Hindawi Publishing Corporation Journal of Applied Mathematics Volume 2013, Article ID 213659, 7 pages http://dx.doi.org/10.1155/2013/213659 Research Article Block Preconditioned SSOR Methods for π»-Matrices Linear Systems Zhao-Nian Pu and Xue-Zhong Wang School of Mathematics and Statistics, Hexi University, Zhangye, Gansu 734000, China Correspondence should be addressed to Xue-Zhong Wang; mail2011wang@gmail.com Received 9 January 2013; Accepted 12 March 2013 Academic Editor: Hak-Keung Lam Copyright © 2013 Z.-N. Pu and X.-Z. Wang. 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. We present a block preconditioner and consider block preconditioned SSOR iterative methods for solving linear system π΄π₯ = π. When π΄ is an π»-matrix, the convergence and some comparison results of the spectral radius for our methods are given. Numerical examples are also given to illustrate that our methods are valid. where π· = diag(π΄ 11 , . . . , π΄ ππ ), −πΏ and −π are strictly block lower and strictly block upper triangular parts of π΄, respectively. Let 0 < π < 2, and 1. Introduction For the linear system π΄π₯ = π, (1) π= where π΄ is an π × π square matrix and π₯ and π are π-dimensional vectors. The basic iterative method for solving (1) is ππ₯π+1 = ππ₯π + π, π = 0, 1, . . . , (2) where π΄ = π − π and π is nonsingular. Thus (2) can be written as π₯π+1 = ππ₯π + π, π = 0, 1, . . . , (3) where π = π−1 π, π = π−1 π. Let us consider the following partition of π΄: π΄ 11 π΄ 12 π΄ 21 π΄ 22 .. π΄ = ( ... . π΄ π1 π΄ π2 ⋅ ⋅ ⋅ π΄ 1π ⋅ ⋅ ⋅ π΄ 2π . d .. ) , π= 1 (π· − ππΏ) π·−1 (π· − ππ) , π (2 − π) 1 ((1 − π) π· + ππΏ) π·−1 ((1 − π) π· + ππ) . π (2 − π) (6) Then, the iteration matrix of the SSOR method for π΄ is given by Lπ = π−1 π = (π· − ππ)−1 π·(π· − ππΏ)−1 (4) ⋅ ⋅ ⋅ π΄ ππ × ((1 − π) π· + ππΏ) π·−1 (7) × ((1 − π) π· + ππ) . where the blocks π΄ ππ ∈ πΆππ ×ππ , π = 1, . . . , π, are nonsingular and π1 + π2 + ⋅ ⋅ ⋅ + ππ = π. Usually we split π΄ into Transforming the original system (1) into the preconditioned form π΄ = π· − πΏ − π, ππ΄π₯ = ππ, (5) (8) 2 Journal of Applied Mathematics then we can define the basic iterative scheme: ππ π₯π+1 = ππ π₯π + ππ, π = 0, 1, . . . , (9) When π΄ is an π-matrix, Alanelli and Hadjidimosin [1] considered the preconditioner π = π + π, where π = diag(πΏ−1 11 , πΌ22 , . . . , πΌππ ) and π is given by π11 π12 −π΄ 21 πΏ−1 11 π22 . .. .. π=( . −1 −π΄ π1 πΏ 11 ππ2 where ππ΄ = ππ − ππ and ππ is nonsingular. Thus (9) can also be written as π₯π+1 = ππ₯π + π, π = 0, 1, . . . , (10) where π = ππ−1 ππ , π = ππ−1 ππ. Similar to the original system (1), we call the basic iterative methods corresponding to the preconditioned system the preconditioned iterative methods. πΌ11 d πΌ + π1 = ( ( Μ−πΏ Μ − π, Μ π1 π΄ = (πΌ + π1 ) (π· − πΏ − π) = π· (13) Μ −πΏ Μ and −π Μ are block diagonally, strictly block lower, where π·, and strictly block upper triangular parts of π1 π΄, respectively. Μ is