1 Lecture5

Math 32a Fall 2003 R.
Palais
Lecture 5
Adjoints of Linear Maps and The Spectral Theorem
5.1 The Dual Vector Space of a Vector Space

If V is a vector space then the vector space L(V, R) of linear maps of V into the one-
dimensional vector space of scalars, R plays an important role in many considerations.
It is called the dual space of V and is denoted by V ∗ . Elements of V ∗ are called linear
functionals on V .
5.1.1 Remark. If v1 , . . . , vn is any basis for V , then (since 1 is a basis for R) it follows
from 4.1.1 that there is a uniquely determined basis `1 , . . . , `n for V ∗ such that `i (vj ) = δji .
This is called the dual basis to v1 , . . . , vn . In particular, V ∗ has the same dimension as V .
5.2 The Self-Duality of Inner-Product Spaces

If V is an inner-product space, then there is an important way to represent elements of
∗
V . Namely, recall that the inner-product is linear in each variable when the other variable
is held fixed. This means that for each vector v ∈ V we can define an element v ∗ : V → R
of V ∗ by the formula v ∗ (x) := hx, vi. Since the inner product is linear in both variables,
it follows that the map v 7→ v ∗ is linear. If v ∗ = 0, then in particular 0 = v ∗ (v) = hv, vi, so
by positive definiteness v = 0, i.e., the kernel of the map v 7→ v ∗ is zero, and since V and
V ∗ have the same dimension, it follows that this map is an isomorphism—i.e., every linear
functional on an inner-product space V can be expressed uniquely as the inner-product
with some fixed element of V . This is often expressed by saying that inner-product spaces
are “self-dual”.
5.2.1 Remark. Note that if ei is an orthonormal basis for V , then e∗i is the dual basis
for V ∗ .
5.3 Adjoint Linear Maps

Now let V and W be finite dimensional inner-product spaces and T : V → W a linear
map. We will next define a linear map T ∗ : W → V called the adjoint of T that satisfies
the identity hT v, wi = hv, T ∗ wi for all v ∈ V and all w ∈ W .
If we fix w in W , then to define T ∗ w we note that the map v → hT v, wi is clearly a
linear functional on V , i.e., an element of V ∗ , so by self-duality there is a uniquely defined
element T ∗ w in V such that hv, T ∗ wi = hT v, wi for all v in V .
We recapitulate the above as a formal definition.
14
Math 32a Fall 2003 R. Palais
5.3.1 Definition. Let V, W be finite dimensional inner-product spaces and T ∈ L(V, W ).

The adjoint of T is the unique element T ∗ ∈ L(W, V ) satisfying the identity:
hv, T ∗ wi = hT v, wi .
Here are some exercises involving adjoints and their basic properties.
. 5.3—Exercise 1. Show that (T ∗ )∗ = T .
. 5.3—Exercise 2. Recall that if Tij is an m × n matrix (i.e., m rows and n columns)

and Sji an n × m matrix, then Sji is called the transpose of Tij if Tij = Sji for 1 ≤ i ≤ m,
1 ≤ j ≤ n. Show that if we choose orthonormal bases for V and W , then the matrix of T ∗
relative to these bases is the transpose of the matrix of T relative to the same bases.
. 5.3—Exercise 3. Show that ker(T ) and im(T ∗ ) are orthogonal complements in V ,

and similarly, im(T ) and ker(T ∗ ) are each other’s orthogonal complements in W . (Note
that by Exercise 1, you only have to prove one of these.)
. 5.3—Exercise 4. Show that a linear operator T on V is in the orthogonal group O(V )

if and only if T T ∗ = I (where I denotes the identity map of V ) or equivalently, if and only
if T ∗ = T −1 .
If T : V → V is a linear operator on V , then T ∗ is also a linear operator on V , so it makes
sense to compare them and in particular ask if they are equal.
5.3.2 Definition. A linear operator on an inner-product space V is called self-adjoint if

