VI.

QUADRATIC FORMS

Martina Bečvářová

Winter semester 2024/2025

Faculty of Transportation Sciences, CTU in Prague

Definition
Let V be a vector space over R, M = {v1 , v2 , . . . , vn } be its basis, A be a symmetric matrix of size
n × n. By a quadratic form on V we mean a real function on V (F : V → R) defined as
t
F (x) = hxiM · A · hxiM , x ∈ V.

Note

1. The symbol hxiM means the coordinates of x with respect to the basis M , so that hxiM =
(x1 , x2 , . . . , xn ) for x = x1 v1 + x2 v2 + · · · + xn vn .
2. The quadratic form F can be written as

a11 a12 a13 . . . a1n

 
x1
 
 a12 a22 a23 . . . a2n 
 x 
F (x) = ( x1 x2 ... xn ) ·  a13 a23 a33 . . . a3n  ·  2 

...
.........................
 
xn
a1n a2n a3n . . . ann

or

n
X
F (x) = aij xi xj = a11 x21 + a22 x22 + · · · + ann x2n + 2a12 x1 x2 + 2a13 x1 x3 + · · · + 2an−1,n xn−1 xn .
i,j=1

Theorem
If V is a vector space over R and F and G are two quadratic forms on V , then
a) F + G is also a quadratic form on V ,
b) αF is also a quadratic form on V for every α ∈ R.

1
2

Examples
1. Consider F (x) = 2x2 + 3y 2 + 2xy − 4xz + 6yz. Decide whether F is a quadratic form on
3
R (with respect to the standard basis). If yes, write its matrix.

Yes, it is a quadratic form (see the analytic expression in the definition) and its matrix is
 
2 1 −2
A= 1 3 3.
−2 3 0

2. Consider F (x) = x2 + y 2 + x + y − xz. Decide whether F is a quadratic form on R3 (with

respect to the standard basis). If yes, write its matrix.

No, it is not a quadratic form (see the analytic expression in the definition). The „problematic
partÿ is x + y.

Examples
1. Consider  
1 2 1
.
1 0 1
Decide whether the matrix A represents a quadratic form on R3 (with respect to the standard
basis).

No, the matrix is not a square matrix.

2. Consider  
1 0 1
2 1 1.
1 0 3
Decide whether the matrix A represents a quadratic form on R3 (with respect to the standard
basis).

No, the matrix is not a symmetric matrix.

3. Consider  
1 0 −1
 0 1 2.
−1 2 −3
Decide whether the matrix A represents a quadratic form on R3 (with respect to the standard
basis). If yes, write its analytic expression.
The matrix is a symmetric matrix. Hence
   
1 0 −1 x
F (x, y, z) = ( x y z ) ·  0 1 2  ·  y  = x2 + y 2 − 3z 2 − 2xz + 4yz
−1 2 −3 z

is a quadratic form.
 3

Definition
By the rank of a quadratic form we mean the rank of its matrix.

Note

1. If A is a regular matrix, then F (x) = 0, if and only if x = o.

2. If A is a non-regular matrix, then there exists x ∈ V , x 6= o, such that F (x) = 0.

Definition
The basis N = {u1 , u2 , . . . , un } of a vector space V is called a polar basis of a quadratic form F ,
if its analytic expression is

a11 0 0 . . . 0 x1
   
 0 a22 0 . . . 0   x2  2 2 2
F (x) = ( x1 x2 ... xn ) ·  ·  = a11 x1 + a22 x2 + · · · + ann xn .
...................... ...
0 0 0 . . . ann xn

Note

1. In this case, the matrix A is diagonal.

2. The form F (x) = a11 x21 + a22 x22 + · · · + ann x2n is called a polar expression of quadratic form F .

Theorem
Let V be a vector space of dimension n over R and let F : V → R be a quadratic form on V .
Then there exists a polar basis {v1 , v2 , . . . , vn } of V of the quadratic form F that is, there exist
b11 , . . . , bnn ∈ R such that

F (x) = b11 x21 + b22 x22 + · · · + bnn x2n ,

Pn
whenever x = i=1 xi vi .

