General Complex Eigenvalues

Section 5.3

Nicholas Moore, PhD

1 A Pattern
Now we consider the case where A has real entries but has complex eigenvalues. In this case, if α + iβ is
an eigenvalue, then because the characteristic equation is a polynomial with real coefficients, the complex
conjugate: α + iβ = α − iβ is also an eigenvalue.

Watch the video below for another (perhaps more satisfying) explanation of this fact.

Video

P ROOF THAT C OMPLEX C ONJUGATE OF AN E IGENVALUE IS AN E IGENVALUE

2 A Change of Coordinates
Suppose A is 2n × 2n with distinct, non-real eigenvalue pairs α j ± iβ j for j = 1, 2, . . . , n and let Vj and Vj be
their associated eigenvectors. Just like the proposition from the previous section, we know that the collection
of eigenvectors is linearly independent.

It will be helpful now to change coordinates. Let

1
W2j−1 = (V + Vj )
2 j
−i
W2j = (Vj − Vj )
2
Note that W2j−1 is the real part of Vj and W2j is the imaginary part of Vj , therefore all the W vectors are
vectors in Rn .

Proposition 1. The vectors W1 , . . . , W2n are linearly independent.

Proof. See the following video

Video

P ROOF THAT W S ARE LINEARLY INDEPENDENT

1
3 The Canonical Form
Let’s multiply these W vectors by A. First,
1
AW2j−1 = ( AVj + AVj )
2
1
= ((α + iβ)Vj + (α − iβ)Vj
2
α iβ
= (Vj + Vj ) + (Vj − Vj )
2 2
= αW2j−1 − βW2j
Similarly
AW2j = βW2j−1 + αW2j
Now construct matrix T with Wj as the columns (that is TEj = Wj . The matrix has real entries and since the
Wj ’s are linearly independent, we know T is invertible. Now, what does T −1 AT look like? Let’s examine
what happens when we multiply by the standard basis vectors:
( T −1 AT ) E2j−1 = T −1 AW2j−1
= T −1 (αW2j−1 − βW2j )
= αE2j−1 − βE2j
where the first line uses the fact that TEj = Wj and the last line uses the fact that Ej = T −1 Wj . Similarly
(you should confirm this yourself) we find that ( T −1 AT ) E2j = βE2j−1 + αE2j . What does this tell us? It tells
us that odd numbered columns (those with index 2j − 1) have a form which is all zeros except for an α in
row 2j − 1 and a − β in row 2j. The even numbered columns (those index 2j) have a form which is all zeros
except for a β in row 2j − 1 and an α in row 2j.

Another way of writing this would be that

 
D1
T −1 AT = 
 .. 
. 
Dn
where each D j is a 2 × 2 matrix of the form
 
αj βj
Dj =
−β j αj

What happens when we have some distinct real eigenvalues and some distinct complex eigenvalues? We
combine the results of this section and the previous section to obtain the following result:

Theorem 1. Suppose that the n × n matrix A has distinct eigenvalues. In particular, suppose that A has k real,
distinct eigenvalues and L distinct complex conjugate pairs of eigenvalues. Then we may choose a linear map T so that
 
λ1
 .. 

 . 

λk
T −1 AT = 
 


 D1 

 . .

 . 
Dl
where the D j are 2 × 2 matrices of the form
 
αj βj
Dj = .
−β j αj