Bilinear and Quadratic form

Bilinear form
Definitions:
Linear functional: Let 𝑉 be a vector space over a filed 𝐹. the linear transformation 𝑓: 𝑉 → 𝐹
defined by 𝑓(𝛼𝑥 + 𝛽𝑦) = 𝛼𝑓(𝑥) + 𝛽𝑓(𝑦), ∀𝑥, 𝑦 ∈ 𝑉, 𝛼, 𝛽 ∈ 𝐹 is called linear functional.

Example: Let 𝑉(𝐹) = {(𝑎𝑖𝑗 )𝑛×𝑛 |𝑎𝑖𝑗 ∈ 𝐹} be a vector space and 𝑓: 𝑉 → 𝐹 is defined by
𝑓(𝐴) = 𝑇𝑟𝑎𝑐𝑒(𝐴), ∀𝐴 ∈ 𝑉. Then 𝑓 is linear functional.
Note: Set of all linear functional on vector space 𝑉(𝐹) is denoted by ℒ(𝑉, 𝐹).
Dual space: The collection of all linear functionals over finite dimensional vector space
𝑉(𝐹) i.e., ℒ(𝑉, 𝐹) form a vector space over the same field 𝐹. This vector space is called dual
space of 𝑉 and is denoted by 𝑉 ∗ .
Note: dim(𝑉 ∗ ) = dim 𝑉.
Important results:
1. Let {𝑣1 , 𝑣2 , … , 𝑣𝑛 } be a basis of 𝑉(𝐹) and define 𝑓𝑖 : 𝑉 → 𝐹 such that 𝑓𝑖 (𝛼1 𝑣1 +
𝛼2 𝑣2 + ⋯ + 𝛼𝑛 𝑣𝑛 ) = 𝛼𝑖 , 𝛼𝑖 ∈ 𝐹, 𝑖 = 1, 2, … , 𝑛 then 𝑓𝑖 is a linear transformation for
all 𝑖 = 1, 2, … , 𝑛 and {𝑓1 , 𝑓2 , … , 𝑓𝑛 } is a basis of 𝑉 ∗ and dim 𝑉 = dim 𝑉 ∗ .
2. Dual basis: Let 𝐵 = {𝑣1 , 𝑣2 , … , 𝑣𝑛 } be a basis of 𝑉(𝐹), 𝑉 ∗ = {𝑓|𝑓 is a linear
1, 𝑖𝑓 𝑖 = 𝑗
functional on 𝑉} and 𝑓𝑖 : 𝑉 → 𝐹 is defined by 𝑓𝑖 (𝑣𝑖 ) = { then 𝑆 =
0, 𝑖𝑓 𝑖 ≠ 𝑗
{𝑓1 , 𝑓2 , … , 𝑓𝑛 } is linearly independent and forms a basis for 𝑉 ∗ and is called dual basis
of 𝑉. 𝐵 is called primal for 𝑆.
Annihilator: Let 𝑊 be a subspace of 𝑉(𝐹). then annihilator of 𝑊, denoted by 𝐴(𝑊) or 𝑊 ⊥
is the set of all linear functionals 𝑓 of 𝑉 such that 𝑓(𝑥) = 0, ∀ 𝑥 ∈ 𝑊.
i.e., 𝐴(𝑊) = 𝑊 ⊥ = {𝑓 ∈ 𝑉 ∗ |𝑓(𝑥) = 0, ∀𝑥 ∈ 𝑊}
Note:
1. 𝑊 ⊥ is a subspace of 𝑉 ∗ .
2. If 𝑊 = {0} then 𝑊 ⊥ = 𝑉 ∗
3. If 𝑊 = 𝑉 then 𝑊 ⊥ = 𝐴(𝑉) = {0} where 0 is the zero functional of 𝑉(𝐹).
Dual map: Let 𝑇: 𝑉 → 𝑊 be liner. For 𝑇 ∈ ℒ(𝑉, 𝑊), the dual map of 𝑇 is a map 𝑇 ∗ : 𝑊 ∗ →
𝑉 ∗ defined by 𝑇 ∗ 𝑔 = 𝑔𝑇, ∀𝑔 ∈ 𝑊 ∗ .
Proposition:
1. Let {𝑣1 , 𝑣2 , … , 𝑣𝑛 } be a basis of a finite dimensional vector space 𝑉 and {𝑓1 , 𝑓2 , … , 𝑓𝑛 }
be the corresponding dual basis of 𝑉 ∗ . For 𝑇 ∈ ℒ(𝑉), the matrix 𝑇 ∗ ∈ ℒ(𝑉 ∗ ) with
respect to {𝑓1 , 𝑓2 , … , 𝑓𝑛 } is the transpose of the matrix 𝑇 with respect to {𝑣1 , 𝑣2 , . . , 𝑣𝑛 }
i.e., 𝑀(𝑇 ∗ ) = 𝑀(𝑇)𝑡 .
𝑉 ∗
2. There exists a natural isomorphism between 𝑊 ⊥ and (𝑊) . So, we can identify linear
𝑉
functional on 𝑊 with elements of 𝑊 ⊥ .
Definition: Let 𝑉 be a vector space over a field 𝐹. A bilinear form on 𝑉 is a function
𝐵: 𝑉 × 𝑉 → 𝐹 that is linear in each variable when the other one is fixed. i.e.,
𝐵(𝑣 + 𝑣 ′ , 𝑤) = 𝐵(𝑣, 𝑤) + 𝐵(𝑣 ′ , 𝑤); 𝐵(𝛼𝑣, 𝑤) = 𝛼𝐵(𝑣, 𝑤), ∀𝑣, 𝑣 ′ , 𝑤 ∈ 𝑉 and 𝛼 ∈ 𝐹
And 𝐵(𝑣, 𝑤 + 𝑤 ′ ) = 𝐵(𝑣, 𝑤) + 𝐵(𝑣, 𝑤 ′ ); 𝐵(𝑣, 𝛼𝑤) = 𝛼𝐵(𝑣, 𝑤), ∀𝑣, 𝑤, 𝑤 ′ ∈ 𝑉 and 𝛼 ∈ 𝐹.
Note:
1. 𝐵 is symmetric if 𝐵(𝑣, 𝑤) = 𝐵(𝑤, 𝑣), ∀𝑣, 𝑤 ∈ 𝑉.
2. 𝐵 is skew-symmetric if 𝐵(𝑣, 𝑤) = −𝐵(𝑤, 𝑣), ∀ 𝑣, 𝑤 ∈ 𝑉.
3. 𝐵 is alternating if 𝐵(𝑣, 𝑣) = 0, ∀𝑣 ∈ 𝑉.
4. Bilinear space is a vector space equipped with a specific choice of bilinear form.
Example: Let 𝑣 = (𝑣1 , 𝑣2 , … , 𝑣𝑛 ), 𝑤 = (𝑤1 , 𝑤2 , … , 𝑤𝑛 ) ∈ ℝ𝑛 then the dot product 𝑣. 𝑤 =
𝑣1 𝑤1 + 𝑣2 𝑤2 + ⋯ + 𝑣𝑛 𝑤𝑛 is a symmetric bilinear form.
Theorem: Let (𝑉, 𝐵) be a bilinear space and 𝑀 be the associated matrix in some basis of 𝑉.
Then
a. 𝐵 is symmetric iff 𝑀𝑇 = 𝑀.
b. 𝐵 is skew symmetric iff 𝑀𝑇 = −𝑀.
c. 𝐵 is alternating iff 𝑀𝑇 = −𝑀 and the diagonal entries of 𝑀 are zero.

