Matemáticas para la Ingeniería 3

Theme 8: Systems of Differential Equations

Francisco Rodríguez Sánchez
Universidad de Málaga

1 Systems of Linear Differential Equations

A system of differential equations of order r express relations between several n functions
x1 (t), x2 (t), . . . , xn (t) and its successive derivatives:
  
(r ) (r ) (r )
F1 t, x1 , x2 . . . , xn , x10 , x20 , . . . , xn0 , x100 , x200 , . . . , xn00 , x1 , x2 , . . . , xn = 0 

  
0 0 0 00 00 00 (r ) (r ) (r ) 
F2 t, x1 , x2 . . . , xn , x , x , . . . , x , x , x , . . . , x , x , x , . . . , xn = 0 

1 2 n 1 2 n 1 2
..
.


 

(r ) (r ) (r )
Fk t, x1 , x2 . . . , xn , x10 , x20 , . . . , xn0 , x100 , x200 , . . . , xn00 , x1 , x2 , . . . , xn = 0




Next problem is solved by a system of differential equations.

x1 x2

m1 m2
k1 k3 k2

Figure 1: Coupled harmonic oscillators.

Example 1.1 (Coupled harmonic oscillators). Suppose two objects joined for three springs
moving in a line. Springs verify the Hooke’s law with respective constants k1 , k2 , k3 , (Fig-
ure 1). The time-dependent functions x1 (t) and x2 (t), determine the distance of each object
to their idle point. The motion is governed by the following second order system of differ-
ential equations:
2

m1 ddtx21 = −k1 x1 + k3 ( x2 − x1 ) 
2
m2 ddtx22 = −k2 x2 + k3 ( x1 − x2 ) 

1.1 Laplace Method for Systems

It is possible to use the Laplace method described in this section to integrate systems of linear
differential equations.

1
Example 1.2. Solve the initial value problem

x 0 + x + 2y = 0
 
x (0) = 1
with
y0 + 2x − 2y = sin t y (0) = 0

Applying Laplace transform,


L( x 0 ) + L( x ) + 2L(y)
=0
 (s + 1)L( x ) + 2L(y) = 1 
⇒ 1
L(y0 ) + 2L( x ) − 2L(y) = L(sin t) 2L( x ) + (s − 2)L(y) =
s2

+1
Solving this algebraic linear system

s3 − 2 s2 + s − 4 2 s2 − s + 1
L( x ) = , L( y ) = −
s4 − s3 − 5 s2 − s − 6 s4 − s3 − 5 s2 − s − 6
For doing Inverse Laplace transform of L( x ), by partial fractions:

22 4 s−7
L( x ) = + −
25 (s + 2) 25 (s − 3) 25 (s2 + 1)

From here
22 −2t 4 cos t − 7 sin t
x (t) = e + e3t −
25 25 25
Similarly, the function y(t) is calculated (exercise).

Example 1.3. Now, we are going to solve the coupled oscillator problem on Example 1.1
for Hooke’s constants k1 = k2 = 1, k3 = 2 and mass values m1 = m2 = 1. Moreover, the
motion starts from repose (initial velocities zero) and the first object is displaced one length
unit. This is expressed by the folowng system with initical conditions:

x100 = − x1 + 2( x2 − x1 )
) 
x1 (0) = 1, x2 (0) = 0
with .
x 00 = − x2 + 2( x1 − x2 ) x10 (0) = x20 (0) = 0
2

Applying the Laplace transform the following algebraic system is obtained

s2 L( x1 ) − s = 2L( x2 ) − 3L( x1 )


s2 L( x2 ) = 2L( x1 ) − 3L( x2 )

with solutions
s3 + 3s 1 s 1 s
L( x1 ) = = · 2 + ·
( x2 + 5)( x2 + 1) 2 s + 5 2 s2 + 1
2s 1 s 1 s
L( x2 ) = 2 2
=− · 2 + · 2
( x + 5)( x + 1) 2 s +5 2 s +1

and the inverse transforms give the solution

√
cos cos t
5t
x1 ( t ) = +
2 √2
cos t cos 5t
x2 ( t ) = − .
2 2

2
2 First Order Systems
Although a first order system of ODEs may be a more general form, usually are written

dx1


 = f 1 (t, x1 , x2 , . . . , xn )



 dt
 dx2


= f 2 (t, x1 , x2 , . . . , xn )
dt
 ..
.




 dxn = f n (t, x , x2 , . . . , xn )



1
dt
and denoting
x = ( x1 , x2 , . . . , xn ) and f = ( f 1 , f 2 , . . . , f n ),
dx
this system can be written = f(t, x). General solutions express parametric family of
dt
curves of R . The named problem of initial values
n

