About this Journal Submit a Manuscript Table of Contents 
ISRN Computational Mathematics
VolumeΒ 2012Β (2012), Article IDΒ 126908, 6 pages
doi:10.5402/2012/126908
Research Article

A New Iterative Algorithm for Solving a Class of Matrix Nearness Problem

Xuefeng Duan1,2Β and Chunmei Li1

1College of Mathematics and Computational Science, Guilin University of Electronic Technology, Guilin 541004, China
2Department of Mathematics, Shanghai University, Shanghai 200444, China

Received 14 September 2011; Accepted 3 October 2011

Academic Editors: T.Β Allahviranloo and K.Β Eom

Copyright Β© 2012 Xuefeng Duan and Chunmei Li. 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

Based on the alternating projection algorithm, which was proposed by Von Neumann to treat the problem of finding the projection of a given point onto the intersection of two closed subspaces, we propose a new iterative algorithm to solve the matrix nearness problem associated with the matrix equations 𝐴 𝑋 𝐡 = 𝐸 , 𝐢 𝑋 𝐷 = 𝐹 , which arises frequently in experimental design. If we choose the initial iterative matrix 𝑋 0 = 0 , the least Frobenius norm solution of these matrix equations is obtained. Numerical examples show that the new algorithm is feasible and effective.

1. Introduction

Denoted by 𝑅 π‘š Γ— 𝑛 be the set of π‘š Γ— 𝑛 real matrices, 𝐴 𝑇 and 𝐴 + be the transpose and Moore-Penrose generalized inverse of the matrix 𝐴 , respectively. For 𝐴 , 𝐡 ∈ 𝑅 π‘š Γ— 𝑛 ,   ⟨ 𝐴 , 𝐡 ⟩ = t r a c e ( 𝐡 𝑇 𝐴 ) denotes the inner product of the matrix 𝐴 and 𝐡 . The induced norm is the so-called Frobenius norm, that is, β€– 𝐴 β€– = ⟨ 𝐴 , 𝐴 ⟩ 1 / 2 ; then 𝑅 π‘š Γ— 𝑛 is a real Hilbert space. Let 𝑀 be a closed convex subset in a real Hilbert space 𝐻 and 𝑒 be a point in 𝐻 ; then the point in 𝑀 nearest to 𝑒 is called the projection of 𝑒 onto 𝑀 and denoted by 𝑃 𝑀 ( 𝑒 ) ; that is to say, 𝑃 𝑀 ( 𝑒 ) is the solution of the following minimization problem (see [1, 2]) m i n π‘₯ ∈ 𝑀 β€– π‘₯ βˆ’ 𝑒 β€– , ( 1 ) that is, β€– β€– 𝑃 𝑀 β€– β€– ( 𝑒 ) βˆ’ 𝑒 = m i n π‘₯ ∈ 𝑀 β€– π‘₯ βˆ’ 𝑒 β€– . ( 2 )

The problem of finding a nearness matrix  𝑋 in a constraint matrix set to a given matrix 𝑋 is called the matrix nearness problem. Because the preliminary estimation 𝑋 is frequently obtained from experiments, it may not satisfy the given restrictions. Hence it is necessary to find a nearness matrix  𝑋 in this constraint matrix set to replace the estimation 𝑋 [3]. In the area of structure design, finite element model updating and control theory, and so forth, the matrix set is always the (constraint) solution set or the least square (constraint) solution set of some matrix equations [4–6]. Thus, the problem mentioned above is also called the matrix nearness problem associated with the matrix equation. Recently, there are many discussions on the matrix nearness problem associated with some matrix equations. For instance, see [4, 6–14].

In this paper, we consider the following problem.

Problem 1. Given matrices   𝐴 ∈ 𝑅 𝑝 Γ— 𝑛 , 𝐡 ∈ 𝑅 𝑛 Γ— π‘ž , 𝐢 ∈ 𝑅 𝑠 Γ— 𝑛 , 𝐷 ∈ 𝑅 𝑛 Γ— 𝑑 , 𝐸 ∈ 𝑅 𝑝 Γ— π‘ž , 𝐹 ∈ 𝑅 𝑠 Γ— 𝑑 , and 𝑋 ∈ 𝑅 𝑛 Γ— 𝑛 , find  𝑋 ∈ Ξ© such that β€– β€–  𝑋 βˆ’ 𝑋 β€– β€– = m i n 𝑋 ∈ Ξ© β€– β€– 𝑋 βˆ’ 𝑋 β€– β€– , ( 3 ) where ξ€½ Ξ© = 𝑋 ∈ 𝑅 𝑛 Γ— 𝑛 ξ€Ύ ∣ 𝐴 𝑋 𝐡 = 𝐸 , 𝐢 𝑋 𝐷 = 𝐹 . ( 4 ) Obviously, Ξ© is the solution set of the matrix equations 𝐴 𝑋 𝐡 = 𝐸 , 𝐢 𝑋 𝐷 = 𝐹 , ( 5 ) and  𝑋 is the optimal approximate solution of (5) to the given matrix 𝑋 . In particular, if 𝑋 = 0 , then the solution  𝑋 of Problem 1 is just the least Frobenius norm solution of the matrix equations (5). It is easy to verify that Ξ© is convex set; then the solution of Problem 1 is unique.

