NUCLEAR SCIENCE AND ENGINEERING: 90, 40-46 (1985) A Resolution of the Stiffness Problem of Reactor Kinetics Yung-An Chao* and Anthony Atard* Carnegie-Metion University, Carnegie Institute of Technology Department of Mechanical Engineering, Schenley Park, Pittsburgh, Pennsylvania Received April 30, 1984 Accepted October 19, 1984 Abstract—The stiffness problem in reactor kinetics is overcome by the stiffness confinement method Jor solving the kinetic equations. The idea is based on the observation that the stiffness characteris ‘ic is present only in the time response of the prompt neutron density, but not in that of the delayed neutron precursors. The method is, therefore, devised to have the stiffness decoupled from the dif- ferential equations for precursors and confined to the one for the prompt neutrons, which can be ana: Iptically solved. Numerical examples of applying the method to a variety of problems confirm that the time siep increment size can be greatly increased and that much computing time can be saved, as compared to other conventional methods. The theory is of general validity and involves no approxi ‘mation other than the discretization of the time variable. u solution au probleme de la “raideur” en cinétique des réacteurs Résumé— Le probleme de la “raideur" en cin ique des réacteurs est maitrisé par la méthode dite de “confinement de ta raideur" (Stiffness Confinement Method, SCM) pour la résolution des équations cinétiques. Lidée est basée sur Vobservation du fait que la caractéristique de “raideur” est présente seulement dans la réponse temporelie de la densité en newirons prompts, et non dans celle des précurseurs des neutrons retardés. C'est pourquoi la méthode est consiruite de telle fagon que la raideur soit découplée des équations differentielles des précurseurs, et limitée a celle des neutrons prompis, qui peut étre résolue analytiquement. Des exemples numériques d'application de la méthode une varieté de problemes confirment que la longueur de Vinerément de tenips peut étre grandement auginentée et le temps de calcul notablement reduit, en conparaison a d'autres méthodes, plus con- ventionelles. La théorie est de validité générale, et wimplique aucune autre approximation que celle de la diserétisation de la variable temps. F Zerlegung des Steitigkeitsproblems in der Reaktorkinetik mmenfassuing — Das Steifigkeitsproblem in der Reaktorkinetik kann durch das Verfahren der Steifigkeiisbeschrankung (Siiffness Confinement Method SCM) zur Losung der kinetischen Gleichun gen tiberwunden werden. Das Konzept geht von der Beobachtung aus, dah die Steifigkeitscharak- feristik nur in ler Zeit-Anbwortfunktion fir die Dichte prompier Neuironen aufirit, jedoch nicht fir die Vorliuferkerne verzogerter Neutronen. Des Verfahren ist daher so angelegt, dap die Steifigkeit von den Differentialgleichungen fiir Vorlauferkerne entkoppeli und auf die Differentialgleichung fir die prompten Newironen beschrankt wird, die dann analytisch gelost werden kann. Zahlenbeispiele fiir die Anwendung des Verfahrens zur Losung einer Vielfalt von Problemen bestatigen, daf die Zeit sohrittweite im Vergleich zu anderen klassischen Verfahren betrdchulich vergréspert und Rechenzeit im groferen Unfange eingespart werden kann. Die Theorie ist allgemeingiltiz, und abgesehen von der Diskretisierung der Zeitvariablen sind keine Approximationen erforderlich. *Permanent address: Westinghouse Electric Corporation, P.O. Box 355, Pittsburgh, Pennsylvania 15230. F 40 anent address: Westinghouse Bettis Atomic Power Laboratory, P.O, Box 79, West Mifflin, PennsylvaniaSTIFFNESS PROBLEM OF REACTOR KINETICS 4 1. INTRODUCTION A well-known, long-standing problem in reactor kinetics is the stiffness arising from the orders of mag- nitude difference between the prompt and delayed neutron lifetimes, which results in the restriction of, very small time step increments in numerical solutions to the Kinetic equations. There have been a number of methods suggested to avoid the difficulty, but they generally involve some approximations or apply effe tively only to certain types of problems. There have also been considerable efforts in developing general purpose numerical integrators for ordinary differential equations that use variable time steps to accommodate sliffiness.