470 CHAPTER 17. BANACH SPACES

This follows from the observation that

Rkλ= (L−λ I)k =

k

∑j=0

(kj

)L j (−λ I)k− j = (−λ )k I+

k

∑j=1

(kj

)L j (−λ I)k− j (17.6.17)

which is a multiple of I−C where C is a compact map. Then by the open mapping theorem,it follows that Rλ is a homeomorphism onto Rk+1

λX = Rk

λX .

What about Rkλ

X ⊕N(Rk

λ

)= X? It only remains to verify that Rk

λX + N

(Rk

λ

)= X

because the only vector in the intersection was shown to be 0. Thus if you have x+ y = 0where x is in one of these and y in the other, then x =−y so each is in both and hence bothare 0. Pick x ∈ X . Then Rk

λx ∈ Rk

λ

(Rk

λX)= Rk

λX . Therefore, Rk

λx = Rk

λ

(Rk

λy)

for some yand so Rk

λ

(x−Rk

λy)= 0. Hence

x−Rkλ

y ∈ N(

Rkλ

)showing that x ∈ Rk

λX +N

(Rk

λ

).

It is obvious that Rkλ

X and N(Rk

λ

)are invariant under L. If λ 0 /∈ Λ\{0} , then L−λ 0I

is one to one and so the compactness of L and Lemma 17.6.2 implies that (L−λ 0I)Xis closed. Hence the open mapping theorem implies L− λ 0I is a homeomorphism onto(L−λ 0I)X . Is this last all of X? There is nothing in the above argument which involved anessential assumption that λ ∈Λ. Hence, repeating this argument, you see that (L−λ 0I)X⊕N (L−λ 0I) = X but N (L−λ 0I) = 0. Hence (L−λ 0I)X = X and so indeed (L−λ 0I) is ahomeomorphism.

For µ ∈Λ,Lx = µx and so |µ|∥x∥ ≤ ∥L∥∥x∥ so |µ| ≤ ∥L∥. Why is Λ at most countableand has only one possible limit point at 0? It was shown that Rλ is a homeomorphism whenrestricted to Rk

λX . It follows that for x ∈ Rk

λX ,∥Rλ x∥> δ ∥x∥ for some δ > 0, this for every

such x ∈ Rkλ

X . Now consider µ close to λ and consider Rµ . Then for x ∈ Rkλ

X ,∥∥Rµ x

∥∥ =∥(Rλ +(λ −µ))x∥ ≥ δ ∥x∥− |λ −µ|∥x∥ > δ

2 ∥x∥ provided |λ −µ| < δ/2. Thus for µ

close enough to λ ,Rµ is one to one on Rkλ

X . But also Rµ is one to one on N(Rk

λ

). Lets see

why this is so. Suppose (L−µI)x = 0 for x ∈ N(Rk

λ

). Then

0 = (L−µI +(µ−λ ) I)k x

= (µ−λ )k x+k

∑j=1

(kj

)(L−µI) j (µ−λ )k− j x

and the second term involving the sum yields 0. Since Rkλ

X ⊕N(Rk

λ

)= X , this shows that

(L−µI) is one to one for µ near λ . It follows that for µ near λ , µ /∈ Λ. Thus the onlypossible limit point is 0. Note that there is no restriction on the size of µ for (L−µI) to beone to one on N

(Rk

λ

).

Why is dim(N(Rk

λ

))< ∞ for each λ ̸= 0. This follows from 17.6.17. Rk

λis a multiple

of I−C for C a compact operator. Hence this is finite dimensional by Lemma 17.6.5.What about N

(Rk

µ

)⊆ Rk

λX for µ an eigenvalue different than λ? Say Rk

µ x = 0. Then,does it follow that x ∈ Rk

λX? From what was just shown

x = y+ z, y ∈ Rkλ

X , z ∈ N(

Rkλ

)

470 CHAPTER 17. BANACH SPACESThis follows from the observation thatre (k\,j mi elk ya kjRe =(L—-an*s= y ( . ) Bean = (-A)"I+ y ( . \u (—AD*! (17.6.17)j=o\ J ja\ Jwhich is a multiple of /—C where C is acompact map. Then by the open mapping theorem,it follows that Ry is a homeomorphism onto RIX = RX .What about Ri X @ N(R‘) = X? It only remains to verify that Ri X +N (RK) =Xbecause the only vector in the intersection was shown to be 0. Thus if you have x+y = 0where x is in one of these and y in the other, then x = —y so each is in both and hence bothare 0. Pick x € X. Then Rix € RS (Ri, X) = RUX. Therefore, Rix = RS (Riy) for some yand so R5 (x—Riy) = 0. Hencex—Riy EN (Ri)showing that x € R|X +N (Rj).It is obvious that Ri X and N(R‘) are invariant under L. If Ao ¢ A\ {0}, then L— Aolis one to one and so the compactness of L and Lemma 17.6.2 implies that (L—Ao/)Xis closed. Hence the open mapping theorem implies L — Ao/ is a homeomorphism onto(L—Aol)X. Is this last all of X? There is nothing in the above argument which involved anessential assumption that A € A. Hence, repeating this argument, you see that (L— Aol) X ®N(L—Aol) =X but N(L—Aol) = 0. Hence (L— Aol) X =X and so indeed (L—Apo/) is ahomeomorphism.For pw € A, Lx = px and so |p| |x|] < ||Z]] ||x]| so |u| < ||Z]|. Why is A at most countableand has only one possible limit point at 0? It was shown that R, is ahomeomorphism whenrestricted to Ri X. It follows that for x € R45 X, ||Rz.x|| > 6 ||x|| for some 6 > 0, this for everysuch x € Ri X. Now consider p close to A and consider Ry. Then for x € Ri X, ||Rux|| =(Ry +(A—)) al) = S|lxl| — [A —m| x > 3 |x| provided [A — | < 8/2. Thus forclose enough to 2, Ry is one to one on Ri X. But also Ry is one to one on N (Ri) . Lets seewhy this is so. Suppose (L— ul) x = 0 for x € N (R45). Then0 = (L—pl+(u—A)l)‘xk A (k k-j(uA) +h ( 5 )e=way! = aysand the second term involving the sum yields 0. Since RX @N (Ri) = X, this shows that(L— 1) is one to one for p near A. It follows that for p near 2, u ¢ A. Thus the onlypossible limit point is 0. Note that there is no restriction on the size of 4 for (L— uJ) to beone to one on N (Ri).Why is dim (N (R5)) < ce for each A 4 0. This follows from 17.6.17. Ri is a multipleof J—C for C a compact operator. Hence this is finite dimensional by Lemma 17.6.5.What about NV (Ri) Cc RLX for p an eigenvalue different than 1? Say Rix = 0. Then,does it follow that x € RUX ? From what was just shownxaytz, yER|X, <EN(Rj)