The matrix equations (5) and its matrix nearness problem have been extensively studied for the past 40 or more years. Wang [15] and Navarra et al. [16] gave some conditions for the existence of a solution and some representations of the general common solution to (5). By the projection theorem and matrix decompositions, Liao et al. [6] gave an analytical expression of the optimal approximate least square symmetric solution of (5). However, these direct methods may be less efficient for the large sparse coefficient matrices due to the limited storages and the speeds of the computers. Therefore, iterative methods for solving the matrix equations (5) have attracted much interests recently. Peng et al. [11] and Chen et al. [7] proposed some iterative methods to compute the symmetric solutions and optimal approximate symmetric solution of (5). An efficient iterative method was presented to solve the matrix nearness Problem 1 associated with the matrix equations (5) in [13]. Ding et al. [17] considered the unique solution of the matrix equations (5) and used gradient-based iterative algorithm to compute the unique solution. The (least square) solution and the optimal approximate (least square) solution of (5), which is constrained as bisymmetric, reflexive, generalized reflexive, generalized centrosymmetric, were studied in [7–10, 12].

The alternating projection algorithm dates back to von Neumann [18], who treated the problem of finding the projection of a given point onto the intersection of two closed subspaces. Later, Bauschke and Borwein [1] extended the analysis of Von Neumann’s alternating projection scheme to the case of two closed affine subspaces. There are many variations and extensions of the alternating projection algorithm, and we can use them to find the projection of a given point onto the intersection of π‘˜ ( π‘˜ β‰₯ 2 ) closed subspaces [19] and π‘˜ ( π‘˜ β‰₯ 2 ) closed convex sets [20, 21]. For a complete discussion on the alternation projection algorithm see Deutsch [2].

In this paper, we propose a new algorithm to solve Problem 1. We state Problem 1 as the minimization of a convex quadratic function over the intersection of two closed affine subspaces in the vector space 𝑅 𝑛 Γ— 𝑛 . From this point of view, Problem 1 can be solved by the alternating projection algorithm. If we choose the initial iterative matrix 𝑋 0 = 0 , the least Frobenius norm solution of the matrix equations 𝐴 𝑋 𝐡 = 𝐸 , 𝐢 𝑋 𝐷 = 𝐹 is obtained. In the end, we use some numerical examples to show that the new algorithm is faster and lower computational cost for each step than the algorithm proposed by Sheng and Chen [13] to solve Problem 1. Especially, the CPU time and iteration steps of our algorithm increase slowly as the dimension of the matrix is increasing; so our algorithm is suitable for large-scale problems.

2. Alternating Projection Algorithm for Solving Problem 1

In this section, we apply the alternating projection algorithm to solve Problem 1. We begin with two lemmas.

Lemma 1 (see [1, Theorem  4.1]). Let 𝑀 1 ξ‚‹ 𝑀 = π‘Ž + 1 , 𝑀 2 ξ‚‹ 𝑀 = 𝑏 + 2 be closed affine subspaces in a Hilbert space 𝐻 and u be a point in 𝐻 . Here, ξ‚‹ 𝑀 1 and ξ‚‹ 𝑀 2 are closed subspaces and π‘Ž , 𝑏 ∈ 𝐻 . If 𝑀 1 ∩ 𝑀 2 β‰  βˆ… , then the sequences { π‘₯ π‘˜ } and { 𝑦 π‘˜ } generated by the alternating projection algorithm π‘₯ 0 = 𝑒 , 𝑦 π‘˜ + 1 = 𝑃 𝑀 1 ξ€· π‘₯ π‘˜ ξ€Έ , π‘₯ π‘˜ + 1 = 𝑃 𝑀 2 ξ€· 𝑦 π‘˜ + 1 ξ€Έ , π‘˜ = 0 , 1 , 2 , … ( 6 ) both converge to the projection of the point 𝑒 onto the intersection of 𝑀 1 and 𝑀 2 , that is, π‘₯ π‘˜ ⟢ 𝑃 𝑀 1 ∩ 𝑀 2 ( 𝑒 ) , 𝑦 π‘˜ ⟢ 𝑃 𝑀 1 ∩ 𝑀 2 ( 𝑒 ) , π‘˜ ⟢ + ∞ . ( 7 )

Lemma 2 (see [22, Theorem  9.3.1]). Let 𝐴 ∈ 𝑅 𝑝 Γ— 𝑛 , 𝐡 ∈ 𝑅 𝑛 Γ— π‘ž and 𝐸 ∈ 𝑅 𝑝 Γ— π‘ž , be known matrices. Then the matrix equation 𝐴 𝑋 𝐡 = 𝐸 has a solution if and only if 𝐴 𝐴 + 𝐸 𝐡 + 𝐡 = 𝐸 , ( 8 ) and the representation of the solution is 𝑋 = 𝐴 + 𝐸 𝐡 + + π‘Œ βˆ’ 𝐴 + 𝐴 π‘Œ 𝐡 𝐡 + , ( 9 ) where π‘Œ ∈ 𝑅 𝑛 Γ— 𝑛 is arbitrary.