(
dx
dt = f ( t, x )
x ( t 0 ) = x 0 ∈ Rn

∂ fi
determines a unique curve (particular solution) when f is a continuous function with ∂x j is
continuous for every i, j.
Sometimes a system of differential equations is expressed in symmetric form

dx1 dx2 dxn dt

= = ··· = =
P1 (t, x1 , . . . , xn ) P1 (t, x2 , . . . , xn ) Pn (t, x1 , . . . , xn ) Q(t, x1 , . . . , xn )

2.1 First Order Linear Systems

A first order system is linear if it is expressed

dx1


 = a11 (t) x1 + a12 (t) x2 + · · · + a1n (t) xn + b1 (t)



 dt
 dx2


= a21 (t) x1 + a22 (t) x2 + · · · + a2n (t) xn + b2 (t)
dt
 ..
.




dx

 n = a (t) x + an2 (t) x2 + · · · + ann (t) xn + bn (t)


n1 1
dt
In matrix form is
x 0 ( t ) = A ( t ) x ( t ) + b ( t ).
A Cauchy problem is to find solutions of the system which verify the so-called initial condi-
tions x1 (t0 ) = x̄1 , x2 (t0 ) = x̄2 , . . . , xn (t0 ) = x̄n . It is expressed as

x0 (t) = A(t)x(t) + b(t), with


(1)
x1 (t0 ) = x̄1 , x2 (t0 ) = x̄2 , . . . , xn (t0 ) = x̄n

Theorem 2.1. Let A(t), b(t), respectively, a n × n-matrix a an n × 1-vector functions with con-
tinuous derivatives in an open interval ( a, b) ∈ R. For t0 ∈ ( a, b), the Cauchy problem (5) has an
unique solution.

3
 Any first order linear system with n variables is equivalent to a linear ODE of grade n.
To see this, consider the ODE

y ( n ) + a 1 ( x ) y ( n −1) + · · · + a n −1 ( x ) y 0 + a n y = p ( x )

and the changes of variable y1 = y, y2 = y0 , …, yn = y(n−1) produce the system

 0
y = y2
 10


 y2 = y3
..

 .
 0

y n = p ( x ) − a n ( x ) y 1 − · · · − a 2 ( x ) y n −1 − a 1 ( x ) y n

Reverse can be done by successive derivatives and replacements.

Example 2.2. Express as an unique ODE the next system
(
x10 = tx1 − x2 + t2
 0      2
x1 t −1 x1 t
⇐⇒ 0 = 2 + (2)
0 2
x2 = (1 − t ) x1 + t x2 . x2 1−t t x2 0

Derivative the first equation and replace x20 from the second equation to obtain

x100 = x1 + t x10 − x20 + 2t =⇒ x100 = t x10 + t x1 − t2 x2 + 2t

Finally, using the first equation again, eliminate x2 from it and you have the second order
linear ODE you were looking for

x100 − (t2 + t) x10 + (t3 − t) x1 = −t4 + 2 t

which is equivalent to the system (6).

2.2 First Order Linear Systems with constant coefficients

A first order system is linear with constant coefficients is expressed
dx

 1 = a11 x1 + a12 x2 + · · · + a1n xn + b1 (t)




 dt
 dx2


= a21 x1 + a22 x2 + · · · + a2n xn + b2 (t)
dt
 ..
.




dx

n

= an1 x1 + an2 x2 + · · · + ann xn + bn (t)


dt
where aij real constants. In matrix form
 0      
x1 ( t ) a11 a12 . . . a1n x1 ( t ) b1 (t)
 x 0 (t)   a21 a22 . . . a2n   x2 (t)   b1 (t) 
 2  
 ..  =    ..  +  ..  ,
   
..
 .   .  .   . 
0
xn (t) an1 an2 . . . ann xn (t) bn ( t )

in short
0
x ( t ) = A x ( t ) + b ( t ). (3)

For solving, we generalize the known method for solving equations of the first order.
First, we solve the associated homogeneous system x0 = Ax. Can be shown that the general
solution is
xh = e At c̃

4
  
c̃1
∞
1  c̃2 
 
where the exponential matrix is defined as e At = ∑ k!
( At)k , and c̃ =  ..  is a column
.
k =0
c̃n
matrix of constants.
In a second step, by any method, we calculate a particular solution x p of system of (7),
and hence the general solution we are seeking is

x = xh + x p

3 How to compute the exponential matrix eAt for n = 2?

 
λ1 0
• If matrix A is diagonalizable, A = = PDP−1 , being D
the diagonal matrix
0 λ2


of (real or complex) eigenvalues and P = v w , the change-of-basis matrix (v, w
eigenvectors, of course). Hence
c
z }| {
At
xh = e c̃ = Pe Dt
P−1 c̃ = Pe Dt c = c1 eλ1 t v + c2 eλ2 t w

• If the 2 × 2 matrix A is not diagonalizable. We know that there exists a unique (real)
eigenvalue λ and a uniqueindependent
 eigenvector v. It is possible to prove that
λ 1
there exists a matrix J = and an invertible matrix called Matrix of Jordan
0 λ
Change of Basis. P = v w , where w is a vector that verifies1 Aw = v + λw, so

that A = PJP−1 .
   