* In this paper, we place more emphasis on the physical properties of the kinetic equations and point out that through a transformation of the equa- tions, the stiffness can be decoupled altogether from the precursor equations and be confined only 0 the single prompt neutron equation, This reformulation of the kinetic theory involves no approximation and is of general validity. We call it the “stiffness confinement method” (SCM), Stiffiness is a basic problem of reactor kinetics, be it point kinetics or space-time kinetics. It is, of course, much more convenient to analyze the problem using, the point kinetic equations, and then generalize it to space-time kinetics, In this paper, we use the point kinetic equations to develop the SCM. The usefulness of the method for practical applications is, however, in space-time kinetic analysis and real-time simulation ‘The basic idea of the method, discussed in detail and demonstrated with a simple analytic example in Sec, IH, is based on the observation that the stiffness char- acteristic is present only in the time response of the prompt neutron density but not in that of delayed new- tron precursors. In Sec. III, we present the formula- tion of the SCM, which decouples the stiffness from, the differential equations for precursor densities and, confines it to the equation for the prompt neutron, density. The precursor densities are first evaluated with the stiffness decoupled precursor equations, and then plugged into the differential equation for the prompt neutron density. The latter equation does contain the stiffness, but can, nevertheless, be analytically solved for the prompt neutron density, given the precursor densities. Numerical examples of applying the method to a variety of problems are presented in Sec. IV. Itis shown that the method can allow much larger steps in, time increment and thus save much computing time as, compared to other conventional methods, Section V cfly summarizes and discusses the outlook of the alization of the SCM to space-time kinetics. IL BASIC IDEAS To clearly explain and demonstrate the basic idea, we confine the discussion throughout this section to the very simple example of a positive step reactivity insertion with only one group of delayed neutron precursor, The analytic solution to the kinetic equa- tions a an (lay and d 5 (ab) can be easily obtained. It is straightforward to show that if p is not very close to 8, the solutions can be simplified to nue) wos) - gee or(?52 4)» (2a) and st) = ego) + exo curs e081) + GP exo) « (2b) which are valid for both p< and p > 8. The general behavior of the functions (1) and ¢(#) is shown in Fig. 1 for the case of p <8. It is important to observe thal there is an abrupt turn in 1(), from a very steep initial slope p/A to a much milder asymptotic slope, reflecting the stiffness characteristic due to the orders of magnitude difference between the time constants A and 1/ present in Eqs. (1). However, the same behav- ior is not exhibited in c(¢), which rises rather smoothly with ¢. This suggests that the function (7) is not sensitive to the stiffness of the problem. This phenomenon can be explained in the following way. If the initial value of 7 is normalized to unity, the initial equilibrium value of c is 6/AX. Although the initial increase in m is very sharp so that the magnitude of 67 may be comparable to one aver an initial time inter~ val, the corresponding initial increase in ¢, dc = én, is nevertheless extremely small compared to 6/AX. ‘Therefore, ¢ increases rather smoothly and can hardly c(t) fe i t t Fig. 1. Responses of the prompt neutron density and the delayed neutron precursor density to a step reactivity insertion,2 CHAO and ATTARD. sense the stiffness as n does. This strongly’ suggests the possiblity of devising a method to avoid the difficulty of stiffness in the evaluation of e(1) and confine it to that of (7) A well-known approximation to avoid the stiffness is the prompt jump (PJ) approximation, which