Lemma 3 (see [22, Theorem  9.3.2]). Given 𝑍 ∈ 𝑅 𝑛 Γ— 𝑛 , set ξ€½ β„œ = 𝑋 ∈ 𝑅 𝑛 Γ— 𝑛 ∣ 𝐴 𝑋 𝐡 = 𝐸 , 𝐴 ∈ 𝑅 𝑝 Γ— 𝑛 , 𝐡 ∈ 𝑅 𝑛 Γ— π‘ž , 𝐸 ∈ 𝑅 𝑝 Γ— π‘ž ξ€Ύ , ( 1 0 ) then the solution  𝑋 of the following problem m i n 𝑋 ∈ β„œ β€– 𝑋 βˆ’ 𝑍 β€– ( 1 1 ) is  𝑋 = 𝑍 + 𝐴 + ( 𝐸 βˆ’ 𝐴 𝑍 𝐡 ) 𝐡 + , ( 1 2 ) that is, β€– β€–  β€– β€– 𝑋 βˆ’ 𝑍 = m i n 𝑋 ∈ β„œ β€– 𝑋 βˆ’ 𝑍 β€– . ( 1 3 )

Now we begin to use the alternating projection algorithm (6) to solve Problem 1. Firstly, we define two sets Ξ© 1 = ξ€½ 𝑋 ∈ 𝑅 𝑛 Γ— 𝑛 ξ€Ύ , Ξ© ∣ 𝐴 𝑋 𝐡 = 𝐸 2 = ξ€½ 𝑋 ∈ 𝑅 𝑛 Γ— 𝑛 ξ€Ύ . ∣ 𝐢 𝑋 𝐷 = 𝐹 ( 1 4 ) It is easy to know that Ξ© = Ξ© 1 ∩ Ξ© 2 , and if the set Ξ© is nonempty, then Ξ© = Ξ© 1 ∩ Ξ© 2 β‰  βˆ… . ( 1 5 ) And by Lemma 2, the sets Ξ© 1 and Ξ© 2 can be equivalently written as Ξ© 1 = ξ€½ = { 𝑋 ∣ 𝐴 𝑋 𝐡 = 𝐸 } 𝑋 = 𝐴 + 𝐸 𝐡 + + π‘Œ βˆ’ 𝐴 + 𝐴 π‘Œ 𝐡 𝐡 + ∣ π‘Œ ∈ 𝑅 𝑛 Γ— 𝑛 ξ€Ύ = 𝐴 + 𝐸 𝐡 + + ξ€½ π‘Œ βˆ’ 𝐴 + 𝐴 π‘Œ 𝐡 𝐡 + ∣ π‘Œ ∈ 𝑅 𝑛 Γ— 𝑛 ξ€Ύ , Ξ© 2 = ξ€½ = { 𝑋 ∣ 𝐢 𝑋 𝐷 = 𝐹 } 𝑋 = 𝐢 + 𝐹 𝐷 + + π‘Œ βˆ’ 𝐢 + 𝐢 π‘Œ 𝐷 𝐷 + ∣ π‘Œ ∈ 𝑅 𝑛 Γ— 𝑛 ξ€Ύ = 𝐢 + 𝐹 𝐷 + + ξ€½ π‘Œ βˆ’ 𝐢 + 𝐢 π‘Œ 𝐷 𝐷 + ∣ π‘Œ ∈ 𝑅 𝑛 Γ— 𝑛 ξ€Ύ . ( 1 6 ) Hence, Ξ© 1 and Ξ© 2 are closed affine subspaces.

After defining the sets Ξ© 1 and Ξ© 2 , Problem 1 can be rewritten as finding  𝑋 ∈ Ξ© = Ξ© 1 ∩ Ξ© 2 , such that β€– β€–  𝑋 βˆ’ 𝑋 β€– β€– = m i n 𝑋 ∈ Ξ© 1 ∩ Ξ© 2 β€– β€– 𝑋 βˆ’ 𝑋 β€– β€– . ( 1 7 ) Noting the equalities (17) and (2), it is easy to find that  𝑋 = 𝑃 Ξ© 1 ∩ Ξ© 2 ξ‚€ 𝑋  . ( 1 8 ) Therefore, Problem 1 can be converted equivalently into finding the projection 𝑃 Ξ© 1 ∩ Ξ© 2 ( 𝑋 ) of a given matrix 𝑋 onto the intersection set Ξ© 1 ∩ Ξ© 2 . Now we will use alternating projection algorithm (6) to compute the projection 𝑃 Ξ© 1 ∩ Ξ© 2 ( 𝑋 ) . Consequently, we can get the solution  𝑋 of Problem 1.

