is obtained by combining solutions of the reduced problem, an initial- value problem, and a terminal-value problem. Error estimates for approximate solutions ...
JOURNAL OF OPTIMIZATION THEORY AND APPLICATIONS: Vol. 99, No. 1, pp. 37-52, OCTOBER 1998
Initial-Value Technique for Singularly-Perturbed Turning-Point Problems Exhibiting Twin Boundary Layers1,2 S. NATESAN3 AND M. RAMANUJAM 4
Communicated by F. E. Udwadia
Abstract. The initial-value technique that was originally developed for solving singularly-perturbed nonturning-point problems (Ref. 1) is used here to solve singularly-perturbed turning-point problems exhibiting twin boundary layers. In this method, the required approximate solution is obtained by combining solutions of the reduced problem, an initialvalue problem, and a terminal-value problem. Error estimates for approximate solutions are obtained. The implementation of the method on parallel architectures is discussed. Numerical examples are presented to illustrate the present technique. Key Words. Singularly perturbed boundary-value problems, asymptotic expansions, boundary layers, turning points, exponentially fitted difference schemes, initial-value techniques, parallel computation.
1. Introduction
Singular perturbation problems with or without turning point(s) are of common occurrence in many branches of applied mathematics such as fluid mechanics, elasticity, quantum mechanics, chemical reactor theory, etc. A few notable examples are boundary-layer problems, WKB problems, modeling of steady and unsteady viscous flow problems with large Reynolds number, and convective heat-transport problems with large Peclet number. In particular, the following singularly-perturbed turning-point problem (TPP) 1The
authors thank the referees for valuable comments. first author acknowledges the Council of Scientific and Industrial Research, New Delhi, India for financial support. 3Research Scholar, Department of Mathematics, Bharathidasan University, Tiruchirapalli, India. 4Professor, Department of Mathematics, Bharathidasan University, Tiruchirapalli, India. 2The
37 0022-3239/98/1000-0037$15.00/0 © 1998 Plenum Publishing Corporation
38
JOTA: VOL. 99, NO. 1, OCTOBER 1998
received much attention in the literature because of the complexity involved in finding uniformly valid asymptotic expansions unlike nonturning-point problems:
where e is a small positive parameter and a, b,f are smooth functions such that a(x) has a simple zero at x = 0, that is,
This type of problem arises, e.g., as a linearized one-dimensional slice of a fluid flow problem having a region of recirculation. In Ref. 2, Abrahamsson derived a priori estimates for solutions of (1). O'Malley (Refs. 3-4), Roos et al. (Ref. 5), Wasow (Ref. 6), Watts (Ref. 7), Bender and Orszag (Ref. 8), Lagerstrom (Ref. 9), and Jindge (Ref. 10) have also studied qualitative aspects of these problems, namely existence, uniqueness, and asymptotic behavior of the solution. Farrell (Ref. 11) obtained a set of general sufficient conditions for a finite-difference scheme to be uniformly accurate for TPPs of the form (1) having a cusp-type boundary layer. Berger et al. (Ref. 12) modified the El-Mistikway and Werle scheme (Ref. 13) for TPPs. Kadalbajoo and Reddy (Ref. 14) presented a novel initial-value technique (IVT) for a class of nonlinear singularly-perturbed two-point BVPs for second-order ordinary differential equations (ODEs) without turning points. In their method, the original BVP is replaced by an asymptotically equivalent initial-value problem (IVP) for first-order ODEs, which is then solved by the Runge-Kutta method. In continuation of this work, Gasparo and Macconi (Ref. 15) used the IVT to solve a more general class of nonlinear