Table of Contents Author Guidelines Submit a Manuscript
Journal of Applied Mathematics
Volume 2012, Article ID 564132, 7 pages
http://dx.doi.org/10.1155/2012/564132
Research Article

A Note on the Eigenvalue Analysis of the SIMPLE Preconditioning for Incompressible Flow

1School of Mathematics and Statistics, Anyang Normal University, Anyang 455000, China
2College of Mathematics, Chengdu University of Information Technology, Chengdu 610255, China

Received 21 November 2011; Accepted 11 January 2012

Academic Editor: Kok Kwang Phoon

Copyright © 2012 Shi-Liang Wu et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

We consider the SIMPLE preconditioning for block two-by-two generalized saddle point problems; this is the general nonsymmetric, nonsingular case where the (1,2) block needs not to equal the transposed (2,1) block, and the (2,2) block may not be zero. The eigenvalue analysis of the SIMPLE preconditioned matrix is presented. The relationship between the two different formulations spectrum of the SIMPLE preconditioned matrix is established by using the theory of matrix eigenvalue, and some corresponding results in recent article by Li and Vuik (2004) are extended.

1. Introduction

Consider the two-by-two generalized saddle point problems𝒜𝑥𝑦𝐴𝐵𝑇𝑥𝑦=𝑓𝑔𝐶𝐷,(1.1) where 𝐴𝑛×𝑛 is nonsingular, 𝐵,𝐶𝑚×𝑛 (𝑚𝑛), 𝐷𝑚×𝑚.

Systems of the form (1.1) arise in a variety of scientific and engineering applications, such as linear elasticity, fluid dynamics, electromagnetics, and constrained quadratic programming [14]. We refer the reader to [5] for more applications and numerical solution techniques of (1.1).

Since the coefficient matrix of (1.1) is often large and sparse, it may be attractive to use iterative methods. In particular, Krylov subspace methods might be used. As known, Krylov subspace methods are considered as one kind of the important and efficient iterative techniques for solving the large sparse linear systems because these methods are cheap to be implemented and are able to fully exploit the sparsity of the coefficient matrix. It is well known that the convergence speed of Krylov subspace methods depends on the eigenvalue distribution of the coefficient matrix [6]. Since the coefficient matrix of (1.1) is often extremely ill-conditioned and highly indefinite, the convergence speed of Krylov subspace methods can be unacceptably slow. In this case, Krylov subspace methods are not competitive without a good preconditioner. That is, preconditioning technique is a key ingredient for the success of Krylov subspace methods in applications.

To efficiently and accurately solve (1.1), Semi-implicit method for pressure linked equations (SIMPLE) were presented in [7] by Patankar. Subsequently, combining the SIMPLE(R) algorithm and Krylov subspace method GCR [8], Vuik et al. [9] proposed the GCR-SIMPLE(R) algorithm for solving (1.1). In this algorithm, the SIMPLE iteration is used as a preconditioner in the GCR method. Numerical experiments show that the SIMPLE(R) preconditioning is effective and competitive.

It is well known that the spectral properties of the preconditioned matrix give important insight in the convergence behavior of the preconditioned Krylov subspace methods. In [10], the eigenvalue analysis was given for the SIMPLE preconditioned matrix with 𝐵=𝐶 and 𝐷=0, and two different formulations spectrum of the preconditioned matrix were derived. The relationship between the two different formulations has been built by using the theory of matrix singular value decomposition. If 𝐵𝐶 and 𝐷0, using matrix singular value decomposition to establish the relationship between the two different formulations is invalid. On this occasion, we present the relationship between the two different formulations by using the theory of matrix eigenvalue and overcome the shortcomings of [10]. Some corresponding results in [10] are extended to two-by-two generalized saddle point problems.

2. Spectral Analysis

For simplicity, 𝜎() denotes the set of all eigenvalues of a matrix, and the diagonal entries of 𝐴 are not equal to zero. If the SIMPLE algorithm is used as preconditioning, it is equivalent to choose the preconditioner 𝒫 as𝒫=1,(2.1) where=𝐼𝑄1𝐵𝑇0𝐼,=𝐴0𝐶𝑅,𝑄=diag(𝐴),𝑅=𝐷+𝐶𝑄1𝐵𝑇.(2.2)

On the nonsingular of 𝒜 and 𝒫 we have the following proposition.