By (6) we can see that the key problems to realize the alternating projection algorithm (6) are how to compute the projections 𝑃 Ξ© 1 ( 𝑍 ) , 𝑃 Ξ© 2 ( 𝑍 ) of a matrix 𝑍 onto Ξ© 1 and Ξ© 2 , respectively. Such problems are perfectly solvable in the following theorems.

Theorem 1. Suppose that the set Ξ© 1 is nonempty. For a given 𝑛 Γ— 𝑛 matrix 𝑍 , we have 𝑃 Ξ© 1 ( 𝑍 ) = 𝑍 + 𝐴 + ( 𝐸 βˆ’ 𝐴 𝑍 𝐡 ) 𝐡 + . ( 1 9 )

Proof. By (1) and (2), we know that the projection 𝑃 Ξ© 1 ( 𝑍 ) is the solution of the following minimization problem m i n 𝑋 ∈ Ξ© 1 β€– 𝑋 βˆ’ 𝑍 β€– , ( 2 0 ) and according to Lemma 3 we know that the solution of the minimization problem (20) is 𝑍 + 𝐴 + ( 𝐸 βˆ’ 𝐴 𝑍 𝐡 ) 𝐡 + . Hence, 𝑃 Ξ© 1 ( 𝑍 ) = 𝑍 + 𝐴 + ( 𝐸 βˆ’ 𝐴 𝑍 𝐡 ) 𝐡 + . ( 2 1 )

Theorem 2. Suppose that the set   Ξ© 2 is nonempty. For a given 𝑛 Γ— 𝑛 matrix 𝑍 , we have 𝑃 Ξ© 2 ( 𝑍 ) = 𝑍 + 𝐢 + ( 𝐹 βˆ’ 𝐢 𝑍 𝐷 ) 𝐷 + . ( 2 2 )

Proof. The proof is similar to that of Theorem 1 and is omitted here.

By the alternation projection algorithm (6) and Theorems 1 and 2, we get a new algorithm to solve Problem 1 which can be stated as follows.

Algorithm 1. One has(1)set  𝐴 = 𝐴 + ,  𝐡 = 𝐡 + ,  𝐢 = 𝐢 + ,  𝐷 = 𝐷 + ;(2)set 𝑋 0 = 𝑋 ;(3)for π‘˜ = 0 , 1 , 2 , 3 , … π‘Œ π‘˜ + 1 = 𝑃 Ξ© 1 ξ€· 𝑋 π‘˜ ξ€Έ = 𝑋 π‘˜ +  𝐴 ξ€· 𝐸 βˆ’ 𝐴 𝑋 π‘˜ 𝐡 ξ€Έ  𝑋 𝐡 , π‘˜ + 1 = 𝑃 Ξ© 2 ξ€· π‘Œ π‘˜ + 1 ξ€Έ = π‘Œ π‘˜ + 1 +  𝐢 ξ€· 𝐹 βˆ’ 𝐢 π‘Œ π‘˜ + 1 𝐷 ξ€Έ  𝐷 , ( 2 3 ) end.
By Lemma 1 and (15) and (16), we get the convergence theorem for Algorithm 1.

Theorem 3. If the set Ξ© is nonempty, then the matrix sequences { 𝑋 π‘˜ } and { π‘Œ π‘˜ } generated by Algorithm 1 both converge to the projection 𝑃 Ξ© 1 ∩ Ξ© 2 ( 𝑋 ) of   𝑋 onto the intersection of Ξ© 1 and Ξ© 2 , that is, 𝑋 π‘˜ ⟢ 𝑃 Ξ© 1 ∩ Ξ© 2 ξ‚€ 𝑋  , π‘Œ π‘˜ ⟢ 𝑃 Ξ© 1 ∩ Ξ© 2 ξ‚€ 𝑋  , π‘˜ ⟢ + ∞ . ( 2 4 )

Proof. If the set Ξ© is nonempty, by (15) we have Ξ© 1 ∩ Ξ© 2 β‰  βˆ… . ( 2 5 ) And noting (16), we know that the sets Ξ© 1 and Ξ© 2 are closed affine subspaces in Hilbert space 𝑅 π‘š Γ— 𝑛 . Hence, by Lemma 1 we derive that the matrix sequences { 𝑋 π‘˜ } and { π‘Œ k } generated by Algorithm 1 both converge to the projection 𝑃 Ξ© 1 ∩ Ξ© 2 ( 𝑋 ) of 𝑋 onto the intersection of Ξ© 1 and Ξ© 2 , that is, 𝑋 π‘˜ ⟢ 𝑃 Ξ© 1 ∩ Ξ© 2 ξ‚€ 𝑋  , π‘Œ π‘˜ ⟢ 𝑃 Ξ© 1 ∩ Ξ© 2 ξ‚€ 𝑋  , π‘˜ ⟢ + ∞ . ( 2 6 )

Combining Theorem 3 and the equalities (18) and (17), we have the following.

