Linear systems

Luis L. Bonilla

Universidad Carlos III de Madrid, 2.1.D17, bonilla@ing.uc3m.es

October 29, 2019

Bonilla (UC3M) Lecture 4 October 29, 2019 1 / 23

 Outline

Outline

Systems of linear first order ODEs.

Autonomous planar linear homogeneous systems.
Inhomogeneous linear systems.

Bonilla (UC3M) Lecture 4 October 29, 2019 2 / 23

 Linear systems

Systems of ODEs
• 3y 00 + 2y 0 − y = 1 =⇒ y 00 + 23 y 0 − 13 y = 13 Define x1 = y , x2 = y 0 . Then

x10 = x2 ,

0
x2 = 3 x1 − 23 x2 + 31 ,
1

is an equivalent system of two ODEs. It can be written in vector form as

     
0 x1 0 1 0
x = A x + f , with x = ,A= 1 2 ,f = 1 .
x2 3 −3 3

• y (n) = F (t, y , y 0 , . . . , y (n−1) ) is equivalent to the system

x10 = x2


x20 = x3
  0
 x1 = f1 (t, x1 , . . . , xn )



... more generally: ...

 0
xn−1 = xn
 0
xn = fn (t, x1 , . . . , xn )


xn0 = F (t, x1 , x2 , . . . , xn−1 , xn )


Bonilla (UC3M) Lecture 4 October 29, 2019 3 / 23

 Linear systems

General properties
Existence, uniqueness, extension and continuity: f (t, x), {∂fi /∂xj },
continuous on a box B in n + 1 space and (t0 , x 0 ) ∈ B. Then the IVP:
x 0 = f (t, x), x(t0 ) = x 0 ,
has a unique solution x(t) on some t-interval containing t0 (plugging
x(t), ODE becomes identity). We can extend the solution to any
interval containing t0 for which the time-state curve lies in B and
extends to the boundary of B as t tends to either endpoint of the
interval. The solution is continuous in the data x 0 and f .
Autonomous systems: f (x), ∂fi /∂xj , continuous on a box S in n
dimensional phase space and (x 0 ) ∈ S. Then the IVP:
x 0 = f (x), x(t0 ) = x 0 ,
has a unique solution x(t) on some t-interval containing t0 . We can
extend the solution to any interval containing t0 for which the state
curve lies in S. The solution is continuous in the data x 0 and f .
Bonilla (UC3M) Lecture 4 October 29, 2019 4 / 23
 Linear systems

Linear systems

Assuming amn (t), Fn (t) continuous on |t − t0 | ≤ δ,

dx1
= a11 (t)x1 + . . . + a1n (t)xn + F1 (t),
dt
...
dxn
= an1 (t)x1 + . . . + ann (t)xn + Fn (t),
dt
IVP with x(t0 ) = x 0 has a unique solution. Equivalently, as a vector ODE:

dx
= A(t) x + F (t),
dt      
x1 a11 (t) . . . a1n (t) F1 (t)
x =  ... , A =  ... , F (t) =  . . . .
xn an1 (t) . . . ann (t) Fn (t)

Bonilla (UC3M) Lecture 4 October 29, 2019 5 / 23

 Linear systems

Homogeneous linear systems. Abel’s formula.

Superposition principle:
 
ϕ1j (t)
dx
= A(t) x, ϕj (t) =  . . . , j = 1, . . . , n, are solutions thereof.
dt
ϕnj (t)

Then x = c1 ϕ1 (t) + . . . + cn ϕn (t) is also a solution.

ϕ1 (t), . . . , ϕn (t) are independent iff their Wronskian is not zero.

W (ϕ1 , . . . , ϕn ) = det(Φ), Φ(t) = (ϕ1 (t), . . . , ϕn (t)) = (ϕij ),

If Φ(t) fundamental matrix of independent solutions,

d
Φ = A(t) Φ, and ln(detΦ) = Tr(ln Φ) imply
dt
d
Abel’s formula: W (ϕ1 , . . . , ϕn ) = [TrA(t)] W (ϕ1 , . . . , ϕn ).
dt
Bonilla (UC3M) Lecture 4 October 29, 2019 6 / 23
 Linear systems

Autonomous homogeneous linear systems.