Proposition 2.1. The matrices 𝒜 and 𝒫, respectively, in (1.1) and (2.1) are nonsingular if and only if the Schur complements (𝐷+𝐶𝐴1𝐵𝑇) and (𝐷+𝐶𝑄1𝐵𝑇), respectively, are nonsingular.

In this paper, we assume that 𝒜 and 𝒫 are nonsingular and that 𝐵 and 𝐶 are of full rank.

Proposition 2.2. If the right preconditioner 𝒫 is defined by (2.1), then the preconditioned matrix is 𝒜=𝒜𝒫1=𝐼𝐼𝐴𝑄1𝐵𝑇𝑅1𝐶𝐴1𝐼𝐴𝑄1𝐵𝑇𝑅10𝐼.(2.3) Therefore, the spectrum of the SIMPLE preconditioned matrix 𝒜 is 𝜎𝒜={1}𝜎𝐼𝐼𝐴𝑄1𝐵𝑇𝑅1𝐶𝐴1.(2.4)

Proof. By simple computations, it is easy to verify that 1=𝐴10𝑅1𝐶𝐴1𝑅1𝜎𝒜=𝐴𝐵𝑇𝐶𝐷𝐼𝑄1𝐵𝑇𝐴0𝐼10𝑅1𝐶𝐴1𝑅1=𝐼𝐼𝐴𝑄1𝐵𝑇𝑅1𝐶𝐴1𝐼𝐴𝑄1𝐵𝑇𝑅1.0𝐼(2.5) Further, it is easy to find that the form of the spectrum of 𝜎(𝒜) is described by (2.4).

By the similarity invariance of the spectrum of the matrix, we have𝜎𝐼𝐼𝐴𝑄1𝐵𝑇𝑅1𝐶𝐴1𝐴=𝜎𝐼1𝑄1𝐵𝑇𝑅1C=𝜎𝐼𝑄1(𝑄𝐴)𝐴1𝐵𝑇𝑅1𝐶=𝜎𝐼𝐽𝐴1𝐵𝑇𝑅1𝐶,(2.6) where the matrix 𝐽=𝑄1(𝑄𝐴) is the Jacobi iteration matrix of the matrix 𝐴. Further, we have the following proposition.

Proposition 2.3. For the SIMPLE preconditioned matrix 𝒜,(1)1 is an eigenvalue with multiplicity at least of 𝑚,(2)the remaining eigenvalues are 1𝜇𝑖, 𝑖=1,2,,𝑛, where 𝜇𝑖 is the 𝑖th eigenvalue of𝑍𝐸𝑥=𝜇𝑥,(2.7) where 𝑍=𝐽𝐴1𝑛×𝑛,𝐸=𝐵𝑇𝑅1𝐶𝑛×𝑛.(2.8)

In fact, we also have the following result.

Proposition 2.4. For the SIMPLE preconditioned matrix 𝒜,(1)1 is an eigenvalue with (algebraic and geometric) multiplicity of 𝑛,(2)the remaining eigenvalues are defined by the generalized eigenvalue problem 𝑆𝑥=𝜆𝑅𝑥,(2.9) where 𝑆=(𝐷+𝐶𝐴1𝐵𝑇) is the Schur complement of the matrix 𝒜.

Proof. Note that 𝒜𝒫1 is the same spectrum as 𝒫1𝒜. So, it is only needed to consider the following generalized eigenvalue problem 𝒜𝑥=𝜆𝒫𝑥,(2.10) where 𝒜=𝐴𝐵𝑇𝐶𝐷,𝒫=𝐴𝐴𝑄1𝐵𝑇𝐶𝐷.(2.11) The generalized eigenvalue problem (2.10) can be written as 𝐴𝐵𝑇𝑢𝑝𝐶𝐷=𝜆𝐴𝐴𝑄1𝐵𝑇𝑢𝑝𝐶𝐷,(2.12) that is, 𝐴𝑢+𝐵𝑇𝑝=𝜆𝐴𝑢+𝐴𝑄1𝐵𝑇𝑝,(2.13)𝐶𝑢𝐷𝑝=𝜆(𝐶𝑢𝐷𝑝).(2.14) From (2.13) and (2.14), it is easy to see that 𝜆=1 is an eigenvalue of (2.12). If the matrix 𝑄1𝐴1 is nonsingular with 𝜆=1 and rank (𝐵𝑇)=𝑚, from (2.13) we have 𝑝=0. Therefore, the eigenvectors corresponding to eigenvalue 1 are 𝜈𝑖=𝑢𝑖0,𝑢𝑖𝑛,𝑖=1,2,,𝑛,(2.15) where {𝑢𝑖}𝑛𝑖 is a basis of 𝑛.
For 𝜆1, from (2.13) we obtain 1𝑢=𝐴1𝜆1𝜆𝐴𝑄1𝐵𝑇𝑝𝐵𝑇𝑝.(2.16) Substituting it into (2.14) yields 𝑆𝑝=𝜆𝑅𝑝,(2.17) where 𝑆=(𝐷+𝐶𝐴1𝐵𝑇) is the Schur complement of the matrix 𝒜.