nonsingular, then (π· Μ − ππΏ) Μ −1 and (π· Μ − ππ) Μ −1 exist If π· and it is possible to define the SSOR iteration matrix for π1 π΄. Namely, Μ π = (π· Μ π· Μ − ππΏ) Μ −1 Μ − ππ) Μ −1 π·( L −1 Μ + ππΏ) Μ π· Μ ((1 − π) π· Μ + ππ) Μ . × ((1 − π) π· ( πΌ + π2 = ( ( d πΌ22 ). πΌππ (12) ) σ΅¨ σ΅¨ β¨π΄ 11 β© − σ΅¨σ΅¨σ΅¨π΄ 12 σ΅¨σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ − σ΅¨σ΅¨π΄ 21 σ΅¨σ΅¨ β¨π΄ 22 β© .. β¨π΄β© = ( ... . σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨ − σ΅¨σ΅¨π΄ π1 σ΅¨σ΅¨ − σ΅¨σ΅¨π΄ π2 σ΅¨σ΅¨σ΅¨ σ΅¨ σ΅¨ ⋅ ⋅ ⋅ − σ΅¨σ΅¨σ΅¨π΄ 1π σ΅¨σ΅¨σ΅¨ σ΅¨ σ΅¨σ΅¨ ⋅ ⋅ ⋅ − σ΅¨σ΅¨π΄ 2π σ΅¨σ΅¨σ΅¨ .. ) . d . (15) ⋅ ⋅ ⋅ β¨π΄ ππ β© Let β¨π΄β© = β¨π·β© − |πΏ| − |π|, where β¨π·β©, −|πΏ|, and −|π| are block diagonally, strictly block lower, and strictly block upper triangular parts of β¨π΄β©, respectively. Notice that the preconditioner of the matrix β¨π΄β© corresponding to π1 is π2 = πΌ + π2 ; namely, −1 σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨ σ΅¨σ΅¨πΌ1 σ΅¨σ΅¨ β¨π΄ 11 β© σ΅¨σ΅¨σ΅¨π΄ 1π σ΅¨σ΅¨σ΅¨ d −1 σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨ σ΅¨σ΅¨πΌπ σ΅¨σ΅¨ β¨π΄ ππ β© σ΅¨σ΅¨σ΅¨π΄ ππ’ σ΅¨σ΅¨σ΅¨ ⋅ ⋅ ⋅ πππ method with preconditioner π1 and give some comparison results of the spectral radius for the case when π΄ is an π»matrix. Let |π΄| denote the matrix whose elements are the moduli of the elements of the given matrix. We call β¨π΄β© = (πππ ) to comparison matrix if πππ = |πππ | for π = π, if πππ = −|πππ | for π =ΜΈ π. For (4), under the previous definition, we have (14) Alanelli and Hadjidimos in [1] showed that the preconditioned Gauss-Seidel, the preconditioned SOR, and the preconditioned Jacobi methods with preconditioner π are better than original methods. Our work in the presentation is to prove convergence of the block preconditioned SSOR πΌ11 −πΌ2 π΄−1 22 π΄ 2π‘ −πΌπ π΄−1 ππ π΄ ππ’ Let (11) with πΏ 11 being the lower triangular matrix in the LU triangular decomposition of π΄ 11 . We consider the preconditioner π1 = πΌ + π1 , where −πΌ1 π΄−1 11 π΄ 1π πΌ22 ⋅ ⋅ ⋅ π1π ⋅ ⋅ ⋅ π2π . d .. ) , −1 σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨ σ΅¨σ΅¨πΌ2 σ΅¨σ΅¨ β¨π΄ 22 β© σ΅¨σ΅¨σ΅¨π΄ 2π‘ σ΅¨σ΅¨σ΅¨ ) ). d πΌππ ) (16) Journal of Applied Mathematics 3 Let π2 β¨π΄β© = (πΌ+π2 )β¨π΄β© = π·−πΏ−π, where π·, −πΏ, and −π are block diagonally, strictly block lower, and strictly block upper triangular parts of π2 β¨π΄β©, respectively. If π· is nonsingular, then (π· − ππΏ)−1 and (π· − ππ)−1 exist and the SSOR iteration matrix for π2 β¨π΄β© is as follows: −1 Lπ = (π· − ππ) π·(π· − ππΏ) −1 −1 (17) × ((1 − π) π· + ππΏ) π· ((1 − π) π· + ππ) . Lemma 6. If π΄ and π΅ are two π × π matrices, then β¨π΄ − π΅β© ≥ β¨π΄β© − |π΅|. Proof. It is easy to see that |πππ − πππ | ≥ |πππ | − |πππ |, for π = π, and −|πππ −πππ | ≥ −|πππ |−|πππ |, for π =ΜΈ π. Therefore, β¨π΄−π΅β© ≥ β¨π΄β©−|π΅| is true. Lemma 7. If π΄ is an π»-matrix with unit diagonal elements, then β β¨π΄β©−1 β∞ > 