Quadratic form
Congruent matrices: A matrix 𝐴 is said to be congruent to a matrix 𝐵 if there exist a non-
singular matrix 𝑃 such that 𝐴 = 𝑃𝑇 𝐵𝑃 and we write is as 𝐴 ≅ 𝐵.
Quadratic form: Let 𝑓: ℝ𝑛 → ℝ be a function. Then the form 𝑓(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = ∑ 𝑎𝑖𝑗 𝑥𝑖 𝑥𝑗
where 𝑎𝑖𝑗 ∈ 𝐹 is called quadratic form of variable 𝑥1 , 𝑥2 , … , 𝑥𝑛 over the filed ℝ.

Example: The function 𝑓(𝑥, 𝑦, 𝑧) = 𝑎𝑥 2 + 𝑏𝑦 2 + 𝑐𝑧 2 + 2ℎ𝑥𝑦 + 2𝑔𝑦𝑧 + 2𝑓𝑦𝑧 can be

written as
𝑎 ℎ 𝑔 𝑥
𝑓(𝑥, 𝑦, 𝑧) = [𝑥 𝑦 𝑧] [ ℎ 𝑏 𝑓 ] [𝑦 ]
𝑔 𝑓 𝑐 𝑧

Theorem: Let 𝑄 be a quadratic form in 𝑛 variable 𝑥1 , 𝑥2 , … , 𝑥𝑛 over a field 𝐹 whose