From Propositions 2.3 and 2.4, two different generalized eigenvalue problems (2.7) and (2.9) have been derived to describe the spectrum of 𝒜. Subsequently, we will investigate the relationship between both spectral formulations for the nonsymmetric case. Here we will make use of the theory of matrix eigenvalue to establish the relationship of the two different formulations spectrum of the SIMPLE preconditioned matrix. To this end, the following lemma is required.

Lemma 2.5 (See [11]). Suppose that 𝑀𝑚×𝑛 and 𝑁𝑛×𝑚 with 𝑚𝑛. Then 𝑁𝑀 has the same eigenvalues as 𝑀𝑁, counting multiplicity, together with an additional 𝑛𝑚 eigenvalues equal to 0.

By (2.7), it follows that𝑍𝐸=𝑍𝑛×𝑛𝐵𝑇𝑛×𝑚𝑅1𝑚×𝑚𝐶𝑚×𝑛𝑛×𝑛,𝑅1𝑚×𝑚𝐶𝑚×𝑛𝑍𝑛×𝑛𝐵𝑇𝑛×𝑚𝑚×𝑚.(2.18) From Lemma 2.5, we have𝑅𝜎(𝑍𝐸)={0}𝜎1𝐶𝑍𝐵𝑇,(2.19) where the eigenvalue 0 is with multiplicity of 𝑛𝑚 and𝜎𝑅1𝐶𝑍𝐵𝑇𝑅=𝜎1𝐶𝑄1(𝑄𝐴)𝐴1𝐵𝑇𝑅=𝜎1𝐶𝐴1𝑄1𝐵𝑇𝑅=𝜎1𝐶𝐴1𝐵𝑇𝐶𝑄1𝐵𝑇𝑅=𝜎1𝐶𝐴1𝐵𝑇+𝐷𝐶𝑄1𝐵𝑇𝑅+𝐷=𝜎1(𝑅𝑆)=𝜎𝐼𝑅1𝑆=1𝜆𝑖,𝑖=1,2,,𝑚.(2.20)

These relations lead to the following proposition.

Proposition 2.6. For two generalized eigenvalue problems (2.7) and (2.9), suppose that 𝜇𝑖𝜎(𝑍𝐸), 𝑖=1,2,,𝑛, and 𝜆𝑖𝜎(𝑅1𝑆), 𝑖=1,2,,𝑚, the relationship between two problems is that 𝜇=0 is an eigenvalue of (2.7) with multiplicity of 𝑛𝑚, which can be denoted as 𝜇𝑚+1=𝜇𝑚+2==𝜇𝑛=0, and that 𝜆𝑖=1𝜇𝑖, 𝑖=1,2,,𝑚, holds for the remaining 𝑚 eigenvalues.