singularly-perturbed BVPs without turning points. Further, Gasparo and Macconi (Ref. 1) used the IVT with a slight modification, and hence solved a class of linear and nonlinear singularly-perturbed BVPs without turning points. The asymptotic nature of the solutions of the singularly perturbed BVPs has been fully exploited by these authors in replacing the given two-point BVP with a terminal-value problem (TVP) and an IVP. The integration of these problems goes in opposite directions, but each problem can be solved independently of the other. In the present paper, we consider the BVP (1) subject to the following conditions:
JOTA: VOL. 99, NO. 1, OCTOBER 1998
39
Under these conditions, the BVP (1) has a unique solution exhibiting two boundary layers, one at x = — l and the other at x=l, and being smooth near x = 0 (Ref. 12). In Ref. 12, it has been proved that
where u is the solution of (1) and MO is the solution of the reduced problem
Further, it is mentioned in Ref. 12 that the number 1/2 in (3) is arbitrary. Also, since a(—1)>0 and a(l)0 and L 0 v(x)>0 for all xeD. Then,
Lemma 2.2. See Ref. 16. Consider the IVP (7). If u(x) is the solution of this IVP, then
To solve (7) numerically, we apply the exponentially fitted difference (EFD) schemes of Refs. 16-18 which are described below. (i)
Explicit EFD Scheme:
JOTA: VOL. 99, NO. 1, OCTOBER 1998
(ii)
41
Implicit EFD Scheme:
Remark 2.1. p i ( p ) is known as a fitting factor. The following lemma is the discrete analogue of Lemmas 2.1 and 2.2. Lemma 2.3. See Ref. 16. Let vi be a mesh function. Suppose that v0>0 and Lhpvi>0,_for all x i eD, p = 1 [ p = 2]. Then, vi>0 for all xieD. Moreover, for XjeD,
The following lemmas of Ref. 16 provide error estimates for the above EFD schemes. Lemma 2.4. See Ref. 16. Let u and ui be the solutions of (7) and (8). Then,
Lemma 2.5. See Ref. 16. Let u and ui be the solutions of (7) and (9). Then,
A well-known method of obtaining a higher-order approximation to the solution of an IVP for first-order ODEs is the Richardson extrapolation technique. Doolan et al. (Ref. 16) gave a necessary and sufficient condition for this technique to provide a higher-order approximation uniformly in e for the IVP (7). In fact, we have the following lemma. Lemma 2.6. See Ref. 16. Let u and uih be the solutions of (7) and (8). Then,
if and only if f/a is constant on D.
42
JOTA: VOL. 99, NO. 1, OCTOBER 1998
Motivated by the IVPs that arise in the IVT to be discussed in the next section, we consider the IVP
where a, b,f are smooth functions such that a ( x ) > a >0, xeD, and b(x) is bounded on D. Remark 2.2. From Lemma 2.2, it is obvious that the solution of (10) is bounded on D. Further, it may be noted that the IVP (10) is a perturbed problem of (7). The following theorems establish a relation between (7) and (10). Theorem 2.1. Let u and w be the solutions of (7) and (10). Then,
Proof.
Let
Then,
Lemma 2.2 yields the desired result. Consider the following explicit EFD scheme for (10):
where
Lemma 2.7. Let vi be a mesh function. Suppose that v 0 >0 and Lh3Vi>0 for all xieD. Then, vi>0 for all X i eD. Moreover, for XjeD,
JOTA: VOL. 99, NO. 1, OCTOBER 1998
Proof.
43
Since
the proof follows from Lemma 2.3. Remark 2.3. is bounded.
D
A consequence of Lemma 2.7 is that the solution of (11)
We have the following discrete analogue of Theorem 2.1, which follows from Lemma 2.7. Theorem 2.2.
Let ui and wi be the solutions of (8) and (11). Then,
Remark 2.4. It may be noted that the fitting factor p i ( p ) appearing in (11a) does not contain the coefficient b. Theorem 2.3.
Proof.
Let w and wi be the solutions of (10) and (11). Then,
By Theorem 2.1, Lemma 2.4, and Theorem 2.2, we have
Similarly, we have the following theorem. Theorem 2.4.
