Sie sind auf Seite 1von 20

Linear Algebra Lecture Notes-II

Vikas Bist

Department of Mathematics
Panjab University, Chandigarh-160014
email: bistvikas@gmail.com
Last revised on March 5, 2018

This text is based on the lectures delivered for the B.Tech students of IIT Bhilai.
These lecture notes are basics of Linear Algebra, and can be treated as a first course
in Linear Algebra. The students are suggested to go through the examples carefully
and attempt the exercises appearing in the text.

C ONTENTS

1 Similarity of matrices 1

2 Eigenvalues and eigenvetors 2

3 Inner product and orthogonality 7

4 Adjoint of a linear transformation 12

5 Unitary similarity 14

6 Bilinear forms 16

7 Symmetric bilinear forms 18


1 SIMILARITY OF MATRICES

1 S IMILARITY OF MATRICES

Let V be a vector space over K of dimension n. Let T be a linear operator on V.1


If B is a fixed basis then we have seen that [T ]B is an n × n matrix. Further T is
injective (hence bijective) if and only if [T ]B is invertible. Here we see how matrix
of a linear operator changes when we change basis.
We see how matrices appear when we compose linear transformations. Let
V, W and X be vector spaces over K. and let T : V −→ W and S : W −→ X be
linear transformations so that S ◦ T is defined. Let B, B 0 and B 00 be ordered bases
of V W and X respectively. Write B = {v1 , . . . , vn }. Then the j-th column of the
matrix B00 [S ◦ T ]B is:

[(S ◦ T )vj ]B00 = B00 [S]B0 [T vj ]B0 = B00 [S]B0 B0 [T ]B [v]B .

Also
[(S ◦ T )vj ]B00 = B00 [(S ◦ T ]B [vj ]B .
Hence on equating these last two equations we have the following:

(1) B00 [(S ◦ T ]B = B00 [S]B0 B0 [T ]B

Now assume that T is a linear operator on V. Let B and B 0 be bases of V. We


write [T ]B for B [T ]B If I is the identity operator on V, then using equation (1):

[T ]B0 = [I ◦ T ]B0 = B0 [I]B B [T ]B0 = B0 [I]B B [T ◦ I]B0 = B0 [I]B [T ]B B [I]B0 .

Also again by (1):


In = [I]B0 = B0 [I]B B [I]B0 .

Thus if P = B [I]B0 , then P −1 = B0 [I]B .

R EMARK 1.1. Note that B0 [I]B is a matrix whose j-th column is the elements
of K when the j-th element of B 0 is expressed as a linear combination of the
elements of B.

P ROPOSITION 1.2. Let V be an n dimensional vector space over a field F, T a


linear operator on V and let B and B 0 be ordered bases of V. Then there is an
invertible matrix P such that [T ]B0 = P −1 [T ]B P.
Square matrices A and B are similar if for some invertible matrix P B =
P −1 AP. The above statement mean that the change of basis is a similarity transform
on the matrices of a linear operator.

P ROPOSITION 1.3. Similar matrices have the same trace and the same
determinants.

1
Recall that a linear operator on V is a linear transformation from V to V.

1
2 EIGENVALUES AND EIGENVETORS

Thus we can define the determinant and the trace of a linear operator T on V
by:
detT := det[T ]B and trT := tr[T ]B ,
where B is any ordered basis of V.

E XAMPLE 1.4. Let T be a linear


 operator whose matrix with respect to a basis
2 1 1
{u1 , u2 , u3 } is A = 0 3 1 . We find the matrix of T with respect to the basis
0 1 3
B 0 = {v1 = u1 + u2 , v2 = u2 + u3 , v3 = u3 + u1 }. Thus
5 1 1
T (v1 ) = T (u1 + u2 ) = T (u1 ) + T (u2 ) = 3u1 + 3u2 + u3 = v1 + v2 + v3
2 2 2
T (v2 ) = T (u2 + u3 ) = T (u2 ) + T (u3 ) = 2u1 + 4u2 + 4u3 = v1 + 3v2 + 3v3
1 1 5
T (v3 ) = T (u3 + u1 ) = T (u3 ) + T (u1 ) = 3u1 + u2 + 3u3 = v1 + v2 + v3
2 2 2
 
5/2 1 1/2
Hence [T ]B0 = 1/2 3 1/2 .

1/2 1 5/2
 
1 0 1
Now B [I]B0 = 1 1 0 . If we call this matrix P, then P −1 [T ]B0 P = [T ]B .
0 1 1
Verify that [T ]B0 and [T ]B have the same determinant and the same rank.

2 E IGENVALUES AND EIGENVETORS

D EFINITION 2.1. Let T be a linear operator on V. Then λ ∈ K is an eigenvalue


of T if there is a non-zero v ∈ V such that T v = λv. The vector v is called an
eigenvector corresponding to λ.

We have seen that a linear operator corresponds to a matrix and conversely, so


to keep the things simple, we basically deal with square matrices. Restating the
above definition for matrices:
Let A be an n × n matrix. A non-zero column vector v is called an eigenvalue of A
if there is a number λ such that Av = λv. The number λ is called an eigenvector
of A.
Thus an eigenvalue is a number λ such that the homogeneous system

(A − λIn )x = 0

has a non-zero solution. This happens if and only if rank(A − λIn ) < n, that is,
A − λIn is not invertible which is equivalent to det(A − λIn ) = 0. This means
that eigenvalues of A are precisely those values λ for which the matrix (A − λIn )

2
2 EIGENVALUES AND EIGENVETORS

has the zero determinant. In other words eigenvalues are the roots of the following
monic polynomial:
cA (x) := det(xIn − A).
This polynomial is called the characteristic polynomial of A.
Note that eigenvectors corresponding to an eigenvalue are not unique. In fact
if u is an eigenvector, then αu is also an eigenvector corresponding to λ for any
α ∈ K,

E XERCISE 2.2. Let A ∈ K n×n . Then λ ∈ K is an eigenvalue if and only if


ker(A − λIn ) 6= {0}. Any non-zero vector of ker(A − λIn ) is an eigenvector
corresponding to λ.