corte sponds to the ascending component of Eq. (2a), as shown in Fig. I. A popular way of deriving the PJ approximation isto assume that dn/dt << [(p ~8)/A]n because of the smallness of A, so that Eg. (1a) reduces to de ; BA * 2 which, when substituted for (1) of Eq. (1b), leads to es al Bap @ In this equation, the prompt neutron lifetime disap- pears, and so does the stiffness. Solving Eq. (4) for py and substituting it for c in Eq. (3), one finds 6) which is exactly the first term of the right side of Eq. (2a), namely, the PJ component. In this PS ap: proximation, both rp, and cp, rise smoothly in time without any abrupt change, indicating the absence of stiffness, which is, however, compensatory with the penalty Of ignoring the initial response of n(¢). Since c(t) is not sensitive to the stiffness, however, ¢py() is 4 good approximation to e(1) throughout all values of {, including small ¢ as well. Therefore, if instead of Substituting cp) for ¢ in Eq. (3), we substitute it for ¢ in Eq, (1a) and solve Eq. (1a) for n(r), we should expect an improvement on mips in the small f region, Remarkably, the result turns out to be exactly Eq. (2a). This is @ good example showing that, indeed, it is possible to decouple the stiffness from the evaluation of e(¢) and confine it to that of m(1), while still re- covering the complete solution for 1(). It is well known that the PJ approximation breaks down for » 2 8, where Ea. (3) becomes absurd. How- ever, it is enlightening to see that in the above exam- ple this difficulty with the PJ approximation can be overcome by interpreting Eq. (3) in a different way ‘We note that it is always possible to express n(t) as an exponential explfw(0’) dt] and rewrite Eq, (La) as pase Ceeee wi) + (B= aya ©) where v(t) varies with time. Since Eq. (1) for this simple problem with one group of delayed neutrons is easily solved, the variation of w(r) can be explicitly analyzed. For p<, w(0) is negligible compared to (B=p)/A, except for the initial instant where it equals /A. If (0) is neglected in Eq, (6), it reduces to Ea. (3) The neglect of w(s) in the PJ approximation implies two assumptions: First, that w(!) <|p—8)/A for asymptotic f, and second that the initial transient around = 0, where w = p/A, spans only a very short, time interval. For p = 8 or p >, these assumptions are no longer valid, and w(¢) must be retained, It is, important to realize, however, that the rigorous differ ential equation restored from Eq. (4) by retaining wD, a. 8 ; ai‘ lawn +d—— |) wD is still stitiness decoupled. For 1=0, Aw(1) = p and the denominator of the equation is around 8, so that the characteristic time period is of the order of 1/3 ot longer, rather than A/p. For large 1, w(1) decreases from p/A to its asymptotic value, which depends on p. For p<, this asymptotic value of Aw() is negligible in the denominator of the equation that the asymptotic characteristic time period is again of the order of 1/2. When p=8, the asymptotic value of w(t) can be shown, by solving the inhour equation of Eq, (1) as (8X/A)"®, Substituting this for w(r) in Eq. (7) gives the asymptotic characteristic time period of the equa- tion as (A/3A)", which is still much longer than A/B. OF course, beyond the prompt criticality, the dominating time constant for c(t) in the asymptotic region eventually becomes 7p as well. But the use of Eq. (7) will still remove stiffness around = 0 and, as wwe will see, in practical applications it significantly speeds up the computation, even for p= 8 HL. SCM In Sec. II, we demonstrated the basic idea with the simple example of the PJ approximation to the step insertion problem. Here, we develop the SCM for the general case without any approximation, except for time discretization, We want to solve the following set of kinetic equations with an arbitrary reactivity function: 42-8 a A n+ One, (8a) and Me (8b) Introducing a set of “reduced” precursor density func tions ¢,(1) through cin =6eene|f wear © and defining tw. waxiliary functions w(s) and w(1), as in Eqs. (13) and (14), we can rewrite Eq. (8) as—B.