Linear Algebra [1]
5.3 Similarity and Diagonalization
• Diagonalization Revisited
Thm. [A] A : n × n matrix.
A is diagonalizable if and only if it has eigenvectors
X1, X2, · · · , Xn s.t. P = X1 X2 · · · Xn is invertible.
In this case, P −1AP = D = diag(λ1, λ2, · · · , λn), where λi
is the eigenvalue of A corresponding to Xi.
Thm. [A’] A : n × n matrix.
A is diagonalizable if and only if Fn has a basis
{X1, X2, · · · , Xn} of eigenvectors of A.
Kyu-Hwan Lee
Linear Algebra [2]
Thm. [B] Let X1, X2, · · · , Xk be eigenvectors corresponding
to distinct eigenvalues λ1, λ2, · · · , λk of A. Then
{X1, X2, · · · , Xk } is linearly independent.
Proof. Assume that {X1, X2, · · · , Xk } is linearly dependent.
We can find j s.t. {X1, X2, · · · , Xj−1} is linearly
independent, and {X1, X2, · · · , Xj } is linearly dependent.
Then we have
(∗) a1X1 + a2X2 + · · · + aj Xj = O,
where not all ai’s are zero, and in particular aj 6= 0.
Multiplying (∗) by A from the left, we get
a1λ1X1 + a2λ2X2 + · · · + aj λj Xj = O.
Kyu-Hwan Lee
Linear Algebra [3]
On the other hand, multiplying (∗) by λj , we obtain
a1λj X1 + a2λj X2 + · · · + aj λj Xj = O.
Subtracting two equations, we have
a1(λ1−λj )X1+a2(λ2−λj )X2+· · ·+aj−1(λj−1−λj )Xj−1 = O,
and a1(λ1 − λj ) = a2(λ2 − λj ) = · · · = aj−1(λj−1 − λj ) = 0.
Since λi’s are distinct, we have
a1 = a2 = · · · = aj−1 = 0,
and aj Xj = 0, aj = 0, a contradiction.
Therefore, {X1, X2, · · · , Xk } is linearly independent. 2
Kyu-Hwan Lee
Linear Algebra [4]
Cor. [B’] If A is an n×n matrix with n distinct eigenvalues,
then A is diagonalizable.
Fact. If one chooses linearly independent sets of
eigenvectors corresponding to distinct eigenvalues, and
combines them into a single set, then that combined set
will be linearly independent.
Kyu-Hwan Lee
Linear Algebra [5]
Def. An eigenvalue λ of A is said to have multiplicity m if
it occurs m times as a root of cA(x).
Def. The set
Eλ(A) = {X ∈ Fn|AX = λX}
of λ-eigenvectors is a subspace of Fn called the eigenspace
of A corresponding to λ.
Note that an eigenspace Eλ(A) is merely the null space of
λI − A.
Kyu-Hwan Lee
Linear Algebra [6]
Thm. [C] A : n × n matrix.
A is diagonalizable if and only if dim Eλ(A) is equal to the
multiplicity of λ for every eigenvalue λ of A.
Proof. (⇒) We omit it.
(⇐) Let λ1, λ2, · · · , λk be distinct eigenvalues. Assume
that dim Eλi (A) is equal to the multiplicity of λi for each
i = 1, 2, · · · , k. Choose a basis Bi of Eλi (A) for each λi.
Let B = B1 ∪ B2 ∪ · · · ∪ Bk . Then |B| = n and B is linearly
independent from Fact. Thus B is a basis of Fn, and A is
diagonalizable by Thm A’. 2
Kyu-Hwan Lee
Linear Algebra [7]
Thm. [C’] A : n × n matrix.
A is diagonalizable if and only if every eigenvalue λ of
multiplicity m yields m basic solutions of the equation
(λI − A)X = O.
Fact. Let λ be an eigenvalue of multiplicity of m of A.
Then
dim Eλ(A) ≤ m.
Kyu-Hwan Lee
Linear Algebra [8]
• Diagonalization Algorithm
Let A be an n × n matrix.
1. Find all the eigenvalues λ of A.
2. For each λ, compute the basic solutions of (λI − A)X = O.
If there are n basic solutions in total, A is diagonalizable.
3. Construct the matrix P whose columns are (scalar multiples
of) basic solutions.
4. P −1AP is diagonal. (P is invertible.)
Kyu-Hwan Lee
Linear Algebra [9]
0 0 1
Eg. A = 0 1 2, cA(x) = x(x − 1)2.
0 0 1
For λ = 1,
1 0 −1 1 0 0
λI − A = 0 0 −2 ⇒ 0 0 1 .
0 0 0 0 0 0
A is not diagonalizable.
1 2 3
B = 0 4 5 is diagonalizable.
0 0 6
Kyu-Hwan Lee
Linear Algebra [10]
• Similar Matrices
Def. A, B: n × n matrices
We say that A and B are similar if B = P −1AP for some
invertible P . We will write A ∼ B for similar matrices A
and B.
1 −2 4 0
Eg. and are similar.
−3 2 0 −1
2
3 1
Indeed, for P = , we have
−1 1
1 −2 4 0
P −1 P = .
−3 2 0 −1
Kyu-Hwan Lee
Linear Algebra [11]
Observations :
1. A is diagonalizable if and only if A is similar to a diagonal
matrix.
2. Assume that A and B are similar. Then A−1 ∼ B −1,
AT ∼ B T , Ak ∼ B k . If one of A and B is diagonalizable,
then the other is also diagonalizable.
3. If A is diagonalizable, then A−1, AT and Ak are also
diagonalizable.
Def. Let A = [aij ]. The trace of an n × n matrix A is
defined by
n
X
trA = aii = a11 + a22 + · · · + ann.
i=1
Kyu-Hwan Lee
Linear Algebra [12]
Prop.
1. tr(A + B) = trA + trB,
2. tr(kA) = ktrA,
3. tr(AT ) = trA,
4. tr(AB) = tr(BA).
Proof. A = [aij ], B = [bij ], AB = [cij ], and BA = [dij ].
n
X n X
X n
tr(AB) = cii = aik bki
i=1 i=1 k=1
Xn Xn n
X
= bkiaik = dkk = tr(BA).
k=1 i=1 k=1
Kyu-Hwan Lee
Linear Algebra [13]
Thm. If A ∼ B, then A and B have the same determinant,
rank, trace, characteristic polynomial, and eigenvalues.
Proof. Let B = P −1AP for some invertible P .
det B = det(P −1AP ) = det P −1 det A det P = det A.
trB = tr(P −1AP ) = tr((AP )P −1) = trA.
cB (x) = det(xI − B) = det(P −1xIP − P −1AP )
= det[P −1(xI − A)P ] = det(xI − A) = cA(x).
rankB = rank(P −1AP ) = rank(AP ) = rankA.
2
Kyu-Hwan Lee