E XAMPLE 2.3.
 Find the  eigenvalues and the corresponding eigenvectors for the
3 1 1
matrix: A =  1 3 1  .
1 1 3
The characteristic equation of A is:

x−3 −1 −1
−1 = (x − 5)(x − 2)2 .

det(xI3 − A) = −1 x − 3
−1 −1 x − 3

Thus the eigenvalues are 2 and 5.


To find eigenvectors corresponding to eigenvalue
 5, we need to
 find solution of
−2 1 1
the system (A − 5I3 )x = 0. Now A − 5I3 is  1 −2 1  . Now to solve
1 1 −2
this we reduce this matrix to row echelon from. Interchanging the first and the third
row and then adding first and the second row to the third row we have:
 
1 1 −2
 1 −2 1 
0 0 0

Now subtracting the first row from the second we have:


 
1 1 −2
 0 −3 3 
0 0 0

This is he row echelon form and so the rank of A − 5I3 is 2 thus nullity is 1. Thus
there is one fundamental solution and that is given by a solution of

x1 + x2 − 2x3 = 0, x1 − x3 = 0.

3
2 EIGENVALUES AND EIGENVETORS

 
1
Hence solution is x1 = x2 = x3 or u  1  , where u ∈ K. Thus we can take
1
 
1
eigenvector corresponding to 5 as  1  .
1
Eigenvector corresponding
  to 2 is a solution of the system (A − 2I3 )x = 0.
1 1 1
Now A − 2I3 =  1 1 1  and the row echelon form for this matrix is
1 1 1
 
1 1 1
 0 0 0 
0 0 0

a matrix of rank 1 and so nullity 2. This means that there are 2 fundamental solutions
that is 2 linearly independent eigenvectors corresponding to eigenvalue 2.
Thus eigenvector corresponding to 2 is satisfying the equation x1 + x2 + x3 = 0
or x3 = −x1 − x2 . Thus any solution of the system is
     
x1 1 0
 x2  = x1  0  + x2  1 
−x1 − x2 −1 −1
  
1 0
The two linearly independent eigenvectors can be taken as  0  and  1  .
−1 −1

Note that the characteristic polynomial for an n × n matrix is of degree n.


Thus an n × n matrix cannot have more than n distinct eigenvalues. The
0 2
matrix may not have eigenvalues. For example the matrix A =
−1 0
2
has characteristic polynomial x − 2. This has no eigenvalues when A is
considered as a matrix in Q. But A has eigenvalues when considered as a
matrix in R.

Note that if A ∈ Cn×n , then A always have eigenvalues.

Procedure to find eigenvalues and eigenvectors


(i). Find the characteristic polynomial of A.
(ii). Find all its roots, these are eigenvalues.
(iii). For each eigenvalue λ, solve the system (A − λIn )x = 0.

4
2 EIGENVALUES AND EIGENVETORS

 
2 1 −1
E XAMPLE 2.4. Consider the matrix A =  0 2 1  . The characteristic
0 1 2
polynomial is (x − 1)(x − 3)(x − 2) Eigenvalues are 1, 2, 3.
Eigenvector
 corresponding
  by solving the system (A − I3 )x = 0,
to 1:obtained
1 1 −1 x1 0
that is,  0 1 1   x2  =  0  . Thus x1 +x2 −x3 = 0 and x2 +x3 = 0.
0 1 1 x3 0
 
2
Thus x2 = −x3 and x1 = 2x3 . An eigenvector corresponding to 1 is  −1  .
1
Eigenvector
 corresponding
 to 2:
   obtained by solving the system (A−2I3 )x = 0,
0 1 −1 x1 0
that is,  0 0 1   x2  =  0  . Thus we have x2 = x3 = 0. Hence an
0 1 0 x3 0
 
1
eigenvector corresponding to 2 is  0  .
0
Eigenvector
 corresponding
 to   by solving the system (A−3I3 )x = 0,
3: obtained
−1 1 −1 x1 0
that is,  0 −1 1   x2 = 0  . Thus we have x2 = x3 and x1 = 0.
 
0 1 −1 x3 0
 
0
Hence an eigenvector corresponding to 3 is  1  . •
1

P ROPOSITION 2.5. Let A be an n × n matrix. If A has k distinct eigen-


values, then eigenvectors corresponding to these eigenvalues are linearly
independent.

Proof. Suppose that λ1 , . . . λk be distinct eigenvalues of A with corresponding


eigenvectors u1 , . . . , uk . Thus Aui = λi ui , for i = 1, . . . , k.
Let α1 u1 + . . . αm um = 0 and let r be largest index such that αr 6= 0. Then
multiplying by (A − λ1 In ) on the both sides we have:

(λ2 − λ1 )u2 + · · · + (λk − λ1 )uk = 0.

Next multiplying by (A − λ2 In ) and continuing upto (A − λr−1 In ), we have

(λr − λ1 )(λr − λ2 ) . . . (λr − λr−1 )αr ur = 0.