Definition
Let F be a quadratic form on a vector space V . Then F

a) is called positive definite, if F (x) > 0 for all non-zero x,

b) is called positive semidefinite, if F (x) ≥ 0 for all x,
c) is called negative definite, if F (x) < 0 for all non-zero x,
d) is called negative semidefinite, if F (x) ≤ 0 for all x,
e) is called indefinite, if F (x) > 0 for some x and F (y) < 0 for some y.
4

Theorem
Let F (x) = a11 x21 + a22 x22 + · · · + ann x2n be a polar expression of a quadratic form on a vector
space V . Then F is

a) positive definite, if and only if aii > 0 for all i = 1, . . . , n,

b) positive semidefinite, if and only if aii ≥ 0 for all i = 1, . . . , n,
c) negative definite, if and only if aii < 0 for all i = 1, . . . , n,
d) negative semidefinite, if and only if aii ≤ 0 for all i = 1, . . . , n,
e) indefinite, if and only if aii > 0 for some i = 1, . . . , n and ajj < 0 for some j = 1, . . . , n.

Definition
By a signature of a quadratic form we mean the 3-tuple of numbers which expresses the
number p of positive entries aii , the number q of negative entries aii and the number r of
zero entries aii in the polar expression of the quadratic form F . We write the signature as
sgn F = (p, q, r).

Note
Let F (x) = a11 x21 + a22 x22 + · · · + ann x2n be a polar expression of a quadratic form on a vector
space V . Then F is

a) positive definite, if and only if its signature is sgn F = (n, 0, 0),

b) positive semidefinite, if and only if its signature is sgn F = (k, 0, m), where k, m ∈ N and k+m = n,
c) negative definite, if and only if its signature is sgn F = (0, n, 0),
d) negative semidefinite, if and only if its signature is sgn F = (0, l, m), where l, m ∈ N and l+m = n,
e) indefinite, if and only, if its signature is sgn F = (k, l, m), where k, l, m ∈ N and k + l + m = n.

Theorem (the so called Sylvester’s criterium)

The signature of a quadratic form is independent on the choice of a polar basis.

Methods how to find a polar expression of a quadratic form

Our objective now is to obtain a polar form for real symmentric matrices.

∗ ∗ ∗ ∗

I. Process of squares completing

Example
Consider the quadratic form F : R3 → R given by

F (x, y, z) = x2 − 2xy + 4yz − 2y 2 + z 2 .

Find the signature of the quadratic form.

 5

By process of „completing the squaresÿ it is readily seen that

F (x, y, z) = x2 − 2xy + 4yz − 2y 2 + z 2 = (x − y)2 − y 2 + (z + 2y)2 − 4y 2 − 2y 2 =

= (x − y)2 − 7y 2 + (z + 2y)2 .
The expression in a polar form is

F (x′ , y ′ , z ′ ) = (x′ )2 − 7(y ′ )2 + (z ′ )2 .

The polar form is of rank 3 and its singnature is (2, 1, 0), hence a quadratic form is indefinite.

Alternatively, we can work with matrices. The matrix of F is

 
1 −1 0
 −1 −2 2
0 2 4

and the Jordan canonical matrix is

 
1 0 0
0 −7 0  .
0 0 1

∗ ∗ ∗ ∗

II. Symmetric elemetary matrix operations

A short theory

Definition
A matrix A is called orthogonal if A−1 exists and is equal to At . If there exists an orthogonal
matrix U such that B = U t AU = U −1 AU , then we say that B is orthogonally similar to A.

Theorem
If A is a square matrix over R, then A is orthogonally similar to a diagonal matrix if and only
if A is symmetric.

