Linear Transformations & Canonical Forms: Department of Foundation Year Institute of Technology of Cambodia
Linear Transformations & Canonical Forms: Department of Foundation Year Institute of Technology of Cambodia
Linear Transformations & Canonical Forms: Department of Foundation Year Institute of Technology of Cambodia
2019–2020
1 Linear Transformations
4 Cayley-Hamilton Theorem
1 Linear Transformations
4 Cayley-Hamilton Theorem
Definition 1
Let V, W ∈ V K . A mapping L : V → W is said to be a linear
transfromation or linear if
1 u, v ∈ V : L(u + v) = L(u) + L(v)
2 v ∈ V, α ∈ K : L(αv) = αL(v).
Definition 2
Let V, W ∈ V K . We denote L(V, W ) the set of all linear
transformation from V into W . That is,
L(V, W ) = {L : V → W | L is linear}
Theorem 1
1 If L ∈ L(V, W ), then L(0) = 0.
2 L ∈ L(V, W ) if and only if for v1 , v2 ∈ V and α ∈ K:
Definition 3
Let L ∈ L(V, W ).
1 The image of V1 ⊂ V is defined by
L (V1 ) = {L (v) : v ∈ V1 }.
2 The range or the image of L is defined by
Im (L) = R(L) = L (V ) = {L (v) ∈ W : v ∈ V } .
3 The pre-image of W1 ⊂ W is defined by
L−1 (W1 ) = {v ∈ V : L (v) = w ∈ W1 } .
4 The kernel or null space of L is defined by
Ker (L) = N(L) = L−1 ({0}) = {v ∈ V : L (v) = 0} .
Definition 4
1 A map f : D → R is said to be one-to-one (or injective) if and
only if
x1 , x2 ∈ D : x1 6= x2 =⇒ f (x1 ) 6= f (x2 ).
2 A map f : D → R is said to be onto (or surjective) if and only if
∀y ∈ R, ∃x ∈ D : y = f (x).
3 A map f : D → R is said to be one-to-one correspondence (or
bijective) if and only if it is one-to-one and onto.
4 Let L ∈ L(V, W ). If L is one-to-one and onto then L is called
isomorphism between V and W .
5 L ∈ L(V ) is called an endomorphism on V .
6 Let L ∈ L(V ). If L is one-to-one and onto then L is called an
automorphism on V .
Theorem 2
Let L ∈ L(V, W ). Then
1 If V1 is a subspace of V , then L (V1 ) is a subspace of W . In
particular, Im(L) is a subspace of W .
2 If W1 is a subspace of W , then L−1 (W1 ) is a subspace of V . In
particular, Ker(L) is a subspace of V .
Definition 5
Let L ∈ L(V, W ). If Ker(L) and Im(L) are finite-dimensional. We
define,
1 The nullity of L, denoted by null(L), to be the dimension of
Ker(L). That is, null(L) = dim(Ker(L)).
2 The rank of L, denoted by rank(L), to be the dimension of Im(L).
That is, rank(L) = dim(Im(L)).
Theorem 4
Let L ∈ L(V, W ). If dim V = dim W < ∞, then L is one-to-one if and
only if it is onto.
Theorem 5
Let L ∈ L(V, W ). If dim V < ∞, then
Theorem 6
If V ∈ VnK . Then V and Kn are isomorphic.
CANONICAL FORMS ITC 9 / 56
Linear Transformations
Theorem 7
Let V ∈ VnK , W ∈ VmK and L ∈ L(V, W ). If BV = {v1 , . . . , vn } and
BW = {w1 , . . . , wm } are ordered bases for V and W respectively, then
there is an m × n matrix, denoted by [L]B W
BV , (called the matrix
representing L with respect to the ordered bases BV and BW )
such that
[L (v)]BW = [L]BW
BV [v]BV , v ∈ V.
where
[L]BW
BV = [L(v1 )]BW [L(v2 )]BW . . . [L(vn )]BW
Theorem 8
Let B1 and B2 be two ordered bases for V ∈ VnK , S be the transition
matrix from B2 to B1 , and L ∈ L (V ). If A = [L]B1 and B = [L]B2 ,
then B = S −1 AS.
Definition 6
Let A, B ∈ Mn (K). The matrix A is said to be similar to B if there
is an invertible matrix S such that B = S −1 A S. In this case, we write
A ∼ B.
Theorem 9
Let A, B, C ∈ Mn (K). Then,
1 A ∼ B ⇐⇒ B ∼ A.
2 A ∼ B and B ∼ C =⇒ A ∼ C.
1 Linear Transformations
4 Cayley-Hamilton Theorem
Definition 7
1 A polynomial p (λ) of degree n is said to be splitted over K if
there are scalars c 6= 0, λ1 , . . . , λn (not necessarily distinct) in K
such that
p (λ) = c (λ − λ1 ) · · · (λ − λn ) .