This is a contradiction as (λk − λ1 )(λk − λ2 ) . . . (λk − λk−1 )αk 6= 0 and ur


being a a characteristic vector is non-zero too. Hence u1 , . . . , uk are linearly
independent.

5
2 EIGENVALUES AND EIGENVETORS

The reader can verify that eigenvectors in the Example 2.4 are actually linearly
independent.
Let A ∈ K n×n and λ be an eigenvalue of A. Assume that u1 , . . . , uk ∈ K n are
linearly independent eigenvectors corresponding to λ. Then extend {u1 , . . . , uk } to
a basis of K n : {u1 , u2 , . . . , uk , uk+1 , . . . , un }. Let P ∈ K n×n with j-th column
as uj . Then P is invertible. Then consider the matrix P −1 AP, For j = 1, . . . , k,
the j-th column is
(P −1 AP )ej = (P −1 A)P ej = P −1 Auj = (P −1 (λuj ) = λP −1 uj
= λP −1 P ej = λej .
 
λIk X
Thus P −1 AP = . It follows that if K n has a basis consisting of
0 Y
eigenvectors of A then P −1 AP is a diagonal matrix.
A matrix A that is similar to a diagonal matrix is called diagonalizable matrix.

A ∈ K n×n is diagonalizable if and only if K n has a basis consisting only of


the eigenvectors of A.

The following is immediate.


P ROPOSITION 2.6. If A ∈ K n×n has n distinct characteristic vectors, then A is
diagonalizable.

Recall that if A is an n × n matrix, then adj(A) is the adjoint of A such that


A adj(A) = det(A) In .
Now replacing the matrix A by the matrix xIn − A we have
(xIn − A) adj(xIn − A) = det(xIn − A) In = cA (x)In ,
where cA (x) is the characteristic polynomial of A. Now adj(xIn − A) is a matrix
with entries consisting of cofactors of xIn − A. Thus the entries are polynomials of
degree at most n − 1. Thus we can write:
adj(xIn − A) = B0 + b1 x + · · · + Bn−1 xn−1 ,
where each Bj is an n × n matrix. Thus:
(xIn − A)(B0 + b1 x + · · · + Bn−1 xn−1 ) = cA (x)In .
Equating the coefficients of the powers of x:
−AB0 = a0 In
B0 − AB1 = a1 In
..
.
Bn−1 − ABn−1 = an−1 In
Bn−1 = In .

6
3 INNER PRODUCT AND ORTHOGONALITY

Now pre-multiply the second equation by A, the third by A2 , . . . , the (n − 1)-th


(the last) by An and add all the resulting equations. We have2

a0 In + a1 A + · · · + an−1 An−1 + An = 0.

Hence cA (A) = 0. This proves the following important result called the Cayley
Hamilton Theorem.

P ROPOSITION 2.7 (Cayley-Hamilton). If A is an n × n matrix and cA (x) is


its characteristic polynomial, then cA (A) = 0.

P ROPOSITION 2.8. A is invertible if and only if cA (0) 6= 0.


Proof. Let cA (x) = c1 + c1 x + · · · + cn−1 xn−1 + xn . Then cA (0) = 0 means that
c0 = 0. Now c0 = 0 if and only if one of the roots of the characteristic polynomial
is 0, that is, one of the the eigenvalues of A is 0. This implies that there is a non-zero
column vector u such that Au = 0. Hence rank(A) < n. This is equivalent to that
A is not invertible.

The Cayley Hamilton Theorem can be used for finding the inverse of a matrix.
The following example illustrates this fact.
 
4 2 2
E XAMPLE 2.9. Let A =  3 3 2  . Then cA (x) = x3 − 7x2 + 14x − 8.
−3 −1 0
Since the constant term of cA (x) is non-zero, A is invertible. By Cayley-Hamilton
Theorem: A3 − 7A2 + 14A − 8I3 = 0. Now multiplying by A−1 , we have:
A2 − 7A + 14I3 − 8A−1 = 0. Hence
1
A−1 = (A2 − 7A + 14I3 ).
8

3 I NNER PRODUCT AND ORTHOGONALITY

From here onwards F will denote either the field R or C.

D EFINITION 3.1. An inner product on V is a mapping V × V to F which maps


an ordered pair x, y to (x, y) ∈ F such that the following properties hold.
(i).(x, x) ≥ 0 and (x, x) = 0 if and only if x = 0.
(ii). (αx + βy, z) = α(x, z) + β(y, z) for all x, y, x ∈ V and α, β ∈ F.
(iii). (y, x) = (x, y) for all x, y, x ∈ V.3
A vector space V over F equipped with an inner product is called an inner product
space.
2
For a polynomial p(x) = p0 + p1 x + · · · + pk xk , we write p(A) = p0 In + p1 A + · · · + pk Ak .
3
z denotes the complex conjugate of z.

7
3 INNER PRODUCT AND ORTHOGONALITY

The following are consequences of definition:


(1) (x, αy + βz) = (αy + βz, x) = α(y, x) + β(z, x) = α(x, y) + β(x, z) for all
x, y, z ∈ V and α, β ∈ F.
(2) α(x, y) = (αx, y) = (x, αy) for all x, y ∈ V and α ∈ F.
E XERCISE 3.2. Show that in an inner product space V : (i) (x, 0) = 0 for all x ∈ V.
(ii) If x ∈ V such that (x, y) = 0 for all y ∈ V, then x = 0.
E XERCISE 3.3. Show that for fixed w ∈ V, the mapping f : V → F given by
f (x) = (x, w) is a linear form.
E XAMPLE 3.4. 1. On Cn the inner product (x, y) := y ∗ x is called the standard
inner product. The corresponding standard inner product on Rn is clearly:
(x, y) := y t x.