Theorem 4. If the set Ξ© is nonempty, then the matrix sequence { 𝑋 π‘˜ } and { π‘Œ π‘˜ } generated by Algorithm 1 both converge to the unique solution of Problem 1. Moreover, if the initial matrix 𝑋 0 = 𝑋 = 0 , then the matrix sequence { 𝑋 π‘˜ } and { π‘Œ π‘˜ } both converge to the least Frobenius norm solution of the matrix equations 𝐴 𝑋 𝐡 = 𝐸 , 𝐢 𝑋 𝐷 = 𝐹 .

3. Numerical Experiments

In this section, we give some numerical examples to illustrate that the new algorithm is feasible and effective to solve Problem 1. All programs are written in MATLAB 7.8. We denote E r r o r = β€– 𝐸 βˆ’ 𝐴 𝑋 𝐡 β€– + β€– 𝐹 βˆ’ 𝐢 𝑋 𝐷 β€– , ( 2 7 ) and use the practical stopping criterion E r r o r ≀ 1 . 0 Γ— 1 0 βˆ’ 1 0 .

Example 1. Consider Problem 1 with βŽ› ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ ⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ , βŽ› ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ ⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ , βŽ› ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ ⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ , βŽ› ⎜ ⎜ ⎜ ⎜ ⎝ ⎞ ⎟ ⎟ ⎟ ⎟ ⎠ , βŽ› ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ ⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ , βŽ› ⎜ ⎜ ⎜ ⎜ ⎝ ⎞ ⎟ ⎟ ⎟ ⎟ ⎠ . 𝐴 = 1 3 1 3 1 3 βˆ’ 7 3 βˆ’ 7 3 1 1 6 1 1 6 1 1 βˆ’ 5 5 βˆ’ 5 5 βˆ’ 5 9 4 9 4 9 1 3 1 3 1 𝐡 = βˆ’ 1 4 βˆ’ 1 4 βˆ’ 1 5 βˆ’ 1 5 βˆ’ 1 5 βˆ’ 1 βˆ’ 2 βˆ’ 1 βˆ’ 2 βˆ’ 1 3 9 3 9 3 7 βˆ’ 8 7 βˆ’ 8 7 𝐸 = 1 1 7 1 8 1 1 7 1 8 1 1 7 βˆ’ 6 5 βˆ’ 1 0 βˆ’ 6 5 βˆ’ 1 0 βˆ’ 6 5 5 8 5 9 0 5 8 5 9 0 5 8 5 βˆ’ 6 5 βˆ’ 1 0 βˆ’ 6 5 βˆ’ 1 0 βˆ’ 6 5 4 5 5 7 0 4 5 5 7 0 4 5 5 1 1 7 1 8 1 1 7 1 8 1 1 7 𝐢 = 3 βˆ’ 5 3 βˆ’ 5 7 βˆ’ 4 3 βˆ’ 2 9 5 3 βˆ’ 5 3 βˆ’ 5 7 𝐷 = 5 4 βˆ’ 1 5 1 4 βˆ’ 2 3 5 βˆ’ 2 5 3 3 5 βˆ’ 1 3 βˆ’ 1 5 3 βˆ’ 6 3 2 3 βˆ’ 6 1 1 9 7 1 7 1 9 𝐹 = 3 0 7 5 3 9 2 7 4 5 7 5 1 1 0 2 7 5 1 4 3 9 9 1 6 5 2 7 5 3 0 7 5 3 9 2 7 4 5 7 5 ( 2 8 ) Here we use o n e s ( 𝑛 ) and z e r o s ( 𝑛 ) to stand for 𝑛 Γ— 𝑛 matrix of ones and zeros. It is easy to verify that 𝑋 = o n e s ( 5 ) is a solution of the matrix equations (5); that is to say, the set Ξ© is nonempty. Therefore we can use Algorithm 1 to solve Problem 1.

