18.

700 - LINEAR ALGEBRA, DAY 17

ADJOINT, SELF-ADJOINT, AND NORMAL OPERATORS
THE SPECTRAL THEOREM

SAM SCHIAVONE

C ONTENTS
I. Pre-class Planning 1
I.1. Goals for lesson 1
I.2. Methods of assessment 1
I.3. Materials to bring 1
II. Lesson Plan 2
II.1. Last time 2
II.2. 7A: Adjoint, Self-Adjoint, and Normal Operators, cont. 2
II.3. 7B The Spectral Theorem 6

I. P RE - CLASS P LANNING
I.1. Goals for lesson.

(1) Students will learn properties of the adjoint of a linear operator.

(2) Students will learn the definition of self-adjoint and normal operators.
(3) Students will learn the statements of the real and complex Spectral Theorems.

I.2. Methods of assessment.

(1) Student responses to questions posed during lecture

(2) Student responses to worksheet

I.3. Materials to bring. (1) Laptop + adapter (2) Worksheets (3) Chalk
1
 II. L ESSON P LAN
(0:00)
Announcements: • Midterm Exam 2: Wednesday, November 13th in class • Extra office
hours: Friday, November 8th, 1 - 2pm. • Exam review session: Nov 12 (Tue) 19:00 - 21:00,
2-361
II.1. Last time.
• Proved properties of orthogonal complements.
• Defined orthogonal projection onto a subspace.
• Showed that, given a vector v, the closest point of a subspace U to v is projU (v).
• Defined linear functionals.
• Proved the Riesz Representation Theorem.
• Defined the adjoint of a linear operator.
II.2. 7A: Adjoint, Self-Adjoint, and Normal Operators, cont. For today, let V and W be
nonzero finite-dimensional inner product spaces over F.
Definition 1. Given T ∈ L(V, W ), the adjoint of T is the unique linear map T ∗ ∈ L(W, V )
such that
⟨ T (v), w⟩ = ⟨v, T ∗ (w)⟩ (*)
for all v ∈ V and all w ∈ W.
Proposition 2. Suppose T ∈ L(V, W ).
(i) (S + T )∗ = S∗ + T ∗ for all S ∈ L(V, W ).
(ii) (λT )∗ = λT ∗ for all λ ∈ F.
(iii) ( T ∗ )∗ = T.
(iv) Let U be a finite-dimensional inner product space. Then (ST )∗ = T ∗ S∗ for all S ∈
L(W, U ).
(v) I ∗ = I.
(vi) If T is invertible, then T ∗ is also invertible, and ( T ∗ )−1 = ( T −1 )∗ .
Proof. Suppose v ∈ v, w ∈ W, and λ ∈ F.
(i) By definition,
⟨(S + T )(v), w⟩ = ⟨v, (S + T )∗ (w)⟩ .
Now
⟨(S + T )(v), w⟩ = ⟨S(v), w⟩ + ⟨ T (v), w⟩ = ⟨v, S∗ (w)⟩ + ⟨v, T ∗ (w)⟩
= ⟨v, S∗ (w) + T ∗ (w)⟩ = ⟨v, (S∗ + T ∗ )(w)⟩ .
(ii) Similar.
(iii)
⟨ T ∗ (w), v⟩ = ⟨v, T ∗ (w)⟩ = ⟨ T (v), w⟩ = ⟨w, T (v)⟩ .
(iv) Given S ∈ L(W, U ) and u ∈ U, then
⟨(ST )(v), u⟩ = ⟨S( T (v)), u⟩ = ⟨ T (v), S∗ (u)⟩ = ⟨v, T ∗ (S∗ (u))⟩ .
(v) Exercise.
(vi) Apply ∗ to the equations T −1 T = I and TT −1 = I and then apply the two previous
parts.
2
 □
Proposition 3. Suppose T ∈ L(V, W ). Then
(i) ker( T ∗ ) = (img( T ))⊥ ;
(ii) img( T ∗ ) = ker( T )⊥ ;
(iii) ker( T ) = (img( T ∗ ))⊥ ;
(iv) img( T ) = (ker( T ∗ ))⊥ .
Proof. (i) (⊆): Given w ∈ ker( T ∗ ), then 0 = T ∗ (w). Given x ∈ img( T ), then x = T (v)
for some v ∈ V. Then
⟨ x, w⟩ = ⟨ T (v), w⟩ = ⟨v, T ∗ (w)⟩ = ⟨v, 0⟩ = 0 .
Thus w ∈ (img( T ))⊥ .
(⊇): Similar.
(ii) Replace T by T ∗ in the previous part.
(iii) Take the orthogonal complement of (i).
(iv) Take the orthogonal complement of (ii).
□
Q: After having chosen bases, how does the matrix of T ∗ relate to the matrix of T?
Definition 4. Let A ∈ Mm×n (F). The conjugate transpose of A, denoted A∗ , is defined by
( A∗ )ij = ( At )ij = A ji .
Proposition 5. Let T ∈ L(V, W ), let E := (e1 , . . . , en ), and F := ( f 1 , . . . , f m ) be orthonormal
bases for V and W, respectively. Then
∗
E [T ]F = (F [ T ]E )∗ .
Proof. Recall that the kth column of F [ T ]E is [ T (ek )]F . Since F is orthonormal, we have
T ( e k ) = ⟨ T ( e k ), f 1 ⟩ f 1 + · · · ⟨ T ( e k ), f m ⟩ f m ,
so  
⟨ T ( e k ), f 1 ⟩
[ T (ek )]F =  ..
.
 