λ 0 0 1
Matrix J can be expressed J = + = D + N.
0 λ 0 0
Proposition 3.1. If matrices A and B commute, i.e. A · B = B · A, then

e A+ B = e A e B .

That property does not hold for no commuting matrices.

Matrices Dt and Nt commute therefore

c
z }| {
xh = e At c̃ = Pe Jt P−1 c̃ = Pe Dt e Nt c.
   
eλt 0 1 t
But e Dt = and e Nt = ∑∞
k =0
1
k! ( Nt )
k = because ( Nt)2 = 0, so
0 eλt 0 1
 
Dt c1 + c2 t
xh = Pe = (c1 + c2 t)eλt v + c2 eλt w.
c2

How to find a particular solution x p ? Similar methods to one-equation case are used.
The particular case of system of two equations will be studied below.
1 Vector w is found by solving the algebraic system ( A − λI )w = v.

5
3.1 Examples
Here are three examples in different situations.
Example 3.2 (Non-homogeneous, diagonalizable matrix). Solve the Cauchy problem

dx


 = 4x − 2y + 1  
 dt

dy
= 3x − y + t  


 dt
with x (0) = 1, y(0) = 0


First step. Solve the associated homogeneous system

x 0 = 4x − 2y
  0   
x 4 −2 x
⇐⇒ =
y0 = 3x − y y0 3 −1 y

4−λ −2
The characteristic polynomial is = λ2 − 3λ + 2 = (λ − 2)(λ − 1).
3 −1 − λ
The eigenvectors are:
  
2 −2 v1
• For λ = 2, = 0 =⇒ v = (1, 1).
3 −3 v2
  
3 −2 w1
• For λ = 1, = 0 =⇒ w = (2, 3).
3 −2 w2
The general solution of the homogeneous associates is
 −1   
et 0
    t  
At 2 1 2 1 c̃1 2 1 e 0 c1
xh (t) = e c̃ = =
3 1 0 e2t 3 1 c̃2 3 1 0 e 2t c2
   
2 1
= c1 e t + c2 e2t
3 1

therefore
(
xh (t) = 2c1 et + c2 e2t
yh (t) = 3c1 et + c2 e2t

Second step. Find a particular solution using the indeterminate coefficients method. Sup-
posing
 
x p = at + b
xp =
y p = ct + d

is a solution, then



 a = −1
 
 b = −1
a = (4a − 2c)t − 2d + 4b + 1
=⇒
c = (3a − c + 1)t + t − d + 3b 
 c = −2

 d = −1


Third step. General solution of the system is

(
x (t) = c1 et + 2c2 e2t − t − 1
y(t) = c1 et + 3c2 e2t − 2t − 1.

6
 Replacing the initial conditions (t = 0) it produces the next linear algebraic system
 (
c1 + 2c2 − 1 = 1 c1 = 4
=⇒
c1 + 3c2 − 1 = 0 c2 = −1
and, therefore, the solution is
x (t) = 4e2t − 2et − t − 1
y(t) = 4e2t − 3et − 2t − 1

Example 3.3 (Homogeneous, two different complex eigenvalues). Integrate the differential
system equation (
x 0 − 2x + y = 0
y0 − x − 2y = 0
and solve the Cauchy problem with x (0) = y(0) = 2.  
2 −1
This differential equations system is homogeneous and its matrix is with
1 2
eigenvalues:
• λ1 = 2 − i, eigenvector v = (1, i ).
• λ1 = 2 + i, eigenvector v = (1, −i ).
The general solution is
x (t) = c1 e(2−i)t + c2 e(2+i)t = e2t [(c1 + c2 ) cos t + i (c2 − c1 ) sin t]
y(t) = ic1 e(2−i)t − ic2 e(2+i)t = e2t [(c1 + c2 ) sin t + i (c1 − c2 ) cos t.]
For solving the Cauchy problem replace the initial conditions, then
 
c1 + c2 = 2 c1 = 1 − i
=⇒
ic1 − ic2 = 2 c2 = i + 1
and hence
x (t) = 2e2t (cos t − sin t)
y(t) = 2e2t (cos t + sin t)
Example 3.4 (Homogeneous, non diagonalizable matrix). Find the general solution of the
differential equations system )
dx
dt = x − y
dy
dt = x + 3y

1 − λ −1
Diagonalize the matrix: = λ2 − 4 λ + 4 = ( λ − 2)2 .
1 3−λ
The unique eigenvalue is λ = 2 and the unique independent eigenvector is v = (1, −1).
To compute the second vector w, solve the algebraic system
    
−1 −1 w1 1
( A − 2I )w = v =⇒ = =⇒ w1 + w2 = −1
1 1 w2 −1
By simplicity, we choose w = (−1, 0), and
   
