safe design of tubular
nonadiabatic packed bed reactors, however, the relation
ay ae parametric sensitivity vs multiplicity has not been
.f= k5=z=0.
obvious so far. The key for understanding this relation is
the axial dispersion process.
This model will be called a general model. After neglec- The importance of axial dispersion depends on the
ting the axial dispersion term (i.e. Pe,.= = Pe,+m) a type of reactor, it is well known that for isothermal[l9]
radial dispersion model results. Finally, on neglecting the and adiabatic191 long enough reactors axial dispersion
radial dispersion term (i.e. Pe, = PeTem) an axial dis- effect can be neglected since the exit concentration ap-
persion model is obtained. The plug flow model (also the proaches equilibrium concentration. For a nonadiabatic
one-dimensional model) is represented only by con- packed bed reactor it was also proposed to omit this term
vection terms. because for large L is PeyX and Pe;-+m[19]. This
conclusion has found acceptance by others[l3,16] and
SOLUTION OF EQUATIONS thus a generally accepted criterion for neglecting the
The general model is described by a coupled set of axial dispersion term was based on the reactor length.
nonlinear elliptic partial differential equations. A For a very common case of a strong exothermic reac-
numerical solution of a set of elliptic partial differential tion taking place in a long packed bed the derivatives
equations has been considered to be a formidable may be extremely steep and hence the terms
task[7]. However, recently some methods have been
devised to circumvent the inherent difficulties. Young de d*@ dy dy
and Finlayson [ 141 pointed out that the double collocation - F5 -pz$ - and w
P
method is very efficient. According to Finlaysons Pe; PeTX Pe,
recommendation six interior collocation points in both
axial and radial directions are sufficient. The resulting set are of the type m/m. Evidently, these terms cannot be
of nonlinear finite-difference equations can be solved by neglected and for a rigorous analysis must be considered.
the Newton-Raphson technique. The false transient Since an analytical approach via the singular pertur-
method was also proposed to solve these elliptic bation technique can be very complicated we have per-
equations[S]. formed the analysis of this problem numerically. As a
Here the method suggested by us recently has been test example we have chosen the CO oxidation reaction.
:dopted[16]. Radial derivatives in eqns (I) and (2) have The correlations, parameter values, etc. are reported in
been replaced by the orthognoal collocation formulas. Table 1.
The resulting set of nonlinear ordinary differential equa- In Fig. 1 a dependence of exit conversion vs inlet
tions (boundary value problem) has been approximated temperature is drawn for L =0.05-2.4m. This figure
by three-points finite difference approximations. After an indicates that the region of multiple steady states does
appropriate ordering of equations and variables a band not disappear even for L = 2.4. For this reactor length
matrix results, i.e. we may easily apply the Newton- PeYX= 2000. This figure reveals that the ignition and
Raphson method combined with an efficient band matrix extinction temperatures do not change for L = 0.6 m and
algorithm. For practical calculation two interior collo- L = 2.4 m. (The ignition temperature is drawn also by a
cation points were used for approximation of radial dashed line, in the picture y(l)crTO we cannot visually
derivatives while in the axial direction 150 points were distinguish the ignition point!) The axial temperature
employed. profiles corresponding to different positions in the hys-
The equations representing the axial dispersion model
can be integrated backwards by a shooting technique, Table 1. The values of the variables and parameters
however, for higher values of Peclet numbers serious (onedimensional model)
difficulties occur [l 11.For Pe > 500 the shooting method
is apparently not suited for practical calculations and a G, = 9.26 lo-* kg m-* WC-
finite-difference approach must be used. Depending on Y& = 0.03
T,E(40;130C),T,=T,,B,=O
the value of the Peclet number we have used different B E{20; 10)
number of mesh points ranging from 150 to 900. For yE(20; 15)
these calculations a double precision arithmetics is
necessary (the calculations have been performed on the L[m] Pe, Pe$ Da( To) B(G)
digital computer ICL-7214). The finite difference method 0.05 39.66 3.27 0.18- 18.0 1.06 - 0.93
was successful also for very high values of Pe(Pe== 0.15 118.98 9.81 0.54 - 54.0 3.2 - 2.9
2000). Apparently the multiple shooting method can be 0.60 475.92 39.24 2.16-216.0 12.8- 11.6
superior to the finite-difference approach[l7]. For a sys- 2.40 1903.7 157.0 8.64-864.0 52.5-46.0
tematic analysis of a particular problem we feel that the rcO = 366.4exp ( - 63% .9/ T)f(y)
selection technique should be K?o(l -Y)
adaptive mesh f(Y) = 1+ 105.1(1- y)Y&
usedl25,26].
 A revision of multiplicity and parametric sensitiviey concepts 1607

Fig. 2. Temperature profiles near the extinction point (figures

indicate a position in the hysteresis loop, see Fig. 1).