characteristic is not 2. Then by a suitable linear transformation to new variables 𝑦1 , 𝑦2 , … , 𝑦𝑛
we can get 𝑄 = 𝑐1 𝑦12 + 𝑐2 𝑦22 + ⋯ + 𝑐𝑛 𝑦𝑛2 for 𝑐1 , 𝑐2 , … , 𝑐𝑛 ∈ 𝐹.
Value class of quadratic form:
Let 𝑥 = (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) ∈ ℝ𝑛 . Then a quadratic form 𝑓: ℝ𝑛 → ℝ is said to be
1. Positive definite if 𝑓(𝑥) > 0, ∀𝑥 ≠ 0.
Example: 𝑓: ℝ3 → ℝ defined by 𝑓(𝑥) = 𝑥12 + 𝑥22 + 𝑥32 .
2. Positive semi-definite if 𝑓(𝑥) ≥ 0 i.e., ∃𝑥 ≠ 0 such that 𝑓(𝑥) = 0.
Example: 𝑓: ℝ2 → ℝ defined by 𝑓(𝑥) = (𝑥1 − 𝑥2 )2.
3. Negative definite if 𝑓(𝑥) < 0, ∀𝑥 ≠ 0
Example: 𝑓: ℝ3 → ℝ defined by 𝑓(𝑥) = −𝑥12 − 𝑥22 − 𝑥32 .
 4. Negative semi-definite if 𝑓(𝑥) ≤ 0 i.e., ∃𝑥 ≠ 0 such that 𝑓(𝑥) = 0.
Example: 𝑓: ℝ3 → ℝ defined by 𝑓(𝑥) = −(𝑥1 − 𝑥2 )2 − 𝑥32 .
5. Indefinite if ∃𝑥, 𝑦 ∈ ℝ𝑛 such that 𝑓(𝑥)𝑓(𝑦) < 0
Example: 𝑓: ℝ2 → ℝ defined by 𝑓(𝑥) = 𝑥12 − 𝑥22 .
Value class of quadratic form when it is in matrix form:

Let 𝑓(𝑥) = 𝑥 𝑇 𝐴𝑥 where 𝑥 = (𝑥1 , 𝑥2 , … , 𝑥𝑛 )𝑇 , 𝐴 = [𝑎𝑖𝑗 ] is a quadratic form. Then

1. Quadratic form is positive definite iff all the principal minors are positive.
2. Quadratic form is negative definite iff the principal minors change their sign
alternatively starting with negative.
3. A 2 × 2 matrix is positive definite iff Trace and determinant of 𝐴 both are positive.
4. If Trace of a matrix is zero then the matrix is indefinite
Proposition: Let 𝑓(𝑥) = 𝑥 𝑇 𝐴𝑥 be a quadratic form. Let 𝑃 be a non-singular transformation
such that 𝑥 = 𝑃𝑦, 𝑥, 𝑦 ∈ ℝ𝑛 then range of 𝑓(𝑥) and 𝑓(𝑦) are same. Also 𝐴 and 𝐵 = 𝑃𝑇 𝐴𝑃
have the same value class.
Method to find value class when eigenvalues of the matrix are known:

Let 𝑓(𝑥) = 𝑥 𝑇 𝐴𝑥 where 𝑥 = (𝑥1 , 𝑥2 , … , 𝑥𝑛 )𝑇 , 𝐴 = [𝑎𝑖𝑗 ]𝑛×𝑛 is a quadratic form. Let 𝜆𝑖 be

eigenvalues of 𝐴.
1. 𝐴 is positive definite if 𝜆𝑖 > 0, ∀𝑖 = 1, 2, … , 𝑛.
2. 𝐴 is positive semi-definite if 𝜆𝑖 ≥ 0 and |𝐴| = 0 i.e., at least one eigenvalue is zero.
3. 𝐴 is negative definite if 𝜆𝑖 < 0, ∀𝑖 = 1, 2, … , 𝑛.
4. 𝐴 is negative semi-definite if 𝜆𝑖 ≤ 0 and |𝐴| = 0 i.e., at least one eigenvalue is zero.
5. 𝐴 is indefinite if ∃𝜆𝑖 , 𝜆𝑗 such that 𝜆𝑖 𝜆𝑗 < 0.

