Lecture 34 : Self-adjoint operator

Definition 34.1. A bounded linear operator T on a Hilbert space H is said to be self-adjoint if

T ∗ = T , and is said to be normal if T ∗ T = T T ∗ . T is said to be unitary if T ∗ T = T T ∗ = I.

It is easy to see that if T is self-adjoint or unitary, then T is normal.

Example: Let A = [aij ]n×n be a matrix of order n with complex entries. Then A defines a linear
operator on Cn . Note that if we consider the vectors of Cn as column vectors, then for all x, y ∈ Cn ,

⟨Ax, y⟩ = y T Ax and ⟨x, A∗ y⟩ = (A∗ y)T x = y T (A∗ )T x.

This gives that

y T Ax = y T (A∗ )T x =⇒ (A∗ )T = A =⇒ A∗ = (A)T .

Proposition 34.1. Let T be a bounded linear operator on a Hilbert space H. Then the following
statements hold:

(i) If T is self-adjoint, then ⟨T x, x⟩ is real for all x ∈ H.

(ii) If H is complex and ⟨T x, x⟩ is real for all x ∈ H, then T is self-adjoint.

(iii) The product of two self-adjoint operators is self-adjoint if and only if both commute.

Proof. (i) Suppose that T is self-adjoint. Then for all x ∈ H,

⟨T x, x⟩ = ⟨x, T x⟩ = ⟨T x, x⟩.

Thus, ⟨T x, x⟩ is real.

(ii) Suppose that H is complex and ⟨T x, x⟩ is real for all x ∈ H. Then

⟨T x, x⟩ = ⟨T x, x⟩ = ⟨x, T x⟩ = ⟨T ∗ x, x⟩ for all x ∈ H.

Thus,
⟨(T − T ∗ )x, x⟩ ≥ 0 for all x ∈ H.
Since H is complex, we conclude that T = T ∗ .

(iii) Note that (ST )∗ = T ∗ S ∗ = T S. It is now obvious that ST is self-adjoint if and only if
ST = T S.

Theorem 34.1. Let (Tn )n≥1 be a sequence of bounded self-adjoint operators on a Hilbert space H.
If Tn → T as n → ∞, then T is self-adjoint.

Proof. Note that

∥T − T ∗ ∥ ≤ ∥T − Tn ∥ + ∥Tn − Tn∗ ∥ + ∥Tn∗ − T ∗ ∥ = ∥T − Tn ∥ + 0 + ∥Tn − T ∥ → 0 as n → ∞.

Thus, T ∗ = T .

Proposition 34.2. Let U be a unitary operator on a Hilbert space H. Then the following are true:

62
 (i) U is an isometry, i.e., ∥U x∥ = ∥x∥ for all x ∈ H.

(ii) ∥U ∥ = 1, provided H ̸= {0}.

(iii) U −1 is unitary.

(iv) U is normal.

(v) If V is another unitary on H, then U V is unitary.

(vi) A bounded linear operator T on a complex Hilbert space is unitary if and only if T is an
isometry and surjective.

Proof. (i) ∥U x∥2 = ⟨U x, U x⟩ = ⟨U ∗ U x, x⟩ = ∥x∥2 .

(ii) Follows from (i).

(iii) Since U ∗ = U −1 , it follows that U −1 is unitary.

(iv) Already noted.

(v) Suppose that V is a unitary on H. Then

(U V )∗ U V = V ∗ U ∗ U V = V ∗ V = I.

Similarly, we can show that U V (U V )∗ = I. Thus, U V is unitary.

(vi) Suppose that T is isometry and surjective. Then

⟨T ∗ T x, x⟩ = ⟨T x, T x⟩ = ⟨x, x⟩ for all x ∈ H.

Thus, T ∗ T = I (as H is complex). Also,

T T ∗ = T T ∗ T T −1 = T T −1 = I.

Hence T is unitary. Converse is obvious.

63
Lecture 35 : Positive operator

Definition 35.1. Let T be a bounded self-adjoint operator on a Hilbert space H. We say that T is
positive if ⟨T x, x⟩ ≥ 0 for all x ∈ H. If T1 , T2 are bounded self-adjoint operators on H, then we say
that T1 ≤ T2 if T2 − T1 is positive. For a positive operator T , we write T ≥ 0.

