Scribd
Upload
145 views20 pages

Tridiagonal Matrices: G Erard MEURANT

This document discusses tridiagonal matrices and various properties and factorizations of tridiagonal matrices including: - Tridiagonal matrices are similar to symmetric tridiagonal matrices and therefore have real eigenvalues. - Cholesky, LU, and UL factorizations can be used to factor tridiagonal matrices. - The eigenvalues of tridiagonal matrices strictly interlace between successive matrices. - Closed form expressions can be derived for the inverse of tridiagonal matrices. - The inverse of a tridiagonal matrix can be computed incrementally as new rows and columns are added.

Uploaded by

human3178
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
145 views20 pages

Tridiagonal Matrices: G Erard MEURANT

This document discusses tridiagonal matrices and various properties and factorizations of tridiagonal matrices including: - Tridiagonal matrices are similar to symmetric tridiagonal matrices and therefore have real eigenvalues. - Cholesky, LU, and UL factorizations can be used to factor tridiagonal matrices. - The eigenvalues of tridiagonal matrices strictly interlace between successive matrices. - Closed form expressions can be derived for the inverse of tridiagonal matrices. - The inverse of a tridiagonal matrix can be computed incrementally as new rows and columns are added.

Uploaded by

human3178
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 20

Tridiagonal matrices

Grard MEURANT e

October, 2008

Similarity Cholesky factorizations Eigenvalues Inverse

Similarity
Let

1 1 Tk =

1 2 .. .

2 .. . .. . k1 k

k2 k1 k1

and i = i , i = 1, . . . , k 1

Proposition
Assume that the coecients j , j = 1, . . . , k 1 are dierent from zero and the products j j are positive. Then, the matrix Tk is similar to a symmetric tridiagonal matrix. Therefore, its eigenvalues are real.

Proof. 1 Consider Dk Tk Dk which is similar to Tk , Dk diagonal matrix with diag. el. j Take j1 1 , j = 2, . . . k 1 = 1, j2 = j1 1 Let

1 1 Jk =

1 2 .. .

2 .. . .. . k1 k

k2 k1 k1

where the values j , j = 1, . . . , k 1 are assumed to be nonzero

Proposition
2 det(Jk+1 ) = k+1 det(Jk ) k det(Jk1 )

with initial conditions det(J1 ) = 1 ,


2 det(J2 ) = 1 2 1 .

The eigenvalues of Jk are the zeros of det(Jk I ) The zeros do not depend on the signs of the coecients j , j = 1, . . . , k 1 We suppose j > 0 and we have a Jacobi matrix

Cholesky factorizations
Let k be a diagonal matrix with diagonal elements j , j = 1, . . . , k and 1 l1 1 .. .. Lk = . . lk2 1 lk1 1 Jk = Lk k LT k 1 = 1 , j = j
2 j1

l1 = 1 /1 lj = j /j , j = 2, . . . , k 1

j1

, j = 2, . . . , k,

The factorization can be completed if no j is zero for j = 1, . . . , k 1 This does not happen if the matrix Jk is positive denite, all the elements j are positive and the genuine Cholesky factorization can be obtained from k Jk = LC (LC )T k k with LC = Lk k k
1/2

which is 1 2 . .. . k1
k1 k1 k2 k2

1 1 LC = k

..

The factorization can also be written as Jk = LD 1 (LD )T k k k with 1 1 LD = k 2 .. . .. . k1 k1 k

k2

Clearly, the only elements we have to compute and store are the diagonal elements j , j = 1, . . . , k To solve a linear system Jk x = c, we successively solve LD y = c, k (LD )T x = k y k

The previous factorizations proceed from top to bottom (LU) We can also proceed from bottom to top (UL) 1 Jk = LT Dk Lk k with d1 1 Lk =
(k)

d2 ..
(k)

..