Theorem 10
Let L1 , L2 ∈ L (V ) such that L1 L2 = L2 L1 and A, B ∈ Mn (K) such
that AB = BA. If p, q ∈ K [X] then
and
p (A) q (B) = q (B) p (A) .
Definition 8
Let B be a basis for V ∈ VnK , and L ∈ L (V ). We define the
determinant of L by det (L) = det ([L]B ).
Theorem 11
The determinant of L ∈ L (V ) does not depend on a basis for V ∈ VnK .
Definition 9
Let L ∈ L(V ).
1 A scalar λ ∈ K is said to be an eigenvalue of L if there exists a
nonzero vector x ∈ V such that L (x) = λx. The nonzero vector x
is called an eigenvector corresponding to the eigenvalue λ. The
pair (λ, x) is called an eigenpair for L.
2 The set of all eigenvalues of L is called the spectrum of L.
3 The eigenspace corresponding to an eigenvalue λ is defined as
Definition 10
Let A ∈ Mn (K).
1 A scalar λ ∈ K is said to be an eigenvalue of A if there exists a
nonzero vector x ∈ Kn such that Ax = λx. The nonzero vector x
is said to be an eigenvector associated to λ.
2 The set of all eigenvalues of A is called spectrum of A, we write
spect(A).
3 The eigenspace corresponding to an eigenvalue λ is defined as
Definition 11
Let V ∈ V K , L ∈ L (V ), A ∈ Mn (K).
1 A subspace S of V is called an L-invariant subspace of V if
L (S) ⊆ S.
2 A subspace S of Kn is called an A-invariant subspace of Kn if
{Ax : x ∈ S} ⊆ S.
Theorem 12
Let V ∈ V K and λ ∈ K be an eigenvalue of L ∈ L (V ). Then the
eigenspace Eλ is a subspace of V and it is L-invariant. If V ∈ VnK , then
the eigenspace Eλ is a finite dimensional subspace of V . The dimension
of this eigenspace is called geometric multiplicity of λ denoted
gm (λ).
Theorem 13
Let λ ∈ K be an eigenvalue of A ∈ Mn (K). Then the eigenspace Eλ is
a subspace of Kn and it is A-invariant. The dimension of this
eigenspace is called geometric multiplicity of λ denoted gm (λ).
Theorem 14
Let L ∈ L(V ). If dim V < ∞ and B is an ordered basis for V . Then λ
is an eigenvalue of L if and only if λ is an eigenvalue of [L]B .
CANONICAL FORMS ITC 18 / 56
Eigenvalues and Eigenvectors
Theorem 15
Let V ∈ VnK , L ∈ L (V ), B be a basis for V , A = [L]B , and λ ∈ K be a
scalar. The following statements are equivalent.
1 λ is an eigenvalue of L.
2 (A − λI) [x]B = 0 has a nontrivial solution.
3 The eigenspace Eλ 6= {0}.
4 det (L − λI) = 0.
Theorem 16
Let A ∈ Mn (K) and λ ∈ K. The following statements are equivalent.
1 λ is an eigenvalue of A.
2 (A − λI) x = 0 has a nontrivial solution.
3 The eigenspace Eλ 6= {0}.
4 det (A − λI) = 0.
CANONICAL FORMS ITC 19 / 56
Eigenvalues and Eigenvectors
Definition 12
Let V ∈ VnK , L ∈ L (V ), B be a basis for V , A = [L]B , and λ ∈ K. The
polynomial pL (λ) = det (L − λI) = det (A − λI) does not depend on a
basis for V . This polynomial is called characteristic polynomial of
L (also of A) and pL (λ) = det (A − λI) = 0 is called characteristic
equation from L (also from A).
Theorem 17
Let A, B ∈ Mn (K). If B ∼ A, then the two matrices have the same
characteristic polynomial and consequently both have the same
spectrum.
Theorem 19
Let V ∈ VnK , L ∈ L (V ), A ∈ Mn (K) and p be a polynomial.
1 If λ is an eigenvalue of L then p (λ) is an eigenvalue of p (L).
2 If λ is an eigenvalue of A then p (λ) is an eigenvalue of p (A).
Theorem 20
Let V ∈ VnK and L ∈ L (V ). Let λ be an eigenvalue of L. Then
1 ≤ gm (λ) ≤ am (λ).