.
⟨ T ( e k ), f 1 ⟩
Thus
(F [ T ]E ) jk = ⟨ T (ek ), f j ⟩ .
Similarly
(E [ T ∗ ]F ) jk = ⟨ T ∗ ( f k ), e j ⟩ = ⟨ f k , T (e j )⟩ = ⟨ T (e j ), f k ⟩ .
Thus
(E [ T ∗ ]F ) jk = (F [ T ]E )kj = (F [ T ]E )tjk .
□
3
II.2.1. Self-adjoint operators.
Definition 6. An operator T ∈ L(V ) is self-adjoint if T = T ∗ .
Lemma 7. If E is an orthonormal basis for V, then T is self-adjoint iff [ T ∗ ]E = ([ T ]E )∗ .
Example 8. Let T ∈ L(F2 ) be the linear operator such that
 
2 i
[ T ]E = .
−i 7
T is self-adjoint.
Remark 9. The adjoint of a linear operator is analogous to the complex conjugate of a
complex number.
Proposition 10. Let T ∈ L(V ) be self-adjoint. Then every eigenvalue of T is real.
Proof. Suppose λ ∈ F is an eigenvalue of T, so T (v) = λv for some 0 ̸= v ∈ V. Then
⟨ T (v), v⟩ = ⟨λv, v⟩ = λ < v, v⟩ = λ∥v∥2
⟨ T (v), v⟩ = ⟨v, T (v)⟩ = ⟨v, λv⟩ = λ⟨v, v⟩ = λ∥v∥2 .
Since v ̸= 0, then ∥v∥2 ̸= 0, so λ = λ. Thus λ ∈ R. □
Proposition 11. Suppose V is an inner product space over C and T ∈ L(V ). Then T is self-
adjoint iff ⟨ T (v), v⟩ ∈ R for all v ∈ V.
Proof. (⇒): Assume T is self-adjoint, so T = T ∗ . Given v ∈ V, then
⟨ T ∗ (v), v⟩ = ⟨v, T ∗ (v)⟩ = ⟨ T (v), v⟩ .
Then
0 = ⟨0(v), v⟩ = ⟨( T − T ∗ )(v), v⟩ = ⟨ T (v), v⟩ − ⟨ T ∗ (v), v⟩ = ⟨ T (v), v⟩ − ⟨ T (v), v⟩ .
Thus ⟨ T (v), v⟩ is real.
(⇐): Exercise. (Similar.) □

II.2.2. Normal operators.

Definition 12. An operator T ∈ L(V ) is normal if T commutes with its adjoint, i.e.,
TT ∗ = T ∗ T .
Example 13. Let T ∈ L(F2 ) whose matrix with respect to the standard basis is
 
2 −3
.
3 2
Since its matrix with respect to the standard basis (which is orthonormal) is not symmet-
ric, then T is not self-adjoint. However, T is normal. [Compute TT ∗ and T ∗ T.]
Proposition 14. Suppose T ∈ L(V ). Then T is normal iff ∥ T (v)∥ = ∥ T ∗ (v)∥ for all v ∈ V.
4
Proof. Suppose T is normal. Then T ∗ T = TT ∗ , so given v ∈ V,

⟨ TT ∗ (v), v⟩ = ⟨ T ∗ (v), T ∗ (v)⟩ = ∥ T ∗ (v)∥2

⟨ T ∗ T (v), v⟩ = ⟨ T (v), T (v)⟩ = ∥ T (v)∥2 .

One can show that each of these steps is reversible, so the reverse implication is also
true. □
Proposition 15. Suppose T ∈ L(V ) is normal. Then
(i) ker( T ) = ker( T ∗ );
(ii) img( T ) = img( T ∗ );
(iii) V = ker( T ) ⊕ img( T );
(iv) T − λI is normal for all λ ∈ F;
(v) if v ∈ V and λ ∈ F, then T (v) = λv iff T ∗ (v) = λv.
Proof. (i) Given v ∈ V, then

v ∈ ker( T ) ⇐⇒ ∥ T (v)∥ = 0 ⇐⇒ ∥ T ∗ (v)∥ = 0 ⇐⇒ v ∈ ker( T ∗ )

where the middle equality holds by the previous proposition.