Some remarks on Proposition 2.6 are given as follows.(i)In [10], the relationship between two different formulations spectrum of the preconditioned matrix with 𝐵=𝐶 and 𝐷=0 was built by using the theory of matrix singular value decomposition, but for the nonsymmetric case, the above strategy is invalid. Whereas, using the theory of matrix eigenvalue not only establishes the relationship between the two different formulations, but also overcomes the shortcomings of [10]. In this way, Propositions 2.22.6 can be regarded as the extension of Propositions  2–5 [10].(ii)In [10], the diagonal entries of matrix 𝐴 must be positive. But, in this paper, the diagonal entries of 𝑄 are only not equal to zero. Clearly, this assumption is weaker than that of [10]. If the diagonal entries of matrix 𝐴 are complex and not equal to zero, then the diagonal entries of 𝑄 take the absolute diagonal entries of 𝐴. This idea is based on an absolute diagonal scaling technique, which is cheaply easy to implement, reducing computation times and amount of memory.(iii)Recently, although Li et al. in [12] discussed the SIMPLE preconditioning for the generalized nonsymmetric saddle point problems and provided some results above the spectrum of the SIMPLE preconditioned matrix, some conditions of the supporting propositions may be defective. In fact, if 𝐴 is nonsingular with rank (𝐵𝑇) = rank (𝐶)=𝑚, then 𝑅 and 𝒫 may be singular. For a counterexample, we take 𝐶=[10], 𝐴=𝑄=2𝐼 and 𝐵=[01], then 𝑅=𝐶𝑄1𝐵𝑇=0. That is, this paper corrects some results in [12].(iv)In fact, 𝑄 is not necessary the diagonal entries of 𝐴; in this case, the diagonal entries of 𝐴 can be equal to zero. In actual implements, the choice of matrix 𝑄 is that the eigenvalue of the generalized eigenvalue problem (2.9) is close to one; Krylov subspace methods such as GMRES will converge quickly.

3. Conclusion

In this paper, the SIMPLE preconditioner for the nonsymmetric generalized saddle point problems is discussed. The relationship of the two different formulations spectrum of the SIMPLE preconditioned matrix has been built by using the theory of matrix eigenvalue.

Acknowledgments

The authors would like to thank Editor Professor Phoon and two anonymous referees for their helpful suggestions, which greatly improve the paper. This research of this author is supported by NSFC Tianyuan Mathematics Youth Fund (11026040).

References

  1. O. Axelsson and V. A. Barker, Finite Element Solution of Boundary Value Problems, Computer Science and Applied Mathematics, Academic Press Inc., Orlando, Fla, USA, 1984. View at Zentralblatt MATH
  2. F. Brezzi and M. Fortin, Mixed and Hybrid Finite Element Methods, vol. 15, Springer, New York, NY, USA, 1991. View at Publisher · View at Google Scholar · View at Zentralblatt MATH
  3. H. C. Elman, D. J. Silvester, and A. J. Wathen, Finite Elements and Fast Iterative Solvers, Oxford University Press, Oxford, UK, 2003.
  4. W. Zulehner, “Analysis of iterative methods for saddle point problems: a unified approach,” Mathematics of Computation, vol. 71, no. 238, pp. 479–505, 2002. View at Publisher · View at Google Scholar · View at Zentralblatt MATH
  5. M. Benzi, G. H. Golub, and J. Liesen, “Numerical solution of saddle point problems,” Acta Numerica, vol. 14, pp. 1–137, 2005. View at Publisher · View at Google Scholar · View at Zentralblatt MATH
  6. Y. Saad, Iterative Methods for Sparse Linear Systems, PWS Publishing Company, Boston, Mass, USA, 1996.
  7. S. V. Patankar, Numerical Heat Transfer and Fluid Flow, McGraw-Hill, New York, NY, USA, 1980.
  8. S. C. Eisenstat, H. C. Elman, and M. H. Schultz, “Variational iterative methods for nonsymmetric systems of linear equations,” SIAM Journal on Numerical Analysis, vol. 20, no. 2, pp. 345–357, 1983. View at Publisher · View at Google Scholar · View at Zentralblatt MATH
  9. C. Vuik, A. Saghir, and G. P. Boerstoel, “The Krylov accelerated SIMPLE(R) method for flow problems in industrial furnaces,” International Journal for Numerical Methods in Fluids, vol. 33, no. 7, pp. 1027–1040, 2000. View at Google Scholar
  10. C. Li and C. Vuik, “Eigenvalue analysis of the SIMPLE preconditioning for incompressible flow,” Numerical Linear Algebra with Applications, vol. 11, no. 5-6, pp. 511–523, 2004. View at Publisher · View at Google Scholar · View at Zentralblatt MATH
  11. R. A. Horn and C. R. Johnson, Matrix Analysis, Cambridge University Press, Cambridge, Mass, USA, 1990.
  12. L. J. Li, Z. J. Zhang, and L. Li, “A note on the SIMPLE preconditioner for nonsymmetric saddle point problems,” Journal of Information and Computing Science, vol. 5, pp. 153–160, 2010. View at Google Scholar