T ∗ = T , i.e., if hT v1 , v2 i = hv1 , T v2 i for all v1 , v2 ∈ V .
Note that by Exercise 3 above, self-adjoint operators are characterized by the fact that
their matrices with respect to an orthonormal basis are symmetric.
. 5.3—Exercise 5. Show that if W is a linear subspace of the inner-product space V ,

then the orthogonal projection P of V on W is a self-adjoint operator on V .
5.3.3 Definition. If T is a linear operator on V , then a linear subspace U ⊆ V is called

a T -invariant subspace if T (U ) ⊆ U , i.e., if u ∈ U implies T u ∈ U .
5.3.4 Remark. Note that if U is a T -invariant subspace of V , then T can be regarded

as a linear operator on U by restriction, and clearly if T is self-adjoint, so is its restriction.
. 5.3—Exercise 6. Show that if T : V → V is a self-adjoint operator, and U ⊆ V is a

T -invariant subspace of V , the U ⊥ is also a T -invariant subspace of V .
5.4 Eigenvalues and Eigenvectors of a Linear Operator.

In this section, T : V → V is a linear operator on a real vector space V . If λ is a real
number, then we define the linear subspace Eλ (T ) of V to be the set of v ∈ V such that
T v = λv. In other words, if I denotes the identity map of V , then Eλ (T ) = ker(T − λI).
15
Of course the zero vector is always in Eλ (T ). If Eλ (T ) contains a non-zero vector, then

we say that λ is an eigenvalue of T and that Eλ (T ) is the λ-eigenspace of T . A non-zero
vector in Eλ (T ) is called an eigenvector of T belonging to the eigenvector λ. The set of all
eigenvalues of T is called the spectrum of T (the name comes from quantum mechanics)
and it is denoted by Spec(T ).
. 5.4—Exercise 1. Show that a linear operator T on V has a diagonal matrix in a

particular basis for V if and only if each element of the basis is an eienvector of T , and
that then Spec(T ) consists of the diagonal elements of the matrix.
The following is an easy but very important fact.
Theorem. If T is a self-adjoint linear operator on an inner-product space V and λ1 , λ2

are distinct real numbers, then Eλ1 (T ) and Eλ2 (T ) are orthogonal subspaces of V . In
particular, eigenvectors of T that belong to different eigenvalues are orthogonal.
. 5.4—Exercise 2. Prove this theorem. (Hint: Let vi ∈ Eλi (T ), i = 1, 2. You must

show that hv1 , v2 i = 0. Start with the fact that hT v1 , v2 i = hv1 , T v2 i.)
A general operator T on a vector space V need not have any eigenvalues, that is, Spec(T )
may be empty. For example a rotation in the plane (by other than π or 2π radians) clearly
has no eigenvectors and hence no eigenvalues. On the other hand self-adjoint operators
always have at least one eigenvector. That is:
Spectral Lemma. If V is an inner-product space of positive, finite dimension and if

T : V → V is a self-adjoint operator on V , then Spec(T ) is non-empty, i.e., T has at least
one eigenvalue and hence at least one eigenvector.
We will prove this result later after some preparation. But next we show how it leads to
an easy proof of the extremely important:
Spectral Theorem for Self-Adjoint Operators. If V is an inner inner-product space

of posiive, finite dimension and T : V → V is a self-adjoint operator on V , then V has an
orthonormal basis consisting of eigenvectors of T . In other words, T has a diagonal matrix
in some orthonormal basis for V .
PROOF. We prove this by induction on the dimension n of V . If n = 1 the theorem is

trivial, since any non-zero vector in V is clearly an eigenvector, Thus we can assume that
the theorem is valid for all self-adjoint operators on inner-product spaces of dimension
less than n. By the Spectral Lemma, we can find at least one eigenvector w for T . Let
e1 = w/ kwk and let W be the one-dimensional space spanned by w. The fact that w is
an eigenvector implies that W is a T -invariant linear subspace of V , and by 5.3, so is W ⊥ .
Since W ⊥ has dimension n − 1, by the inductive hypothesis T restricted to W ⊥ has an
orthonormal basis e2 , . . . .en of eigenvectors of T , and then e1 , . . . , en is an orthonormal
basis of V consisting of eigenvectors of T .
16
5.5 Finding One Eigenvector of a Self-Adjoint Operator