Autonomous homogeneous linear system: ẋ(t) = A x(t).
Letting x(t) = X e λt , get eigenvalue problem:

A X = λX .

If all eigenvalues λj are different,

n
X
x(t) = cj e λj t Φj , Φj eigenvectors, cj are constant.
j=1

Example. x 0 = x + y , y 0 = 3x − y . Let x1 = x, x2 = y .
   
0 1 1  1−λ 1  = λ2 − 4 =⇒ λ = ±2.

x = x. 0 =  
3 −1 3 −1 − λ 

Φ1 = (1, 1), Φ2 = (1, −3); x = c1 e 2t + c2 e −2t , y = c1 e 2t − 3c2 e −2t . The

origin (0, 0) is a saddle point; see below.
Bonilla (UC3M) Lecture 4 October 29, 2019 7 / 23
 Autonomous planar linear homogeneous systems

Two dimensional (planar) systems

 
a b
A= , with τ =TrA = a + d, ∆ = detA = ad − bc.
c d
Characteristic equation for eigenvalues (λ1 + λ2 = τ , λ1 λ2 = ∆)
√
2 τ + (−1)j τ 2 − 4∆
λ − τ λ + ∆ = 0 =⇒ λj = , j = 1, 2.
2

Bonilla (UC3M) Lecture 4 October 29, 2019 8 / 23

 Autonomous planar linear homogeneous systems

Saddle point
∆ < 0 =⇒ λ1 < 0 < λ2 . Separatrices along eigenvector directions

Bonilla (UC3M) Lecture 4 October 29, 2019 9 / 23

 Autonomous planar linear homogeneous systems

Asymptotically stable node

τ < 0, ∆ > 0, τ 2 − 4∆ > 0 =⇒ λ1 < λ2 < 0.

Bonilla (UC3M) Lecture 4 October 29, 2019 10 / 23

 Autonomous planar linear homogeneous systems

Centers or asymptotically stable spiral points

τ
τ ≤ 0, ∆ > 0, τ 2 − 4∆ = −4Ω2 < 0 =⇒ λj = 2 ± iΩ.

τ = 0 (center) τ < 0 (stable spiral point) τ > 0 (unstable spiral point)

Bonilla (UC3M) Lecture 4 October 29, 2019 11 / 23

 Autonomous planar linear homogeneous systems

Borderline cases

τ 2 − 4∆ = 0.
A = τ2 I , star point: all directions are eigendirections;
otherwise degenerate node: only one eigendirection.

Bonilla (UC3M) Lecture 4 October 29, 2019 12 / 23

 Autonomous planar linear homogeneous systems

τ
Solution for degenerate node, τ 2 = 4∆, λ = 2

We use undetermined coefficients. x(t) = e τ t/2 (u + v t). Inserting this in

x 0 = A x, we find
 τ 
A − I v = 0,
2 
 τ
A − I u = v.
2
Then v = cψ and (a11 − τ2 )u1 + a12 u2 = cψ1 (from 2nd equation). This
gives u2 as function of u1 and c. Setting u1 = d and
ψ = (1, ( τ2 − a11 )/a12 ), the result is
    
τ t/2 t 1
x(t) = e c τ −2a11 1 +d τ −2a11
2a12 t + a12 2a12

Bonilla (UC3M) Lecture 4 October 29, 2019 13 / 23

 Autonomous planar linear homogeneous systems

Solution for degenerate node. Problem 4.1.7

x 0 = A x with
 
1 −1
A= , with τ = 4, ∆ = 4, λ = 2 (double).
1 3

x(t) = e 2t (u + v t) yields
     
−1 −1 v1 −1 −1 u1
= 0, = v.
1 1 v2 1 1 u2

Then v = c (1, −1) and −u1 − u2 = c (from 2nd equation). This gives
u2 = −c − u1 . Setting u1 = d, the result is
    
2t t 1
x(t) = e c +d
−t − 1 −1

Bonilla (UC3M) Lecture 4 October 29, 2019 14 / 23

 Autonomous planar linear homogeneous systems

Problem 4.1.7. Abel’s formula

From the eigenvalue and eigenvector, one solution is
 
2t 1
e .
−1

As TrA = 4, Abel’s formula W 0 = 4W gives W = ce 4t . Then

 x1 e 2t 
 
4t 2t 4t 2t
 x2 −e 2t  = c e =⇒ −(x1 + x2 )e = c e =⇒ x2 = −x1 − c e .
 