Fig. 3. Temperature profiles near the ignition point (figures

indicate a position in the hysteresis loop, see Fii. 1).
To. 72
Fig. 1. Dependence of exit conversion on inlet temperature. temperature difference between the extinction and igni-
Regions of multiplicity; ----, ignition and extinction. tion point amounts to 40C. This observation is
extremely important for control of industrial packed bed
teresis loop are drawn in Figs. 2 and 3. From these reactors. For certain type of catalysts the jump into a
figures may be inferred that the shape of temperature diGsion regime need not result in a deactivation of the
profiles calculated for different lengths is almost catalytic packing. To adjust the kinetic regime again it is
identical, only near the exit they are modtied by the exit necessary to lower the inlet temperature below the value
boundary condition. The temperature profiles 8 and 8* of the extinction point. This hysteresis behavior of long
drawn in Fig. 3 indicate that for a long packed bed we nonadiabatic packed beds has been experimentally
can have either the kinetic regime 8 or a diffusion regime observed[ZO].
8. Both regimes result in an almost complete exit con-
version. In the kinetic regime the reaction proceeds (ii) Behavior of a general model
almost isothermally and the hot spot temperature is very In a very important paper Young and Finlayson[l4]
low. For the diffusion regime the reaction occurs at the indicated that the experimental data presented by
reactor intet and a high hot spot temperature results. Schuler et a1.[21] cannot be reconciled with a model
After increasing the inlet temperature (To> 110C) the excluding axial dispersion. They have developed a cri-
kinetic regime cannot exist anymore and only the terion making it possible to predict situations when axial
diffusion regime must be expected. For the axial dis- dispersion is important in nonadiabatic packed bed reac-
persion model this jump is interpreted as an ignition tors. However, they did not test this criterion in the case
temperature. For the plug flow model this process is of multiple steady states.
referred to as the high parametric sensitivity. Obviously Here we wish to reexamine the question of the im-
in the kinetic regime the values of the first and second portance of axial dispersion in nonadiabatic reactors
derivatives of concehtration and temperature are low and where the effects of both axial and radial dispersion are
the piston flow description describes very precisely the considered. The data used for calculation are reported in
situation. As soon as the plug flow model indicates a high Table 2.
parametric sensitivity region the axial derivatives are In Fig. 4 a dependence of exit conversion on inlet
very large and as a result the dispersion term must be temperature is drawn. We may note that the prediction
included. The plug flow model describes reasonably the by all three models result in an almost identical ignition
low parametric sensitivity regime (i.e. also the kinetic temperature. The values of extinction are different for
regime), however, is not capable of predicting the hys- both models, the general model results in a much lower
teresis phenomena. The hysteresis lqop exists also in extinction temperature because of overheating the in-
very long reactors, for the particular case of Fig. 1 the ternal part of the reactor. A comparison of radial tem-
1608 J. PuszYljsKtet al.

Table 2. The values of the parameters and variables (two-

I
I
dimensional model)t I
I I

i :
Pe, = 227.86
Pe$ = 20.0 4oc
Bi- 3.0
R = 0.0115 m
L=O.lSm
rc=I_jR cl

tThe values of other parametersare presented in Table I. ;

2oc

Fig. 5. Radial temperature profiles (drawn for axial hot spot

temperature position). A.. ., inlet temperature TM; B.. ., inlet
temperature ToB.

I I ,
100 r,, Oc
OJ 5o ft
4 O8

Fig. 4. Dependence of exit conversion on inlet temperature.