2t 1 2t −1
x = ( c1 + c2 t ) e + c2 e .
−1 0
Hence
x = (c1 + c2 t) e2t − c2 e2t
y = − (c1 + c2 t) e2t

7
4 How to compute the exponential matrix eAt for n ≥ 3?
For a homogeneous system x0 = Ax the solution is

xh = e At c

is obtained in a similar way as in the case of dimension 2. The most significant differences
occur in cases where the matrix A is not diagonalizable.
The Jordan matrices for dimension greater than 2 involve many more situations than in
the two-dimensional case previously studied. For this reason, we will not develop it in this
course and I refer to the very large literature of linear algebra.
There is a method, called the Putzer Algorithm, for computing the exponential e At that
avoids the complexity of Jordan matrices.

Lemma 4.1 (Cayley-Hamilton theorem). Given a matrix A of dimension n × n with character-

istic polynomial p(λ) = (−1)n (λn + a1 λn−1 + · · · + an−1 λ + an ), then

An + a1 An−1 + . . . an−1 A + an I = 0.

In other words, if λ1 , λ2 , . . . λn are the eigenvalues of A, then

( A − λ1 I )( A − λ2 I ) · · · ( A − λn I ) = 0 (4)

Theorem 4.2 (Putzer Algorithm). Given a matrix A with eigenvalues λ1 , λ2 , . . . , λn (not neces-
sarily distinct), then
n −1
e At = ∑ r k + 1 ( t ) Mk ,
k =0

being: 
 M0 = I,

k
 Mk =
 ∏ ( A − λ j I ) = M j −1 ( A − λ k I ), k = 1, 2, . . . n − 1
j =1

and the functions rk (t) are the solutions of the following first order equations system
 0   
r1 λ1 0 0 ... 0 r1

 r1 (0) = 1,
 r2   1 λ2 0 . . . 0   r2  

     r2 (0) = 0,

 r3 
  =0 1 λ3 . . . 0    r3  with initial conditions
  
 ..   .. ..
.. .. . . ..   ..   .
.  . . . . .  .  


rn (0) = 0.

rn 0 ... 0 1 λn rn
−1
Proof. Let S(t) = ∑nk= 0 rk +1 ( t ) Mk , then
−1
• S(0) = ∑nk= 0 rk +1 (0) Mk = M0 = I.
−1
• A · S(t) = r1 (t) A + ∑nk= 1 rk+1 AMk , but AMk = ( A − λk +1 I + λk +1 I ) Mk = Mk +1 +
λk+1 Mk , therefore

A · S = λ1 r1 I + (r1 + λ2 r2 ) M1 + · · · + (rn−1 + λn rn ) Mn−1 + 

r n
Mn

because Mn = 0 by Cayley-Hamilton Theorem.

• S0 (t) = r10 (t) I + r20 (t) M1 + · · · + rn0 (t) Mn−1 , therefore

S0 = λ1 r1 I + (r1 + λ2 r2 ) M1 + · · · + (rn−1 + λn rn ) Mn−1

8
and, hence S(t) = e At .

Example 4.3. Let us go to solve the system in exercise 6.4 using the Putzer algorithm.
 0     
x 1 −1 x 1 −1
Remember = . Then A = , with eivengalues λ = 2
y 1 3 y 1 3
(double).    
1 0 −1 −1
M0 = , M1 = A − 2I =
0 1 1 1
The functions are obtained from
 0  (
r1 (t) = e2t ,
   
r1 2 0 r1 2r1
= = ; r1 (0) = 1, r2 (0) = 0 =⇒
r2 1 2 r2 r1 + 2r2 r2 (t) = te2t

Hence
− (c1 + c2 ) te2t + c1 e2t
   
At x
e · c = r1 M0 + r2 M1 = =
(c1 + c2 ) te2t + c2 e2t y
 
0 −1 −1
Example 4.4. Calculate the exponential e At for the matrix A = −2 1 −1.
−2 2 2
First, calculate the eigenvalues λ1 = −1, λ2 = 2 (double), not diagonalizable. So,
applying the Putzer algorithm, obtain
       
1 0 0 1 −1 −1 2 −2 0 0 0 0
M0 = 0 1 0 , M1 = −2 2 −1 , M2 =  2 −2 0 , M3 = 0 0 0
0 0 1 −2 2 3 −6 6 0 0 0 0

and
e2t e−t te2t e2t e−t
r1 ( t ) = e t , r2 ( t ) = − , r3 ( t ) = − + .
3 3 3 9 9
Finally, the exponential matrix is
 −t −t

2te2t 2t 2t 2t e−t e2t
3 + e9 + 8e9 − 2te3 − e9 + e9 3 − 3
e At = r1 M0 + r2 M1 + r3 M2 =
 2te2t 8e2t 8e−t 2t 2t −t e−t e2t 
 3 − 9 + 9 − 2te3 + 8e9 + e9 3 − 3
.
−2te2t 2te2t e2t