___ ner away Boot = FMIGU (0 sw=[Zn4 ofero| [aera] . aD and d ou) Sn Hm ‘i 12) a TM + S10) (12) where w(1) is defined by d w= inn . 13 (= Finn (13) The function w(1) is defined in the same way as that of Eq. (6) and provides the key mechanism of the SCM. The function w(/), however, has nothing to do with stiffness decoupling and is not really required theoretically. Since an exponential behavior is often characteristic for first-order differential equations, however, a proper choice of u(t) may make ¢;(¢) vary more slowly in time and thus expedite the numerical calculation. We therefore choose the following u(¢) d w= “ins 4 (1) = GSW + aay where S(¢) is defined by Eq. (11) as the sum over all dest) Equations (10) through (14) form the complete set of kinetic equations for the SCM. The initial condi- tions to be satisfied are m0) =0, (say w(0) = 0(0)/A, (130) (0) = 20 + (15e) and (0) = m08,/A0, (isa) To solve the equations numerically, we start with setting w and w in Eq. (10) at their initial values and solve Eq. (10) for ¢; by discretizing the equation in Having obtained ¢;, we calculate S(1) with Eq. (11) and solve Eq. (12) analytically for n. Then, we use Eq. (13) to reevaluate w(s), plug it back into Eq. (10), and repeat the process until w converges (requiring, only two or three iterations). Calculation for the cur rent time step is then finished with an evaluation of the ‘output value of w and w via Eqs. (13) and (14). After ward, we predict the input values of w and w for the next time step by a linear extrapolation from their out- put values in the previous and current time steps, and repeat the whole process of calculation for the next time step. It should be emphasized that within each time step, there is an iteration to convergence on w but STIFFNESS PROBLEM OF REACTOR KINETICS, 3 no iteration on uw, because u is not required by the theory of the SCM and is, in principle, an arbitrary independent function chosen only to’ expedite the computation. The above procedure is devised to benefit two fea- tures. One advantage is that the discretized time step in solving Eq. (10) can be enlarged because of the decoupling of stiffness due to the use of w(t). The other advantage is that in solving Eq. (12), although the stiffness is present, the solution can be obtained analytically in an integral form within each time step where S(7) has been computed from Eqs. (10) and (11). Making the approximation (consistent to finite difference) that within each time step (1) and &,(1) vary linearly, this analytical integral solution can be explicitly evaluated in a closed form. It is also worth mentioning that due to the particularly simple struc: ture of Eq. (10), the matrix equation resulting from finite differencing Eq. (10) is analytically invertible. As a result, we not only have large time increment steps but very speedy computations within each time step also. To save space, the time-discretized equations used for coding are not explicitly presented here, but it is straightforward to derive them from Eqs. (10) through (14), IV, NUMERICAL TESTS We have tested the SCM with four types of prob: lems and compared the results against those obtained with other methods. The four types of problems are step reactivity insertion, ramp input, sinusoidal input, ‘and zigzag ramp input of reactivity. For step and ramp reactivity insertions, only cases of positive insertion are presented, because the ones with negative insertion are easy problems and can always be computed very ai rately. The examples are described separately. IV.A, Step Reactivity Insertion We consider a kinetic problem with the following parameters: A = 0.00002 s, 6 = 0.007, 8; = (0.000266, 0.001491, 0.001316, 0.002849, 0.000896, 0.000182), and A, (5!) = (0.0127, 0.0317, 0.115, 0.311, 1-4, 3.87). Four step insertions are considered: two prompt subcritical p = 0.003 and 0.0055, one prompt critical 0.007, and one prompt supercritical p = 0.008. The values of (1) obtained with the SCM are compared (see Table I) to those obtained with a code based on the so-called “Henry's d-weighting method,” which modifies finite difference equations by introducing tac- tically chosen weighting functions.? The