Академический Документы
Профессиональный Документы
Культура Документы
In this paper higher order convergent methods for computing square roots of
nonsingular complex matrices are derived. These methods are globally convergent
and are based on eigenvalue shifting and powering. Specifically, it is shown for
each positive integer r G 2, a convergent method of order r can be developed.
These algorithms can be used to compute square roots of general nonsingular
complex matrices such as computing square roots of matrices with negative
eigenvalues. Q 1997 Academic Press
1. INTRODUCTION
0 l 0 l
cos Ž u . sin Ž u .
Assume that A s l I, then the family l
2
½ sin Ž u . ycos Ž u . 5 0 F u - 2p forms an
l 2
infinite set of square roots of A. On the other hand, if A s 1
, then
0 l2
matrices may not exist. For example, when l s 0 in the last matrix no
square root exists. From this observation, it is obvious that for 2 = 2
matrices, the equation B 2 s A / 0 has a solution if and only if A has a
nonzero eigenvalue.
To understand the structure of solutions of the equation B 2 s A for
b b
m s 2, let B s b11 b12 , and A s aa1121 aa1222 such that B 2 s A. Then we
21 22
¡b q b12 b 21 y a11
11 b 11
~
F Ž b11 , b12 , b 21 , b 22 . s
b11 b12 q b12 b 22 y a12
Ž 1.
b 21 b11 q b 22 b 21 y a21
¢ b 21 b12 q b 22 b 22 y a22 .
2 b11 b 21 b12 0
F Ž b11 , b12 , b 21 , b 22 . b12 b11 q b 22 0 b12
Js s .
Ž b11 , b12 , b 21 , b 22 . b 21 0 b11 q b 22 b 21
0 b 21 b12 2 b 22
0 0
b11 b12
B0 s 0 0
b 21 b 22
"'l2 y rs r
½ s .'l y rs
2 5 r, s g C.
and
r
r
Bkq 1 s Ý ž Ary2/ly1 2 lq1 l
Bk A . Ž 3.
ls0
2l q 1 k
Then there exists an a g C such that Bk is nonsingular for all sufficiently large
k. Set X k s Bky1A k , then the initial guess A 0 s aIm and B0 s Im the sequence
X k con¨ erges to a square root W of A. Moreo¨ er,
r
X kq 1 " W s By1
kq1 Bk Ž X k " W . ,
r
Ž 4.
i.e., if the sequence X k con¨ erges, it is rth order con¨ ergent to W. Addition-
ally,
lim k ª` Ay1
kq1 A k s I
r
and y1
lim k ª` Bkq1 Bkr s I.
s A kq 1 " Bkq1W ,
where the last equality follows from Ž2. and Ž3.. Hence Ž5. is true for the
integer k q 1. This shows that Ž5. is true for each nonnegative integer k.
SQUARE ROOTS OF COMPLEX MATRICES 397
or equivalently
r
X kq 1 " W s By1
kq1 Bk Ž X k " W . .
r
3. ANALYSIS OF CONVERGENCE
FŽ Z. y FŽ j . F K 5 Z y j 5
r
Ck " D k'A s Ž X k " 'A . , Ž 11 .
where C k [ C Ž X k . s Ý rls 0 r
ž /X
2l
ry 2 l l
k A, and Dk [ DŽ Xk . s
Ý rls0 2 l q
ž r
/ 1
X kry2 ly1Al. Since DŽ'A . s Ž2'A . ry1 is nonsingular, D k is
nonsingular for all sufficiently large k. It follows that there exists a
constant S G 0 such that 5 Dy1 k
5 F S for all sufficiently large k. Therefore
5 X kq 1 y 'A 5 F S 5 X k y 'A 5 r. This proves the r th order convergence of
Xk . Q.E.D.
There are many formulas in the literature that approximate square roots
of numbers and matrices using continued fraction expansion w4x. In this
section, we express the results of Theorem 3 in continued fraction form.
W s aI q Ž A y a2 I . 2 aI q Ž A y a2 I .
½
y1 y1
2 aI q Ž A y a2 I . 2 aI q ??? 4 y1 4 5 . Ž 12 .
SQUARE ROOTS OF COMPLEX MATRICES 401
A y a2
'A saq , Ž 13 .
2 a q Ž A y a2 . r Ž 2 a q Ž A y a2 . r Ž 2 a q ??? . .
which in the case a s 1 becomes that of w4x. Note that when A is positive,
Ž13. converges for any a g C with nonzero real part. The free parameter
a provides some flexibility in choosing the initial guess in that the closer a
is to 'A the more accelerated is the convergence.
5. COMPUTATIONAL RESULTS
8 q 15i y1 y 3i y4 y 9i
3 3 3
y1 y 3i 5 q 9i y1 y 3i
As .
3 3 3
y4 y 9i y1 y 3i 8 q 15i
3 3 3
which agrees with the exact square root to five decimal places, i.e.,
5 X5,2 2 y A 5 F s O Ž10y6 .. Comparable accuracy can also be achieved using
only four iterations with a third order method with the same initial guess
in which case we obtain 5 X 4,2 3 y A 5 F s O Ž10y6 ..
EXAMPLE 2. Consider the 4 = 4 matrix w2x
5 4 1 1
4 5 1 1
As .
1 1 4 2
1 1 2 4
The eigenvalues of this matrix are lŽ A. s 0.03, 3.03, y1.97 " i4 . Apply-
ing the iteration of Theorem 3 with r s 3 and X 0 s Ž1 q i . I we obtain
1 1 y2
A s y1 2 1 ,
0 1 y1
which has the eigenvalues lŽ A. s y1, 1, 24 , i.e., this matrix has a nega-
tive eigenvalue. Thus the initial matrix X 0 s aI should be chosen so that
404 MOHAMMED A. HASAN
6. CONCLUSION
ACKNOWLEDGMENT
The author thanks the referees for their helpful remarks and suggestions which improved
the quality of this work.
SQUARE ROOTS OF COMPLEX MATRICES 405
REFERENCES
1. N. J. Higham, Computing real square roots of a real matrix, Linear Algebra Appl. 88r89
Ž1987., 405]430.
2. N. J. Higham, Newton’s method for the matrix square root, Math. Comp. 46, No. 174
Ž1986., 537]549.
3. E. D. Denman, Roots of real matrices, Linear Algebra Appl. 36 Ž1981., 133]139.
4. L. S. Shieh and N. Chahin, A computer-aided method for the factorization of matrix
polynomials, Appl. Math. Comput. 2 Ž1976., 63]94.
5. E. D. Denman and A. N. Beavers, The matrix sign function and computation of systems,
Appl. Math. Comput. 2 Ž1976., 63]94.
6. A. Bjorck and S. Hammarling, A Schur method for the square root of a matrix, Linear
Algebra Appl. 52r53 Ž1983., 127]140.
7. W. D. Hoskins and D. J. Walton, A fast method of computing the square root of a matrix,
IEEE Trans. Automat. Control AC-23, No. 3 Ž1978., 494]495.
8. W. D. Hoskins and D. J. Walton, A fast, more stable method for computing the pth root
of positive definite matrices, Linear Algebra Appl. 26 Ž1979., 139]163.
9. P. Henrici, ‘‘Applied and Computational Complex Analysis,’’ Vol. 1, Wiley, New York,
1974.
10. A. S. Householder, ‘‘The Numerical Treatment of a Single Nonlinear Equation,’’ Mc-
Graw]Hill, New York, 1970.
11. M. A. Hasan, Higher order convergent algorithms for computing nth roots of complex
matrices, submitted for publication.