Spectral Theorems in Euclidean and Hermitian Spaces: 12.1 Normal Linear Maps
Spectral Theorems in Euclidean and Hermitian Spaces: 12.1 Normal Linear Maps
Spectral Theorems in Euclidean and Hermitian Spaces: 12.1 Normal Linear Maps
631
632 CHAPTER 12. SPECTRAL THEOREMS
p
In both cases we let kuk = hu, ui and the map
u 7! kuk is a norm.
Our first goal is to show that for every normal linear map
f : E ! E (where E is a Euclidean space), there is an
orthonormal basis (w.r.t. h , i) such that the matrix
of f over this basis has an especially nice form:
we have
f (u) = u µv and f (v) = µu + v,
where j 2 C.
where i 2 R.
4. orthogonal i↵
A A> = A>A = In.
652 CHAPTER 12. SPECTRAL THEOREMS
where i 2 R.
654 CHAPTER 12. SPECTRAL THEOREMS
Bi = {z 2 C | |z i| cond(P ) k Ak},
Note that the matrix A(✏) from the beginning of the sec-
tion is not normal.
664 CHAPTER 12. SPECTRAL THEOREMS
x>Ax
R(A)(x) = > , x 2 Rn, x 6= 0.
x x
i µi n m+i , i = 1, . . . , m.
668 CHAPTER 12. SPECTRAL THEOREMS
1 µ1 2 µ2 · · · µn 2 n 1 µn 1 n,
a genuine interlacing.
k µk k+n r , k = 1, . . . , r. (⇤)
12.5. RAYLEIGH RATIOS AND THE COURANT-FISCHER THEOREM 671
sion k, then
x>Ax
k = max min
W 2Vn k+1 x2W,x6=0 x> x
x>Ax
k = min max .
W 2Vk x2W,x6=0 x> x
2. If i + j = k + n, then
k (A + B) i (A) + j (B).
If i = 1 and j = k, we obtain
1 (A) + k (B) k (A + B),
and if i = k and j = n, we obtain
k (A + B) k (A) + n (B),
k (A) k (A + B) k = 1, . . . , n.