2. A generalization of the standard inner product. Let A be an invertible n × n


matrix. Then the following is also an inner product:

(x, y)A := (Ay)∗ (Ax), for all x, y ∈ V.

3. On F m×n , the space of m × n matrices with entries from F, the following is


an inner product:
(A, B) = tr(B ∗ A).
If n = 1 this reduces to the standard inner product of vectors. It is for this
reason it is referred as the standard inner product of matrices.

4. Let V = Rn [x], the space of polynomials with coefficients in R and of degree


at most n. Then there is an inner product on V given by:
Z 1
(p, q) = p(t)q(t) dt.
0

R EMARK 3.5. We normally deal here with the standard inner product space
Rn . Whenever we say inner product space F n , without mentioning inner
product, we mean the standard inner product space.

If V is an inner product space, for x ∈ V, define the norm of x :


p
kxk := (x, x).

A vector x such that kxk = 1 is called a unit vector. The following is an important
inequality. Vectors x, y in an inner product space are called orthogonal if (x, y) =
0.
 
1
3
E XERCISE 3.6. In R find all vectors orthogonal to 1 .

1

8
3 INNER PRODUCT AND ORTHOGONALITY

A subset {u1 , . . . , uk } is an orthogonal set if (ui , uj ) = 0, for i 6= j, and is


called an orthonormal set if it is orthogonal and very vector is a unit vector, that is,
(
1 if i = j
(ui , uj ) = δij = .
0 if i 6= j

E XAMPLE 3.7. The standard basis of F n is an orthonormal basis.

P ROPOSITION 3.8. Every orthonormal set of an inner product space is


linearly independent

Proof. Let {u1P


, . . . , uk } be an orthonormal set. Suppose that α1 u1 + · · · + αk uk =
0. Then αl = ( ki=1 αi ui , ul ) = 0 for each l.

Our next result shows that every finite dimensional inner product space has an
orthonormal basis.

P ROPOSITION 3.9 (Gram Schmidt orthogonalization procedure). Let {x1 , . . . , xk }


be an ordered linearly independent set. Then there is an orthonormal set {u1 , . . . , uk }
such that for each hx1 , . . . , xl i = hu1 , . . . , ul i for all l = 1, . . . , k.
1
Proof. Let u1 = x1 . Suppose that mutually orthonormal set of vectors
kx1 k
{u1 , . . . , ul } have been constructed such that hx1 , . . . , xl i = hu1 , . . . , ul i. Define
l
X
yl+1 = xl+1 − (xl+1 , ui )ui .
i=1

1
Then (yl+1 , ui ) = 0 for all i = 1, . . . , l. Now let ul+1 = yl+1 . then clearly,
kul+1 k
{u1 , . . . , ul+1 } is an orthonormal set.
Since from the above equation it follows that xl+1 ∈ hyl+1 , u1 , . . . ul i =
hul+1 , u1 , . . . , ul i, and so hx1 , . . . , xl+1 i ⊆ hu1 , . . . , ul+1 i.
Also from above equation, it follows that
l
!
1 X
ul+1 = xl+1 − (xl+1 , ui )ui .
kyl+1 k
i=1

Therefore ul+1 ∈ hu1 , . . . , ul , xl+1 i = hx1 , . . . , xl+1 i. Hence hu1 , . . . , ul+1 i ⊆


hx1 , . . . , xl+1 i.

9
3 INNER PRODUCT AND ORTHOGONALITY

Procedure for generating an orthonormal set.


Given a linearly independent set (ordered): {x1 , . . . , xk }.
x1
Write u1 = . (Normalize the vector)
kx1 k
u1 , . . . , ui are done, then define
yi+1 = xi+1 − ((xi+1 .u1 )u1 + · · · + (xi+1 , ui )ui ),
yi+1
Write ui+1 = .
|kyi+1 k
{u1 , . . . , uk } is an orthonormal set.

     
 0 1 1 
E XAMPLE 3.10. Find an orthonormal basis for R3 from a given basis 1 , 0 , 1 .
1 1 0
 
       
0 1 1 0
1  
x1 = 1 , x2 = 0 , x3 = 1 . u1 = 2 1 .
      √
1 1 0 1
      r  1 
1 0 1
2  1
y2 = x2 − (x2 , u1 )u1 = 0 − 12 1 = −1/2 . u2 = −
3 12
1 1 1/2 2
      
1 0 1 1
y3 = x3 − (x3 , u1 )u1 − (x3 , u2 )u2 = 1 − 12 1 − 31 − 12  = 23  1  .
1
0 1 2 −1
 
1
1
u3 = √  1  .
2 3 −1
      
 0 q 1 1 
Hence orthonormal basis is √12 1 , 23 − 21  , 2√ 1 
3
1 .
1
1 −1
 
2

E XERCISE 3.11.PShow that if {u1 , . . . , un } is an orthonormal basis for V and


v ∈ V, then v = ni=1 (v, ui )ui .

P ROPOSITION 3.12. Let V be an inner product space over F. Let S ⊆ V.


Define S ⊥ = {u ∈ V : (u, x) = 0 for all x ∈ S}. Then S ⊥ is a subspace of
V.

