434 CHAPTER 18. LINEAR TRANSFORMATIONS

columns is orthogonal to the rows of Ci since CiB j = 0 if i ̸= j. Therefore,

S−1AS =

C1...

Cr

A(

B1 · · · Br

)=

C1...

Cr

( AB1 · · · ABr

)

=

C1AB1 0 · · · 0

0 C2AB2 · · · 0... 0

. . . 00 · · · 0 CrABr

and Crk ABrk is an rk× rk matrix.

What about the eigenvalues of Crk ABrk ? The only eigenvalue of A restricted to Vλ kis

λ k because if Ax = µx for some x ∈Vλ kand µ ̸= λ k, then

(A−λ kI)rk x = (A−µI +(µ−λ k) I)rk x

=rk

∑j=0

(rk

j

)(µ−λ k)

rk− j (A−µI) j x = (µ−λ k)rk x ̸= 0

contrary to the assumption that x ∈ Vλ k. Suppose then that Crk ABrk x = λx where x ̸= 0.

Why is λ = λ k? Let y = Brk x so y ∈Vλ k. Then

S−1Ay = S−1AS



0...x...0

=



0...

Crk ABrk x...0

= λ



0...x...0

and so

Ay = λS



0...x...0

= λy.

Therefore, λ = λ k because, as noted above, λ k is the only eigenvalue of A restricted to Vλ k.

Now let Pk =Crk ABrk . ■The above theorem contains a result which is of sufficient importance to state as a

corollary.

Corollary 18.4.3 Let A be an n× n matrix and let Dk denote a basis for the generalizedeigenspace for λ k. Then {D1, · · · ,Dr} is a basis for Cn.