1. Proof. Let β¨π΄β© = πΌ − π΅, from β¨π΄β© being an π-matrix; then π΅ ≥ 0 and π(π΅) < 1, and thus, we have 2. Preliminaries A matrix π΄ is called nonnegative (positive) if each entry of π΄ is nonnegative (positive). We denote it by π΄ ≥ 0 (π΄ > 0). Similarly, for π-dimensional vector π₯, we can also define π₯ ≥ 0 (π₯ > 0). Additionally, we denote the spectral radius of π΄ by π(π΄). π΄π denotes the transpose of π΄. A matrix π΄ = (πππ ) is called a π-matrix if for any π =ΜΈ π, πππ ≤ 0. A π-matrix is a nonsingular π-matrix if π΄ is nonsingular and π΄−1 ≥ 0, If β¨π΄β© is a nonsingular π-matrix , then π΄ is called an π»-matrix. π΄ = π − π is said to be a splitting of π΄ if π is nonsingular, π΄ = π − π is said to be regular if π−1 ≥ 0 and π ≥ 0, and weak regular if π−1 ≥ 0 and π−1 π ≥ 0, respectively. Some basic properties on special matrices introduced previously are given to be used in this paper. ∞ β¨π΄β©−1 = ∑ π΅π ≥ πΌ and then β β¨π΄β©−1 β∞ > 1. 3. Convergence Results π π Let ππ = (1, . . . , 1)π ∈ π ππ , π = 1, 2, . . . , π, π = (π1π , . . . , ππ ) , π −1 π π π π = (π1 , . . . , ππ ) = β¨π΄β© π, ππ = (0, . . . , 0) ∈ π ππ , where π and π are partitioned in accordance with the block partitioning of the matrix π΄, and let σ΅© −1 σ΅¨ σ΅¨ σ΅© π π = σ΅©σ΅©σ΅©σ΅©β¨π΄ ππ β© σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨ ππ σ΅©σ΅©σ΅©σ΅©∞ , Lemma 1 (see [2]). Let A be a π-matrix. Then the following statements are equivalent. βπ = (a) π΄ is an π-matrix. 1 , σ΅© σ΅©σ΅© σ΅© π π (2σ΅©σ΅©β¨π΄β©−1 σ΅©σ΅©σ΅©σ΅©∞ − 1) (21) π = 1, 2, . . . , π. (b) There is a positive vector π₯ such that π΄π₯ > 0. Theorem 8. Let π΄ be a nonsingular H-matrix; if |πΌπ | < βπ , π = 1, 2, . . . , π, then π1 π΄ is also an H-matrix. (c) π΄−1 ≥ 0. (d) All principal submatrices of π΄ are π-matrices. (e) All principal minors are positive. Lemma 2 (see [3, 4]). Let π΄ be an π-matrix and let π΄ = π − π be a weak regular splitting. Then π(π−1 π) < 1. Proof. From π΄ being an π»-matrix, we have π > 0, and ππ ≤β β¨π΄β©−1 β∞ ππ . Let ((π1 π΄)ππ ) Lemma 3 (see [2]). Let π΄ and π΅ be two π × π matrices with 0 ≤ π΅ ≤ π΄. Then π(π΅) ≤ π(π΄). ΜΈ π, π = 1, 2, . . . , π, π =ΜΈ π, π΄ −πΌ π΄−1 ππ π΄ ππ π΄ ππ , π=π, = { ππ π −1 π΄ ππ −πΌπ π΄ ππ π΄ ππ π΄ ππ , π = π, π, π = 1, 2, . . . , π, π =ΜΈ π. (22) Lemma 4 (see [5]). If π΄ is an π»-matrix, then |π΄−1 | ≤ β¨π΄β©−1 . Lemma 5 (see [6]). Suppose that π΄ 1 = π1 − π1 and π΄ 2 = π2 − π2 are weak regular splitting of monotone matrices π΄ 1 and π΄ 2 , respectively, such that π2−1 ≥ π1−1 . If there exists a positive vector π₯ such that 0 ≤ π΄ 1 π₯ ≤ π΄ 2 π₯, then for the monotone norm associated with π₯, σ΅©σ΅© −1 σ΅©σ΅© σ΅© σ΅© σ΅©σ΅©π1 π1 σ΅©σ΅© ≤ σ΅©σ΅©σ΅©π2−1 π2 σ΅©σ΅©σ΅© . (18) σ΅© σ΅©π₯ σ΅© σ΅©π₯ In particular, if π1−1 π1 has a positive Perron vector, then π (π1−1 π1 ) ≤ π (π2−1 π2 ) . (20) πΎ=0 (19) Moreover if π₯ is a Perron vector of π1−1 π1 and strict inequality holds in (18), then