5 First Order Systems

Although a first order system of ODEs may be a more general form, usually are written

dx1


 = f 1 (t, x1 , x2 , . . . , xn )



 dt
 dx2


= f 2 (t, x1 , x2 , . . . , xn )
dt
 ..
.




dx

 n = f n (t, x , x2 , . . . , xn )


1
dt
and denoting
x = ( x1 , x2 , . . . , xn ) and f = ( f 1 , f 2 , . . . , f n ),

9
 dx
this system can be written = f(t, x). General solutions express parametric family of
dt
curves of R . The named problem of initial values
n

(
dx
dt = f ( t, x )
x ( t 0 ) = x 0 ∈ Rn

∂ fi
determines a unique curve (particular solution) when f is a continuous function with ∂x j is
continuous for every i, j.
Sometimes a system of differential equations is expressed in symmetric form

dx1 dx2 dxn dt

= = ··· = =
P1 (t, x1 , . . . , xn ) P1 (t, x2 , . . . , xn ) Pn (t, x1 , . . . , xn ) Q(t, x1 , . . . , xn )

5.1 First Order Linear Systems

A first order system is linear if it is expressed

dx1


 = a11 (t) x1 + a12 (t) x2 + · · · + a1n (t) xn + b1 (t)
 dt



 dx2


= a21 (t) x1 + a22 (t) x2 + · · · + a2n (t) xn + b2 (t)
dt
 ..
.




dx

 n = a (t) x + an2 (t) x2 + · · · + ann (t) xn + bn (t)


n1 1
dt
In matrix form is
x 0 ( t ) = A ( t ) x ( t ) + b ( t ).
A Cauchy problem consists in finding solutions of the system satisfying the so-called initial
conditions x1 (t0 ) = x̄1 , x2 (t0 ) = x̄2 , . . . , xn (t0 ) = x̄n . It is expressed as

x0 (t) = A(t)x(t) + b(t), with


(5)
x1 (t0 ) = x̄1 , x2 (t0 ) = x̄2 , . . . , xn (t0 ) = x̄n

Theorem 5.1. Let A(t), b(t), respectively, a n × n-matrix a an n × 1-vector functions with con-
tinuous derivatives in an open interval ( a, b) ∈ R. For t0 ∈ ( a, b), the Cauchy problem (5) has an
unique solution.

Any first order linear system with n variables is equivalent to a linear ODE of order n.
To see this, consider the ODE

y ( n ) + a 1 ( x ) y ( n −1) + · · · + a n −1 ( x ) y 0 + a n y = p ( x )

and the changes of variable y1 = y, y2 = y0 , …, yn = y(n−1) produce the system

 0
 y1 = y2
 y 0 = y3


2
..

 .
 0

y n = p ( x ) − a n ( x ) y 1 − · · · − a 2 ( x ) y n −1 − a 1 ( x ) y n

Converse can be made by successive derivatives and replacements.

10
Example 5.2. Express as an unique ODE the next system
(
x10 = tx1 − x2 + t2
 0      2
x1 t −1 x1 t
⇐⇒ = + (6)
x20 = (1 − t) x1 + t2 x2 . x20 1 − t t2 x2 0

Derive the first equation and replace x20 from the second equation to obtain

x100 = x1 + t x10 − x20 + 2t =⇒ x100 = t x10 + t x1 − t2 x2 + 2t

Finally, using the first equation again, eliminate x2 from this, and you have obtain the
searched second order linear ODE

x100 − (t2 + t) x10 + (t3 − t) x1 = −t4 + 2 t

equivalent to the system (6).

5.2 First Order Linear Systems with constant coefficients

A first order system is linear with constant coefficients is expressed

dx1


 = a11 x1 + a12 x2 + · · · + a1n xn + b1 (t)



 dt
 dx2


= a21 x1 + a22 x2 + · · · + a2n xn + b2 (t)
dt
 ..
.




dx

n

= an1 x1 + an2 x2 + · · · + ann xn + bn (t)


dt
where aij real constants. In matrix form

x10 (t)
      
a11 a12 . . . a1n x1 ( t ) b1 (t)
 x 0 (t)   a21 a22 . . . a2n   x2 (t)   b1 (t) 
 2  
 ..  =    ..  +  ..  ,
   
..
 .   .  .   . 
0
xn (t) an1 an2 . . . ann xn (t) bn ( t )

in short
0
x ( t ) = A x ( t ) + b ( t ). (7)

For solving, we generalize the known method for solving equations of the first order.
First, we solve the associated homogeneous system x0 = Ax. Can be shown that the general
solution is
xh = e At c̃
 
c̃1
∞
1  c̃2 
 
where the exponential matrix is defined as e At = ∑ ( At)k , and c̃ =  .  is a column
k =0
k!  .. 
c̃n
matrix of constants.
In a second step, by any method, we calculate a particular solution x p of system of (7),
and hence the general solution we are seeking is

x = xh + x p

11
6 How to compute the exponential matrix eAt for n = 2?
 