Remark: We may define a partial ordering on the set of all bounded self-adjoint operators on a
Hilbert space. In fact, we set T1 ≤ T2 if and only if T2 − T1 ≥ 0. Further, it is easy to see that the
sum of two positive operators is positive.

Theorem 35.1. Let S and T be positive operators on a Hilbert space H. If ST = T S, then ST is

positive.

Proof. If S = 0 or T = 0, then clearly ST is positive. Assume that both S and T are non-zero.
Define a sequence (Sn )n≥1 of operators on H as follows:

S
S1 = , and Sn+1 = Sn − Sn2 for all n ≥ 1.
∥S∥

We claim that
0 ≤ Sn ≤ I for all n ≥ 1. (35.31)
We prove Eq.(35.31) by induction. For n = 1, clearly S1 ≥ 0 as S ≥ 0. Further, for all x ∈ H,

1 ∥Sx∥∥x∥ ∥S∥∥x∥2
⟨S1 x, x⟩ = ⟨Sx, x⟩ ≤ ≤ = ∥x∥2 = ⟨Ix, x⟩.
∥S∥ ∥S∥ ∥S∥

Thus, S1 ≤ I. Suppose that (1) is true for n. Note that

⟨Sn2 (I − Sn )x, x⟩ = ⟨(I − Sn )Sn x, Sn x⟩ ≥ 0 for all x ∈ H.

Thus,
Sn2 (I − Sn ) ≥ 0.
Similarly, we can also show that
Sn (I − Sn )2 ≥ 0.
Adding them, we obtain

0 ≤ Sn2 (I − Sn ) + Sn (I − Sn )2 = Sn − Sn2 = Sn+1 .

Since I − Sn ≥ 0 and Sn2 ≥ 0, we get

0 ≤ I − Sn + Sn2 = I − Sn+1 .

This proves Eq.(35.31).

Observe that

S1 = S12 + S2 = S12 + S22 + S3 = · · · = S12 + · · · + Sn2 + Sn+1 .

Since Sn+1 > 0, we get

S12 + · · · + Sn2 = S1 − Sn+1 ≤ S1 . (35.32)

64
Thus, for all x ∈ H,
n
X n
X n
X
∥Sj x∥2 = ⟨Sj x, Sj x⟩ = ⟨Sj2 x, x⟩ ≤ ⟨S1 x, x⟩.
j=1 j=1 j=1
P∞
Since n is arbitrary, we get that j=1 ∥Sj x∥2 is convergent. Hence, ∥Sn x∥ → 0 or Sn x → 0 as
n → ∞. By Eq.(35.32),
 
Xn
 Sj2  x = (S1 − Sn+1 )x → S1 x as n → ∞.
j=1

S
Note that all Sj commute with T , as they are sums and products of S1 = ∥S∥ . Thus, we get
n
X n
X
⟨ST x, x⟩ = ∥S∥⟨T S1 x, x⟩ = ∥S∥ lim ⟨T Sj2 x, x⟩ = ∥S∥ lim ⟨T Sj x, Sj x⟩ ≥ 0.
n→∞ n→∞
j=1 j=1

Thus, ST ≥ 0.