Then the first ODE x10 = x1 − x2 yields

x10 = 2x1 + c e 2t =⇒ e −2t (x10 − 2x1 ) = c =⇒ e −2t x1 = ct + d. We find
x1 = (ct + d)e 2t , x2 = −c(t + 1)e 2t − de 2t . The result is the same as
before:
 
ct + d
x(t) = e 2t .
−c(t + 1) − d

Bonilla (UC3M) Lecture 4 October 29, 2019 15 / 23

 Autonomous planar linear homogeneous systems

Solution for stable spiral point. Problem 4.1.3

x 0 = A x with
 
1 −1
A= , with τ = −2, ∆ = 2, λ = −1 ± i.
5 −3

Eigenvector
    
2−i −1 X1 1
= 0, =⇒ X = .
1 −2 − i X2 2−i

Solution (c = a + ib is an arbitrary complex number, a, b are real):

(a + ib)e it
 
−t
x(t) = e Re
(a + ib)(2 − i)e it
 
−t a cos t − b sin t
= e .
(2a + b) cos t + (a − 2b) sin t

Bonilla (UC3M) Lecture 4 October 29, 2019 16 / 23

 Autonomous planar linear homogeneous systems

Problem 4.1.3 using undetermined coefficients

x 0 = A x with eigenvalues λ = −1 ± i. Insert x = e −t (a cos t + b sin t),
which produces

(A + I )a = b, (A + I )b = −a. Note (A + I )2 + I = 0.

Then a is arbitrary and we solve

    
2 −1 b1 −a1
(A + I )b = =
5 −2 b2 −a2
b1 = 2a1 − a2 , b2 = 5a1 − 2a2 . Solution is:
 
−t a1 cos t + (2a1 − a2 ) sin t
x(t) = e
a2 cos t + (5a1 − 2a2 ) sin t
 
−t a cos t − b sin t
= e ,
(2a + b) cos t + (a − 2b) sin t
provided a1 = a, a2 = 2a + b, b = a2 − 2a1 .
Bonilla (UC3M) Lecture 4 October 29, 2019 17 / 23
 Inhomogeneous linear systems

Variation of parameters x 0 = A(t)x + F (t)

Fundamental matrix Ψ(t) = (ψ 1 (t), . . . , ψ n (t)), s.t., Ψ0 (t) = A(t) Ψ(t).
Then we insert a solution of the homogeneous system,
n
X
x(t) = yj ψ j (t) = Ψ(t)y
j=1

for yj = yj (t) in the inhomogeneous system:

Ψ(t)y 0 (t) + 
Ψ0 0 −1

(t) − A(t)Ψ(t)y

(t)y  (t) = F (t) =⇒ y (t) = Ψ (t)F (t).
  

Integrating and substituting in x(t), we find the solution of the IVP
x 0 (t) − A(t)x(t) = F (t), x(0) = x 0 :
Z t
x(t) = Ψ(t) Ψ−1 (s)F (s)ds + Ψ(t)x 0 .
0

Bonilla (UC3M) Lecture 4 October 29, 2019 18 / 23

 Inhomogeneous linear systems

Example: x 0 = Ax + F

et
   
−1 2
F (t) = , A= , with τ = −3, ∆ = −4. (1)
0 3 −2
Eigenvalues of A are λ1 = −4, with eigenvector ψ 1 = (2, −3) and λ2 = 1
with ψ 2 = (1, 1). The fundamental matrix is
2e −4t e t
 4t
−e 4t
  
−1 1 e
Ψ(t) = , and its inverse is Ψ (t) =
−3e −4t e t 5 3e −t 2e −t
For F (t) as in (1), we find
dy 1 e 5t 1 (e 5t − 1)/5
   