In this section we will outline the strategy for finding an eigenvector v for a self-adjoint
operator T on an inner product space V . Actually carrying out this strategy will involve
some further preparation. In particular, we will need first to review the basic facts about
differential calculus in vector spaces (i.e., “multivariable calculus”).
We choose an orthonormal basis e1 , . . . , en for V , and let Tij denote the matrix of T in
this basis.
We define a real-valued function F on V by F (x) = hT x, xi. F is called the quadratic
form on V defined by T . The name comes from the following fact.
Pn
. 5.5—Exercise 1. Show that if x = x1 e1 + · · · + xn en , then F (x) = i,j=1 Tij xi xj ,
∂F
Pn
and using the symmetry of Tij deduce that ∂xi
= 2 k=1 Tik xk .
We will denote by S(V ) the set {x ∈ V | kxk = 1}, i.e., the “unit sphere” of normalized
vectors. Note that if x = x1 e1 + · · · + xn en , then x ∈ S(V ) if and only if x21 + · · · + x2n = 1,
so if we define G : V → R by G(x) := x21 + · · · + x2n − 1, then S(V ) = {x ∈ V | G(x) = 0}.
∂F
Pn
5.5.1 Remark. Now Tij = ∂x i
= 2 k=1 Tik xk is just twice the i-th component of T x in
∂G
the basis ej . On the other hand ∂x i
= 2xi , which is twice the i-th component of x in this
basis. It follows that a point x of S(V ) is an eigenvector of T provided there is a constant
∂F ∂G
λ (the eigenvalue) such that ∂x i
(x) = λ ∂x i
(x). If you learned about constrained extrema
and Lagrange multipliers from advanced calculus, this should look familiar. Let me quote
the relevant theorem,
Theorem on Constrained Extrema. Let F and G be two differentiable real-valued

functions on Rn , and define a “constraint surface” S ⊆ Rn by S = {x ∈ V | G(x) = 0}.
Let p ∈ S be a point where F assumes its maximum value on S, and assume that the
partial derivatives of G do not all vanish at p. Then the partial derivatives of F and of
G are proportional at p, i.e., there is a constant λ (the “Lagrange Multiplier”) such that
∂F ∂G
∂xi (x) = λ ∂xi (x) for i = 1, . . . , n.
(We will sketch the proof of the Theorem on Constrained Extrema in the next lecture.)
Thus, to find an eigenvector of T , all we have to do is choose a point of S(V )
where the quadratic form F (x) = hT x, xi assumes its maximum value on S(V ).
17

1 Lecture5

Uploaded by

Copyright:

Available Formats

1 Lecture5

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 Lecture5

Uploaded by

Copyright:

Available Formats

Math 32a Fall 2003 R.

5.1 The Dual Vector Space of a Vector Space

5.2 The Self-Duality of Inner-Product Spaces

5.3 Adjoint Linear Maps

5.3.1 Definition. Let V, W be finite dimensional inner-product spaces and T ∈ L(V, W ).

. 5.3—Exercise 1. Show that (T ∗ )∗ = T .

. 5.3—Exercise 2. Recall that if Tij is an m × n matrix (i.e., m rows and n columns)

. 5.3—Exercise 3. Show that ker(T ) and im(T ∗ ) are orthogonal complements in V ,

. 5.3—Exercise 4. Show that a linear operator T on V is in the orthogonal group O(V )

5.3.2 Definition. A linear operator on an inner-product space V is called self-adjoint if

. 5.3—Exercise 5. Show that if W is a linear subspace of the inner-product space V ,

5.3.3 Definition. If T is a linear operator on V , then a linear subspace U ⊆ V is called

5.3.4 Remark. Note that if U is a T -invariant subspace of V , then T can be regarded

. 5.3—Exercise 6. Show that if T : V → V is a self-adjoint operator, and U ⊆ V is a

5.4 Eigenvalues and Eigenvectors of a Linear Operator.

Of course the zero vector is always in Eλ (T ). If Eλ (T ) contains a non-zero vector, then

. 5.4—Exercise 1. Show that a linear operator T on V has a diagonal matrix in a

Theorem. If T is a self-adjoint linear operator on an inner-product space V and λ1 , λ2

. 5.4—Exercise 2. Prove this theorem. (Hint: Let vi ∈ Eλi (T ), i = 1, 2. You must

Spectral Lemma. If V is an inner-product space of positive, finite dimension and if

Spectral Theorem for Self-Adjoint Operators. If V is an inner inner-product space

PROOF. We prove this by induction on the dimension n of V . If n = 1 the theorem is

5.5 Finding One Eigenvector of a Self-Adjoint Operator

Theorem on Constrained Extrema. Let F and G be two differentiable real-valued

You might also like