Consider the following implicit EFD scheme for (10):
where
Let w and wi be the solutions of (10) and (12). Then,
44
JOTA: VOL. 99, NO. 1, OCTOBER 1998
2.2. Initial-Value Technique. See Ref. 1. Consider the following singularly-perturbed second-order BVP:
where a, b, f are sufficiently smooth functions such that
Under these assumptions, the BVP (13) has a unique solution exhibiting a boundary layer of width O(e) at x = c. In order to get a numerical solution of (13) by the IVT, we proceed as follows (Ref. 1). Consider the following problems:
and
Then, we have
In this way, it is possible to obtain via (16) an 0(e)-approximation to u(x) by solving problems (14) and (15). These problems can be solved independently. Since the right-hand side of the ODE (15a) is zero, the IVP (15) can be solved by applying the Richardson extrapolation technique to yield a higher-order approximation (Lemma 2.6).
3. Error Estimates
Consider the BVP (13). We now state the maximum principle and uniform stability result for this BVP. Lemma 3.1. See Ref. 16. Let y be the smooth function satisfying Bcy(c)>0, Bdy(d)>0, and Ly(x)0, compute v02i by integrating (25) on [1/2, 1]. Step 4. For i>0, compute ui by
It is obvious that the parallelism is inherent in the above algorithm. Hence, separate processors can be assigned for Step 2 and Step 3, which lead to the following two-processor scheme. Parallel Scheme.
Task 1. Perform Step 2 on Processor P1. Task 2. Perform Step 3 on Processor P2. Task 3. Perform Step 4 by using the results of previous tasks. Remark 4.1. For a discussion of the reduced problem (4), one may refer to Pages 54, 59, 68 of Ref. 2 and to Refs. 3 and 5.
5. Numerical Examples
In this section, we present some numerical examples to illustrate the present method. For convenience, we take the interval [0,1] as the domain of definition of the ODE instead of [-1,1], and consequently x= 1/2 will be the turning point. The numerical computations have been carried out on the DEC Micro Vax II Computer system at Bharathidasan University, Tiruchirapalli, India. Example 5.1. Consider the following singularly-perturbed TPP (Ref. 9, p. 67):
JOTA: VOL. 99, NO. 1, OCTOBER 1998
49
The exact solution is given by
Here, u0(x) = 0; v01 and v02 are respectively the solutions of the following problems
and
The numerical results are presented in Table 1. Example 5.2. Consider the following TPP:
whose asymptotic expansion solution is given by
Table 2 presents computational results of the problem. Remark 5.1. The IVP (20) and the TVP (25) are solved here by the implicit EFD scheme (12). One may also apply the explicit EFD scheme (11) or the Richardson extrapolation procedure for the explicit EFD scheme (11). Our experiences with the above Examples 5.1 and 5.2 show that these schemes yield almost the same results.