λ1 0
• If matrix A is diagonalizable, A = = PDP−1 , being D
the diagonal matrix
0 λ2


of (real or complex) eigenvalues and P = v w , the change-of-basis matrix (v, w
eigenvectors, of course). Hence
c
z }| {
At
xh = e c̃ = Pe Dt
P−1 c̃ = Pe Dt c = c1 eλ1 t v + c2 eλ2 t w

• If matrix A is not diagonalizable. We know that there exists an unique (real) eigen-
value λ and an unique independent
 eigenvector v. It is possible to prove that there
λ 1
exists a matrix2 J = and a invertible matrix, called Jordan change-of-basis ma-
0 λ
trix P = v w , being w a vector verifying3 Aw = v + λw, such that A = PJP−1 .


   
λ 0 0 1
Matrix J can be expressed J = + = D + N.
0 λ 0 0
Proposition 6.1. If matrices A and B commute, i.e. A · B = B · A, then

e A+ B = e A e B .

That property does not hold for no commuting matrices.

Matrices Dt and Nt commute therefore

c
z }| {
xh = e At c̃ = Pe Jt P−1 c̃ = Pe Dt e Nt c.
   
eλt 0 1 t
But e Dt = and e Nt = ∑∞ 1
k =0 k! ( Nt )
k = because ( Nt)2 = 0, so
0 eλt 0 1
 
Dt c1 + c2 t
xh = Pe = (c1 + c2 t)eλt v + c2 eλt w.
c2

How to find a particular solution x p ? Similar methods to one-equation case are used.
The particular case of system of two equations will be studied below.

6.1 Examples
We present three examples in different situations.

Example 6.2 (Non-homogeneous, diagonalizable matrix). Solve the Cauchy problem


dx


 = 4x − 2y + 1


 dt

dy
= 3x − y + t  


 dt
with x (0) = 1, y(0) = 0

2 Matrix J is called Matrix of Jordan.
3 Vector w is found solving the algebraic system ( A − λI )w = v.

12
First step. Solve the associated homogeneous system

x 0 = 4x − 2y
  0   
x 4 −2 x
⇐⇒ =
y0 = 3x − y y0 3 −1 y

4−λ −2
The characteristic polynomial is = λ2 − 3λ + 2 = (λ − 2)(λ − 1).
3 −1 − λ
The eigenvectors are:
  
2 −2 v1
• For λ = 2, = 0 =⇒ v = (1, 1).
3 −3 v2
  
3 −2 w1
• For λ = 1, = 0 =⇒ w = (2, 3).
3 −2 w2
The general solution of the homogeneous associates is
 −1   
et 0
    t  
At 2 1 2 1 c̃1 2 1 e 0 c1
xh (t) = e c̃ = =
3 1 0 e2t 3 1 c̃2 3 1 0 e 2t c2
   
2 1
= c1 e t + c2 e2t
3 1

therefore
(
xh (t) = 2c1 et + c2 e2t
yh (t) = 3c1 et + c2 e2t

Second step. Find a particular solution using the indeterminate coefficients method. Sup-
posing
 
x p = at + b
xp =
y p = ct + d

is a solution, then



 a = −1
 
 b = −1
a = (4a − 2c)t − 2d + 4b + 1
=⇒
c = (3a − c + 1)t + t − d + 3b  c = −2


 d = −1


Third step. General solution of the system is

(
x (t) = c1 et + 2c2 e2t − t − 1
y(t) = c1 et + 3c2 e2t − 2t − 1.

Replacing the initial conditions (t = 0) it produces the next linear algebraic system
 (
c1 + 2c2 − 1 = 1 c1 = 4
=⇒
c1 + 3c2 − 1 = 0 c2 = −1

and, therefore, the solution is

x (t) = 4e2t − 2et − t − 1

y(t) = 4e2t − 3et − 2t − 1

13
Example 6.3 (Homogeneous, two different complex eigenvalues). Integrate the differential
system equation (
x 0 − 2x + y = 0
y0 − x − 2y = 0
and solve the Cauchy problem with x (0) = y(0) = 2.  
2 −1
This differential equations system is homogeneous and its matrix is with
1 2
eigenvalues:

• λ1 = 2 − i, eigenvector v = (1, i ).

• λ1 = 2 + i, eigenvector v = (1, −i ).

The general solution is

x (t) = c1 e(2−i)t + c2 e(2+i)t = e2t [(c1 + c2 ) cos t + i (c2 − c1 ) sin t]

y(t) = ic1 e(2−i)t − ic2 e(2+i)t = e2t [(c1 + c2 ) sin t + i (c1 − c2 ) cos t.]