CANONICAL FORMS ITC 21 / 56
Eigenvalues and Eigenvectors
Theorem 21
Let pA (λ) be the characteristic polynomial and mA (λ) be the minimal
polynomial of A.
1 The polynomial mA (λ) divides every polynomial that has A as a
zero.
2 The polynomials pA (λ) and mA (λ) have the same irreducible
factors over K. Consequently, pA (λ) and mA (λ) have the same
zeros in K.
3 If pA (λ) = (λ1 − λ)n1 · · · (λk − λ)nk where λ1 , . . . , λk are distinct,
then there exist integers m1 , . . . , mk such that 1 ≤ mj ≤ nj ,
j = 1, 2, . . . , k and
mA (λ) = (λ − λ1 )m1 · · · (λ − λk )mk .
Theorem 23
Let V ∈ VnK , L ∈ L(V ), A = [L]B and pL (λ) be the characteristic
polynomial of L. Suppose that V = S1 ⊕ · · · ⊕ Sk where Si is an
L-invariant subspace of V for each i = 1, 2, . . . , k with bases
B1 , . . . , Bk , respectively and Ai = [LSi ]Bi , and pLSi (λ) is the
characteristic polynomial for LSi (i = 1, 2, . . . , k). Then
1 A is similar to the block diagonal matrix with diagonal blocks
A1 , . . . , A k .
2 pL (λ) = pLS1 (λ) . . . pLSk (λ) .
CANONICAL FORMS ITC 23 / 56
Eigenvalues and Eigenvectors
Theorem 24
Let V ∈ VnK , L ∈ L(V ). If the minimal polynomial of L is
Theorem 26
Let λ1 , . . . , λk be k different eigenvalues of L ∈ L(V ) (also of
A ∈ Mn (K)). Then
Theorem 27
Let λ be an eigenvalue of A ∈ M(K) and k be the multiplicity of λ.
Then
1 ≤ dim Eλ ≤ k.
CANONICAL FORMS ITC 25 / 56
Contents
1 Linear Transformations
4 Cayley-Hamilton Theorem
Definition 13
Let V ∈ VnK , L ∈ L (V ).
1 L is said to be diagonalizable in K if there is an ordered basis B
such that [L]B is diagonal.
2 L is said to be triangularizable in K if there is an ordered basis
B such that [L]B is a triangular matrix.
Definition 14
Let A ∈ Mn (K).
1 A is said to be diagonalizable in K if A ∼ D, where D is a
diagonal matrix.
2 A is said to be triangularizable in K if A ∼ T , where T is a
triangular matrix T .
Theorem 28
Let V ∈ VnK , L ∈ L (V ), B be an ordered basis for V , and A = [L]B .
Then
1 L is diagonalisable over K if and only if A is diagonalisable over K.
2 L is triangularizable over K if and only if A is triangularizable over
K.
Theorem 29
Let V ∈ VnK , L ∈ L (V ), and Eλ1 , . . . , Eλk be eigenspaces of L. The
following assertions are equivalent.
1 L is diagonalizable over K.
2 There exists a basis of V formed by the eigenvectors of L.
3 V = Eλ1 ⊕ · · · ⊕ Eλk .
4 n = dim Eλ1 + · · · + dim Eλk
CANONICAL FORMS ITC 28 / 56
Diagonalization and Triangularization
Theorem 30
Let A ∈ Mn (K) and λ1 , λ2 , . . . , λk be k distinct eigenvalues of A. The
following assertions are equivalent.
1 A is diagonalizable in K.
2 There exists a basis of Kn formed by the eigenvectors of A.
3 Kn = Eλ1 ⊕ · · · ⊕ Eλk .
4 n = dim Eλ1 + · · · + dim Eλk
Theorem 31
Let V ∈ VnK . L ∈ L (V ) is diagonalizable over K if and only if
1 The characteristic polynomial pL (λ) splits over K.
2 Every eigenvalue λ of L, am (λ) = gm (λ).
Theorem 32
A ∈ Mn (K) is diagonalizable over K if and only if
1 The characteristic polynomial pA (λ) splits over K.
2 Every eigenvalue λ of A, am (λ) = gm (λ).
Theorem 33
Let V ∈ VnK . L ∈ L (V ) is diagonalisable over K if and only if its
minimal polynomial is a product of district linear polynomials.
Theorem 34
If A ∈ Mn (R) and A is symmetric, then there exists an orthogonal
matrix P such that P −1 AP is diagonal (i.e. A can be diagonalized by
an orthogonal matrix).
Theorem 35
Let V ∈ VnK and L ∈ L (V ) with the characteristic polynomial pL (λ).
The following assertions are equivalent.
1 L is triagularizable over K.
2 pL (λ) splits over K.
