Submissions for assignment 1

WEEK 1: “Rì viu” about Euclidean vector spaces

Quách Đăng Duy - K45 TTA

1 Bilinear maps
Definition 1.1. Let V be a real vector space, a map

η :V ×V →R

is called a bilinear map iff η(au + bv, w) = aη(u, w) + bη(v, w) for all a, b ∈ R, u, v, w ∈ V . (linearity with
each variable)
Definition 1.2. Let η : V × V → R be a bilinear map, then η is said to be
ˆ Symmetric if

η(u, v) = η(v, u), ∀u, v ∈ V.

ˆ Positive if

η(u, u) ≥ 0, ∀u ∈ V.

ˆ Definite if

η(u, u) = 0, ⇔ u = 0.

Definition 1.3. A bilinear map which symmetric, positive definite on V is called a Inner product on V ,
denoted h., .i.
Example 1.1. Usual dot product on Rn , (which we have learned on High school) is an inner product.
Example 1.2. Let V = C[a, b] (the space of real-valued continuous functions on [a, b]). We can check that
Z b
hf, gi = f (x)g(x) dx, f, g ∈ C[a, b]
a

define a inner product on C[a, b].

Definition 1.4. Let V be a Euclidean space with pinner product h., .i. Then we call the Norm of a vector
u is a non-negative real number, defined by kuk = hu, ui.

2 Orthogonal, Orthonomal, Gram-Schmidt process

Definition 2.1. A set {v1 , v2 , . . . , vk } of the Euclidean vector space V is said to be orthogonal iff

hvi , vj i = 0, if i 6= j.

Definition 2.2. A set {e1 , e2 , . . . , ek } of the Euclidean vector space V is said to be orthogonal iff
(
0, if i 6= j,
hei , ej i =
1, if i = j.

1
Definition 2.3. An orthogonal basis - OGB (respectively orthonormal basis -ONB) is a basis which also
is an orthogonal (respectively orthonormal) set.

Definition 2.4 (Gram-Schmidt Process to build an OGB, ONB from a basis of an Euclidean vector space
V ). Let (v1 , v2 , . . . , vk ) be a basis of V , we define

e1 = v1
hv2 , e1 i
e2 = v2 − e1
he1 , e1 i
hv3 , e1 i hv3 , e2 i
e3 = v3 − e1 − e2
he1 , e1 i he2 , e2 i
..
.
hvk , e1 i hvk , ek−1 i
ek = vk − e1 − · · · − ek−1 .
he1 , e1 i hek−1 , ek−1 i

Then we obtain {e1 , e2 , · · · , ek } is an OGB of V .

ei
Continue define ui = for all i = 1, k then we get that {ui }ki=1 is an ONB of V .
kei k
Definition 2.5. Let U, V be two vector subspaces of Euclidean vector space W .

ˆ We say α ∈ W is orthogonal with U , denote α ⊥ U, if α ⊥ u for all u ∈ U .

ˆ We say U orthogonal with V , denoted U ⊥ V , if

u ⊥ v, ∀u ∈ U, ∀v ∈ V.

Remark 2.1. U + V is a direct sum (direct orthogonal sum), denoted U ⊕⊥ V .

Definition 2.6 (Orthogonal Complement). Let U be a vector subspace of V , then

U ⊥ = {v ∈ V |v ⊥ U }

is said to be the Orthogonal Complement of U in V .

Proposition 2.1. If U ⊆ V , dim V < +∞, then

ˆ (U ⊥ )⊥ = U ,

ˆ V = U ⊕⊥ U ⊥ .

Definition 2.7 (Orthogonal projection). Suppose dim U < +∞, U ⊆ V . The orthogonal projection of V
onto U is the operator PU ∈ L(V ) defined as follows: For v ∈ V , write v = u + w, where u ∈ U and
w ∈ U ⊥ . Then PU v = u.

Theorem 2.1 (Properties of PU ).

ˆ PU ∈ L(V ),

ˆ PU u = u for every u ∈ U ,

ˆ PU w = 0 for every w ∈ U ⊥ ,

2
 ˆ imPU = U ,

ˆ ker PU = U ⊥ ,

ˆ v − PU v ∈ U ⊥ ,

ˆ PU2 = PU ,

ˆ kPU vk ≤ kvk,

ˆ for every ONB e1 , . . . , em of U ,

m
X
PU v = hv, ei iei .
i=1

Remark 2.2 (Minimizing the distance to a subspace). Suppose dim U < +∞, U ⊆ V, v ∈ V, u ∈ U . Then

kv − PU vk ≤ kv − uk.

The inequality above is an equality iff u = PU v.

3 Operators on Inner Product Spaces

3.1 Adjoint and Self-adjoint
Definition 3.1 (adjoint T ∗ ). Suppose T ∈ L(V, W ). The adojoint of T is the function T ∗ : W → V such
that

hT v, wi = hv, T ∗ wi

for every v ∈ V and every w ∈ W .

