*This preview shows
page 1. Sign up
to
view the full content.*

**Unformatted text preview: **ies of A. Consequently, the eigenvalues of A
must vary continuously with the entries of A. Caution! Components of an
eigenvector need not vary continuously with the entries of A —e.g., consider
x = ( −1 , 1)T as an eigenvector for A = 0 1 , and let → 0.
0 D
E Example 7.1.4
Spectral Radius. For square matrices A, the number T
H ρ(A) = max |λ|
λ∈σ (A) is called the spectral radius of A. It’s not uncommon for applications to
require only a bound on the eigenvalues of A. That is, precise knowledge of
each eigenvalue may not called for, but rather just an upper bound on ρ(A)
is all that’s often needed. A rather crude (but cheap) upper bound on ρ(A)
is obtained by observing that ρ(A) ≤ A for every matrix norm. This is
true because if (λ, x) is any eigenpair, then X = x | 0 | · · · | 0 n×n = 0, and
λX = AX implies |λ| X = λX = AX ≤ A X , so Y
P IG
R |λ| ≤ A for all λ ∈ σ (A) . (7.1.12) This result is a precursor to a stronger relationship between spectral radius
and norm that is hinted at in Exercise 7.3.12 and developed in Example 7.10.1
(p. 619). O
C The eigenvalue bound (7.1.12) given in Example 7.1.4 is cheap to compute,
especially if the 1-norm or ∞ -norm is used, but you often get what you pay
for. You get one big circle whose radius is usually much larger than the spectral
67
radius ρ(A). It’s possible to do better by using a set of Gerschgorin circles as
described below.
67 S. A. Gerschgorin illustrated the use of Gerschgorin circles for estimating eigenvalues in 1931,
but the concept appears earlier in work by L. L´vy in 1881, by H. Minkowski (p. 278) in 1900,
e
and by J. Hadamard (p. 469) in 1903. However, each time the idea surfaced, it gained little
attention and was quickly forgotten until Olga Taussky (1906–1995), the premier woman of
linear algebra, and her fellow German emigr` Alfred Brauer (1894–1985) became captivated
e
by the result. Taussky (who became Olga Taussky-Todd after marrying the numerical analyst
John Todd) and Br...

View
Full
Document