For solving the Cauchy problem replace the initial conditions, then
 
c1 + c2 = 2 c1 = 1 − i
=⇒
ic1 − ic2 = 2 c2 = i + 1

and hence

x (t) = 2e2t (cos t − sin t)

y(t) = 2e2t (cos t + sin t)

Example 6.4 (Homogeneous, non diagonalizable matrix). Find the general solution of the
differential equations system )
dx
dt = x − y
dy
dt = x + 3y

1 − λ −1
Diagonalize the matrix: = λ2 − 4 λ + 4 = ( λ − 2)2 .
1 3−λ
The unique eigenvalue is λ = 2 and the unique independent eigenvector is v = (1, −1).
To compute the second vector w solve the algebraic system
    
−1 −1 w1 1
( A − 2I )w = v =⇒ = =⇒ w1 + w2 = −1
1 1 w2 −1

By simplicity, we choose w = (−1, 0), and

   
2t 1 2t −1
x = ( c1 + c2 t ) e + c2 e .
−1 0

Hence

x = (c1 + c2 t) e2t − c2 e2t

y = − (c1 + c2 t) e2t

14
7 How to compute the exponential matrix eAt for n ≥ 3?
For a homogeneous system x0 = Ax the solution

xh = e At c

is obtained in a similar way to the dimension 2 case. The most significant differences are
produced in cases where matrix A is not diagonizable.
The Jordan matrices for dimension higher than 2 involves a many more situations than
in the two-dimensional case previously studied. For that reason, we are not going to develop
it in this course and I refer to the very wide literature of linear algebra.
There exists a method, called Putzer Algorithm for calculating the exponential e At bypass-
ing the complexity of the Jordan matrices.

Lemma 7.1 (Cayley-Hamilton theorem). Given a matrix A of dimension n × n with character-

istic polynomial p(λ) = (−1)n (λn + a1 λn−1 + · · · + an−1 λ + an ), then

An + a1 An−1 + . . . an−1 A + an I = 0.

In other words, if λ1 , λ2 , . . . λn are the eigenvalues of A, then

( A − λ1 I )( A − λ2 I ) · · · ( A − λn I ) = 0 (8)

Theorem 7.2 (Putzer Algorithm). Given a matrix A with eigenvalues λ1 , λ2 , . . . , λn (not neces-
sarily distinct), then
n −1
e At = ∑ r k + 1 ( t ) Mk ,
k =0

being: 
 M0 = I,

k
 Mk =
 ∏ ( A − λ j I ) = M j −1 ( A − λ k I ), k = 1, 2, . . . n − 1
j =1

and the functions rk (t) are the solutions of the following first order equations system
 0   
r1 λ1 0 0 ... 0 r1

 r1 (0) = 1,
 r2   1 λ2 0 . . . 0   r2  

     r2 (0) = 0,

 r3 
  =0 1 λ3 . . . 0    r3  with initial conditions
  
 ..   .. ..
.. .. . . ..   ..   .
.  . . . . .  .  


rn (0) = 0.

rn 0 ... 0 1 λn rn
−1
Proof. Let S(t) = ∑nk= 0 rk +1 ( t ) Mk , then
−1
• S(0) = ∑nk= 0 rk +1 (0) Mk = M0 = I.
−1
• A · S(t) = r1 (t) A + ∑nk= 1 rk+1 AMk , but AMk = ( A − λk +1 I + λk +1 I ) Mk = Mk +1 +
λk+1 Mk , therefore

A · S = λ1 r1 I + (r1 + λ2 r2 ) M1 + · · · + (rn−1 + λn rn ) Mn−1 + 

r n
Mn

because Mn = 0 by Cayley-Hamilton Theorem.

• S0 (t) = r10 (t) I + r20 (t) M1 + · · · + rn0 (t) Mn−1 , therefore

S0 = λ1 r1 I + (r1 + λ2 r2 ) M1 + · · · + (rn−1 + λn rn ) Mn−1

15
and, hence S(t) = e At .

Example 7.3. Let us go to solve the system in exercise 6.4 using the Putzer algorithm.
 0     
x 1 −1 x 1 −1
Remember = . Then A = , with eivengalues λ = 2
y 1 3 y 1 3
(double).    
1 0 −1 −1
M0 = , M1 = A − 2I =
0 1 1 1
The functions are obtained from
 0  (
r1 (t) = e2t ,
   
r1 2 0 r1 2r1
= = ; r1 (0) = 1, r2 (0) = 0 =⇒
r2 1 2 r2 r1 + 2r2 r2 (t) = te2t

Hence
− (c1 + c2 ) te2t + c1 e2t
   
At x
e · c = r1 M0 + r2 M1 = =
(c1 + c2 ) te2t + c2 e2t y
 
0 −1 −1
Example 7.4. Calculate the exponential e At for the matrix A = −2 1 −1.
−2 2 2
First, calculate the eigenvalues λ1 = −1, λ2 = 2 (double), not diagonalizable. So,
applying the Putzer algorithm, obtain
       