Definition 3.2 (Matrix of T ∗ ). Let T ∈ L(V, W ). Suppose S = {e1 , . . . , en } is an ONB of V , S 0 =

{f1 , . . . , fm } is and ONB of W and A = [aij ] is the matrix of T wrt S and S 0 , then the matrix of T ∗ wrt
those bases is

B = [bij ] = ĀT

which means bij = aji .

Definition 3.3 (Self-adjoint (a.k.a Hermitan) operator). An operator T ∈ L(V ) is called self-adjoint if
T = T ∗ . In other words, T ∈ L is self-adjoint if and only if

hT v, wi = hv, T wi

for all v, w ∈ V .

3
3.2 Isometry a.k.a Orthogonal Transformation
Definition 3.4. An operator T ∈ L(V ) is called an Isometri if

kT vk = kvk ∀v ∈ V.

Definition 3.5. The matrix A of an isometry is called an orthogonal matrix.

Theorem 3.1. Let A ∈ Mn (R), then TFAE

ˆ A is orthogonal.

ˆ A is invertible and A−1 = AT .

ˆ Rows of A is an orthonomal set in Rn with the usual dot product, that is

A.AT = In .

4 Quadratic forms
4.1 Quadratic forms and Matrices of quadratic forms
Definition 4.1. Suppose η is an bilinear symmetric form on V . The map q : V → R defined by

h(α) = η(α, α)

is called a quadratic form corresponding with η on V .

Remark 4.1. η can also be defined by q, in fact

1
η(α, β) = [η(α + β, α + β) − η(α, α) − η(β, β)]
2
1
= [q(α + β) − q(α) − q(β)].
2
A such bilinear form η is called the polar form of the quadratic form q.

Definition 4.2 (Matrix of η (q)). Let {α1 , α2 , . . . , αn } be a basis of V , then we say A = [aij ]n×n =
[η(αi , αj )]n×n is the matrix of η (or q) in {α1 , α2 , . . . , αn }.

Proposition 4.1. Suppose A (respectively B) is matricx of the bilinear map η (or q) in the basis S =
{α1 , α2 , . . . , αn } (respectively {S 0 = β1 , β2 , . . . , βn }) of V . If C is the transition matrix from S to S 0 then
we have

B = C T AC.

4.2 Write quadratic forms as its standard and normal form(s)

Definition 4.3. A basis α1 , α2 , . . . , αn of V is called a η-orthogonal if

η(αi , αj ) = 0 ∀i 6= j.

4
Its follows that the matrix of η in this basis is diagonal, which also means the coordinate expression of the
corresponding quadratic form H of η in that basis have the form
n
X
H(α) = = ai x2i
i=1
a1 x21 + · · · + ap x2p − ap+1 x2p+1 − · · · − ap+q x2p+q , (1)

where ai > 0(0 < i ≤ p + q ≤ n).

(1) is called the standard form of q.
√
In (1), we set yi = ai xi , (i = 1, p + q), yi = xi (i > p + q) and then get that
2
H(α) = y12 + · · · + yp2 − yp+1 2
− · · · − yp+q . (2)

(2) is called the normal form of q.

Theorem 4.1 (Sylvester Theorem about the inertia indices).

ˆ The number p in the formula (2) is called the positive inertia indices of H.

ˆ The number q in the formula (2) is called the negative inertia indices of H.

ˆ The pair (p, q) is the inertia indices of H.

ˆ The difference p − q is called the signature of H.

Theorem 4.2 (Lagrange Method). Suppose the quadratic form q is given in some basis (α1 , α2 , . . . , αn )
by
X
q(α) = aij xi xj (aij = aji ).
ij

Then we consider three cases:

ˆ Case 1. Suppose aii 6= 0 for some i, say, a11 6= 0. Then

n
X a1i
q = a11 (x21 + 2x1 xi ) + (other terms which does not contain x1 )
i=2
a 11
n
X a1i 2
= a11 (x1 + xi ) + (a quadratic form of x2 , . . . , xn )
a
i=2 11
n
X
= a11 y12 + bkl yk yl (bkl = blk )
k,l=2

where
a1i
 Pn
 y1 = x1 + xi ,
i=2
a11
y = xk (k = 2, 3, . . . , n)
k

Continue this process we can inductively find the standard (and so the normal) form of q.

5
 ˆ Case 2. All aii = 0, (i = 1, . . . , n) but aij 6= 0(i 6= j), say, a12 6= 0. We can use a non-singular
substitution as follow

 x1 = y 1 + y 2

x2 = y 1 − y 2

xk = yk (k = 3, . . . , n),


we have

2a12 x1 x2 = 2a12 (y12 − y22 ).

Then
X X
q= aij xi xj = bij yi yj
ij ij

where the coefficient of y12 is 2a12 6= 0, which leads us back to Case 1.

ˆ Case 3. All aij = 0(i, j = 1, . . . , n), then q has the standard form in any basis of V .

Remark 4.2. Since the matrix A of any quadratic form q is a symmetric matrix, which always diagonal-
izable, we can also find the standard form of q by orthogonal diagonalizing the matrix of q.