Theorem 35.2. Let (Tn )n≥1 be a sequence of bounded self-adjoint linear operators on a complex
separable Hilbert space H such that
Tn ≤ Tn+1 and Tn ≤ K for all n ≥ 1,
where K is a bounded self-adjoint operator on H. Suppose that Ti Tj = Tj Ti for all i, j ≥ 1 and
Tj K = KTj for all j ≥ 1. Then there exists a bounded self-adjoint operator T on H such that
Tn x → T x for all x ∈ H, and T ≤ K.
Proof. Let Sn = K − Tn for all n ≥ 0. Then Sn ≥ 0 for all n ≥ 1. Let m < n. Then
2
Sm − Sm Sn = (Sm − Sn )Sm = (Tn − Tm )(K − Tm ) ≥ 0
by the preceding theorem. Similarly, we get
Sn Sm − Sn2 = Sn (Sm − Sn ) = (K − Tn )(Tn − Tm ) ≥ 0.
Thus,
Sn2 ≤ Sn Sm ≤ Sm
2
for m < n.
This gives that
∥Sn x∥2 = ⟨Sn x, Sn x⟩ = ⟨Sn2 x, x⟩ ≤ ⟨Sn Sm x, x⟩ ≤ ⟨Sm
2
x, x⟩ = ∥Sm x∥2 ≥ 0.
Hence, for a fixed x, (∥Sn x∥2 )n≥1 is a monotonic decreasing sequence of non-negative numbers, and
thus convergent. Also,
∥Sn x − Sm x∥2 = ⟨(Sn − Sm )x, (Sn − Sm )x⟩ = ∥Sn x∥2 − 2⟨Sn Sm x, x⟩ + ∥Sm x∥2
≤ ∥Sm x∥2 − ∥Sn x∥2 → 0 as m, n → ∞.
Thus, (Sn x)n≥1 is Cauchy for each fixed x ∈ H. Consequently, (Tn x)n≥1 is Cauchy for each fixed
x ∈ H. Since H is complete, let y ∈ H be such that Tn x → y. Define T : H → H as
T x = lim Tn x.
n→∞

Then T is bounded by the uniform boundedness principle. Clearly, T is self-adjoint as each Sn is

self-adjoint. Since Tn ≤ K, it follows that T ≤ K.

65
Lecture 36 : Positive square root

Definition 36.1. Let T be a positive operator on a complex Hilbert space H. Then a bounded self-
adjoint operator A is said to be a square root of T if A2 = T . If A ≥ 0, then A is called a positive
square root of T and is denoted by T 1/2 .
Theorem 36.1. Let T be a positive operator on a complex Hilbert space H. Then T has a unique
positive square root A. Further, A commutes with every operator on H which commutes with T .
T
Proof. If T = 0, then A = 0. Assume that T ̸= 0. Set Q := ∥T ∥ . Then

∥T x∥
⟨Qx, x⟩ ≤ ∥x∥ ≤ ∥x∥2 = ⟨Ix, x⟩, for all x ∈ H.
∥T ∥
Thus, Q ≤ I. If B is the unique positive square root of Q, then
 2
∥T ∥1/2 B = ∥T ∥B 2 = ∥T ∥Q = T.

Thus, ∥T ∥1/2 B is the positive square root of T . Hence without loss of generality, we may assume
that T ≤ I. Define
1
A0 = 0 and An+1 = An + (T − A2n ), n ≥ 0.
2
We claim that
An ≤ An+1 and An ≤ I for all n ≥ 0. (36.33)
We prove Eq.(36.33) by induction. Clearly A0 ≤ I and A0 ≤ A1 . Assume that Eq.(36.33) is true
for n. Since I − An ≥ 0 and I − T ≥ 0, we obtain that
1 1 1
0 ≤ (I − An )2 + (I − T ) = I − An − (T − A2n ) = I − An+1 .
2 2 2
Also,
 
1 2 1 2 1
An+2 − An+1 = An+1 + (T − An+1 ) − An − (T − An ) = (An+1 − An ) I − (An + An+1 ) ≥ 0.
2 2 2
By preceding theorem, there exists a bounded self-adjoint operator A such that An x → Ax for all
x ∈ H. Note that
1
An+1 x − An x = (T x − A2n x) → 0 as n → ∞.
2
2 2
Thus, An x → T x, and hence, T x = A x for all x ∈ H. Since An ≥ A0 = 0, it follows that A ≥ 0.
Let S be an operator which commutes with T . Then SAn = An S for all n ≥ 0. Thus, SAn x = An Sx
for all x ∈ H and n ≥ 0. Letting n → ∞, we get SA = AS.
For uniqueness Let A and B be positive square roots of T . Then

A2 = B 2 = T.

Also,
BT = BB 2 = B 2 B = T B.
Hence, AB = BA. Let x ∈ H and y = (A − B)x. Then we get

⟨Ay, y⟩ + ⟨By, y⟩ = ⟨(A + B)y, y⟩ = ⟨(A2 − B 2 )x, y⟩ = 0.