Proof. If u, w ∈ S ⊥ , then for α, β ∈ F : (αu + βw, x) = α(u, x) + β(w, x) = 0


for all x ∈ S. Hence αu + βw ∈ S ⊥ , and so S ⊥ is a subspace of V.

E XERCISE 3.13. Let V be an inner product space and let X, Y ⊆ V. Prove that:
(i) If X ⊆ Y, then Y ⊥ ⊆ X ⊥ .
(ii) S ⊥ = hSi⊥ .
(iii) If X is a basis of a subspace W of V, then X ⊥ = W ⊥ .

10
3 INNER PRODUCT AND ORTHOGONALITY

(iv) If W is a subspace of V, then (W ⊥ )⊥ = W.


Caution: (iv) is not true if X is a subset of V as X ⊥ is a subspace.
P ROPOSITION 3.14. Let w1 , . . . , wk be linearly independent in an inner product
space V. Then dim{w1 , . . . , wk }⊥ = dim(V ) − k.
Proof. Let dimV = n. Let W = hw1 , . . . , wk i. and let {u1 , . . . , uk } be an or-
thonormal basis of W. Then {w1 , . . . , wk }⊥ = W ⊥ = hu1 , . . . , uk i⊥ . Extend
u1 , . . . , uk to u1 , . . . , uk , xk+1 , . . . , xn so that this is a basis of V. Now by Gram
Schmidt the P take the corresponding orthonormal basis u1 , . . . , uk , uk+1 , . . . , un .
Now if v = ni=1 αi ui ∈ W ⊥ , then αj = (v, uj ) = 0 for j = 1, . . . , k. Hence
v ∈ huk+1 , . . . , un i. Therefore, dim W ⊥ = n − k.

We leave it to the student to verify the following.

Let W be a subspace of an inner product space V and let {w1 , . . . , wk be an


orthonormal basis for W. Then for any v ∈ V :
k
X
v− (v, wj )wj ∈ W ⊥ .
i=1

The vector ki=1 (v, wj )wj ∈ W is called the orthogonal projection of v


P
onto W (and along W ⊥ .)

  

 x 1 

 x2  x1 + x2 + x3 + x4 = 0 
E XAMPLE 3.15. Let W =   :   be a subspace of
x 2x1 − x3 − x4 = 0
 3

 

x4

R4 . Find W ⊥ . Also find the orthogonal projection of e1 along W.
    ⊥    
* 1 2 + * 1 2 +
1  0  1
. Thus W ⊥ = y1 =  0
   
Observe that W =  1 , −1
  
1 , y2 = −1 .
1 −1 1 −1
These two vectors are already orthogonal. Thus an orthonormal basis for W ⊥ is
{ 12 y1 , √16 y2 }. Now e1 = 41 y1 + 31 y2 .

E XERCISE 3.16. Let W be a subspace of V. Show that W ∩ W ⊥ = {0}.


P ROPOSITION 3.17. Let W be a subspace of V. Then to each v ∈ V there are
unique vectors v1 and v2 such that v = v1 + v2 , v1 ∈ W and v2 ∈ W ⊥ .
Proof. Let {w1 , . . . , wk } be an orthonormal basis of W. Then for v2 = v −
P k ⊥ k
i=1 (v, wj )wj ∈ W . Thus v = v1 + v2 where v1 = sumi=1 (v, wj )wj ∈ W.
If v = v1 + v2 = v1 + v2 v1 v1 ∈ W, v2 v2 ∈ W ∗ ⊥, then v1 − v10 = v20 − v2 ∈
0 0 0 0

W ∩ W ⊥ = {0}. Hence uniqueness.

11
4 ADJOINT OF A LINEAR TRANSFORMATION

Let V be an inner product space. Let W be a subspace of V. The above


proposition proves that for given v ∈ V unique vectors v1 ∈ W and v2 ∈
W ⊥ such that v = v1 + v2 . Thus we have a mapping

PrW : V → V given by Prw (v) = v1 .

called the orthogonal projection of V onto W. Thus if v ∈ V, then


P rW (v) ∈ W and so v − P rW (v) ∈ W ⊥ . Also if {w1 , . . . , wk } is an
orthonormal basis of W, then by uniqueness it follows that
k
X
P rW (v) = (v, wi )wi .
i=1

E XERCISE 3.18. (i) Verify that P rW is a linear operator on V.


(ii) Observe that P rW (w) = w for all w ∈ W. Thus P rW 2 (v) = P r (v) for all
W
2
v ∈ V, that is P rW = P rW .
(iii) Let X = {w1 , . . . , wn } be an orthonormal basisof V such
 that {w1 , . . . , wk }
Ir 0
is an orthonormal basis of W. Show that [P rW ]X = .
0 0

4 A DJOINT OF A LINEAR TRANSFORMATION

Exercise 3.3 shows that if V is an inner product space, then for fixed y ∈ V, the
map: v 7→ (v, x) is a linear functional. The next result states that the converse also
holds.

P ROPOSITION 4.1. [Riesz representation theorem] Let V be an n dimensional inner


product space and let f be a linear functional on V. Then there is a unique y ∈ V
such that f (x) = (x, y) for all x ∈ V.