Let 𝑋 0 = 𝑋 = z e r o s ( 5 ) . After 5 iterations of Algorithm 1, we get the optimal approximate solution  𝑋 β‰ˆ 𝑋 5 = βŽ› ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ ⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ βˆ’ 0 . 6 8 1 7 0 . 7 8 1 3 βˆ’ 0 . 5 5 0 3 0 . 9 6 3 7 0 . 7 8 0 8 0 . 0 1 5 2 0 . 8 1 8 5 βˆ’ 0 . 2 8 6 6 0 . 9 6 9 9 0 . 8 1 8 1 0 . 0 1 1 3 0 . 8 1 7 8 βˆ’ 0 . 2 9 1 7 0 . 9 6 9 8 0 . 8 1 7 3 0 . 6 0 9 1 0 . 9 2 8 0 0 . 4 8 9 3 0 . 9 9 8 0 0 . 9 2 7 8 0 . 9 4 9 7 0 . 9 9 0 7 0 . 9 3 4 2 0 . 9 9 8 5 0 . 9 9 0 7 , ( 2 9 ) which is also the least Frobenius norm solution of the matrix equations (5), and its residual error β€– β€– E r r o r β‰ˆ 𝐸 βˆ’ 𝐴 𝑋 5 𝐡 β€– β€– + β€– β€– 𝐹 βˆ’ 𝐢 𝑋 5 𝐷 β€– β€– = 2 . 7 8 Γ— 1 0 βˆ’ 1 1 . ( 3 0 ) By concrete computations, we know that the distance from 𝑋 to the solution set Ξ© is m i n 𝑋 ∈ Ξ© β€– β€– 𝑋 βˆ’ 𝑋 β€– β€– = β€– β€–  𝑋 βˆ’ 𝑋 β€– β€– β‰ˆ β€– β€– 𝑋 5 βˆ’ 𝑋 β€– β€– = 3 . 9 0 5 7 . ( 3 1 )

Let 𝑋 0 = 𝑋 = o n e s ( 5 ) . After 6 iterations of Algorithm 1, we get the optimal approximate solution  𝑋 β‰ˆ 𝑋 6 = βŽ› ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ ⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ 5 . 7 4 6 7 1 . 8 7 4 7 7 . 2 0 1 3 1 . 1 4 5 2 1 . 8 7 7 0 4 . 9 3 9 2 1 . 7 2 5 9 6 . 1 4 6 3 1 . 1 2 0 5 1 . 7 2 7 8 4 . 9 5 4 8 1 . 7 2 8 8 6 . 1 6 6 8 1 . 1 2 0 9 1 . 7 3 0 7 2 . 5 6 3 6 1 . 2 8 8 1 3 . 0 4 2 7 1 . 4 7 8 1 . 2 8 8 9 1 . 2 0 1 3 1 . 0 3 7 1 1 . 2 6 3 0 1 . 0 0 6 2 1 . 3 7 2 , ( 3 2 ) and its residual error β€– β€– E r r o r β‰ˆ 𝐸 βˆ’ 𝐴 𝑋 6 𝐡 β€– β€– + β€– β€– 𝐹 βˆ’ 𝐢 𝑋 6 𝐷 β€– β€– = 3 . 8 9 Γ— 1 0 βˆ’ 1 1 . ( 3 3 ) By concrete computations, we know that the distance from 𝑋 to the solution set Ξ© is m i n 𝑋 ∈ Ξ© β€– β€– 𝑋 βˆ’ 𝑋 β€– β€– = β€– β€–  𝑋 βˆ’ 𝑋 β€– β€– β‰ˆ β€– β€– 𝑋 6 βˆ’ 𝑋 β€– β€– = 1 6 . 0 9 0 2 . ( 3 4 )

Example 1 shows that Algorithm 1 is feasible to solve Problem 1.

Example 2. Consider Problem 1 with 𝐴 = r a n d ( 1 0 0 , 𝑛 ) , 𝐡 = r a n d ( 𝑛 , 1 5 0 ) , 𝐸 = 𝐴 βˆ— o n e s ( 𝑛 ) βˆ— 𝐡 , 𝐢 = r a n d ( 7 0 , 𝑛 ) , 𝐷 = r a n d ( 𝑛 , 1 2 0 ) , 𝐹 = 𝐢 βˆ— o n e s ( 𝑛 ) βˆ— 𝐷 , ( 3 5 ) where r a n d ( 𝑠 , 𝑑 ) stand for 𝑠 Γ— 𝑑 random matrix. Let the given matrix 𝑋 = z e r o s ( 𝑛 ) . It is easy to verify that 𝑋 = o n e s ( 𝑛 ) is the solution of the matrix equations (5); that is, the set Ξ© is nonempty; therefore, we can use Algorithm 1 and the following algorithm proposed by Sheng and Chen [13] to solve Problem 1.