numbers pre- sented in Table I are computed with variable time steps. Each time step is chosen as a fraction of the instantaneous period 7, except for the initial time step of the SCM. The initial time step of the SCM is cho sen to be much larger to emphasize the effect of stiff ness decoupling. We see that the SCM results are“ CHAO an ATTARD TABLE | Comparison of the SEM Against # Weighting for Step Reactive sertons Variable ar | nital Ar i ° Method 8 ‘© m) em [em [eas | 6 weighing 0 000333 eos) za0neet | 0.003 SCM id 0.01 8.0324 2 Xactr som | 2 vaas | tes | 4 weighing a0 cowie [sis | 420541 coos | SCN a on S087 | a a0aue EXACT S100 | 4.00501 gos | reoss | r=es | 6 yeighing 20 coos [aso | saaees | 2ceue oa | SeM a oo tse | Saswra | 207211 EXACT fame | suoea | ost gos | reoas | rats i 6 veigbing 0 oooors [ean | Lanse sec aoe | Sex a oo aoe | angers | Gisteas | Exact cao | tstonss | eae Read as 2.8076 x 10! good as or better than those of @ weighting even though SCM uses time steps much larger than those of 6 weighting IV.B. Ramp Input of Reactivity We consider two examples of ramp input —one extremely fast and the other moderately fast, The first example is taken from Ref. 4, where a comparative study is made on the application of two variants of the weighted residue method, of the linear and the quad: ratic trial functions, respectively. The response of reactor core at 0,001 s after a ramp input of reactiv~ TABLE Il Comparison of SCM Versus Weighted Residue Method for Extremely Fast Ramp 8 | ase Not available | ity at the rate of $100/s is calculated (with six groups of delayed neutron), In Table II, we compare the results of applying the SCM to this example to those given in Ref. 4. The superiority of the SCM is obvious, The second example is a $0.1/s ramp input to a reactor core. The values of physical parameters are the same as those of the step insertion examples of Sec. IV.A. The values of n(1) obtained with the SCM are compared (see Table 11}) to those obtained with the @- weighting method. Again the SCM works far better. IV.C. Sinusoidal Input of Reactivity The test example is taken from Ref. 1, where the point kinetic equations with one group of delayed neu- tron and a reactivity input posin(rt/T) are solved for a comparative study with three different methods — ‘Adler’s method, Hasen’s method, and the quadratic weighted residue method. The one-group delayed neutron parameters are \ = 0.077 s~! and 8 = 0.0079. Two prompt neutron lifetimes are considered, A= 10-8 and 10~? s, The reactivity amplitude po is varied from 0.001808 (23 cents) to 0.005333 (68 cents), and the half period Tis also varied correspondingly from 350 to 50s. The time interval size used for time dis- retization varies from 0.1 t0 5 s, It was concluded inSTIFFNESS PROBLEM OF REACTOR KINETICS 48 TABLE Il ‘Comparison of SCM Versus 9. W ighting for Moderately Fast Ramp + | @ Weighting @ Weighting | @ Weighting ScM scm | amit, | SMES | AEBS | ay (verte ane [> | a 13382 Ta388 Taam ram 0.90) 1 o4 | 2. 2.2286 2.2290 2.2284, 2.2267 (0.80) ¢| aa 33 sa ran (0 aa ye | BB | Si | lane rn 9 | 4.8816+2 4.878142 4.8698+2 (0.11) | a | shst3 torts sSoiss | ters G09 Here, A¢= min (7/10,1 8), initial A¢ = 0.1 s. The may parenthesis Read as 4.2781 x 10! Ref. I that the accuracy of the methods of Adler and Hansen tends to be more variant, while the weighted residue method works very well uniformly through all the cases studied. The weighted residue method gener- ally works the best except for the case of py = 68 cents, T= 50 s, and A=10~% s, where Adler's method works better ‘When SCM is applied to this example, we find that it consistently gives better results than all of the three methods for all the cases. For comparison, in Table IV we list the errors of the four different methods in their predictions on the peak power for the two cases of the smallest and largest pq considered in Ref. 1, using the largest time increment size given there, namely, Ar=5 and 0.5 s, respectively. The errors associated with the three methods other than SCM are estimated from graphs in Ref. 1 IV.D. Zigzag Ramp ‘The zigzag ramp is an example taken