(ii) By a previous result, img( T ) = (ker( T ∗ ))⊥ and (ker( T ))⊥ = img( T ∗ ). Then

img( T ) = (ker( T ∗ ))⊥ = (ker( T ))⊥ = img( T ∗ )

by part (i).
(iii) We have

V = (ker( T )) ⊕ (ker( T ))⊥ = ker( T ) ⊕ img( T ∗ ) = ker( T ) ⊕ img( T ) .

(iv) Exercise.
(v) [Leave as exercise if necessary.] Suppose v ∈ V and λ ∈ F. By the previous part,

∥( T − λI )(v)∥ = ∥( T − λI )∗ (v)∥ = ∥( T ∗ − λI )(v)∥ .

Thus T (v) = λv iff T ∗ (v) = λv.

□
Proposition 16. Suppose T ∈ L(V ) is normal. Then the eigenvectors of T with distinct eigen-
values are orthogonal.
Proof. [Leave as exercise if necessary.] Suppose α ̸= β are eigenvalues of T with corre-
sponding eigenvectors u and v, so T (u) = αu and T (v) = βv. Then

0 = ⟨ T (u), v⟩ − ⟨ T (u), v⟩ = ⟨ T (u), v⟩ − ⟨u, T ∗ (v)⟩ = ⟨αu, v⟩ − ⟨u, βv⟩

= α⟨u, v⟩ − β⟨u, v⟩ = (α − β)⟨u, v⟩ .

Since α − β ̸= 0, then ⟨u, v⟩ = 0. □

5
II.3. 7B The Spectral Theorem. Let V be a finite-dimensional C-vector space and T ∈
L(V ).
• There exists a basis B of V such that [ T ]B is upper triangular.
• If T is diagonalizable, then there exists a basis B of V such that [ T ]B is diagonal. In
this case, B consists of eigenvectors of T.
Q: Now let V be a finite-dimensional inner product space. When does V have an
orthonormal basis E consisting of eigenvectors of T?
A:
• For F = C, when T is normal.
• For F = R, when T is self-adjoint.
Throughout today, let V be a finite-dimensional inner product space over F.
Theorem 17 (Spectral Theorem over C). Suppose F = C and T ∈ L(V ). TFAE.
(i) T is normal.
(ii) There is an orthonormal basis B of V such that [ T ]B is diagonal.
(iii) There is an orthonormal basis of V consisting of eigenvectors of T.
Proof. We have already seen (b) ⇐⇒ (c), so it remains to show (a) ⇐⇒ (b).
(a) =⇒ (b): Assume T is normal. Since minpoly( T ) splits into degree 1 factors, then
there is an orthonormal basis E := (e1 , . . . , en ) of V such that [ T ]E is upper triangular.
 
a11 a12 · · · a1n
 a22 · · · a2n 
[ T ]E =  ..  .
 
. .
 . . 
ann
Since E is orthonormal, then
 
a11
 a12 a22 
[ T ∗ ]E = ([ T ]E )∗ =  .. .. .
 