Algorithm 2. One has(1)input 𝐴 , 𝐡 , 𝐢 , 𝐷 , 𝐸 , 𝐹 and 𝑋 0 ;(2)calculate 𝑅 0 = 𝐸 βˆ’ 𝐴 𝑋 0 π‘Ÿ 𝐡 , 0 = 𝐹 βˆ’ 𝐢 𝑋 0 𝑃 𝐷 , 0 = 𝐴 𝑇 𝑅 0 𝐡 𝑇 , 𝑄 0 = 𝐢 𝑇 π‘Ÿ 0 𝐷 𝑇 ; ( 3 6 ) (3)if 𝑅 π‘˜ = 0 , π‘Ÿ 0 = 0 , then stop; else, π‘˜ ∢ = π‘˜ + 1 ; (4)calculate 𝑋 π‘˜ = 𝑋 π‘˜ βˆ’ 1 + β€– 𝑅 π‘˜ βˆ’ 1 β€– 2 + β€– π‘Ÿ π‘˜ βˆ’ 1 β€– 2 β€– 𝑃 π‘˜ βˆ’ 1 + 𝑄 π‘˜ βˆ’ 1 β€– 2 ξ€· 𝑃 π‘˜ βˆ’ 1 + 𝑄 π‘˜ βˆ’ 1 ξ€Έ , 𝑅 π‘˜ = 𝐸 βˆ’ 𝐴 𝑋 π‘˜ π‘Ÿ 𝐡 , π‘˜ = 𝐹 βˆ’ 𝐢 𝑋 π‘˜ 𝑃 𝐷 , π‘˜ = 𝐴 𝑇 𝑅 π‘˜ 𝐡 𝑇 + β€– 𝑅 π‘˜ β€– 2 + β€– π‘Ÿ π‘˜ β€– 2 β€– 𝑅 π‘˜ βˆ’ 1 β€– 2 + β€– π‘Ÿ π‘˜ βˆ’ 1 β€– 2 , 𝑄 π‘˜ = 𝐢 𝑇 π‘Ÿ π‘˜ 𝐷 𝑇 + β€– 𝑅 π‘˜ β€– 2 + β€– π‘Ÿ π‘˜ β€– 2 β€– 𝑅 π‘˜ βˆ’ 1 β€– 2 + β€– π‘Ÿ π‘˜ βˆ’ 1 β€– 2 𝑄 π‘˜ βˆ’ 1 ; ( 3 7 ) (5)go to step 3.

It is easy to see that Algorithm 1 has lower computational cost for each step in the comparison with Algorithm 2. Experiments show that Algorithm 1 and Algorithm 2 are feasible to solve Problem 1. We list the iteration steps (denoted by IT), CPU time (denoted by CPU), residual error (denoted by ERR), and the distance β€– 𝑋 I T βˆ’ 𝑋 β€– (denoted by DIS) in Table 1.

tab1
Table 1

From Table 1, we can see that Algorithm 1 outperforms Algorithm 2 in iteration step and CPU time. Therefore our algorithm is faster than the algorithm proposed by Sheng and Chen [13]. Especially, the CPU time and iteration steps of our algorithm increase slowly as the dimension 𝑛 is increasing; so our algorithm is suitable for large-scale problems.

4. Conclusion

The alternating projection algorithm dates back to von Neumann [18], who treated the problem of finding the projection of a given point onto the intersection of two closed subspace. In this paper, we first apply the alternating projection algorithm to solve Problem 1, which occurs frequently in experimental design [23]. If we choose the initial matrix 𝑋 0 = 0 , the least Frobenius norm solution of the matrix equations 𝐴 𝑋 𝐡 = 𝐸 , 𝐢 𝑋 𝐷 = 𝐹 can be obtained. Numerical examples show that the new algorithm is faster and lower computational cost for each step than the algorithm proposed by Sheng and Chen [13] to solve Problem 1. Especially, the CPU time and iteration steps of the new algorithm increase slowly as the matrix’s dimension is increasing; so the alternating projection algorithm is suitable for large-scale matrix nearness problems.

Acknowledgments

The work was supported in part by National Science Foundation of China (11101100; 10861005) and Natural Science Foundation of Guangxi Province (0991238; 2011GXNSFA018138).