6. Discussion In the literature, the initial-value technique has been applied to solve various singularly-perturbed BVPs for second-order ODEs without turning points as reported in Section 1 of this paper. Although we have not derived a similar technique for turning-point problems, we have used the IVT (originally developed for nonturning-point problems) to solve the present turningpoint problem, by suitably dividing the interval [-1,1] into three disjoint subintervals, namely [-1, -1/2], (-1/2,1/2), [1/2, 1]. That is, we have applied the IVT on the intervals [-1, -1/2] and [1/2,1]. To solve the IVP
JOTA: VOL. 99, NO. 1, OCTOBER 1998
50
Table 1. Numerical results for Example 5.1. Error
Numerical solution e = H=0.0001
0.0010 0.0020 0.1000 0.2000 0.2500
1.000000 0.818912 0.670644 0.549242 0.449835 0.368434 0.301776 0.247187 0.202481 0.165867 0.135880 0.000048 0.000000 0.000000 0.000000 0.000000
0.0E + 00 1.3E-04 2.3E-04 2.7E-04 2.9E-04 3.0E-04 2.9E-04 2.8E-04 2.6E-04 2.4E-04 2.2E-04 2.6E-07 5.3E-12 0.0E + 00 0.0E + 00 0.0E + 00
0.3000 0.4000 0.5000 0.6000 0.7000
0.000000 0.000000 0.000000 0.000000 0.000000
0.7500 0.8000 0.9000 0.9980 0.9990 1-10e
0.000000 0.000000 0.000000 0.000000 0.000048 0.135880 0.165867 0.202481 0.247187 0.301776 0.368434 0.449835 0.549242 0.670644 0.818912 1.000000
Nodal points 0.000
1e 2e
3e 4e 5e 6e 7e
8e 9c l0e
1-9e -8e
1-7e
1-6e 1-5e 1-4e 1-3e l-2e 1-le 1.000
Numerical solution e = H=-0.001
Error
Numerical solution 6 = H=0.00001
Error
1.000000 0.818749 0.670352 0.548855 0.449380 0.367935 0.301252 0.246656 0.201955 0.165356 0.135390 0.000046 0.000000 0.000000 0.000000 0.000000
0.0E + 00 1.3E-05 2.2E-05 2.7E-05 2.9E-05 3.0E-05 2.9E-05 2.8E-05 2.6E-05 2.4E-05 2.2E-05 3.4E-12 4.8E-16 0.0E + 00 0.0E + 00 0.0E + 00
1.000000 0.818733 0.670323 0.548816 0.449334 0.367885 0.301200 0.246603 0.201902 0.165305 0.135341 0.000045 0.000000 0.000000 0.000000 0.000000
0.0E + 00 1.3E-06 2.2E-06 2.7E-06 2.9E-06 3.0E-06 2.9E-06 2.8E-06 2.6E-06 2.4E-06 2.2E-06 3.3E-13 4.2E-17 0.0E+00 0.0E+00 0.0E+ 00
0.0E+00 0.0E+00 0.0E + 00 0.0E + 00 0.0E+00
0.000000 0.000000 0.000000 0.000000 0.000000
0.0E + 00 0.0E + 00 0.0E + 00 0.0E + 00 0.0E + 00
0.000000 0.000000 0.000000 0.000000 0.000000
0.0E+00 0.0E+00 0.0E+00 0.0E + 00 0.0E+ 00
0.0E+00 0.0E+00 0.0E + 00 5.3E-12 2.6E-07 2.2E-04 2.4E-04 2.6E-04 2.8E-04 2.9E-04 3.0E-04 2.9E-04 2.7E-04 2.3E-04 1.3E-04 0.0E+00
0.000000 0.000000 0.000000 0.000000 0.000046 0.135390 0.165356 0.201955 0.246656 0.301252 0.367935 0.449380 0.548855 0.670352 0.818749 1.000000
0.0E + 00 0.0E+00 0.0E+00 4.8E-16 3.4E-12 2.2E-05 2.4E-05 2.6E-05 2.8E-05 2.9E-05 3.0E-05 2.9E-05 2.7E-05 2.2E-05 1.3E-05 0.0E + 00
0.000000 0.000000 0.000000 0.000000 0.000045 0.135341 0.165305 0.201902 0.246603 0.301200 0.367885 0.449334 0.548816 0.670323 0.818733 1.000000
0.0E+00 0.0E+00 0.0E+00 4.2E-17 3.3E-13 2.2E-06 2.4E-06 2.6E-06 2.8E-06 2.9E-06 3.0E-06 2.9E-06 2.7E-06 2.2E-06 1.3E-06 0.0E+00
(20) and the TVP (25) that arise in this technique, we have used the EFD schemes, because they seem to be more appropriate to these problems as they are straightforward, provide good results, and take less time. The proposed method is easy to implement as one has to solve only an IVP and a TVP for first-order ODEs. Also, from the numerical results, the present method seems to be efficient and accurate. This method is suitable for parallel architectures, and hence the computation time is further reduced. It may be noted that we have also provided necessary error estimates for the present method.