strict inequality holds in (19). Then (β¨π1 π΄β© π)π = β¨π΄ ππ − πΌπ π΄−1 ππ π΄ ππ π΄ ππ β© ππ π σ΅¨ σ΅¨σ΅¨ σ΅¨ − ∑ σ΅¨σ΅¨σ΅¨σ΅¨π΄ ππ − πΌπ π΄−1 ππ π΄ ππ π΄ ππ σ΅¨σ΅¨ ππ π =ΜΈ π,π σ΅¨σ΅¨ σ΅¨ σ΅¨ − σ΅¨σ΅¨σ΅¨σ΅¨π΄ ππ − πΌπ π΄−1 ππ π΄ ππ π΄ ππ σ΅¨σ΅¨ ππ π σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨ σ΅¨σ΅¨ σ΅¨ σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨ π΄ π΄ π ≥ β¨π΄ ππ β© ππ − σ΅¨σ΅¨σ΅¨πΌπ σ΅¨σ΅¨σ΅¨ σ΅¨σ΅¨σ΅¨σ΅¨π΄−1 − ∑ σ΅¨ σ΅¨ σ΅¨ σ΅¨ σ΅¨ ππ σ΅¨ σ΅¨ ππ σ΅¨ σ΅¨ ππ σ΅¨ π σ΅¨ σ΅¨ π =ΜΈ π,π π σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨ σ΅¨σ΅¨ σ΅¨ σ΅¨ σ΅¨ − ∑ σ΅¨σ΅¨σ΅¨πΌπ σ΅¨σ΅¨σ΅¨ σ΅¨σ΅¨σ΅¨σ΅¨π΄−1 ππ σ΅¨σ΅¨ σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨ σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨ ππ π =ΜΈ π,π 4 Journal of Applied Mathematics σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨ σ΅¨ σ΅¨ σ΅¨σ΅¨ σ΅¨ − σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨ ππ − σ΅¨σ΅¨σ΅¨πΌπ σ΅¨σ΅¨σ΅¨ σ΅¨σ΅¨σ΅¨σ΅¨π΄−1 ππ σ΅¨σ΅¨ σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨ σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨ ππ that (24) is a weak regular splitting; from Lemma 2, we know that π(LΜ π ) < 1. Since −1 σ΅¨ σ΅¨σ΅¨ σ΅¨ σ΅¨ σ΅¨ ≥ ππ − σ΅¨σ΅¨σ΅¨πΌπ σ΅¨σ΅¨σ΅¨ β¨π΄ ππ β© σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨ σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨ ππ σ΅¨σ΅¨ Μ σ΅¨σ΅¨ σ΅¨σ΅¨Lπ σ΅¨σ΅¨ = σ΅¨ σ΅¨ π −1 σ΅¨ σ΅¨ σ΅¨ σ΅¨σ΅¨ σ΅¨ − ∑ σ΅¨σ΅¨σ΅¨πΌπ σ΅¨σ΅¨σ΅¨ β¨π΄ ππ β© σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨ σ΅¨σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨σ΅¨ ππ σ΅¨σ΅¨ σ΅¨σ΅¨(π· Μ Μ −1 Μ Μ −1 σ΅¨σ΅¨ − ππ) π·(π· − ππΏ) σ΅¨ Μ + ππΏ) Μ π·−1 ((1 − π) π· Μ + ππ) Μ σ΅¨σ΅¨σ΅¨ × ((1 − π) π· σ΅¨σ΅¨ π =ΜΈ π,π −1 σ΅¨ σ΅¨σ΅¨ σ΅¨ σ΅¨ σ΅¨ − σ΅¨σ΅¨σ΅¨πΌπ σ΅¨σ΅¨σ΅¨ β¨π΄ ππ β© σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨ σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨ ππ σ΅¨σ΅¨ −1 −1 Μ −1 π) Μ (πΌ − ππ· Μ Μ −1 πΏ) = σ΅¨σ΅¨σ΅¨σ΅¨(πΌ − ππ· σ΅¨ −1 σ΅¨ σ΅¨ σ΅¨ σ΅¨ = ππ + σ΅¨σ΅¨σ΅¨πΌπ σ΅¨σ΅¨σ΅¨ β¨π΄ ππ β© σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨ σ΅¨ Μ ((1 − π) πΌ + ππ· Μ −1 π) Μ σ΅¨σ΅¨σ΅¨ × ((1 − π) πΌ + ππ·−1 πΏ) σ΅¨σ΅¨ σ΅¨σ΅¨ −1 σ΅¨ σ΅¨ −1 σ΅¨ Μ −1 π) Μ −1 πΏ) Μ σ΅¨σ΅¨σ΅¨σ΅¨ σ΅¨σ΅¨σ΅¨σ΅¨(πΌ − ππ· Μ σ΅¨σ΅¨σ΅¨σ΅¨ ≤ σ΅¨σ΅¨σ΅¨σ΅¨(πΌ − ππ· σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨σ΅¨ σ΅¨ σ΅¨ σ΅¨ Μ −1 π) Μ σ΅¨σ΅¨σ΅¨σ΅¨ σ΅¨σ΅¨σ΅¨((1 − π) πΌ + ππ· Μ σ΅¨σ΅¨σ΅¨ × σ΅¨σ΅¨σ΅¨σ΅¨((1 − π) πΌ + ππ·−1 πΏ) σ΅¨σ΅¨ σ΅¨ σ΅¨σ΅¨ π σ΅¨ σ΅¨ σ΅¨ σ΅¨ × (− ∑ σ΅¨σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨σ΅¨ ππ − σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨ ππ π =ΜΈ π + β¨π΄ ππ β© ππ − β¨π΄ ππ β© ππ ) Μ σ΅¨σ΅¨σ΅¨ Μ −1 |π|) Μ Μ −1 σ΅¨σ΅¨σ΅¨σ΅¨πΏ ≤ (πΌ − πβ¨π·β© (πΌ − πβ¨π·β© σ΅¨ σ΅¨σ΅¨) −1 −1 σ΅¨ σ΅¨ σ΅¨ σ΅¨ = ππ + σ΅¨σ΅¨σ΅¨πΌπ σ΅¨σ΅¨σ΅¨ β¨π΄ ππ β© σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨ (ππ − 2ππ ) σ΅© −1 σ΅¨ σ΅¨ σ΅¨ σ΅¨ σ΅© ≥ ππ − σ΅¨σ΅¨σ΅¨πΌπ σ΅¨σ΅¨σ΅¨ (2σ΅©σ΅©σ΅©σ΅©β¨π΄β©−1 σ΅©σ΅©σ΅©σ΅©∞ − 1) β¨π΄ ππ β© σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨ ππ σ΅© σ΅© σ΅¨ σ΅¨ ≥ ππ − σ΅¨σ΅¨σ΅¨πΌπ σ΅¨σ΅¨σ΅¨ π π (2σ΅©σ΅©σ΅©σ΅©β¨π΄β©−1 σ΅©σ΅©σ΅©σ΅©∞ − 1) ππ Μ −1 σ΅¨σ΅¨σ΅¨ Μ σ΅¨σ΅¨σ΅¨ Μ σ΅¨σ΅¨σ΅¨ Μ −1 σ΅¨σ΅¨σ΅¨σ΅¨πΏ × ((1−π) πΌ+πβ¨π·β© σ΅¨ σ΅¨σ΅¨) ((1−π) πΌ+πβ¨π·β© σ΅¨σ΅¨πσ΅¨σ΅¨) = π (LΜ π ) (26) > ππ . (23) Therefore, β¨π1 π΄β© is an π-matrix, and then π1 π΄ is an π»matrix. Theorem 9. If π΄ is a nonsingular π»-matrix with unit diagonal elements, 0 < π ≤ 1 and |πΌπ | < βπ , π = 1, 2, . . . , π. Then Μ π ) < 1. π(L Μ − |πΏ| Μ − |π| Μ is Proof. From Theorem 8, we know β¨π1 π΄β© = β¨π·β© an π-matrix; if we let β¨π1 π΄β© = 1 Μ −1 (β¨π·β© Μ σ΅¨σ΅¨σ΅¨σ΅¨) Μ σ΅¨σ΅¨σ΅¨σ΅¨) β¨π·β© Μ − π σ΅¨σ΅¨σ΅¨σ΅¨π Μ − π σ΅¨σ΅¨σ΅¨σ΅¨πΏ (β¨π·β© σ΅¨ σ΅¨ σ΅¨ σ΅¨ π (2 − π) 1 Μ σ΅¨σ΅¨σ΅¨ Μ −1 Μ + π σ΅¨σ΅¨σ΅¨σ΅¨πΏ ((1 − π) β¨π·β© σ΅¨ σ΅¨σ΅¨) β¨π·β© π (2 − π) Μ σ΅¨σ΅¨σ΅¨ Μ + π σ΅¨σ΅¨σ΅¨σ΅¨π × ((1 − π) β¨π·β© σ΅¨ σ΅¨σ΅¨) , Μ π ) ≤ π(|L Μ π |) ≤ π(LΜ π ) < 1. then, by Lemma 3, π(L 4. Comparison Results of Spectral Radius Theorem 10. Let π΄ be a nonsingular π»-matrix with unit diagonal elements, 0 < π ≤ 1 and |πΌπ | < βπ , π = 1, 2, . . . , π. Then π2 β¨π΄β© is an π-matrix and π(Lπ ) < 1. Proof. Similar to the proof of Theorems 8 and 9, it is easy to get the proof of this theorem. In what follows we will give some comparison results on the spectral radius of preconditioned SSOR iteration matrices with different preconditioner. Let Μ −π Μ β¨π΄β© = π − = (24) 1 (β¨π·β© − π |πΏ|) β¨π·β©−1 (β¨π·β© − π |π|) π (2 − π) 1 − ((1 − π) β¨π·β© + π |πΏ|) β¨π·β©−1 π (2 − π) then the SSOR iteration matrix for β¨π1 π΄β© is as follows: (27) × ((1 − π) β¨π·β© + π |π|) , σ΅¨σ΅¨ Μ σ΅¨σ΅¨ −1 Μ σ΅¨σ΅¨σ΅¨ −1 Μ Μ − π σ΅¨σ΅¨σ΅¨σ΅¨π Μ σ΅¨ σ΅¨ LΜ π = (β¨π·β© σ΅¨ σ΅¨σ΅¨) β¨π·β© (β¨π·β© − π σ΅¨σ΅¨πΏσ΅¨σ΅¨) Μ σ΅¨σ΅¨σ΅¨ Μ −1 Μ + π σ΅¨σ΅¨σ΅¨σ΅¨πΏ × ((1 − π) β¨π·β© σ΅¨ σ΅¨σ΅¨) β¨π·β© Μ σ΅¨σ΅¨σ΅¨ Μ + π σ΅¨σ΅¨σ΅¨σ΅¨π × ((1 − π) β¨π·β© σ΅¨ σ΅¨σ΅¨) . −1 where (25) Μ β¨π·β© Μ − π|πΏ| Μ and Since β¨π1 π΄β© is an π-matrix; we have β¨π·β©, Μ Μ β¨π·β© − π|π| are π-matrices; by simple calculation, we obtain Μ= π 1 (β¨π·β© − π |πΏ|) β¨π·β©−1 (β¨π·β© − π |π|) , π (2 − π) Μ= π 1 ((1 − π) β¨π·β© + π |πΏ|) β¨π·β©−1 π (2 − π) × ((1 − π) β¨π·β© + π |π|) . (28) Journal of Applied Mathematics 5 Then the SSOR iteration matrix for β¨π΄β© is as follows: It follows that −1 Μπ = π Μ Μ π L = (β¨π·β© − π |π|)−1 β¨π·β© (β¨π·β© − π |πΏ|)−1 × ((1−π) β¨π·β©+π |πΏ|) β¨π·β©−1 ((1−π) β¨π·β©+π |π|) , σ΅© −1 σ΅©σ΅© σ΅©σ΅© −1 σ΅©σ΅© σ΅©σ΅©π πσ΅©σ΅© ≤ σ΅©σ΅©σ΅©π Μ π Μ σ΅©σ΅© . (35) σ΅©σ΅©π₯ σ΅©σ΅©π₯ σ΅©σ΅© σ΅©σ΅© Μ−π Μ is a weak regular splitting, there exists As β¨π΄β© = π a positive perron vector π¦; by Lemma 5, the following inequality holds: (29) and let −1 1 (π· − ππΏ) π· (π· − ππ) = π (2 − π) −1 1 ((1 − π) π· + ππΏ) π· − π (2 − π) (30) × ((1 − π) π· + ππ) , where π= −1 1 (π· − ππΏ) π· (π· − ππ) , π (2 − π) −1 1 ((1 − π) π· + ππΏ) π· ((1 − π) π· + ππ) . π (2 − π) (31) Then the AOR iteration matrix for π2 β¨π΄β© is (17). Theorem 11. If π΄ is a nonsingular π»-matrix with unit diagonal elements, 0 < π ≤ 1 and |πΌπ | < βπ , π = 1, 2, . . . , π. Then Μ π ). π(Lπ ) ≤ π(L Proof. Since β¨π΄β© is a nonsingular π-matrix, by Theorem 10, π2 β¨π΄β© is a nonsingular π-matrix, and thus β¨π΄β© and π2 β¨π΄β© are two monotone matrices. From β¨π΄β© and π2 β¨π΄β© being π-matrices, we can get β¨π·β©, Μ and π are π-matrices, together with π·, π, ((1−π) πΌ+πβ¨π·β©−1 |πΏ|) β¨π·β©−1 ((1−π) πΌ+πβ¨π·β©−1 |π|) > 0, −1 −1 −1 ((1 − π) πΌ + ππ· πΏ) π· ((1 − π) πΌ + ππ· π) > 0. (32) Μ−π Μ and π2 β¨π΄β© = π − π are two We obtain that β¨π΄β© = π weak regular splittings. By simple calculation, we have π= ≤ −1 1 (π· − ππΏ) π· (π· − ππ) π (2 − π) 1 Μ (β¨π·β© − π |πΏ|) β¨π·β©−1 (β¨π·β© − π |π|) = π π (2 − π) (33) −1 (36) Μπ) . π (Lπ ) ≤ π (L (37) that is, π2 β¨π΄β© = π − π π= −1 Μ , Μ −1 π) π (π π) ≤ π (π −1 Μ ≥ 0; letting π₯ = β¨π΄β©−1 π > 0, then and thus π ≥ π −1 Μ −1 ≥ 0, we (π2 β¨π΄β© − β¨π΄β©)π₯ = (πΌ + π2 )π > 0; since π ≥ π have −1 −1 Μ −1 π) Μ π₯. Μ −1 β¨π΄β© π₯ = (πΌ− π π (π2 β¨π΄β©) π₯ = (πΌ−π π) π₯ ≥ π (34) When π΄ is a nonsingular π-matrix, we have π΄ = β¨π΄β©. If πΌπ > 0, π = 1, 2, . . . , π, then π2 = π1 . Furthermore, we have Μ π and Lπ = L Μ π ; therefore, we get the following Lπ = L result. Corollary 12. Let π΄ be a nonsingular π-matrix with unit diagonal elements, 0 < πΌπ < |βπ |, π = 1, 2, . . . , π, and 0 < π ≤ 1. Then Μ π ) ≤ π (L Μ π ) = π (Lπ ) . π (Lπ ) = π (L (38) Theorem 13. Let π΄ be a nonsingular π»-matrix with unit diagonal elements, 0 < π ≤ 1 and |πΌπ | < βπ , π = 1, 2, . . . , π. Then π(LΜ π ) ≤ π(Lπ ). Proof . Let β¨π1 π΄β© = 1 Μ −1 (β¨π·β© Μ σ΅¨σ΅¨σ΅¨σ΅¨) β¨π·β© Μ σ΅¨σ΅¨σ΅¨ Μ − π σ΅¨σ΅¨σ΅¨σ΅¨πΏ Μ − π σ΅¨σ΅¨σ΅¨σ΅¨π (β¨π·β© σ΅¨ σ΅¨ σ΅¨ σ΅¨σ΅¨) π (2 − π) 1 Μ −1 Μ σ΅¨σ΅¨σ΅¨σ΅¨) β¨π·β© Μ + π σ΅¨σ΅¨σ΅¨σ΅¨πΏ ((1 − π) β¨π·β© σ΅¨ σ΅¨ π (2 − π) Μ σ΅¨σ΅¨σ΅¨ Μ + π σ΅¨σ΅¨σ΅¨σ΅¨π × ((1 − π) β¨π·β© σ΅¨ σ΅¨σ΅¨) . − (39) Then the SSOR iteration matrix for β¨π1 π΄β© is LΜ π which is defined in the proof of Theorem 9, and let π2 β¨π΄β© = −1 1 (π· − ππΏ) π· (π· − ππ) π (2 − π) − −1 1 ((1 − π) π· + ππΏ) π· π (2 − π) (40) × ((1 − π) π· + ππ) . Then the SSOR iteration matrix for π2 β¨π΄β© is (17). It is easy to know that the previous two splittings are weak regular splittings. Furthermore, by Lemma 6, we have the following result, for any π, π = 1, 2, . . . , π, Μ ππ β© = β¨π΄ ππ − πΌπ π΄−1 π΄ ππ π΄ ππ β© β¨π· ππ −1 σ΅¨ σ΅¨σ΅¨ σ΅¨ σ΅¨ σ΅¨ ≥ β¨π΄ ππ β© − σ΅¨σ΅¨σ΅¨πΌπ σ΅¨σ΅¨σ΅¨ β¨π΄ ππ β© σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨ σ΅¨σ΅¨σ΅¨π΄ ππ σ΅¨σ΅¨σ΅¨ = β¨π·ππ β© . (41) From β¨π1 π΄β© and π2 β¨π΄β© being two π-matrices, we have Μ −1 ≤ π· −1 0 ≤ β¨π·β© (42) 6 Journal of Applied Mathematics Table 1: Comparison of spectral radius with preconditioner π1 . π, π Μπ) π(L 0.5636 0.6030 0.6120 0.3650 