Example
Consider the quadratic form F on R3 given by

F (x, y, z) = x2 + 2xy + 2xz + 2y 2 + 2z 2 + 2yz

expressed with respect to the canonical basis. Find a polar basis, polar expression and signature
of the quadratic form.
6

In order to a diagonal matrix, we will introduce a special matrix. In its first part, there is
a matrix of our quadratic form; in each row of the second part there are vectors from a basis of R3 .
In our example, we may chose the canonical (standard) basis, therefore {(1, 0, 0), (0, 1, 0), (0, 0, 1}.
We will transform our matrix by symmetric elementary operations, thus we will make the same
elementary operations to rows as well as to columns in the first part of our „special matrixÿ.

The matrix of F is  
1 1 1
1 2 1.
1 1 2
By process of „symmetric elemetary operationsÿ it is readily seen that

   
1 1 1 | 1 0 0 1 1 1 | 1 0 0
1 2 1 | 0 1 0  ∼  0 −1 0 | 1 −1 0 ∼
1 1 2 | 0 0 1 0 0 −1 | 1 0 −1
 
1 0 0 | 1 0 0
∼  0 1 0 | 1 −1 0 .
0 0 1 | 1 0 −1

In the rows of the second part of the last matrix, we can see the polar basis B = {(1, 0, 0),
(1, −1, 0), (1, 0, −1)}. In the first part of the last matrix, we see the diagonal matrix which is the
Jordan canonical matrix of the original matrix of the quadratic form considered, so that
 
1 0 0
D = 0 1 0,
0 0 1

hence the new analytic expression of our quadratic form with respect to B is

F (x′ , y ′ , z ′ ) = (x′ )2 + (y ′ )2 + (z ′ )2

and its signature is (3, 0, 0). Our quadratic form is positive definite.

Note
The above described method is a suitable tool in linear algebra because it gives us a new
analytic expression, a signature as well as a polar basis.

∗ ∗ ∗ ∗
III. Method of general upper minors

A short theory

We will use what we learnt in the chapter Determinants.

 7

Theorem
Let A be the matrix of a quadratic form and let us denote by det A1 , det A2 , . . . , det An = det A
the sequence of general upper determinants of a matrix A. Then

1. if all entries of the sequence of general upper determinants are positive, then the quadratic form
is positive definite.
2. if the sings of entries of the sequence of general upper determinants alternate, which means
−, +, −, +, . . . , then the quadratic form is negative definite.

Examples
1. Consider the quadratic form F : R3 → R given by the matrix
 
1 1 1
1 2 1.
1 1 2

Determine the signature of F .

We calculate the general upper determinants. We obtain the following sequence:

1 1 1
1 1
det A1 = |1| = 1, det A2 = = 1, det A3 = 1 2 2 = 1.
1 2
1 1 2

As we see, all entries of our sequence are positive, therefore our quadratic form is positive
definite and its signature is (3, 0, 0).

2. Consider the quadratic form F : R3 → R given by the matrix

 
−2 1 0
 1 −1 0.
0 0 −3

Determine the signature of F .

We calculate the general upper determinants. We obtain the following sequence:

−2 1 0
−2 1
det A1 = | − 2| = −2, det A2 = = 1, det A3 = 1 −1 0 = −3.
1 −1
0 0 −3

As we see, the signs of the sequence of general upper determinants alternate (they are −, +, −),
hence the quadratic form is negative definite and its signature is (0, 3, 0).
8

3. Consider the quadratic form F : R3 → R given by the matrix

 
1 0 1
0 1 2.
1 2 0

Determine the signature of F .

We calculate the general upper determinants. We obtain the following sequence

1 0 1
1 0
det A1 = |1| = 1, det A2 = = 1, det A3 = 0 1 2 = −5.
0 1
1 2 0

As we see, the quadratic form is indefinite and its signature cannot be determined by using
the method of general upper minors.

Note
The method described above is useful in mathematical analysis (determination of relative
maxima and minima of functions f : Rn → R; if the form is positive definite at x, then f has
a relative minimum at x, and if the form is negative definite at x, then f has a relative maximum
at x). We will study these applications in the lecture Caculus I.