.
(k)

k2 dk1 (k) k1 dk
(k)

and Dk a diagonal matrix with elements dj


(k) dk

= k ,

(k) dj

= j

j2 dj+1
(k)

, j = k 1, . . . , 1

From LU and UL factorizations we can obtain all the so-called twisted factorizations of Jk
T Jk = Mk k Mk

Mk is lower bidiagonal at the top for rows with index smaller than l and upper bidiagonal at the bottom for rows with index larger than l 2 j1 1 = 1 , j = j , j = 2, . . . , l 1 j1 k = k , j = j j2 j+1 , j = k 1, . . . , l + 1

2 l1 2 l = l l l1 l+1

Eigenvalues
The eigenvalues of Jk are the zeros of det(Jk I ) det(Jk I ) = 1 () k () = d1 () dk () This shows that k () = det(Jk I ) , det(Jk1 I ) d1 () =
(k) (k) (k)

det(Jk I ) det(J2,k I )

Theorem

The eigenvalues i of Jk+1 strictly interlace the eigenvalues of Jk (k+1) (k) (k+1) (k) (k) (k+1) 1 < 1 < 2 < 2 < < k < k+1 (Cauchy interlacing theorem)

(k+1)

Proof. Eigenvector x = (y )T of Jk+1 corresponding to Jk y + k e k = y k yk + k+1 = Eliminating y from these relations, we obtain
2 (k+1 k ((e k )T (Jk I )1 e k )) = k 2 k+1 k j=1

j2 j
(k)

=0

where j is the last component of the jth eigenvector of Jk (k) The zeros of this function interlace the poles j

Inverse

Theorem
There exist two sequences of numbers {ui }, {vi }, i = 1, . . . , k such that u1 v1 u1 v2 u1 v3 . . . u1 vk u1 v2 u2 v2 u2 v3 . . . u2 vk 1 Jk = u1 v3 u2 v3 u3 v3 . . . u3 vk . . . . .. . . . . . . . . . u1 vk u2 vk u3 vk . . . uk vk Moreover, u1 can be chosen arbitrarily, for instance u1 = 1 see Baranger and Duc-Jacquet; Meurant 1 1 Hence, we just have to compute Jk e 1 and Jk e k

Jk v = e 1 Use UL factorization of Jk v1 = For vk J k u = e k use LU factorization uk = 1 , k vk ukj = (1)j kj k1 , j = 1, . . . , k 1 kj k vk 1


(k) d1

vj = (1)j1

1 j1 d1 dj
(k) (k)

, j = 2, . . . , k

Theorem
The inverse of the symmetric tridiagonal matrix Jk is characterized as (k) (k) dj+1 dk 1 ji (Jk )i,j = (1) i j1 , i, j > i i k
1 (Jk )i,i

d dk = i+1 , i i k
(k) (k)

(k)

(k)

Proof. ui = (1)(i+1)

d1 dk 1 1 i1 i k

1 The diagonal elements of Jk can also be obtained using twisted factorizations

Theorem
Let l be a xed index and j the diagonal elements of the corresponding twisted factorization Then 1 1 (Jk )l,l = l
1 In the sequel we will be interested in (Jk )1,1 1 (Jk )1,1 =

1 d1
(k)

1 Can we compute (Jk )1,1 incrementally?

Theorem
1 1 (Jk+1 )1,1 = (Jk )1,1 +

(1 k )2 (1 k )2 k+1 k e k k+1

Proof. Jk+1 = Jk k (e k )T

1 The upper left block of Jk+1 is the inverse of the Schur complement 1 2 k k k T e (e ) Jk k+1

Inverse of a rank-1 modication of Jk

Use the ShermanMorrison formula (A + xy T )1 = A1 This gives Jk


2 k k k T e (e ) k+1 1 1 = Jk + 1 1 (Jk e k )((e k )T Jk ) k+1 2 k 1 (e k )T Jk e k

A1 xy T A1 1 + y T A1 x

1 Let l k = Jk e k 1 1 (Jk+1 )1,1 = (Jk )1,1 + 2 k k (l1 )2 2 k k+1 k lk

k l1 = (1)k1

1 k1 , 1 k

k lk =

1 k

To simplify the formulas, we note that


2 k k+1 k lk = k+1 2 k = k+1 k

1 We start with (J1 )1,1 = 1 = 1/1 and c1 = 1 2 t = k k ,

k+1 = k+1 t,

k+1 =

1 , k+1

ck+1 = t ck k

This gives
1 1 (Jk+1 )1,1 = (Jk )1,1 + ck+1 k+1

J. Baranger and M. Duc-Jacquet, Matrices tridiagonales symtriques et matrices factorisables, RIRO, v 3, e (1971), pp 6166 G.H. Golub and C. Van Loan, Matrix Computations, Third Edition, Johns Hopkins University Press, (1996) G. Meurant, A review of the inverse of tridiagonal and block tridiagonal matrices, SIAM J. Matrix Anal. Appl., v 13 n 3, (1992), pp 707728

You might also like