−1
=Ψ F = =⇒ y = , (3)
dt 5 3 5 3t
which satisfies y (0) = 0 and yields the particular solution
(e − e −4t ) + 35 te t
 2 t 
x(t) = Ψ y = 25 . (4)
3
− 25 (e t − e −4t ) + 35 te t
Bonilla (UC3M) Lecture 4 October 29, 2019 19 / 23
 Inhomogeneous linear systems

Undetermined coefficients for the example

We use x(t) = (b + ct)e t , because λ2 = 1. We get
 
1
(b + c + ct)e t = A(b + ct)e t + et .
0
Thus Ac = c (so c = µψ 1 is the eigenvector corresponding to λ2 = 1), and
   
1 1
c =µ , (A − I )b = c − .
1 0
The latter equation gives 2(b2 − b1 ) = µ − 1, −3(b2 − b1 ) = µ. Then
− 23 = µ−1 3
µ =⇒ µ = 5 and b1 − b2 = 1/5. Thus
 3t
+ b2 + 15

t 5
x(t) = e 3t ,
5 + b2
3 1
and the general solution of the system is (b2 = − 25 , k1 = − 25 previously)
−4t
 3t 1   t   
x(t) = e t 5 + 5 + b2
e
+ k1
2e
, b2 , k1 arbitrary const.
3t
5 e t −3e −4t
Bonilla (UC3M) Lecture 4 October 29, 2019 20 / 23
 Inhomogeneous linear systems

Laplace transform for the example with x(0) = 0

Let X (s) = L{x(t)}(s), L{e t } = 1/(s − 1). We get
   
1 1 1 −1 1
s X = AX + =⇒ X (s) = (sI − A) .
s −1 0 s −1 0
Since
   
−1 1 s +2 2 1 s +2 2
(sI − A) = 2 =
s + 3s − 4 3 s +1 (s − 1)(s + 4) 3 s +1

     3  
1 1 1 1 1 2 1
(sI − A)−1 = − + 5
,
s −1 0 25 s − 1 s + 4 −3 (s − 1)2 1
and (using Laplace transform table) the solution of the IVP is Eq. (4):
   
1 t −4t 2 3 t 1
x(t) = (e − e ) + te .
25 −3 5 1

Bonilla (UC3M) Lecture 4 October 29, 2019 21 / 23

 Inhomogeneous linear systems

Reduction to normal modes

Let A = R −1 D R, where D is a diagonal matrix with entries d1 ,. . . , dn .
x = R v yields

x 0 = R v̇ = A R v + F (t) =⇒ v̇ = R −1 A R v + R −1 F (t) = D v + R −1 F (t),

that is
n
X
v̇ (t) = D v + R −1 F (t), v̇i = di vi + (R −1 )ij Fj (t),
j=1

for i = 1, . . . , n. The solutions are

n
X Z t
di t −1
vi (t) = ci e + (R )ij e di (t−s) Fj (s) ds, (5)
j=1 0
n
X n
X Z t
dk t −1
xi (t) = (R)ik ck e + (R)ik (R )kj e dk (t−s) Fj (s) ds. (6)
k=1 j,k=1 0

Bonilla (UC3M) Lecture 4 October 29, 2019 22 / 23

 Inhomogeneous linear systems

Reduction to normal modes for the example

Transformation matrices: R = Ψ(0) and R −1 = Ψ−1 (0), the diagonal
components of D are d1 = −4, d2 = 1. The components of the vector
R −1 F (t) are 15 e t and 35 e t . Then normal mode equations are

dv1 1 dv2 3
= −4v1 + e t , = v2 + e t .
dt 5 dt 5
Solving these equations, we find the normal modes
 
−4t 1 t dv2 3t t
v1 = e c1 + e , = c2 + e , thereby yielding
25 dt 5

et
   
−4t t 2 −4t 3t 3
x1 = 2e c1 + e c2 + + 3t , x2 = −3c1 e + c2 + − et .
5 5 5 25
1
With c1 = − 25 , c2 = 0, this equation becomes (4).
Bonilla (UC3M) Lecture 4 October 29, 2019 23 / 23