Conversion into quadratic form: Let 𝐴 = (𝑎𝑖𝑗 ) be a symmetric matrix and 𝐵 = 𝑃𝑇 𝐴𝑃.
1, 𝑖 = 𝑗
Then 𝐵 is canonical form of 𝐴 if 𝑎𝑖𝑗 = 𝑘𝛿𝑖𝑗 where 𝑘 = −1 or 0 or 1 and 𝛿𝑖𝑗 = { .
0, 𝑖 ≠ 𝑗
6 −2 2
Question: Convert the quadratic form of a matrix 𝐴 = [−2 3 −1]
2 −1 3
Solution:
𝐴 = 𝐼𝐴𝐼
6 −2 2 1 0 0 1 0 0
[−2 3 −1] = [0 1 0] 𝐴 [ 0 1 0]
2 −1 3 0 0 1 0 0 1
𝑅2 → 3𝑅2 + 𝑅1 ; 𝐶2 → 3𝐶2 + 𝐶1
6 0 2 1 0 0 1 0 0
[0 21 −1] = [0 3 0 ] 𝐴 [0 3 0]
2 −1 3 0 0 1 0 0 1
𝑅3 → 3𝑅3 − 𝑅1 ; 𝐶3 → 3𝐶3 − 𝐶1
 6 0 0 1 0 0 1 1 −1
[0 21 −3 ] = [ 1 3 0] 𝐴 [0 3 0]
0 −3 −33 −1 0 3 0 0 3
𝑅3 → 7𝑅3 + 𝑅2 ; 𝐶3 → 7𝐶3 + 𝐶2
6 0 0 1 0 0 1 1 −1
[0 21 0 ]=[ 1 3 0 ] 𝐴 [0 3 3]
0 0 −1638 −6 3 21 0 0 21
𝑅1 𝑅2 𝑅3 𝐶1 𝐶2 𝐶3
𝑅1 → , 𝑅2 → , 𝑅3 → ; 𝐶1 → , 𝐶2 → , 𝐶3 →
√6 √21 √1638 √6 √21 √1638
1 1 1 6
0 0 −
1 0 0 √6 √6 √21 √1638
1 3 3 3
[0 1 0 ]= 0 𝐴 0
√21 √21 √21 √1638
0 0 −1 6 3 21 21
[− √1638 √1638 √1638] [0 0
√1638 ]
Which is of the form 𝐵 = 𝑃𝑇 𝐴𝑃
1 0 0 𝑦1
So 𝑓(𝑦) = 𝑦 𝐵𝑦 = [𝑦1
𝑇 𝑦2 𝑦3 ] [0 1 0 ] [𝑦2 ] = 𝑦12 + 𝑦22 − 𝑦32
0 0 −1 𝑦2
Properties:
Rank (𝒓): The rank of a quadratic form is equal to the number of non-zero eigenvalues of the
matrix of quadratic form.
Index (𝒑): The index of a quadratic form is equal to the number of positive eigenvalues of
the matrix of quadratic form.
Note: Number of negative eigenvalues = total number of positive eigenvalues – number of
positive eigenvalues = 𝑟 − 𝑝
Signature (𝒔): The signature of a quadratic form is equal to the difference between the
number of positive eigenvalues and the number of negative eigenvalues of the quadratic
form. So
𝑠 = 𝑝 − (𝑟 − 𝑝) = 2𝑝 − 𝑟
Example: Find the rank, index and signature of the quadratic form 𝑓(𝑥, 𝑦, 𝑧) = 3𝑥 2 + 5𝑦 2 +
3𝑧 2 − 2𝑥𝑦 − 2𝑦𝑧 + 2𝑥𝑧
Solution:
𝑓(𝑥, 𝑦, 𝑧) = 3𝑥 2 + 5𝑦 2 + 3𝑧 2 − 2𝑥𝑦 − 2𝑦𝑧 + 2𝑥𝑧
3 −1 1 𝑥
= [𝑥 𝑦 𝑧] [−1 5 −1] [𝑦]
1 −1 3 𝑧
3 −1 1
Eigenvalues of the matrix [−1 5 −1] is given by
1 −1 3
 3−𝜆 −1 1
| −1 5−𝜆 −1 | = 0
1 −1 3−𝜆
⇒ (3 − 𝜆)(15 − 3𝜆 − 5𝜆 + 𝜆2 − 1) − 1(−1 + 3 − 𝜆) + 1(1 − 5 + 𝜆) = 0
⇒ (3 − 𝜆)(𝜆2 − 8𝜆 + 14) + 𝜆 − 2 + 𝜆 − 4 = 0
⇒ 3𝜆2 − 𝜆3 − 24𝜆 + 8𝜆2 + 42 − 14𝜆 + 2𝜆 − 6 = 0
⇒ 𝜆3 − 11𝜆2 + 36𝜆 − 36 = 0
⇒ (𝜆 − 6)(𝜆 − 3)(𝜆 − 2) = 0
So, eigenvalues are 2, 3, 6.
Therefore rank (𝑟) = 3, index (𝑝) = 3 and signature (𝑠) = 2𝑝 − 𝑟 = 6 − 3 = 3.
Value of class of a quadratic form by rank and signature:
A quadratic form 𝑓: ℝ𝑛 → ℝ is
1. Positive definite iff 𝑠 = 𝑟 = 𝑛
2. Positive semi-definite iff 𝑠 = 𝑟 < 𝑛
3. Negative definite iff −𝑠 = 𝑟 = 𝑛
4. Negative semi-definite iff −𝑠 = 𝑟 < 𝑛
5. Indefinite iff |𝑠| ≠ 𝑟.