∗ ∗ ∗ ∗

IV. Method of eigenvalues and eigenvectors

Theorem
Each real symmetric matrix is diagonalizable. All eigenvalues of a symmetric matrix are real.
If u is an eigenvector associated with the eigenvalue λi and v is an eigenvector associated with
the eigenvalue λj and λi 6= λj , then the vectors u and v are orthogonal, that is, u · v = 0.

Note
It should observed that the polar basis contains eigenvectors, so that it is an orthogonal basis.
Eigenvalues are on the diagonal of the Jordan canonical form.

Example
1. Consider the quadratic form F : R3 → R given by the matrix
 
0 2 3
A = 2 0 0.
3 0 0

Find its polar basis and signature.

 9

We calculate eigenvalues and eigenvectors:

 
−λ 2 3
det (A − λE) =  2 −λ 0  = −λ(λ2 − 13) = 0.
3 0 −λ

As we see, the eigenvalues are

√ √
λ1 = 0, λ2 = 13, λ3 = − 13.

Hence the Jordan canonical form of the matrix A is

 
0 √ 0 0
0 13 √ 0 ,

0 0 − 13

√ √
its polar expression is F (x′ , y ′ , z ′ ) = 0(x′ )2 + 13(y ′ )2 − 13(z ′ )2 and its signature is (1, 1, 1).
Now, we can calculate three independent and orthogonal eigenvectors.

     
0 2 3 x 0
(A − 0 · E) · v1t = ot =  2 0 0 · y  = 0 ⇒ v1 = (0, 3, −2),
3 0 0 z 0

 √     
√ − 13 √ 2 3 x 0 √
(A − 13 · E) · v2t = ot =  2 − 13 √ 0  · y  = 0 ⇒ v2 = ( 13, 2, 3),
3 0 − 13 z 0

√     
√ 13 √ 2 3 x 0 √
(A + 13 · E) · v3t = ot =  2 13 √ 0  · y  = 0 ⇒ v3 = (− 13, 2, 3).
3 0 13 z 0

√ √
The polar basis is {(0, 3, −2), ( 13, 2, 3), (− 13, 2, 3)}.

Exercises

1. Decide whether f is a quadratic form. If yes, write its matrix and calculate its signature.
a)
f (x1 , x2 , x3 ) = x1 2 + x2 2 + 2x3 2 + 4x1 x2 + 2x1 x3 + 2x2 x3 .

b)
f (x1 , x2 , x3 ) = x1 2 − 2x2 2 + x3 2 + 2x1 x2 + 4x1 x3 + 2x2 x3 .

c)
f (x1 , x2 , x3 ) = x1 2 − 3x2 2 − 2x1 x2 + 2x1 x3 − 6x2 x3 .
10

d)
f (x1 , x2 , x3 , x4 ) = x1 x2 + x1 x3 + x1 x4 + x2 x3 + x2 x4 + x3 x4 .

e)
f (x1 , x2 , x3 , x4 ) =

x1 2 + 2x2 2 + x4 2 + 4x1 x2 + 4x1 x3 + 2x1 x4 + 2x2 x3 + 2x2 x4 + 2x3 x4 .

f)
f (x1 , x2 , x3 ) = x1 2 + 5x2 2 − 4x3 2 + 2x1 x2 − 4x1 x3 .

g)
f (x1 , x2 , x3 ) = 4x1 2 + x2 2 + x3 2 − 4x1 x2 + 4x1 x3 − 3x2 x3 .

h)
f (x1 , x2 , x3 ) = 2x1 x2 + x1 x3 + x2 x3 .

i)
f (x1 , x2 , x3 ) = 2x1 2 + 18x2 2 + 8x3 2 − 12x1 x2 + 8x1 x3 − 27x2 x3 .

j)
f (x1 , x2 , x3 ) = −12x1 2 − 3x2 2 + −12x3 2 + 12x1 x2 + −24x1 x3 + 8x2 x3 .