51
JOTA: VOL. 99, NO. 1, OCTOBER 1998
Table 2. Initial-value technique profiles for Example 5.2. Nodal points
Numerical solution e=H=0.001
0.000
Error
Numerical solution e = H-0.0001
Error
Numerical solution e=H =0.00001
Error
0.0E + 00 4.8E-06 1.1E-05 1.6E-05 2.1E-05 2.6E-05 2.9E-05 3.1E-05 3.2E-05 3.3E-05 3.2E-05 4.2E-11 4.8E-21 0.0E+00 0.0E+00 0.0E + 00
1.000000 0.818729 0.670317 0.548807 0.449323 0.367872 0.301185 0.246586 0.201884 0.165281 0.135319 -0.000200 -0.040000 -0.200000 -0.400000 -0.500000
0.0E + 00 4.8E-07 1.1E-06 1.6E-06 2.1E-06 2.6E-06 2.9E-06 3.1E-06 3.2E-06 3.3E-06 3.2E-06 4.2E-12 0.0E+00 0.0E + 00 0.0E + 00 0.0E + 00
0.0010 0.0020 0.1000 0.2000 0.2500
1.000000 0.818579 0.670025 0.548374 0.448742 0.367134 0.300281 0.245505 0.200617 0.163824 0.133658 -0.02000 -0.040000 -0.200000 -0.400000 -0.500000
0.0E+00 0.0E + 00 0.0E+00
1.000000 0.818716 0.670291 0.548768 0.449270 0.367805 0.301103 0.246488 0.201769 0.165151 0.135168 -0.002000 -0.040000 -0.200000 -0.400000 -0.500000
0.3000 0.4000 0.5000 0.6000 0.7000
-0.600000 -0.800000 -1.000000 -1.200000 -1.400000
0.0E+00 0.0E+00 0.0E+00 0.0E + 00 0.0E + 00
-0.600000 -0.800000 -1.000000 -1.200000 -1.400000
0.0E+ 00 0.0E + 00 0.0E + 00 0.0E+00 0.0E+00
-0.600000 -0.800000 -1.000000 - .200000 - .400000
0.0E+00 0.0E+00 0.0E+00 0.0E+00 0.0E + 00
0.7500 0.8000 0.9000 0.9980 0.9990 1-10e l-9e l-8e 1-7e 1-6e 1-5e 1-4e 1-3e 1-2e 1-1e .0000
-1.500000 -1.600000 -1.800000 -1.941053 -1.591994 -1.591994 -1.502303 -1.392710 -1.258809 -1.095217 -0.895362 -0.651213 -0.352965 0.011360 0.456392 1.000000
0.0E+ 00 0.0E+00 0.0E+00 2.4E-17 3.3E-09 6.6E-04 6.9E-04 7.0E-04 7.1E-04 7.1E-04 6.8E-04 6.3E-04 5.5E-04 4.3E-04 2.5E-04 0.0E + 00
-1.500000 -1.600000 -1.800000 -1.997999 -1.996000 -1.593794 -1.503923 -1.394150 -1.260069 -1.096297 -0.896262 -0.651933 -0.353505 0.011000 0.456212 1.000000
0.0E + 00 0.0E + 00 0.0E + 00 1.2E-19 1.3E-11 6.9E-05 7.2E-05 7.4E-05 7.5E-05 7.4E-05 7.1E-05 6.6E-05 5.7E-05 4.4E-05 2.6E-05 0.0E+00
- .500000 - .600000 - .800000 - .997999 - .996000 - .593974 - .504085 - .394294 - .260195 - .096405 -0.896352 -0.652005 -0.353559 0.010964 0.456194 1.000000
0.0E + 00 0.0E+00 0.0E+00 1.2E-19 1.3E-I4 7.0E-06 7.2E-06 7.4E-06 7.5E-06 7.4E-06 7.2E-06 6.6E-06 5.8E-06 4.5E-06 2.6E-06 0.0E+00
\e 2f
3e 4f
5e 6e 7e
8e
9e l0e
0.0E + 00 4.8E-05
1.1E-04 1.6E-04 2.1E-04 2.5E-04 2.9E-04 3.1E-04 3.2E-04 3.3E-04 3.2E-04 4.7E-10 3.7E-18
References 1. GASPARO, M. G., and MACCONI, M., Initial-Value Methods for Second-Order Singularly Perturbed Boundary- Value Problems, Journal of Optimization Theory and Applications, Vol. 66, pp. 197-210, 1990. 2. ABRAHAMSSON, L. R., A Priori Estimates for Solutions of Singular Perturbations with a Turning Point, Studies in Applied Mathematics, Vol. 56, pp. 51-69, 1977.