..
 . . . 
a1n a2n · · · ann
We will show that [ T ]E is diagonal. Then
T (e1 ) = a11 e1
T ∗ (e1 ) = a11 e1 + a12 e2 + · · · + a1n en .
Since e1 , . . . , en is orthonormal, then
∥ T (e1 )∥2 = | a11 |2
∥ T ∗ (e1 )∥2 = | a11 |2 + | a12 |2 + · · · + | a1n |2
= | a11 |2 + | a12 |2 + · · · + | a1n |2
by the Pythagorean Theorem. Since T is normal, then these are equal by a previous result.
Subtracting, then
0 = | a12 |2 + · · · + | a1n |2
so 0 = a12 = · · · = a1n . [Update matrices for [ T ]E and [ T ∗ ]E by filling in 0s.]
6
 Now we have
T (e2 ) = a22 e2
T ∗ (e2 ) = a22 e2 + · · · a2n en
so
∥ T (e2 )∥2 = | a22 |2
∥ T ∗ (e2 )∥2 = | a22 |2 + | a23 |2 + · · · + | a2n |2
= | a22 |2 + | a23 |2 + · · · + | a2n |2 .
By analogous reasoning, then 0 = a23 = · · · = a2n . Proceeding similarly, we find that
aij = 0 for all i ̸= j. Thus [ T ]E is diagonal.
(b) =⇒ (a): Assume there is an orthonormal basis E of V such that [ T ]E is diagonal.
Then [ T ∗ ]E = ([ T ]E )∗ , which is also diagonal. Since diagonal matrices commute, then
[ TT ∗ ]E = [ T ]E [ T ∗ ]E = [ T ∗ ]E [ T ]E = [ T ∗ T ]E ,
so TT ∗ = T ∗ T. (Recall that [·]E : L(V ) → Mn×n (F) is an isomorphism.) Thus T is
normal. □
Theorem 18 (Spectral Theorem over R). Suppose F = R and T ∈ L(V ). TFAE.
(i) T is self-adjoint.
(ii) There is an orthonormal basis B of V such that [ T ]B is diagonal.
(iii) There is an orthonormal basis of V consisting of eigenvectors of T.
We’ll need some preliminary results to prove the theorem.
Lemma 19. Suppose T ∈ L(V ) is self-adjoint, and b, c ∈ R with b2 − 4c < 0. Then
T 2 + bT + cI
is an invertible operator.
Skip if necessary. Suppose 0 ̸= v ∈ V. By Cauchy-Schwarz, we have
|⟨bT (v), v⟩| ≤ ∥bT (v)∥∥v∥ = |b|∥ T (v)∥∥v∥
⇐⇒ −|b|∥ T (v)∥∥v∥ ≤ ⟨bT (v), v⟩ ≤ |b|∥ T (v)∥∥v∥ .
Since T is self-adjoint, then
⟨( T 2 + bT + cI )(v), v⟩ = ⟨ T 2 (v), v⟩ + b⟨ T (v), v⟩ + c⟨v, v⟩
= ⟨ T (v), T (v)⟩ + b⟨ T (v), v⟩ + c∥v∥2
≥ ∥ T (v)∥2 − |b|⟨ T (v), v⟩ + c∥v∥2
|b|∥v∥ 2 |b|2 ∥v∥2
 
= ∥ T (v)∥ − − + c ∥ v ∥2
2 4
|b|∥v∥ 2 4c − b2
 
= ∥ T (v)∥ − + ∥v∥ > 0 .
2 | 4 }
{z
| {z }
≥0 >0
7
Thus ( T 2 + bT + cI )(v) ̸= 0. This shows that ker( T 2 + bT + cI ) = {0}, so this operator is
injective. Since the domain and codomain are both V, then this implies that it is invertible.
□
Proposition 20. Suppose T ∈ L(V ) is self-adjoint. Then
minpoly( T ) = (z − λ1 ) · · · (z − λm )
for some λ1 , . . . , λm ∈ R.
Proof. Case 1: F = C. Then minpoly( T ) splits into degree 1 factors. Recall that the roots
of minpoly( T ) are the eigenvalues of T. Since T is self-adjoint, by a previous result, all its
eigenvalues are real. Thus minpoly has the desired form.
Case 2: F = R. Then minpoly( T ) factors into a product of degree 1 and degree 2
factors:
minpoly( T ) = (z − λ1 ) · · · (z − λm )(z2 + b1 z + c1 ) · · · (z2 + b N z + c N )
for some λ1 , . . . , λm ∈ R and some b1 , . . . , b N , c1 , . . . , c N ∈ R with bk2 − 4ck < 0 for all
k = 1, . . . , N. Goal: N = 0. Since this is minpoly( T ), then
( T − λ1 ) · · · ( T − λm )( T 2 + b1 T + c1 ) · · · ( T 2 + b N T + c N ) = 0 . (∗)
For contradiction, suppose N > 0. Then ( T 2 + b N T + c N ) is invertible by the previous
result, so multiplying both sides of (∗) by its inverse, we have
( T − λ1 ) · · · ( T − λm )( T 2 + b1 T + c1 ) · · · ( T 2 + b N −1 T + c N −1 ) = 0 .
But this has degree strictly smaller than minpoly( T ), contradiction. Thus N = 0. □
Proof of Spectral Theorem over R. We have already seen (b) ⇐⇒ (c).
(a) =⇒ (b): Assume T is self-adjoint. Since minpoly( T ) splits into degree 1 factors by
the previous result, then there exists an orthonormal basis E of V such that [ T ]E is upper
triangular. Since T is self-adjoint, then
([ T ]E )t = ([ T ]E )∗ = [ T ∗ ]E = [ T ]E .
Now ([ T ]E )t is lower triangular, so we must have that [ T ]E is diagonal.
(b) =⇒ (a): Assume there exists an orthonormal basis E of V such that [ T ]E is diagonal.
Then
[ T ]E = ([ T ]E )t = ([ T ]E )∗
so T = T ∗ . Thus T is self-adjoint. □