66
Thus, ⟨Ay, y⟩ = 0 and ⟨By, y⟩ = 0. Let C be the positive square root of A. Then

0 = ⟨Ay, y⟩ = ⟨C 2 y, y⟩ = ⟨Cy, Cy⟩ = ∥Cy∥2 .

That is, Cy = 0 ⇒ C 2 y = Ay = 0. Similarly, By = 0, and hence (A − B)y = 0. Note that

∥Ax − Bx∥2 = ⟨(A − B)2 x, x⟩ = ⟨(A − B)y, x⟩ = 0 for all x ∈ H.

This proves that A = B.

67
Lecture 37 : Orthogonal projections

Theorem 37.1. A bounded linear operator P : H → H on a Hilbert space H is an orthogonal

projection if and only if P is self-adjoint and idempotent.
Proof. Suppose that P is an orthogonal projection onto a closed subspace Y of H. Then H = Y ⊕Y ⊥ .
If x ∈ H, then x = y + z, where y ∈ Y and z ∈ Y ⊥ . Note that
P 2 x = P (P x) = P (y) = y = P x.
Thus, P 2 = P . Further, let x1 , x2 ∈ H. Then x1 = y1 + z1 and x2 = y2 + z2 . Hence,
⟨P x1 , x2 ⟩ = ⟨y1 , y2 + z2 ⟩ = ⟨y1 , y2 ⟩ = ⟨x1 , P x2 ⟩.
This shows that P ∗ = P .
Conversely, suppose that P 2 = P and P ∗ = P . Then for each x ∈ H, we can write
x = P x + (I − P )x.
Note that for x, y ∈ H,
⟨P x, (I − P )y⟩ = ⟨x, P (I − P )y⟩ = ⟨x, 0⟩ = 0.
Thus, P (H) ⊥ (I − P )(H). Also,
(I − P )P x = P x − P 2 x = 0, for all x ∈ H.
Hence P (H) ⊆ N (I − P ). If (I − P )x = 0, then P x = x. That is, x ∈ P (H). Thus,
N (I − P ) ⊆ P (H),
Consequently, P (H) = N (I − P ). Hence, Y = P (H) is closed. Finally,
P (P x) = P 2 x = P x for all x.

Proposition 37.1. If P is an orthogonal projection on a Hilbert space H, then the following are
true:
(i) ⟨P x, x⟩ = ∥P x∥2 for all x ∈ H.
(ii) P ≥ 0.
(iii) ∥P ∥ = 1 if P (H) ̸= {0}.
Proof. For x ∈ H, note that
⟨P x, x⟩ = ⟨P x, P x⟩ = ∥P x∥2 ≥ 0.
This proves (i) and (ii). To see (iii), assume that P (H) ̸= {0}. Note that for all x ∈ H,
∥P x∥2 = ⟨P x, x⟩ ≤ ∥P x∥∥x∥ =⇒ ∥P x∥ ≤ ∥x∥ if P x ̸= 0.
Hence ∥P ∥ ≤ 1. Also ∥P x∥ = 1 if x ∈ P (H) \ {0}. This proves that
∥P ∥ = 1 if P (H) ̸= {0}.

68
Theorem 37.2. Let P1 , P2 be orthogonal projections on a Hilbert space H. Then P1 P2 is an
orthogonal projection if and only if P1 P2 = P2 P1 . In this case, P1 P2 is an orthogonal projection
onto P1 (H) ∩ P2 (H). Further, two closed subspaces M1 , M2 of H are orthogonal if and only if
the corresponding projections satisfy PM1 PM2 = 0, where PMj is an orthogonal projection onto
Mj , j = 1, 2.
Proof. Suppose that P1 P2 = P2 P1 . Let P = P1 P2 . Then clearly P ∗ = P . Also,

P 2 = (P1 P2 )(P1 P2 ) = P12 P22 = P1 P2 .

Therefore, by preceding theorem, P1 P2 is an orthogonal projection. It is left as an exercise to verify

that P1 P2 (H) = P1 (H) ∩ P2 (H).
Conversely, if P1 P2 is an orthogonal projection, then

P1 P2 = (P1 P2 )∗ = P2∗ P1∗ = P2 P1 .