Region of multiplicity: ----, general model; -v-., axial dispersion
model;...., piston flow model.

perature profiles for a diffusion regime which occur near

the extinction and ignition points (see TOA and ToB in Pig. 6. Temperature and conversion profiles in a nonadiabatic
Fii. 4) are dr;iwn in Fig. 5. The profile B is more or less bed. Diffusion regime. (GO= 9.26 lo-* kg me2set-I, To = IWC,
parabolical while the profile A exhibits a deformated 3%CO.).
shape because of overheating in the axis. We may note
that for the profile A the radial temperature difference is
much higher than the adiabatic temperature rise. This
fact is caused by different values of the Peclet numbers
for heat and mass. A detailed information on axial and
radial profiles is shown in Figs. 6 and 7.
Based on these results a conclusion may be drawn that
the ignited profiles occurring within the hysteresis loop
must be calculated by means of a general model. For
these conditions the radial dispersion model cannot pre-
dict multiplicity of steady states. The axial dispersion
model, because of nonparabolic radial temperature
profiles, does not yield reliable results, too.
Fig. 7. Tempbrature and conversion proliles in a nonadiabatic
DISCUSSION
5. CONCLUSION hND bed. Diffusion regime. (Go =9.26 kg rn- set-, To = 2(PC, 3%
The object of the present paper has been to establish a CO.).
relation between multiplicity and parametric sensitivity
in packed bed nonadiabatic reactors. It has been shown neglected only after a careful deliberation. For design of
that multiple steady states may occur also in very long industrial packed bed nonadiabatic reactors following
tubular packed bed reactors. The parametrically sensitive procedure is proposed.
regime indicates that two stable steady states in a (i) Solve the one-phase one-dimensional model and
catalytic reactor may be expected. The important term test the region of parametric sensitivity by the Bar-
representing axial heat and mass dispersion can be kelews method[2] or by a procedure based on varia-
 A revision of multiplicity and parametric sensitivity concepts
tional variables [6]. For a reactor operating in a regime of
low parametric sensitivity the one-dimensional one-phase
description yields satisfactory results.
(ii) For regimes having higher values of parametric
sensitivity use the axial dispersion model. This model
makes it possible to calculate the hysteresis phenomena.
If hysteresis loop does not occur the radial dispersion
model must be used for calculation.
(iii) The value of the extinction point temperature
calculated by means of an axial dispersion model, if
necessary, must be revised by a general model.
Frequently, the value calculated from the axial dis-
persion mode1 is not reliable. This procedure is necessary
only if a jump in a diffusion regime does not result in a
catalyst damage and a re-adjusting of a kinetic regime
operation is possible.
For calculation in the regime of the low parametric
sensitivity the one-phase model yields reliable results.
Recently this conclusion has been experimentally
verified by Emig et a1.[22]. It has been shown
theoretically[23] that gas-to-solid heat and mass transfer
does not play a very important role in the reactor design
and that for long tubular reactors the agreement between
the one-phase and two-phase models is satisfactory. The
gas-to-solid heat and mass transfer may be important, of
course, in the diffusion part of the hysteresis loop_
However, the two-phase general model has not been
analyzed so far.
The possibility of existence of multiple steady states in
nonadiabatic long tubular reactors is a typical feature of
wall cooled exothermic fixed bed reactors. Of course, the
occurrence of this effect is influenced by a number of
factors as: (i) intensity of cooling, (ii) contact time, (iii)
magnitude of activation energy and heat evolution, and
(iv) reaction kinetics.
For adiabatic [ lo] and quasiadiabatic [l l] conditions the
multiple steady states disappear, for very low and long
contact time low temperature and high temperature
regimes exist, respectively. High values of the activation
energy and the reaction heat support the multiplicity
effects. As a result; for hydrogenation reactions which