0.4510 0.4345 0.2387 0.3494 0.3569 0.3674 π 100 200 500 100 200 500 100 200 500 1000 π = 0.8 π = 0.6 π = 0.9 and then −1 −1 Μ −1 σ΅¨σ΅¨σ΅¨σ΅¨π Μ σ΅¨σ΅¨σ΅¨ Μ −1 σ΅¨σ΅¨σ΅¨ Μ σ΅¨σ΅¨σ΅¨ = (πΌ − πβ¨π·β© σ΅¨ σ΅¨σ΅¨) (πΌ − πβ¨π·β© σ΅¨σ΅¨πΏσ΅¨σ΅¨) Μ π ) ≤ π (LΜ π ) ≤ π (Lπ ) ≤ π (L Μ π ) < 1. π (L σ΅¨σ΅¨ Μ σ΅¨σ΅¨ Μ −1 σ΅¨σ΅¨σ΅¨σ΅¨π Μ σ΅¨σ΅¨σ΅¨ Μ σ΅¨σ΅¨πΏσ΅¨σ΅¨) ((1−π) πΌ+πβ¨π·β© × ((1−π) πΌ+πβ¨π·β© σ΅¨ σ΅¨ σ΅¨ σ΅¨σ΅¨) −1 = Lπ . (43) Therefore, by Lemma 3, π(LΜ π ) ≤ π(Lπ ). For randomly generated nonsingular π»-matrices for π = 100, 200, 500, 1000 with π1 = π2 = ⋅ ⋅ ⋅ = ππ = 5, we have determined the spectral radius of the iteration matrices of SSOR method mentioned previously with preconditioner π1 . We report the spectral radius of the corresponding iteration matrix by π. The π parameters πΌπ , π = 1, 2, ..., π, are taken from the π equal-partitioned points of the interval [0, 1]. We take π΄−1 11 π΄ 12 π13 ⋅⋅⋅ π΄−1 22 π΄−1 11 π΄ 23 d d ⋅⋅⋅ d d ππ−1,π−2 π΄−1 π−1,π−1 ππ2 (44) 5. Numerical Example −1 σ΅¨ σ΅¨σ΅¨ −1 σ΅¨ Μ σ΅¨σ΅¨ −1 Μσ΅¨ σ΅¨σ΅¨) (πΌ − ππ· σ΅¨σ΅¨σ΅¨πΏ ≤ (πΌ − ππ· σ΅¨σ΅¨σ΅¨σ΅¨π σ΅¨ σ΅¨ σ΅¨σ΅¨) −1 σ΅¨ σ΅¨σ΅¨ Μ σ΅¨σ΅¨) ((1 − π) πΌ + ππ·−1 σ΅¨σ΅¨σ΅¨σ΅¨π Μ σ΅¨σ΅¨σ΅¨ × ((1 − π) πΌ + ππ· σ΅¨σ΅¨σ΅¨σ΅¨πΏ σ΅¨ σ΅¨ σ΅¨σ΅¨) −1 π΄−1 ππ π΄ π1 π(Lπ ) 0.6288 0.7059 0.7103 0.4770 0.5769 0.5609 0.3602 0.4899 0.5019 0.5173 Theorem 14. Let π΄ be a nonsingular π»-matrix with unit diagonal elements, 0 < π ≤ 1 and |πΌπ | < βπ , π = 1, 2, . . . , π. Then σ΅¨σ΅¨ Μ σ΅¨σ΅¨ Μ −1 σ΅¨σ΅¨ Μ σ΅¨σ΅¨ Μ Μ σ΅¨σ΅¨πΏσ΅¨σ΅¨) β¨π·β© ((1−π) β¨π·β©+π σ΅¨σ΅¨πσ΅¨σ΅¨) × ((1−π) β¨π·β©+π σ΅¨ σ΅¨ σ΅¨ σ΅¨ π21 ( .. ( . π1 = ( ( π ( π−1,1 Μπ) π(L 0.8999 0.9473 0.9847 0.8530 0.9240 0.9835 0.8087 0.9009 0.9731 0.9840 Combining the previous Theorems, we can obtain the following conclusion. σ΅¨σ΅¨ Μ σ΅¨σ΅¨ −1 Μ σ΅¨σ΅¨σ΅¨ −1 Μ Μ − π σ΅¨σ΅¨σ΅¨σ΅¨π Μ σ΅¨ σ΅¨ LΜ π = (β¨π·β© σ΅¨ σ΅¨σ΅¨) β¨π·β© (β¨π·β© − π σ΅¨σ΅¨πΏσ΅¨σ΅¨) π΄−1 11 π(LΜ π ) 0.7404 0.7698 0.7844 0.5923 0.6507 0.6766 0.6647 0.5800 0.6284 0.6316 π(Lπ ) 0.8635 0.9195 0.9751 0.7906 0.8832 0.9730 0.7512 0.8491 0.9561 0.9738 ⋅⋅⋅ ππ,π−1 π1π .. . ) .. ) . ). −1 π΄ π−1,π−1 π΄ π−1,π ) ) π΄−1 ππ ) ( For π1 , we make two groups of experiments. In Figure 1, we test the relation between π and π, when π = 100, π = 0.6, where “×”, “+”, “∗”, “⋅” and “β” denote the spectral radius of β¨π΄β©, π2 β¨π΄β©, β¨π1 π΄β©, π΄, and π1 π΄, respectively. In Table 1, the Μ π ), π(Lπ ), π(LΜ π ), π(L Μ π ), and meaning of notations π(L π(Lπ ) denotes the spectral radius of π1 π΄, π2 β¨π΄β©, β¨π1 π΄β©, β¨π΄β©, and π΄, respectively. From Figure 1 and Table 1, we can conclude that the spectral radius of the preconditioned SSOR method with (45) preconditioner π1 is the best among others, which further illustrates that, Theorem 14 is true. Acknowledgments The authors express their thanks to the editor Professor HakKeung Lam and the anonymous referees who made much useful and detailed suggestions that helped