Suppose that M1 ⊥ M2 . Then M1 ∩ M2 = {0}, and hence, by previous part, PM1 PM2 = 0.
Conversely, suppose that PM1 PM2 = 0. Then for every x ∈ M1 and y ∈ M2 , we get

⟨x, y⟩ = ⟨PM1 x, PM2 y⟩ = ⟨x, PM1 PM2 y⟩ = 0.

Thus, M1 ⊥ M2 .

Theorem 37.3. Let P1 , P2 be orthogonal projections on a Hilbert space H. Then the following are
true:
(i) P1 + P2 is an orthogonal projection if and only if P1 (H) ⊥ P2 (H).

(ii) If P1 + P2 is an orthogonal projection, then

(P1 + P2 )(H) = P1 (H) ⊕ P2 (H).

Proof. (i) Let P = P1 + P2 . Suppose that P is an orthogonal projection. Then we must have
P 2 = P . Hence we get

(P1 + P2 )2 = P12 + P1 P2 + P2 P1 + P22 = P1 + P2 + P1 P2 + P2 P1 .

Thus, we get
P1 P2 + P2 P1 = 0.
Multiplying by P2 from the left we get

P2 P1 P2 + P22 P1 = 0.

Multiplying by P2 from the right,

P2 P1 P2 + P2 P1 P2 = 0 ⇒ 2P2 P1 P2 = 0 ⇒ P2 P1 P2 = 0.

Putting this in above, we get P2 P1 = 0. Hence P1 (H) ⊥ P2 (H). Conversely, if P1 (H) ⊥ P2 (H),
then P1 P2 = P2 P1 = 0. Then it is clear that

(P1 + P2 )2 = P1 + P2 and (P1 + P2 )∗ = P1 + P2 .

Hence P1 + P2 is an orthogonal projection.

69
(ii) Suppose that P = P1 + P2 is an orthogonal projection. Then P x = P1 x + P2 x for all x ∈ H,
and hence, P (H) ⊆ P1 (H) ⊕ P2 (H). Let x ∈ P1 (H) ⊕ P2 (H). Then x = x1 + x2 , where
x1 ∈ P1 (H), x2 ∈ P2 (H). Thus,

P x = P1 (x1 + x2 ) + P2 (x1 + x2 ) = x1 + x2 = x.

Hence x ∈ P (H). Consequently, P (H) = P1 (H) ⊕ P2 (H).

70
Lecture 38 : Spectrum of a bounded linear operator

Definition 38.1. Let T be a bounded linear operator on a complex normed linear space X. A point
λ ∈ C is said to be a regular value of T if (T − λI)−1 exists and is bounded. The set of all regular
values of T is called the resolvent set of T and is denoted by ρ(T ). The complement of the resolvent
set of T is called the spectrum of T and is denoted by σ(T ).
The spectrum of T has several disjoint parts which are defined as follows:

Point spectrum or discrete spectrum or eigenvalues: The point spectrum of T , denoted by

σp (T ), contains all λ ∈ σ(T ) for which (T − λI)−1 does not exist. Equivalently, λ ∈ σp (T ) if and
only if (T − λI) is not one-one.

Continuous spectrum: The continuous spectrum of T , denoted by σc (T ), contains all λ ∈ σ(T )

for which the range of (T − λI) is dense, (T − λI)−1 exists but (T − λI)−1 is unbounded.

Residual spectrum: The residual spectrum of T , denoted by σr (T ), contains all λ ∈ σ(T ) for
which the range of (T − λI) is not dense but (T − λI)−1 exists which may be bounded or not.
It is easy to see that σp (T ), σc (T ) and σr (T ) are mutually disjoint subsets of σ(T ). Further,

σ(T ) = σp (T ) ∪ σc (T ) ∪ σr (T ).

Remark: The spectrum of an operator on a finite dimensional normed linear space (i.e., the spec-
trum of a matrix) is equal to the point spectrum. In case of infinite dimensional normed linear space,
the spectrum of an operator is much larger than the point spectrum, in general.