1 0 0 1 −1 −1 2 −2 0 0 0 0
M0 = 0 1 0 , M1 = −2 2 −1 , M2 =  2 −2 0 , M3 = 0 0 0
0 0 1 −2 2 3 −6 6 0 0 0 0

and
e2t e−t te2t e2t e−t
r1 ( t ) = e t , r2 ( t ) = − , r3 ( t ) = − + .
3 3 3 9 9
Finally, the exponential matrix is
 −t −t

2te2t 2t 2t 2t e−t e2t
3 + e9 + 8e9 − 2te3 − e9 + e9 3 − 3
e At = r1 M0 + r2 M1 + r3 M2 =
 2te2t 8e2t 8e−t 2t 2t −t e−t e2t 
 3 − 9 + 9 − 2te3 + 8e9 + e9 3 − 3
.
−2te2t 2te2t e2t

16
Exercises
Exercise 1
Let be x = x (t) e y = y(t). Solve the following system of differential equations using the
Laplace method: (
x0 = y
con x (0) = 0, y(0) = 1.
y0 = 2x − 2y
√ √ √ √
3 −t
Solutions: x (t) = √1 e−t sinh( 3t), y(t) = e−t cosh( 3t) − 3 e sinh( 3t).
3

Exercise 2
Solve the following systems of differential equations using the Laplace method:
x 0 + y0 = t

a) with x (0) = 3, x 0 (0) = −2, y(0) = 0.
x 00 − 2y = e−t
3x 0 + y + 2x = 1

b) with x (0) = y(0) = 0.
x 0 + 4y0 + 3y = 0
Solutions:
 √  √ √   2
√ 
cos 2t 5 2 sin 2t 2 1−e− 3 t cos 62 t
 1 −
 x (t) = 3 e +
 t − +t+ 5 
 x (t) =

6 6 2 2
a) √  √   √  b) √ 2t
√ 
 5 2 sin 2t cos 2t  − 2e− 3 sin 62t
− 13 e−t + 12
 
y(t) = − y(t) =
 
6 6 4

Exercise 3
Find homogeneous systems of first-order differential equations with one of their functions
as a set of solutions:
a) y = Ce x + De2x . b) y = Ax2 + Bx + C + D sin x + E cos x.
Determinate if the matrix of these systems are diagonalizable.

Exercise 4
(
x 0 + y = sin 2t
Find the general solution of the system .
y0 − x = cos 2t
1 1
Solution: x = c1 cos t + c2 sin t − 3 cos 2t; y = −c2 cos t + c1 sin t + 3 sin 2t.

Exercise 5
(
x 0 = 3x + 5y
Solve the system with x (0) = 2, y(0) = 5.
y0 = −2x − 8y
Solutionn: x = 5e2t − 3e−7t ; y = −e2t + 6e−7t .

Exercise 6
Let x = x (t), y = y(t). Find the general solution of the second-order differential equation
system by transforming it into an equivalent first-order system and using the matrix of the
system.
(
x 00 + y0 = t
y00 − x 0 = 1 − t
Solution: x = 12 t2 + A cos t + B sin t; y = 12 t2 − t + A sin t − B cos t.

17
Exercise 7
(
5x 00 + y0 + 2x = 4 cos t
Solve the second-order system .
3x 0 + y = 8t cos t
Solution: x = c1 cos t + c2 sin t − t2 cos t; y = 3c1 sin t − 3c2 cos t + 14t cos t − 3t2 sin t.

Exercise 8
dx


 =y



 dt
dy

Solve the homogeneous system =x with x (0) = y(0) = z(0) = 1.


 dt
 dz = z



dt
Solution: x (t) = y(t) = z(t) = et .

Exercise 9
Solve the problem of the Example 1.1 at page 1 (Coupled Harmonic Oscillator) for the case
where the masses are m1 = m2 = 1, the Hooke’s constants are k1 = k2 = 1 and k3 = 2 and
with the following initial conditions:

x1 (0) = 0; x10 (0) = 1; x2 (0) = 0; x20 (0) = 0.

√  √ 
sin 5t sin (t) sin (t) sin 5t
Solution: x1 = √ + 2 ; x2 = 2 − √ .
2 5 2 5

Exercise 10
Solve the following non linear systems of differential equations:

dx y2

dx 
=y  dt = x

 

dt

a) b)
 dy y 2  dy x2

 =

 =
dt x dt y


dx dy dz
c) = =
x (y − z) y(z − x ) z( x − y)
Solutions:

a) x = c2 ec1 t y y = c1 c2 ec1 t .
p p
b) x = c2 cosh(2t + c1 ) y y = c2 sinh(2t + c1 ).
(
x + y + z = c1
c) Familia de curvas .
xyz = c2

18