52
JOTA: VOL. 99, NO. 1, OCTOBER 1998
3. O'MALLEY, R. E., Introduction to Singular Perturbations, Academic Press, New York, New York, 1974. 4. O'MALLEY, R. E., Singular Perturbation Methods for Ordinary Differential Equations, Springer Verlag, New York, New York, 1991. 5. Roos, H. G., STYNES, M., and TOBISKA, L., Numerical Methods for Singularly Perturbed Differential Equations, Springer Verlag, New York, New York, 1996. 6. WASOW, W., Linear Turning Point Theory, Springer Verlag, New York, New York, 1985. 7. WATTS, A. M., A Singular Perturbation Problem with a Turning Point, Bulletin of Australian Mathematical Society, Vol. 5, pp. 61-73, 1971. 8. BENDER, C. M., and ORSZAG, S. A., Advanced Mathematical Methods for Scientists and Engineers, McGraw-Hill, New York, New York, 1978. 9. LAGERSTROM, P. A., Matched Asymptotic Expansions, Springer Verlag, New York, New York, 1988. 10. JINDGE, D., Singularly Perturbed Boundary-Value Problems for Linear Equations with Turning Points, Journal of Mathematical Analysis and Applications, Vol. 155, pp. 322-337, 1991. 11. FARRELL, P. A., Sufficient Conditions for the Uniform Convergence of a Difference Scheme for a Singularly Perturbed Turning Point Problem, SIAM Journal on Numerical Analysis, Vol. 25, pp. 618-643, 1988. 12. BERGER, A. E., HAN, H., and KELLOG, R. B., A Priori Estimates and Analysis of a Numerical Method for a Turning Point Problem, Mathematics of Computation, Vol. 4, pp. 465-492, 1984. 13. EL-MISTIKAWY, T. M., and WERLE, M. J., Numerical Method for Boundary Layers with Blowing: The Exponential Box Scheme, AIAA Journal, Vol. 16, pp. 749-751, 1978. 14. KADALBAJOO, M. K., and REDDY, Y. N., Initial-Value Technique for a Class of Nonlinear Singular Perturbation Problems, Journal of Optimization Theory and Applications, Vol. 53, pp. 395-406, 1987. 15. GASPARO, M. G., and MACCONI, M., New Initial-Value Method for Singularly Perturbed Boundary-Value Problems, Journal of Optimization Theory and Applications, Vol. 63, pp. 213-224, 1989. 16. DOOLAN, E. P., MILLER, J. J. H., and SCHILDERS, W. H. A., Uniform Numerical Methods for Problems with Initial and Boundary Layers, Boole Press, Dublin, Ireland, 1980. 17. FARRELL, P. A., Uniform and Optimal Schemes for Stiff Initial-Value Problems, Computers and Mathematics with Applications, Vol. 13, pp. 925-936, 1987. 18. MILLER, J. J. H., Optimal Uniform Difference Schemes for Linear Initial-Value Problems, Computer and Mathematics with Applications, Vol. 12B, pp. 12091215, 1986.