Example 1: For a bounded linear operator T , T −1 may exist but may not be bounded. For
example, consider T : ℓ2 → ℓ2 defined as
 x2 x3 xn 
T (x1 , x2 , . . .) = x1 , , , . . . , , . . . , x = (xn )n≥1 ∈ ℓ2 .
2 3 n
Then T is bounded with ∥T ∥ = 1 (verify!). Note that

T −1 (x1 , x2 , . . .) = (x1 , 2x2 , 3x3 , . . . , nxn , . . .), x = (xn )n≥1 ∈ ℓ2 .

But T −1 is not bounded (verify!).

Lemma 38.1. Suppose that T is a bounded linear operator on a complex Banach space X. If
∥T ∥ < 1, then the operator (I − T )−1 exists and is a bounded linear operator on X. Further,
∞
X
−1
(I − T ) = Tj = I + T + T2 + ···
j=0

Proof. Note that ∥T ∥ < 1, hence

∞
X ∞
X
j
∥T ∥ ≤ ∥T ∥j < ∞.
j=0 j=0

Now for all m > 0,

(I − T ) I + T + T 2 + · · · + T m = I − T m+1 → 0 as m → ∞.

71
Thus, we get
(I − T ) I + T + T 2 + · · · = I + T + T 2 + · · · (I − T ) = I

Thus, (I − T )−1 is bounded and is given by (I − T )−1 = ∞ j

P
j=0 T .

Example 2: Consider T : ℓ2 → ℓ2 defined as

T (x1 , x2 , . . .) = (0, x1 , x2 , . . .), x = (xn )n≥1 ∈ ℓ2 .

Before proceeding towards the computation of σ(T ), let us observe few things. For any bounded
linear operator S on a Hilbert space H,
⊥
ker S ∗ = (Range(S)) . (1)

In fact, let y ∈ ker S ∗ . Then for all x ∈ H,

⟨Sx, y⟩ = ⟨x, S ∗ y⟩ = 0 ⇒ Range(S) ⊥ ker S ∗ ,

⊥
This gives ker S ∗ ⊆ Range(S) . Conversely, suppose that y ⊥ Range(S). In particular, y ⊥
Range(S). Thus, for all x ∈ H,
0 = ⟨y, Sx⟩ = ⟨S ∗ y, x⟩.
Hence, S ∗ y = 0, which proves (1).
The Hilbert space adjoint T ∗ of T is given by

T ∗ (x1 , x2 , . . .) = (x2 , x3 , x4 , . . .).

To see this, note that for all x, y ∈ ℓ2 , we get

⟨T ∗ x, y⟩ = ⟨x, T y⟩ = ⟨(x1 , x2 , . . .), (0, y1 , y2 , . . .)⟩ = x2 y1 + x3 y2 + · · · = ⟨(x2 , x3 , . . .), y⟩.

Thus, T ∗ x = (x2 , x3 , x4 , . . .).

Note that ∥T ∥ = 1, and hence for |λ| > 1,
 
T
(T − λI) = λ −I
λ

is invertible and (T − λI)−1 is bounded as T

λ < 1. Thus,

σ(T ) ⊆ {λ ∈ C : |λ| ≤ 1}.

Suppose λ ∈ σp (T ). Clearly, λ ̸= 0 as T is injective (in fact, it is isometry). Further, for any

0 ̸= λ ∈ C, if (T − λI)x = 0, then

(0, x1 , x2 , . . .) = (λx1 , λx2 , . . .) =⇒ x1 = x2 = · · · = 0.

Thus, (T − λI) is one-one for all λ ∈ C. Hence σp (T ) = ∅.

Let λ ∈ σp (T ∗ ). Then

T ∗ x = λx for some x ∈ ℓ2 =⇒ (x2 , x3 , x4 , . . .) = (λx1 , λx2 , λx3 , . . .)

We get
2 3 n
x2 = λx1 , x3 = λ x1 , x4 = λ x1 , · · · , xn+1 = λ x1 , · · ·

72
Clearly,
∞
X
|xn |2 < ∞ if and only if |λ| < 1.
n=1

Thus, for |λ| < 1,

ker(T ∗ − λI) ̸= {0},
which gives that
⊥
Range(T − λI) ̸= {0}.
Hence (T − λI)−1 exists but Range(T − λI) is not dense. Thus,

σr (T ) = {λ ∈ C : |λ| < 1} .