Theorem 36
Let A ∈ Mn (K) with the characteristic polynomial pA (λ). The
following assertions are equivalent.
1 A is triagularizable over K.
2 pA (λ) splits over K.
1 Linear Transformations
4 Cayley-Hamilton Theorem
Definition 15
Let V ∈ V K , L ∈ L(V ), A ∈ Mn (K), and k > 0 be an integer.
1 The subspace C spanned by x, L (x) , L2 (x) , . . .
x is called the
L-cyclic subspace of V generated by x ∈ V .
x, Ax, A2 x, . . . is called the
2 The subspace C spanned by
x
A-cyclic subspace of Kn generated by x ∈ Kn .
Theorem 37
Let V ∈ VnK , L ∈ L(V ) and S be an L-invariant subspace of V . Then
the characteristic polynomial of LS divides the characteristic
polynomial pL (λ) of L.
Theorem 38
Let V ∈ VnK , L ∈ L(V ), and Cx be an L-cyclic subspace of V generated
by a nonzero vector x with dim Cx = k. Then
x, L (x) , L2 (x) , . . . , Lk−1 (x) is a basis for Cx .
1
Theorem 39 (Cayley-Hamilton)
Let V ∈ VnK , L ∈ L(V ). If pL (λ) is the polynomial characteristic of L
then pL (L) = 0.
Theorem 40 (Cayley-Hamilton)
Let A ∈ Mn (K). If pA (λ) is the polynomial characteristic of A then
pA (A) = 0.
1 Linear Transformations
4 Cayley-Hamilton Theorem
Definition 16
Let V ∈ V K , L ∈ L(V ), A ∈ Mn (K), and k > 0 be an integer.
1 The linear operator L is said to be nilpotent if Lk = 0 for some
positive integer k.
2 The linear operator L is said to be nilpotent of index k if
Lk = 0 and Lk−1 6= 0.
3 The matrix A is said to be nilpotent if Ak = 0 for some positive
integer k.
4 The matrix A is said to be nilpotent of index k if Ak = 0 and
Ak−1 6= 0.
Definition 17
Let V ∈ VnK and λ be an eigenvalue of L ∈ L(V ).
1 The index of λ, denoted index (λ), is the smallest positive integer
k such that rank(L − λI)k = rank(L − λI)k+1
2 A nonzero vector x ∈ V is called a generalized eigenvector of L
corresponding to the eigenvalue λ if (L − λI)p (x) = 0 for some
positive integer p.
3 The generalized eigenspace of L corresponding to λ is defined
by
Definition 18
Let λ be an eigenvalue of A ∈ Mn (K).
1 The index of λ, denoted index (λ), is the smallest positive integer
k such that rank(A − λIn )k = rank(A − λIn )k+1
2 A nonzero vector x ∈ Kn is called a generalized eigenvector of
A corresponding to the eigenvalue λ if (A − λIn )p x = 0 for some
positive integer p.
3 The generalized eigenspace of A corresponding to λ is defined
by
Theorem 41
Suppose the characteristic polynomial of L ∈ L(V ) splits over K and λ
is an eigenvalue with algebraic multiplicity k. Then
1 dim (Gλ (L)) ≤ k.
2 Gλ (L) = N (L − λI)k .
Theorem 42
Suppose the characteristic polynomial of A ∈ Mn (K) splits over K and
λ is an eigenvalue with algebraic multiplicity k. Then
1 dim (Gλ (A)) ≤ k.
2 Gλ (A) = N (A − λIn )k .
Theorem 43
Suppose the characteristic polynomial of L ∈ L(V ) splits over K,
λ1 , . . . , λm are distinct eigenvalues with algebraic multiplicities
k1 , . . . , km , respectively. Then for every x ∈ V ∈ VnK there exist
elements xi ∈ Gλi (L) , 1 ≤ i ≤ m such that
x = x1 + · · · + xm .
Theorem 44
Suppose the characteristic polynomial of A ∈ Mn (K) splits over K,
λ1 , . . . , λm are distinct eigenvalues with algebraic multiplicities
k1 , . . . , km , respectively. Then for every x ∈ Kn there exist elements
xi ∈ Gλi (A) , 1 ≤ i ≤ m such that
x = x1 + · · · + xm .
Theorem 45
Suppose the characteristic polynomial of L ∈ L(V ) splits over K,
λ1 , . . . , λm are distinct eigenvalues with algebraic multiplicities
k1 , . . . , km , respectively, and Bi is an ordered basis for Gλi (L) for
i = 1, 2, . . . , m. Then
1 Bi ∩ Bj = ∅, i 6= j.
2 B = B1 ∪ · · · ∪ Bm is an ordered basis for V ∈ VnK
3 dim (Gλi (L)) = ki for i = 1, 2, . . . , m.
