598 CHAPTER 22. HILBERT SPACES

so A is also continuous. Now (Ax,x) = limn→∞ (Anx,x)≥ 0 so A is positive and it is clearlyalso self adjoint since each An is. From passing to the limit in the definition of An,

Ax = Ax+1m(T x−Amx)

and so T x = Amx. This proves the theorem in the case that T ≤ I. Then if T > I, considerT/∥T∥. T/∥T∥ ≤ I and so there is B such that Bm = T/∥T∥ . Let A = ∥T∥1/m B. Thisproves the existence of the mth root. It is clear that A commutes with every continuouslinear operator that commutes with T because this is true of each of the iterates. In fact,each of these is just a polynomial in T . It remains to verify uniqueness.

Next suppose both A and B are mth roots of T having all the properties stated in the the-orem. Then AB = BA because both A and B commute with every operator which commuteswith T . Then from Proposition 22.6.2,((

Am−1 +Am−2B+ ...+Bm−1)(A−B)x,(A−B)x)≥ 0 (22.24)

Therefore, ((Am−Bm)x,(A−B)x) = (0,(A−B)x) = 0.Now this means

(AkBl (A−B)x,(A−B)x

)= 0 for all k+ l = m− 1 since the sum of

such terms is 0 and each of them is nonnegative. Now this implies(√AkBl (A−B)x,

√AkBl (A−B)x

)= 0

and so√

AkBl (A−B)x = 0⇒ AkBl (A−B)x = 0,k+ l = m−1. Then, using the binomialtheorem,

0 =

=(A−B)m−1

m−1

∑j=0

(m−1

j

)Am−1− jB j (−1) j (A−B)x = (A−B)m x

This clearly implies A = B. To see this, consider m = 7.If m = 7,(A−B)7 x = 0 so

(A−B)8 x = 0

so((A−B)4 x,(A−B)4 x

)= 0 which implies (A−B)4 x = 0 which implies(

(A−B)2 x,(A−B)2 x)= 0

so((A−B)2 x,x

)= 0 which yields ((A−B)x,(A−B)x) = 0, so (A−B) = 0. ■

This next was shown earlier, but this is a nice way to think of it in terms of a squareroot.

Proposition 22.6.4 Let T ∈L2 (H,G). Then if {ek} ,{

f j}

are two orthonormal bases,then ∑k ∥Tek∥2 = ∑k ∥T fk∥2 .

Proof: T ∗T is self adjoint and in L2 (H,H). Therefore,

∑k

∥∥∥√T ∗T ek

∥∥∥2= ∑

k(T ∗Tek,ek) = ∑

k∥Tek∥2

598 CHAPTER 22. HILBERT SPACESso A is also continuous. Now (Ax, x) = limy—oo (Anx,x) > 0 so A is positive and it is clearlyalso self adjoint since each A, is. From passing to the limit in the definition of Aj,1Ax = Ax+ — (Tx—A™x)mand so Tx = Ax. This proves the theorem in the case that T < J. Then if T > J, considerT/||T||. T/|\T|| <Z and so there is B such that B” = T/||T||. Let A = ||T||!/"B. Thisproves the existence of the m'” root. It is clear that A commutes with every continuouslinear operator that commutes with T because this is true of each of the iterates. In fact,each of these is just a polynomial in 7. It remains to verify uniqueness.Next suppose both A and B are m'” roots of T having all the properties stated in the the-orem. Then AB = BA because both A and B commute with every operator which commuteswith T. Then from Proposition 22.6.2,((A™ | +A" B+ ...4+B""!) (A—B)x,(A—B)x) >0 (22.24)Therefore, ((A” — B”) x, (A —B)x) = (0,(A—B)x) =0.Now this means (A‘B! (A — B).x,(A—B)x) =0 for all k+/ =m-—1 since the sum ofsuch terms is 0 and each of them is nonnegative. Now this implies(vat! (A —B)x, VAKBI (A — B)x) =0and so VA‘B! (A — B) x = 0 = A*B! (A—B) x =0,k +1 =m-— 1. Then, using the binomialtheorem,=(A—B)""!— m—\ m—1—j pj J m0=)y j A™~!~JBI (—1)/ (A—B)x = (A—B)"xj=0This clearly implies A = B. To see this, consider m = 7.If m=7,(A—B)'x=0s0(A—B)*x=0so ((A —B)*x,(A— B)*x) = 0 which implies (A — B)* x = 0 which implies((A —B)?x,(A— By x) =0so ((A —B)*x,x) = 0 which yields ((A — B) x, (A —B) x) =0, so (A—B) =0. llThis next was shown earlier, but this is a nice way to think of it in terms of a squareroot.Proposition 22.6.4 Let T € 22 (H,G). Then if {e,} , {fj} are two orthonormal bases,then Yy.||Tex ||” = Le ll Sell’ -Proof: T*T is self adjoint and in % (H,H). Therefore,2y | vrTex|| =) (T*Tex,e~) = Y ||Tex||”k k k