Proof. Since ker f is n − 1 dimensional, (ker f )⊥ is one-dimensional (Proposition


3.14). Let u be a unit vector in (ker f )⊥ .
We show that f (x) = (x, αu) for a suitable α ∈ F. If this is the case, then
f (u) = α or α = f (u). Now we verify that y = f (u)u is the required vector.
Indeed each v = v1 + γu, where v1 ∈ ker f, Now f (v) = γf (u). Also (v, y) =
(v, f (u)u) = f (u). Hence f (v) = (v, y) for all v ∈ V.
Finally, if w ∈ V is another such vector. Then f (v) = (v, y) = (v, w), for all
v ∈ V. But then (v, y − w) = 0 for all v ∈ V and so y = w.

Let V and W be inner product spaces of dimensions n and m respectively. Let


T V → W be a linear transformation. Let y ∈ W be fixed. Then fT : V → F given
by fT (v) = (T v, y) is a linear form. By Proposition 4.1, there is unique vector
x ∈ V such that fT (x) = (v, x). Thus for each y ∈ W there is a unique vector

12
4 ADJOINT OF A LINEAR TRANSFORMATION

x ∈ V. This defines a map: T ∗ : W → V given by T ∗ (y) = x. In other words

(T (v), y) = (v, T ∗ (y)) for all v ∈ V.

This map T ∗ is called the adjoint of T.


P ROPOSITION 4.2. T ∗ : W → V is linear.
Proof. For α, β ∈ F and y, z ∈ W :

(v, T ∗ (αy + βz)) = (T (v), αy + βz) = α(T (v), y) + β(T (v), z)


= α(v, T ∗ (y)) + β(v, T ∗ (z)) = (v, αT ∗ (y) + βT ∗ (z)), for all v ∈ V

Hence T ∗ (αy + βz) = αT ∗ (y) + βT ∗ (z) for all α, β ∈ F and y, z ∈ W.

E XERCISE 4.3. Prove that (T ∗ )∗ = T.


Let V and W be inner product spaces of dimensions n and m respectively. Let
V = {v1 , . . . , vP
n } and W = {w1 , . . . , wm } be orthonormal bases for V and W.
Then T (vj ) = m i=1 (T (vj ), wi )wi . Therefore

W [T ]V is an m × n matrix with (i.j)-th entry as (T (vj ), wi ).

This implies that V [T ∗ ]W is an n×m matrix with (i.j)-th entry as (T ∗ (wj ), vi ).


Now as (T ∗ (wj ), vi ) = (vi , T ∗ (wj ) = (T (vi ), wj ). This proves the following.

The (i, j)-th entry of V [T ∗ ]W is the conjugate of the (j, i)-th entry of W [T ]V .

Thus if A is the matrix of T with respect to some orthonormal bases, then


the matrix of T ∗ is the conjugate transpose of A.

E XERCISE 4.4. Prove the following properties of the adjoint.


(i) If S, T ∈ L(V, W ), then (αS + T )∗ = αS ∗ + T ∗ for α ∈ F.
(ii) (ST )∗ = T ∗ S ∗ .
(iii) If T ∈ L(V ) and T is invertible, then (T ∗ )−1 = (T −1 )∗ .
 
x1  
3 2 x1 + 2x2 + 3x3
E XAMPLE 4.5. Let T : R → R given by T x2 =   . Then
4x1 + 5x2 + 6x3
x3
 
1 2 3
the matrix of with respect to the standard bases is . Therefore the matrix
4 5 5
 
1 4
of T ∗ with respect to the same bases if 2 5 . Hence T ∗ : R2 → R3 given by
3 6
 
  y1 + 4y2
y
T ∗ 1 = 2y1 + 5y2  .
y2
3y1 + 6y2

13
5 UNITARY SIMILARITY

 
x  
3 2 x + ιy
E XERCISE 4.6. Let T : C → C given by T y  = . Find
(i + ι)y + 3z
z
 
1 + ι
T∗ .
1−ι
D EFINITION 4.7. A linear operator on an inner product space V is called self
adjoint if T ∗ = T. Thus T is self adjoint if and only if the matrix of the matrix of
T with respect to orthonormal basis is Hermitian4

5 U NITARY SIMILARITY

Let us assume from here onwards that all matrices have entries in C and Cn
has the standard inner product.

For an n × n matrix, write A∗ for the conjugate transpose of A. A matrix with


orthonormal columns is called a unitary matrix. Thus columns of a unitary matrix
form an orthonormal basis.
Note that the (i, j)-th entry of the matrix A∗ A is the inner (standard) product of
the i-th and j-th columns, it follows that A is unitary if and only if A∗ A = In .

R EMARK 5.1. An n × n matrix such that At A = In is called an orthog-


onal matrix. An orthogonal matrix need not be unitary. For example
1 2 ι

3 ι −2
. However, if A ∈ Rn×n , then A is orthogonal if and only
if unitary.

E XERCISE 5.2. Show that the inverse of a unitary matrix is unitary, and the product
of two unitary matrices is unitary. Show by example that the sum of two unitary
matrices is not unitary.

D EFINITION 5.3. Matrices A, B ∈ Cn×n are called unitary similar if there is a


unitary matrix U such that U ∗ AU = B.
A ∈ Cn×n is unitary diagonalizable if A is unitary similar to a diagonal
matrix.

P ROPOSITION 5.4 (Schur). Let A ∈ Cn×n . Then A is unitary similar to an upper


triangular matrix.

Proof. By induction onn. If n = 1 the statement is obvious. Assume that the


statement is true for all matrices of order less than n. Let λ be an eigenvalue of
A and u be corresponding eigenvector of norm 1.. Let U1 be a unitary matrix
4
A matrix A is Hermitian if it is equal to its conjugate transpose, that is A = At ..