References

  1. H. H. Bauschke and J. M. Borwein, β€œDykstra′s alternating projection algorithm for two sets,” Journal of Approximation Theory, vol. 79, no. 3, pp. 418–443, 1994. View at Publisher Β· View at Google Scholar Β· View at MathSciNet Β· View at Scopus
  2. F. Deutsch, Best Approximation in Inner Product Spaces, Springer, New York, NY, USA, 2001.
  3. N. Higham, Matrix Nearness Problems and Applications, Oxford University Press, London, UK, 1989.
  4. H. Dai and P. Lancaster, β€œLinear matrix equations from an inverse problem of vibration theory,” Linear Algebra and Its Applications, vol. 246, pp. 31–47, 1996. View at Publisher Β· View at Google Scholar Β· View at Zentralblatt MATH Β· View at MathSciNet
  5. M. Friswell and J. Mottorshead, Finite Element Model Updating in Structure Dynamics, Kluwer Academic publisher, Dodrecht, The Netherlands, 1995.
  6. A. P. Liao, Y. Lei, and S. F. Yuan, β€œThe matrix nearness problem for symmetric matrices associated with the matrix equation [ATXA, BTXB] = [C,D],” Linear Algebra and Its Applications, vol. 418, no. 2-3, pp. 939–954, 2006. View at Publisher Β· View at Google Scholar Β· View at MathSciNet
  7. Y. Chen, Z. Peng, and T. Zhou, β€œLSQR iterative common symmetric solutions to matrix equations AXB=E and CXD=F,” Applied Mathematics and Computation, vol. 217, no. 1, pp. 230–236, 2010. View at Publisher Β· View at Google Scholar
  8. J. Cai and G. Chen, β€œAn iterative algorithm for the least squares bisymmetric solutions of the matrix equations A1XB1=C1, A2XB2=C2,” Mathematical and Computer Modelling, vol. 50, no. 7-8, pp. 1237–1244, 2009. View at Publisher Β· View at Google Scholar
  9. M. Dehghan and M. Hajarian, β€œAn iterative algorithm for solving a pair of matrix equations AXB=E and CXD=F over generalized centro-symmetric matrices,” Computers and Mathematics with Applications, vol. 56, no. 12, pp. 3246–3260, 2008. View at Publisher Β· View at Google Scholar
  10. L. Fanliang, H. Xiyan, and Z. Lei, β€œThe generalized reflexive solution for a class of matrix equations (AX=B, XC=D),” Acta Mathematica Scientia, vol. 28, no. 1, pp. 185–193, 2008. View at Publisher Β· View at Google Scholar
  11. Y. X. Peng, X. Y. Hu, and L. Zhang, β€œAn iterative method for symmetric solutions and optimal approximation solution of the system of matrix equations A1XB1=C1, A2XB2=C2,” Applied Mathematics and Computation, vol. 183, no. 2, pp. 1127–1137, 2006. View at Publisher Β· View at Google Scholar Β· View at MathSciNet
  12. Z. H. Peng, X. Y. Hu, and L. Zhang, β€œAn efficient algorithm for the least-squares reflexive solution of the matrix equation A1XB1=C1, A2XB2=C2,” Applied Mathematics and Computation, vol. 181, no. 2, pp. 988–999, 2006. View at Publisher Β· View at Google Scholar Β· View at MathSciNet
  13. X. Sheng and G. Chen, β€œA finite iterative method for solving a pair of linear matrix equations (AXB, CXD) = (CXD),” Applied Mathematics and Computation, vol. 189, no. 2, pp. 1350–1358, 2007. View at Publisher Β· View at Google Scholar Β· View at MathSciNet
  14. S. Yuan, A. Liao, and G. Yao, β€œThe matrix nearness problem associated with the quaternion matrix equation ATXA+BTXB=C,” Journal of Applied Mathematics and Computing, vol. 37, no. 1-2, pp. 133–144, 2011. View at Publisher Β· View at Google Scholar
  15. Q. W. Wang, β€œA system of matrix equations and a linear matrix equation over arbitrary regular rings with identity,” Linear Algebra and Its Applications, vol. 384, no. 1-3, pp. 43–54, 2004. View at Publisher Β· View at Google Scholar Β· View at MathSciNet
  16. A. Navarra, P. L. Odell, and D. M. Young, β€œRepresentation of the general common solution to the matrix equations A1XB1=C1, A2XB2=C2 with applications,” Computers and Mathematics with Applications, vol. 41, no. 7-8, pp. 929–935, 2001. View at Publisher Β· View at Google Scholar Β· View at MathSciNet
  17. J. Ding, Y. Liu, and F. Ding, β€œIterative solutions to matrix equations of the form AiXBi=Fi,” Computers and Mathematics with Applications, vol. 59, no. 11, pp. 3500–3507, 2010. View at Publisher Β· View at Google Scholar
  18. J. von Numann, Functional Operators, vol. 11, Princeton University Press, Princeton, NJ, USA, 1950.
  19. I. Halperin, β€œThe product of projection operators,” Acta Scientiarum Mathematicarum, vol. 23, pp. 96–99, 1962.
  20. R.L. Dykstra, β€œAn algorithm for restricted least-squares regression,” Journal of the American Statistical Association, vol. 78, pp. 837–842, 1983.
  21. R. Escalante and M. Raydan, β€œDykstra's algorithm for a constrained least-squares matrix problem,” Numerical Linear Algebra with Applications, vol. 3, no. 6, pp. 459–471, 1996.
  22. H. Dai, Matrix Theory, Science Press, Beijing, China, 2001.
  23. T. Meng, β€œExperimental design and decision support,” in Expert system the Technology of Knowledge Management and Decision Making for 21st Centry, C. Leondes, Ed., vol. 1, Academic Press, New York, NY, USA, 2001.