We will see later that σ(T ) is a compact subset of C. Hence,

σc (T ) = {λ ∈ C : |λ| = 1}

and thus,
σ(T ) = {λ ∈ C : |λ| ≤ 1} .

73
Lecture 39 : Spectrum of a bounded linear operator continued

Theorem 39.1. Let T be a bounded linear operator on a complex Banach space X. Then ρ(T ) is
an open subset of C. Consequently, σ(T ) is a closed subset of C.

Proof. If ρ(T ) = ∅, it is open. Assume that ρ(T ) ̸= ∅. For a fixed λ0 ∈ ρ(T ) and any λ ∈ C, we have

T − λI = T − λ0 I − (λ − λ0 )I = (T − λ0 I) I − (λ − λ0 )(T − λ0 I)−1
 

Note that I − (λ − λ0 )(T − λ0 I)−1 is invertible if ∥(λ − λ0 )(T − λ0 I)−1 ∥ < 1. That is,
1
|λ − λ0 | < .
∥(T − λ0 I)−1 ∥

Thus, if this holds, T − λI is invertible. Consequently, ρ(T ) is open.

Corollary: The map λ 7→ (T − λI)−1 is analytic on ρ(T ).

Theorem 39.2. The spectrum σ(T ) of a bounded linear operator T on a complex Banach space X
is a non-empty compact set and lies in the disc {λ ∈ C : |λ| ≤ ∥T ∥}.
∥T ∥
Proof. We first show that σ(T ) is compact. To see this, let |λ| > ∥T ∥. Then |λ| < 1. Hence (I − Tλ )
is invertible, and thus, (T − λI) is invertible. Therefore,

{λ ∈ C : |λ| > ∥T ∥} ⊆ ρ(T ) =⇒ σ(T ) ⊆ {λ ∈ C : |λ| ≤ ∥T ∥} .

Thus, σ(T ) is bounded. Since σ(T ) is closed by the preceding theorem, it follows that σ(T ) is
compact.
Now, on contrary, suppose that σ(T ) = ∅. Then ρ(T ) = C. Hence the map λ 7→ (T − λI)−1 is
an entire function. Further, for |λ| > 2∥T ∥, we get
∞  j
−1 1 X ∥T ∥ 1 1
∥(T − λI) ∥≤ = ≤ .
|λ| |λ| |λ| − ∥T ∥ ∥T ∥
j=0

Thus, the map λ 7→ (T − λI)−1 is a bounded entire function, and hence by Liouville’s theorem is
constant. This is a contradiction. Hence σ(T ) is non-empty.

Definition 39.1. Let T be a bounded linear operator on a complex Banach space X. Then sup{|λ| :
λ ∈ σ(T )} is called as the spectral radius of T and is denoted by r(T ).

Gelfand’s Theorem. Let T be a bounded linear operator on a complex Banach space X. Then

r(T ) = lim ∥T n ∥1/n .

n→∞

74
Lecture 40 : Compact operators

Definition 40.1. Let X and Y be normed linear spaces and T : X → Y be a linear operator. Then
T is said to be a compact linear operator if for every bounded subset M of X, T (M ) is compact.

Example: Let X be an infinite dimensional normed linear space. Then the identity operator I on
X is not compact. In fact, take M = {x ∈ X : ∥x∥ ≤ 1}. Then we have already seen that M is not
compact. Hence I(M ) = M = M is not compact.

Proposition 40.1. Every compact linear operator is bounded

Proof. Let T : X → Y be a compact linear operator and M = {x ∈ X : ∥x∥ ≤ 1}. Then T (M ) is

compact, and hence, bounded. Thus,

∥T ∥ = sup ∥T x∥ < ∞.
∥x∥=1

Thus, T is bounded.

Theorem 40.1 (Compactness criterion). Let X and Y be normed linear spaces and T : X → Y
be a linear operator. Then T is compact if and only if it maps every bounded sequence in X onto a
sequence in Y which has a convergent subsequence.