2. Decide whether the matrices A and B represent quadratic forms . If yes, write their analytic
expressions and express their signatures.
a)  
4 2 1  
2 3
A = 2 0 1, B= .
1 2
1 1 2

b)
2 3 0
   
1 0 −1
3 2 0
A =  0 12 1 , B= .
2 0 2
−1 1 2
0 2 1

c)    
0 2 −1 2 3 0
A= 2 0 2 , B = 3 2 0.
−1 2 0 0 2 1

d)    
10 0 1 0 3 1
A= 2 1 2, B =  −3 0 2  .
1 2 1 −1 −2 0

e)    
10 1 2 3 2
A= , B= .
1 2 3 2 0
 11

f)    
0 0 1 0 0 1
A = 0 1 2, B = 3 0 2.
1 2 3 1 2 0

g)  
2 1 1  
0 0 1
A = 1 2 1, B= .
3 0 2
1 1 2

h)  
1 0 −1  
0 0
A= 0 0 1 , B= .
0 0
−1 1 2

i)  
2 0 −1  
1 2
A= 0 1 0 , B= .
0 1
−1 0 1

j)  
11 1 −1  
1 2 2
A =  1 8 1 , B= .
2 2 2
−1 1 1

3. Calculate signatures of the quadratic forms f and g.

a)
f (x1 , x2 , x3 , x4 ) = −4x1 x4 ,
g(x1 , x2 ) = x1 2 − 2x1 x2 + 4x2 2 .

b)
f (x1 , x2 ) = x1 2 + 26x2 2 + 10x1 x2 ,
g(x1 , x2 ) = x1 2 + 56x2 2 + 16x1 x2 .

c)
f (x1 , x2 , x3 ) = 8x1 2 − 28x2 2 + 14x3 2 + 16x1 x2 + 14x1 x3 + 32x2 x3 ,
g(x1 , x2 ) = x1 2 + 4x2 2 + 2x3 2 + 2x1 x2 .

d)
f (x1 , x2 , x3 ) = 2x1 2 + 3x2 2 − x3 2 + 2x1 x2 + 2x2 x3 ,
g(x1 , x2 , x3 ) = 3x1 2 + 2x2 2 + x3 2 − 2x1 x3 .

e)
f (x1 , x2 ) = x1 2 + 4x1 x2 − x2 2 ,
g(x1 , x2 ) = x1 2 + 6x1 x2 + 5x2 2 .
12

f)
f (x1 , x2 , x3 ) = 21x1 2 − 18x2 2 + 6x3 2 + 4x1 x2 + 28x1 x3 + 6x2 x3 ,
g(x1 , x2 , x3 ) = 11x1 2 + 6x2 2 + 6x3 2 − 12x1 x2 + 12x1 x3 − 6x2 x3 .

g)
f (x1 , x2 , x3 ) = 14x1 2 − 4x2 2 + 17x3 2 + 8x1 x2 − 40x1 x3 − 26x2 x3 ,
g(x1 , x2 , x3 ) = 9x1 2 + 6x2 2 + 6x3 2 + 12x1 x2 − 10x1 x3 − 2x2 x3 .

h)
f (x1 , x2 , x3 ) = 3x2 2 + 3x3 2 + 4x1 x2 + 4x1 x3 − 2x2 x3 ,
g(x1 , x2 , x3 ) = 7x1 2 + 7x2 2 + 7x3 2 + 2x1 x2 + 2x1 x3 + 2x2 x3 .

i)
f (x1 , x2 , x3 ) = x1 2 − 2x1 x2 − 2x1 x3 − 2x2 x3 ,
g(x1 , x2 , x3 ) = 3x1 2 + 3x2 2 − x3 2 − 6x1 x3 + 4x2 x3 .

j)
f (x1 , x2 , x3 ) = 11x1 2 + 5x2 2 + 2x3 2 + 16x1 x2 + 4x1 x3 − 20x2 x3 ,
g(x1 , x2 , x3 ) = x1 2 + x2 2 + 5x3 2 − 6x1 x2 − 2x1 x3 + 2x2 x3 .