them to correct some minor errors and improve the quality of the paper. Journal of Applied Mathematics 7 1 0.9 0.8 π 0.7 0.6 0.5 0.4 0.3 0.2 0 0.1 0.2 0.3 0.4 0.5 π 0.6 0.7 0.8 0.9 1 Figure 1: The relation between π and π, when π = 100, π = 0.6. References [1] M. Alanelli and A. Hadjidimos, “Block Gauss elimination followed by a classical iterative method for the solution of linear systems,” Journal of Computational and Applied Mathematics, vol. 163, no. 2, pp. 381–400, 2004. [2] A. Berman and R. J. Plemmons, Nonnegative Matrices in the Mathematical Sciences, vol. 9 of Classics in Applied Mathematics, Society for Industrial and Applied Mathematics (SIAM), Philadelphia, Pa, USA, 1994. [3] W. Li and Z.-y. You, “The multi-parameters overrelaxation method,” Journal of Computational Mathematics, vol. 16, no. 4, pp. 367–374, 1998. [4] R. S. Varga, Matrix Iterative Analysis, Prentice-Hall, Englewood Cliffs, NJ, USA, 1962. [5] L. Yu. Kolotilina, “Two-sided bounds for the inverse of an Hmatrix,” Linear Algebra and Its Applications, vol. 225, pp. 117–123, 1995. [6] M. Neumann and R. J. Plemmons, “Convergence of parallel multisplitting iterative methods for M-matrices,” Linear Algebra and Its Applications, vol. 88-89, pp. 559–573, 1987. Advances in Operations Research Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 Advances in Decision Sciences Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 Mathematical Problems in Engineering Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 Journal of Algebra Hindawi Publishing Corporation http://www.hindawi.com Probability and Statistics Volume 2014 The Scientific World Journal Hindawi Publishing Corporation http://www.hindawi.com Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 International Journal of Differential Equations Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 Volume 2014 Submit your manuscripts at http://www.hindawi.com International Journal of Advances in Combinatorics Hindawi Publishing Corporation http://www.hindawi.com Mathematical Physics Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 Journal of Complex Analysis Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 International Journal of Mathematics and Mathematical Sciences Journal of Hindawi Publishing Corporation http://www.hindawi.com Stochastic Analysis Abstract and Applied Analysis Hindawi Publishing Corporation http://www.hindawi.com Hindawi Publishing Corporation http://www.hindawi.com International Journal of Mathematics Volume 2014 Volume 2014 Discrete Dynamics in Nature and Society Volume 2014 Volume 2014 Journal of Journal of Discrete Mathematics Journal of Volume 2014 Hindawi Publishing Corporation http://www.hindawi.com Applied Mathematics Journal of Function Spaces Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 Optimization Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 Hindawi Publishing Corporation http://www.hindawi.com Volume 2014