Theorem 46
Suppose the characteristic polynomial of A ∈ Mn (K) splits over K,
λ1 , . . . , λm are distinct eigenvalues with algebraic multiplicities
k1 , . . . , km , respectively, and Bi is an ordered basis for Gλi (A) for
i = 1, 2, . . . , m. Then
1 Bi ∩ Bj = ∅, i 6= j.
2 B = B1 ∪ · · · ∪ Bm is an ordered basis for Kn
3 dim (Gλi (A)) = ki for i = 1, 2, . . . , m.
Definition 19
The matrix
λ 1
λ ...
Jb (λ) = ∈ Mb (K)
..
. 1
λ
is called Jordan block of order b corresponding to λ.
Definition 20
The matrix
Jb1 (λ) O ··· O
.. .. ..
O . . .
∈ Mb +···+b (K)
Js (λ) = .
. .. 1 k
.. ..
. O
O ··· O Jbk (λ)
Definition 21
A matrix J ∈ Mn (K) is said to be in Jordan canonical form or a
Jordan matrix if it is made of Jordan segment (along the diagonal).
Namely,
Js1 (λ1 )
J =
..
.
Jsl (λl )
Definition 22
Let V ∈ VnK and L ∈ L(V ). The linear operator L is said to be
jordanizable over K if there is a basis B for V such that [L]B = J is
a matrix in Jordan canonical form. The basis B is called Jordan
canonical basis for V . [L]B = J is a Jordan is called Jordan
canonical form for the linear operator L.
Definition 23
A matrix A ∈ Mn (K) is said to be jordanizable over K if it is
similar to a matrix in Jordan canonical form, that is, there is a
nonsingular matrix S such that J = S −1 AS is a Jordan matrix.
Definition 24
Let V ∈ V K , L ∈ L(V ), and x ∈ Gλ (L) be a generalized eigenvector
corresponding to the eigenvalue λ. Suppose that k is the smallest
positive integer for which (L − λI)k (x) = 0. Then the ordered set
n o
(L − λI)k−1 (x) , (L − λI)k−2 (x) , . . . , (L − λI) (x) , x
Definition 25
Let A ∈ Mn (K) and x ∈ Gλ (A) be a generalized eigenvector
corresponding to the eigenvalue λ. Suppose that k is the smallest
positive integer for which (A − λI)k x = 0. Then the ordered set
n o
(A − λI)k−1 x, (A − λI)k−2 x, . . . , (A − λI) x, x
Theorem 47
Let V ∈ VnK and L ∈ L(V ). Suppose that the characteristic polynomial
pL (λ) splits over K and B is a basis for V such that B is a disjoint
union of cycles of generalized eigenvectors of L. Then
1 For each cycle of generalized eigenvectors C contained in B,
S = span (C) is L-invariant, and [LS ]C is a Jordan block.
2 B is a Jordan canonical basis for V .
Theorem 48
Let V ∈ VnK and L ∈ L(V ), and λ be an eigenvalue of L. Suppose that
B1 , . . . , Bk are cycles of generalized eigenvectors of L corresponding to
λ such that the initial vectors of the Bi s are distinct and form a
linearly independent set. Then the Bi s are disjoint and the union
B1 ∪ · · · ∪ Bk is linearly independent.
Theorem 49
Let V ∈ VnK and L ∈ L(V ), and λ be an eigenvalue of L. Then the
generalized eigenspace Gλ (L) has an ordered basis consisting of a
union of disjoint cycles of generalized eigenvectors corresponding to λ.
Consequently, if the characteristic polynomial pL (λ) splits over K then
L is jordanizable over K.
Theorem 51
Let V ∈ VnC , L ∈ L(V ), A = [L]B with distinct eigenvalues λ1 , . . . , λl .
Then there is a Jordan canonical basis forming a nonsingular matrix P
such that:
Js1 (λ1 )
1 J = P −1 AP =
..
.
Jsl (λl )
2 J has one Jordan segment J
sj (λj ) for each eigenvalue
λj , j = 1, 2, . . . , l.
3 Each Jordan segment Jsj (λj ) is made up of βj = dim N (A − λj I)
Jordan blocks Jb (λj ).
4 The largest Jordan block in Jsj (λj ) is kj × kj , where
kj = index (λj ).
D
a −b 1 0
D= , and I = .
b a 0 1
Theorem 52
If A ∈ Mn (R), then A is similar to a real Jordan canonical form.
CANONICAL FORMS ITC 56 / 56