Proof. Suppose that T is compact and (xn )n≥1 is a bounded sequence in X. Then {T xn : n ≥ 1} is
compact in Y , and therefore, sequentially compact. Hence (T xn )n≥1 has a convergent subsequence.
Conversely, every bounded sequence (xn )n≥1 in X has a subsequence (xnk )k≥1 such that (T xnk )k≥1
is convergent. Let M be any bounded subset of X and (T xn )n≥1 be any sequence in T (M ). Clearly,
(xn )n≥1 ⊂ M is bounded as M is bounded. Therefore, by assumption, there exists a subsequence
(T xnk ) which is convergent (in T (M )). Since T (M ) is dense in T (M ), it follows that T (M ) is
sequentially compact, and hence, T (M ) is compact.

Theorem 40.2. Let X and Y be normed linear spaces and T : X → Y be a linear operator. Then
the following statements are true:

(i) If T is bounded and dim T (X) < ∞, then T is compact.

(ii) If dim X < ∞, then T is compact.

Proof. (i) Suppose that T is bounded and dim T (X) < ∞. Let (xn )n≥1 be any bounded sequence
in X. Then
∥T xn ∥ ≤ ∥T ∥∥xn ∥ for all n ≥ 1.
Hence (T xn )n≥1 is a bounded sequence in T (X), which is a finite-dimensional subspace. There-
fore, {T xn : n ≥ 1} is compact, and by sequential compactness, (T xn )n≥1 has a convergent
subsequence. Thus, T is compact.

(ii) If dim X < ∞, then dim T (X) < ∞. Thus, the result follows from (i).

Theorem 40.3. Let (Tn )n≥1 be a sequence of compact operators from a normed linear space X into
a Banach space Y . If ∥Tn − T ∥ → 0 as n → ∞ for some linear operator T : X → Y , then T is
compact.

75
Proof. Let (xm )m≥1 be any bounded sequence in X. Since T1 is compact, (T1 x1,m )m≥1 is convergent,
where (x1,m )m≥1 is a subsequence of (xm )m≥1 . Again, (x1,m )m≥1 is bounded and T2 is compact, so
(T2 x2,m )m≥1 is convergent, where (x2,m )m≥1 is a subsequence of (x1,m )m≥1 . Continuing in this way,
we get that the sequence (ym )m≥1 = (xm,m )m≥1 is a subsequence of (xm )m≥1 such that for every
n ≥ 1, (Tn ym )m≥1 is convergent, and hence, Cauchy in Y .
Since (xm )m≥1 is bounded, (ym )m≥1 is bounded too. Let ∥ym ∥ ≤ c for all m ≥ 1. Since
∥Tn − T ∥ → 0, there exists n0 ∈ N such that ∥Tn0 − T ∥ < ε. Since (Tn0 ym )m≥1 is Cauchy, there
exists n1 ∈ N such that
∥Tn0 yj − Tn0 yk ∥ < ε for all j, k > n1 .
Hence, for j, k ≥ n1 , we get

∥T yj − T yk ∥ ≤ ∥T yj − Tn0 yj ∥ + ∥Tn0 yj − Tn0 yk ∥ + ∥Tn0 yk − T yk ∥

≤ ∥T − Tn0 ∥∥yj ∥ + ε + ∥Tn0 − T ∥∥yk ∥
≤ εc + ε + εc = (2c + 1)ε.

Thus, (T yj )j≥1 is a Cauchy sequence in Y . Since Y is complete, (T yj )j≥1 is convergent. That is,
(T xm )m≥1 has a convergent subsequence. Hence, T is compact.

Example: Let T : ℓ2 → ℓ2 be defined as

 x2 x3 xn 
T (x1 , x2 , . . .) = x1 , , , . . . , , . . . , x = (xn )n≥1 ∈ ℓ2 .
2 3 n
Then T is compact. To see this, define
 x2 x3 xn 
Tn (x1 , x2 , . . .) = x1 , , , . . . , , 0, 0, . . . , x = (xn )n≥1 ∈ ℓ2 , n ≥ 1.
2 3 n
Then each Tn is a bounded linear compact operator. Further,
∞
X |xj |2 1
∥(T − Tn )x∥2 = ≤ ∥x∥2 .
(j + 1)2 (n + 1)2
j=n+1

Thus, we get
1
∥T − Tn ∥ ≤ → 0 as n → ∞.
n+1
By preceding theorem, T is compact.

76