21 Solving linear systems of ODE with constant coefficients.

Part
II. Complex eigenvalues
Consider
ẏ = Ay, A = [aij ]n×n ∈ Mn (R) (1)
together with the initial condition
y(t0 ) = y 0 ∈ Rn . (2)
From the previous lecture we know that the general solution to (1) has the form
y(t) = C1 v 1 eλ1 t + . . . + Cn v n eλn t ,
where λ1 , . . . , λn are the eigenvalues and v 1 , . . . , v n are the corresponding eigenvectors, provided all
the eigenvalues are distinct. However, even in this simple case we can have complex eigenvalues with
complex eigenvectors. The goal here is to show that we still can choose a basis for the vector space of
solutions such that all the vectors in it are real.
Proposition 1. If y(t) is a solution to (1) then Re y(t) and Im y(t) are also solutions to (1).
Proof. Let y(t) = u1 (t) + iu2 (t), where u1 (t) = Re y(t) and u2 (t) = Im y(t). Plug this into (1) and
use the linearity to find
(u̇1 − Au1 ) + i(u̇2 − Au2 ) = 0,
which implies that
u̇1 = Au1 , u̇2 = Au2 .


Note that both u1 (t) and u2 (t) are real valued solutions, therefore, instead of one complex-valued
solution y(t) we have two real valued solutions. To finalize the argument we need to show that the
new basis, where two complex valued solutions corresponding to λ and λ are substituted with u1 (t)
and u2 (t) is still a linearly independent set. To see this first note that if λ is a complex eigenvalue
with eigenvector v, then λ is an eigenvalue with eigenvector v. This follows from
Av = λv =⇒ Av = λv.
Now we need to show that if {v, v, v 3 , . . . , v n } is a linearly independent set, then {Re v, Im v, v 3 , . . . , v n }
is also a linearly independent set (left as an exercise).
Finally, let us see in details how our new real-valued solution looks like in coordinates. We have
y(t) = veλt is a complex valued solution, here λ = α + iβ, v = v 1 + iv 2 . We find
y(t) = (v 1 + iv 2 )e(α+βi)t
= (v 1 + iv 2 )eαt (cos βt + i sin βt)
= eαt (v 1 cos βt − v 2 sin βt) + ieαt (v 1 sin βt + v 2 cos βt).

Therefore, instead of two complex-valued solutions veλt and veλt we have two real-valued solutions
u1 (t) = eαt (v 1 cos βt − v 2 sin βt),
u2 (t) = eαt (v 1 sin βt + v 2 cos βt).
MATH266: Intro to ODE by Artem Novozhilov, e-mail: [email protected]. Spring 2024

1
Example 2. Solve
ẏ = Ay,
where  
3 −2
A= .
4 −1
First we find the characteristic polynomial

det(A − λI) = λ2 − 2λ + 5,

which means that we have two complex eigenvalues

λ1,2 = 1 ± 2i.

An eigenvector corresponding to λ1 can be found as a solution to

    
3 − 1 − 2i −2 v1 0
= ,
4 −1 − 1 − 2i v2 0

or, in coordinates,

(1 − i)v1 − v2 = 0,
2v1 − (1 + i)v2 = 0.

This is not quite obvious from the first view that the two equations are equivalent, but they are
(multiply the first by 1 + i), hence we have

v2 = (1 − i)v1 ,

⊤
if I take v1 as a free variable. Therefore, any solution to our system is given by the vector v1 , (1−i)v1 ,
and for my eigenvector I can choose, e.g., v1 = 1, therefore,
 
1
v1 = .
1−i

I do not need to look for an eigenvector corresponding to λ2 , because, as it was shown above, v 2 = v 1 .
By making these calculations I proved that my system has two linearly independent solutions
   
1 (1+2i)t 1
y 1 (t) = e , y 2 (t) = e(1−2i)t ,
1−i 1+i

and the general solution is

y(t) = C1 y 1 (t) + C2 y 2 (t),
where C1 , C2 are arbitrary constants. However, this solution is complex-valued. To find a real valued
solution I will follow the algorithm from the first part of the lecture and take Re y 1 (t) and Im y 2 (t)

2
as new two linearly independent real-valued solutions.
   
1 (1+2i)t 1
y 1 (t) = e = et (cos 2t + i sin 2t)
1−i 1−i
 
t cos 2t + i sin 2t
=e
cos 2t + sin 2t + i(− cos 2t + sin 2t)
   
t cos 2t t sin 2t
=e +i e .
cos 2t + sin 2t − cos 2t + sin 2t
| {z } | {z }
u1 (t) u2 (t)

Please do not include i into u2 (t)!

Therefore, the general solution to the problem has the form
   
t cos 2t t sin 2t
y(t) = C1 e + C2 e ,
cos 2t + sin 2t − cos 2t + sin 2t
or
y1 (t) = et (C1 cos 2t + C2 sin 2t),


y2 (t) = et (C1 − C2 ) cos 2t + (C1 + C2 ) sin 2t ,
in coordinates.
Example 3. Solve the IVP    
1 0 0 1
ẏ = 0 1 −1 y, y(0) = 1 .

0 1 1 1
First we find the characteristic polynomial
P3 (λ) = (1 − λ)(λ2 − 2λ + 2),
which means that we have
λ1 = 1, λ2,3 = 1 ± i.
We can choose eigenvector v 1 = (1, 0, 0)⊤ corresponding to λ1 . Similarly, we find that eigenvector
v 2 = (0, i, 1)⊤ can be taken for λ2 . Since we are looking for a real-valued solution, we do not care
about λ3 . We find that
     
0 0 0
y 2 (t) =  i  e(1+2i)t = et − sin t + iet cos t ,
1 cos t sin t
therefore, the general solution to our problem can be written as
       
1 0 0 C1
y(t) = C1 0 et + C2 − sin t et + C3 cos t et = et −C2 sin t + C3 cos t .
0 cos t sin t C2 cos t + C3 sin t
Now we use the initial conditions and find that C1 = C2 = C3 = 1. Thence, the final answer is
 
1
y(t) = et cos t − sin t .
cos t + sin t