14
5 UNITARY SIMILARITY

whose
 firstcolumn is u. Then the first column of P1−1 AP1 is λe1 and so U1∗ AU1 =
λ ∗
. Here * denotes those entries which are of no interest to us.
0 A1

Now by induction hypothesis  thereis an unitary matrix V such that V A1 V is
1 0
upper triangular. Now if U2 = , then U = U1 U2 is unitary and U ∗ AU =
0 V
 
λ ∗
. Hence an upper triangular matrix.
0 V ∗ A1 V
Note that diagonal entries of an upper triangular matrix are eigenvalues. Hence
we have the following statement.

Let A ∈ Cn×n . Then det(A) is the product of diagonal eigenvalues, and the
trace of A is the sum of its eigenvalues.

An immediate consequence of the Proposition 5.4 is the following.

P ROPOSITION 5.5. A unitary matrix is unitary diagonalizable.

E XERCISE 5.6. Let T be an upper triangular n × n matrix such that T T ∗ = T ∗ T.


Then T is a diagonal matrix.

D EFINITION 5.7. An n × n matrix A is normal if AA∗ = A∗ A.

P ROPOSITION 5.8 (Spectral decomposition). An n × n matrix A is unitary diago-


nalizable if and only if A is normal.

Proof. Suppose A is normal matrix. Then by Proposition 5.4, there is a uni-


tary matrix U such that U ∗ A = T, an upper triangular matrix. Now T T ∗ =
(U ∗ AU )(U ∗ AU )∗ = U ∗ (AA∗ )U = U ∗ (A∗ A)U = (U ∗ A∗ U )(U ∗ AU ) = T ∗ T.
Hence T is a diagonal matrix.
Conversely, if U ∗ AU = D, a diagonal matrix, then A = U DU ∗ . Since any two
diagonal matrices commute, DD∗ = D∗ D. Thus AA∗ = (U DU ∗ )(U D∗ U ∗ ) =
U (DD∗ )U ∗ = U (D∗ D)U = A∗ A, and A is normal.

P ROPOSITION 5.9. Let A be a normal matrix. Then:


(i) kAu|| = kA∗ uk for every u ∈ Cn .
(ii) If Au = λu, then A∗ u = λu, where λ ∈ C.
(iii) Eigenvectors corresponding to distinct eigenvalues are orthogonal.

Proof. (i) kAuk2 = (Au)∗ (Au) = u∗ (A∗ A)u = u∗ (AA∗ )u = (A∗ u)∗ (A∗ u) =
kA∗ uk2 . Hence kAu|| = kA∗ uk.
(ii) A − λIn is also normal. Thus using (i): 0 = k(A − λIn )uk = k(A − λIn )∗ uk =
k(A∗ − λIn )uk. Hence A∗ u = λu.
(iii) Let Au = λu and Av = µv, λ 6= µ. Then λv ∗ u = v ∗ (Au) = (A∗ v)∗ u =
(µv)∗ u = µv ∗ u. Hence v ∗ u = 0

15
6 BILINEAR FORMS

 
2 1 1
E XAMPLE 5.10. The matrix 1 2 1 is Hermitian, and so unitary diagonalizable.
1 1 2
The
  eigenvalues
  of this matrix are 1 and 4. Eigenvectorscorresponding
 to to 1 are
1 1 1
−1 and  0  and eigenvector corresponding to 4 is 1. Now by Proposition
0 −1 1
5.9, we need to find an orthonormal
 1 eigenvectors corresponding to 1. These can be
√1 √1

    √
1 1 2 6 3
−1 1 1 
−1 and  1 . Hence U =   √2 √6 √3  .
0 −2 −2 √1
0 √ 6 3

6 B ILINEAR FORMS

D EFINITION 6.1. A bilinear form on V is a mapping β : f × V → K such that


(i) f (αx + βy, z) = αf (x, z) + βf (y, z) for all α, β ∈ K and x, y, z ∈ V.
(ii)f (x, αy + βz) = αf (x, y) + βf (x, z) for all α, β ∈ K and x, y, z ∈ V.

R EMARK 6.2. If f : V × V → K is a bilinear form then:


(i) f (x, 0) = 0 for all x ∈ V and f (0, y) = 0 for all y ∈ V.
(ii) f (αx, y) = αf (x, y) = f (x, αy) for α ∈ K and x, y ∈ V .

E XAMPLE 6.3. 1. Let A ∈ K n×n . Then f : K n ×K n → K given by f (x, y) =


y t Ax is a bilinear form.

2. An inner product on Rn is a bilinear form.

V × V → K. Let BP= {u1 , . . . , un } be a basis of


Consider a bilinear form f :P
V. For x, y ∈ V, we have: x = ni=1 αi ui and y = ni=1 γi ui and thus
n
X
f (x, y) = αi γj f (ui , uj ) = [x]tB A[y]B ,
i,j=1

where A is an n × n matrix whose (i, j)-th entry is f (ui , uj ). The matrix A is called
the matrix of bilinear form f with respect to an ordered basis B, denoted by [f ]B .

E XERCISE 6.4. Show that if f, g : V × V → K are bilinear forms and B is a fixed


basis of V, then [f + g]B = [f ]B + [g]B and [αf ]B = α[f ]B for all α ∈ K.

16
6 BILINEAR FORMS

E XERCISE 6.5. Let B = {u1 , . . . , un } be a basis of V, and φ : B × B → K


be
Pnany map. Then for Pany x, y ∈ V we have unique representations: x =
n
i=1 αi ui and y = i=1 βi ui . Verify that if

n
X
f (x, y) = αi βj φ(ui , uj ),
i,j=1

then f is a bilinear from on V.