from Ref. 5, ‘where the Padé approximation is applied to compute the response of a reactor core to a zigzag ramp input of reactivity of $1/s up t0 0.5 s followed by ~S1/s up to I's, then $1/s again up to 1.5 s. From there it remains TABLE 1V Comparison of Errors of Various Methods for the Case ‘of a Sinusoidal Reactivity Input Weighted po | T | A | Adler | Hansen | Residue | SCM cay | | © |) | co | Cm | Cy 2 [3solio|-so] -os | +02 | -o1 23 |3s0}102 | -4.0| 03 | +402 | -o1 68 | sofi-| -o1) -0s | 406 | -o1 } 68 | sofio-} +01] -15 | +01 | -01 L : xximum Af (s) reached in the quoted time interval is shown in constant at $0.5. The physical parameters in this ex- ample have the following values: A = 0.0005 s, 6 = 0.0075, 8; = (0.000285, 0.0015975, 0.001410, 0.0030525, 0,0009600, 0.000195), and A; (s~') = (0.0127, 0.0317, 0.115, 0.311, 1.40, 3.87). The results obtained with the SCM are compared in Table V against those obtained by the Pade (1,1) (or Crank-Nicholson) method given in Ref. 4, These computations are done with a fixed time step size. We see that the SCM gives much more accurate results, particularly for large time increment steps V. CONCLUSION Recognizing the fact and exploiting the advantage that delayed neutron precursor densities are not sen- sitive {0 the stiffness of the kinetic equations, we have developed the SCM and applied it to a variety of prob- lems and compared it to a number of other methods, The SCM is proven to be consistently much better than other methods in all the cases. It not only can employ much larger time increment steps due to the stiffness confinement, but also computes rapidly for a given time step due (0 its completely analytic formu- lation. The SCM can also be applied straightforwardly to problems, including feedbacks, provided within each time increment step the feedback is linearized. Otherwise, the scheme of stiffness confinement still applies, but the mathematics becomes more involved because of the nonlinearity. The real power of SCM, though, lies in its application to real-time simulation and space-time kinetics, where computation time is still a practical problem. To generalize the method to space-time kinetics, the function w(t) in Eq. (13) becomes spatially dependent and enters the spatial dif fusion equation. The details of the formulation, how- ever, would depend on the additional problem of the type of method used to solve the spatial diffusion46 CHAO and ATTARD TABLE V Comparison of Percent Errors of SCM Versus Padé Approximation for a Zigzag Ramp Reactivity Input [are | Methoa $5 Iss 12208 0.08 0.01 SCM +3.48(—4) 6.60( 4) +5,28(—4) —5.16(—4) —3.32(-4) | Padé ) 3.50(=1) 70(—1) —4.28(~2) 5.10(—2) | 0.10 SCM 2 ~6.7%-2) | +5.23(-2) 5.04 -2) -3.40(—-2) "Read as —4,24 x 10 equation. It would only seem practical to use coarse. mesh methods, In fact, recent development in advanced nodal method is particularly promising in this aspect. We would defer the work to a later pub- lication, ACKNOWLEDGMENTS We would like to acknowledge the very enthusiastic participation of C. R. Conner and R. A. Reynolds at the early stage of the development of this work, It is through these laborious efforts that we were able to arrive at the correct conclusions. REFERENCES 1. D, L. HETRICK, Dynamies of Nuclear Reactors, The University of Chicago Press (1971), 2. C, W. GEAR, Numerical Initial Value Problems in Ordinary Differential Equations, Prentice-Hall, Inc., Engle wood Cliffs, New Jersey (1971) 3. T. A. PORSCHING, “The Numerical Solution of the Reactor Kinetics Equations by Difference Analogs: A Comparison of Methods,” WAPD-TM.S64, U.S. National Bureau of Standards, U.S. Department ‘of Commerce (1966), 4, J. J. KAGANOVE, “Numerical Solution of the One- Group Space-Independent Reactor Kinetics Equations for Neutron Density Given the Excess Reactivity," ANL-6132, Argonne National Laboratory (1960). 5. JAW. da NOBREGA, Nucl. Sci. Eng., 46, 366 (1970) 6. Y.A. CHAO, C, A. SUO, and J. A. PENKROT, “On the Theory of Interface Flux Nodal Method,” Proc, CNS/ ANS Int. Conf. Numerieal Methods in Nuclear Engineer ing, September 6-9, 1983, Montreal, Canada, p. 307 (1983).