are characterized by low values of the activation energy
the effect of multiplicity is suppressed. The effect of the
reaction kinetics may be very important. We may expect
the same trends as for an adiabatic conliguration[lO], for
the second order reaction the phenomenon of multi-
plicity is suppressed, while the Langmuir-Hinshelwood
expressions (where the bounds are the zero and the first
order reaction) enlarge the region of multiplicity.
Recent investigation has indicated that no gross
backmixing effects occur in the fixed bed even at high
Reynolds numbers[27]. Thus the addition of the second
derivative term to allow for dispersion creates
difficulty, since such mode1 does predict backmixing.
Amundson and Aris pointed out[27] that for the iso-
thermal conditions the cell model has all desirable pro-
perties. As a result, a model of a fixed bed may be
represented by a discrete description for axial mass
dispersion and quasicontinuum for axial heat
transfer[28]. Since in this description, which seems to be
in a qualitative ameement with a number of unexplained
1609
phenomena, the axial penetration of mass is suppressed
we may expect (see the results in eqns 29 and 30) that the
region of multiplicity grows up essentially.
Our calculations revealed that the ignition point tem-
peratures may be well approximated by a ptug flow
model and that for these conditions the results from the
axial dispersion model and genera1 model agree satis-
factorily. Since the general model should correspond
to the Deans-Lapidus description a disagreement be-
tween experiments reported by Emig et oL[22] and
Agnew and Potters diagrams[4] indicates that also
the classical Deans-Lapidus description of the
packed bed[5] must be used with care. This finding is in
accord with our former observations that the classical
Deans-Lapidus. model must be improved by an axial
interaction term between neighbouring cells [24] and that
different mechanism for heat and mass transfer must be
accounted for [30].
REFERENCES
fll Bilous 0. and Amundson N. R., A.1Ch.E.J. 19562 117.
[21 ;ykelew C. H., Chem. Engng Prog. Symp. Ser. 195955 (25)
131Dek hf., Cappclli A.. Ferraris G. B. and Collins A., Ing.
Clrinr. Ital. 19662 5, 19662 47, 1966 2 55 (in Italian).
[41 Agnew J. B. and Potter 0. E., Trans. I.Ch.E. 196644 T216.
[5] Deans H. A. and Lapidus L., A.I.Ch.E.J. 1960 6 656.
[6] Hlavtiek V., Ind. cngng Chem. 197062 8.
[7] Lapidus L., Chem. Engng Prog. Symp. Ser. 1961 57 (35) (36).
IS] Carberry J. J. and Wendel M., A.LCh.E.J. I%3 9 120.
[91 Hla&ek V. and Marek hi., C/tern. Engng Sci. I%6 21501.
[IO] HlavaEek V. and Hofmann H., Chem..Engng Sci. 1970 25
187.
(111 Kubitek M., Hofmann A. and Hlav&ek V., Cbem. Engng
Sci. 197934 593.
[I21 Finlayson 3. A. Chem. Engng Sci. 197126 1OSt.
[19 Valstar J. hf.. PhD. Thesis, Technische Hogeschole, Delft,
Netherlands 1969.
[14] Young L. C. and Finlayson A., Ind. Engne Chem. Fundls.
1973 12 412.
[151 Zasmolin A. V., Jablonskij G. C. and Beskov V. C., Theor.
Found. Chem. Techn. 19682 237 (in Russian).
[16] HlavGek V. and Votruba J., Steady Skate Operution of
FTTedBed Reactors and Monolithic SIructures in Chemical
Reactor Theory. A Review (Edited by Lapidus L. and
Amundson N. R.). Prentice Hall, Euglewood Clilfs, New
Jersey 1977.
[17] HlavBZek V., Unpublished results.
[18] Carberry J. J., Can. /. Chem. Engng 1958 36 207.
[19] Beek J., Advan. Chem. Engng 1%23 303.
[20] Puszyriski J., Anita D. and Hlav&?ek V., Chem. Engng Sci. In
press.
[2ll Schuler R. W., Stallings V. P. and Smith J. M. Chem. Engng
Prog. Symp. Ser. 1954 48(4)19.
[22] Emig G., Hofmann H., Hoffmann U. and Rand V.. Chem.
Engng Sci. I980 35 249.
I231 Holodniok M., Hlav&k V.. KubiiLek M. and Hofmann H..
Chem. Engng Sci. In press.
WI Sinkule I.. HlavBZek V.. Votruba J. and Tvrdik I., Chem.
Engng Sci. 197429 689.
a [25] Denny V. E. and Landis R. 8.. J. Cornput. Phys. 1972 9 120.
[26J Russell R. D. and Christiansen J., SIAM J. Numer. Anal.
1978 t5 59.
[27] Sundaresan S., Amundson
N. R. and Aris R.. A.LCkE.J,
198026 529.
[28] HlavsEek V. and van Rompay P.. Chem. Engng Sci.
1981% 1587.
[29] Hlavdeek V., Hofmann H., Votruba J. and KublEek M..
Chem. Engng Set 1973 28 1833.
[30] Sinkule J., Hlav&Zek V. and Votruba J., Chem. Eagng Sci.
1976 31 31.