E XERCISE 6.6. Let T : V → V ∗ be a linear transformation5 . Define f : V × V →


K by f (x, y) = T (y)(x). Show that f is a bilinear from.

P ROPOSITION 6.7. Let B = {u1 , . . . , un } and C = {v1 , . . . , vn } be bases


of V. Let f : V × V → K is a bilinear form on V. Then there is an invertible
matrix P such that
[f ]B = P t [f ]C P.

Proof. Let B = {u1 , . . . , un } and C = {v1 , . . . , vn } be bases of V. Let φ is a


bilinear form on V, and B and C be the matrices of φ with respect to B and C,
respectively. There is an invertible matrix P such that [u]B = P [u]C . Thus for any
x, y ∈ V :

φ(x, y) = [x]tB B[y]B = [x]tC P t BP [y]C = [x]tC C[y]C .

This gives the change of base formula:

[φ]B = P t [φ]C P,

where P is invertible, for given any two ordered basis B and C.

D EFINITION 6.8. A bilinear form f : V ×V → K is called symmetric if f (x, y) =


f (y, x), for all x, y ∈ V, and skew-symmetric if f (x, y) = −f (y, x), for all
x, y ∈ V.

E XERCISE 6.9. Let V be a vector space over K and B be a fixed basis. f : V ×V →


K is a symmetric (skew-symmetric) bilinear form if and only if [f ]B is a symmetric
(skew-symmetric) matrix.

P ROPOSITION 6.10. Let K is a field of characteristic not equal to 2. Then every


bilinear form on V over K is the sum of symmetric and skew-symmetric bilinear
forms.
5
Recall that V ∗ is the space of all linear forms on V.

17
7 SYMMETRIC BILINEAR FORMS

Proof. Define g(x, y) = 21 (f (x, y) + f (y, x)) and h(x, y) = 12 (f (x, y) − f (y, x)).
Then verify that g is symmetric bilinear for on V and h is skew-symmetric bilinear
for on V such that f = g + h.

E XERCISE 6.11. Let K is a field of characteristic not equal to 2. Show that if


f : V × V → K is a skew-symmetric bilinear form then f (x, x) = 0 for all x ∈ V.

7 S YMMETRIC BILINEAR FORMS

In this section we study bilinear forms over R. Let V be an n-dimensional vector


space over R. f : V × V → R be a symmetric bilinear form. Let [f ] be the
matrix of f with respect to the standard basis. Thus f (x, y) = [x]t [f ][y]. Since f
is symmetric bilinear form, [f ] is a symmetric matrix. Since the eigenvalues of [f ]
are all real numbers. Let there be p be the number of positive eigenvalues, q be the
number of negative eigenvalues of [f ] and let r be the rank of the matrix [f ]. Write
s = n − r. there is a orthogonal matrix U such that

U t [f ]U = diag(λ1 , . . . , λp , −λp+1 , . . . , −λp+q , 0 . . . , 0), λi > 0.



Let Q = diag( λ1 , . . . , λp+q , 0 . . . , 0). Then if P = U Q−1 , we have
p

P t [f ]P = diag(Ip , −Iq , 0).

The triplet (p, q, s) is called the index of f , the number p + q is the rank of f ,
and the number p − q is called the signature of f.

Let V be a finite dimension vector space over R.


(i). Every symmetric bilinear from f on V can be represented by a block
diagonal matrix: diag(Ip , −Iq , 0s ), where p, q, s are uniquely determined
by f .
(ii). Two symmetric bilinear forms are equivalent if and only if they have the
same index.
(iii). Two symmetric bilinear forms are equivalent if and only if they have
the same rank and the same signature.

Let A be a symmetric real matrix. Then q : Rn → R given by q(x) = xA x is


called the quadratic form corresponding to A.
A ∈ Rn×n and symmetric is positive semi-definite if xt Ax ≥ 0 for all x ∈
n
R .
The following conditions are equivalent for a symmetric matrix: (i) A is
positive semi-definite.
(ii) All eigenvalues of A are non-negative.
(iii) All principal submatrices have non-negative determinants’
(iv) There is a matrix B such that A = B t B.

18
7 SYMMETRIC BILINEAR FORMS

 
1 1 2
E XAMPLE 7.1. Let A = 1 2 1 . We find a matrix Q so that Qt AQ is diagonal.
2 1 3
The method is we start with A and I3 then the column operation that we apply on
A the same row operation is to be applied on A and the corresponding column
operation is applied on the identity matrix. Finally this way when A transformed to
the diagonal form, then the identity matrix transforms to Q.
Write columns of A and I3 .

1 1 2 1 0 0
1 2 1 0 1 0
2 1 3 0 0 1

−1 times the first column added to the second column, do the same on rows

1 0 2 1 −1 0
0 1 −1 0 1 0
2 1 3 0 0 1

−2 times the first column added to the third column, do the same on rows

1 0 0 1 −1 −2
0 1 −1 0 1 0
0 −1 −1 0 0 1

Add second column to the third column, do the same for rows:

1 0 0 1 −1 −3
0 1 0 0 1 1
0 0 −2 0 0 1
   
1 −1 −3 1 0 0
Hence Q =  0 1 1  and Qt AQ =  0 1 0 .
0 0 1 0 0 −2

19

Das könnte Ihnen auch gefallen