Systems of Differential Equations - Better than the Book, I Hope

1. The Situation: We know that if a population grows by 40% per year that if x1 is the
1
population then dx
dt = 0.4x1 . This is because the change (derivative) is 40% of the original.
So suppose there are two populations, x1 and x2 . Suppose each year x1 grows by 10% of its
own population but suppose it also grows by 120% of x2 s population - perhaps x2 grows a lot
1
and most move to x1 . Then wed have dx
dt = 0.1x1 + 1.2x2 . Suppose in addition x2 grows by
2
40% of x1 s population and by 30% of its own. Then wed have dx
dt = 0.4x1 + 0.3x2 .
Together we call this a system of differential equations:
dx1
= 0.1x1 + 1.2x2
dt
dx2
= 0.4x1 + 0.3x2
dt

A solution is a pair of functions x1 = ... and x2 = ..., both functions of time t, which satisfy
both. How on earth would we find this?
2. Harder Still: What if both populations are additionally affected by some outside influence.
For example perhaps x1 gains et new people each year from some third source and x2 gains 7
new people each year from some third source. Then we have
dx1
= 0.1x1 + 1.2x2 + et
dt
dx2
= 0.4x1 + 0.3x2 + 7
dt

How would we solve this?! Well see!

3. Theorem: There is a basic premise underlying this section which we will mention but not
prove. The proof requires some serious linear algebra.
 
a
and 2
Suppose M is a 2 2 matrix. Suppose 1 is an eigenvalue with some eigenvector
b
 
c
is an eigenvalue with some eigenvector
. Then it turns out that:
d
1



a c 1 0 a c
M=
b d 0 2 b d
Well write
M = P DP





a b
and D = 1
with P =
0
c d

0
2

4. Example of Theorem: For example if we start with



0.1 1.2
M=
0.4 0.3
 
3
then we find the eigenvalues and eigenvectors we get (work omitted) 1 = 0.9 with
and
2
 
2
. Thus we can check that the following is in fact true - check it:
2 = 0.5 with
1
1


 

3 2
0
3 2 0.9
0.1 1.2
=
0 0.5 2 1
2 1
0.4 0.3

5. Thats Crazy Talk! Now What? Suppose we have a system of differential equations of
the form
dx1
= m11 x1 + m12 x2 + q1 (t)
dt
dx2
= m21 x1 + m22 x2 + q2 (t)
dt
We can rewrite this with matrices and vectors as
 dx1  

  
m11 m12 x1
q1 (t)
dt
=
+
dx2
m21 m22 x2
q2 (t)
dt
Then we can define X =
system as

 

x1
m11
, M =
x2
m21

m12
m22

and Q =



q1 (t)
and then rewrite this
q2 (t)

dX
= MX + Q
dt
Now lets play a bit. Define Y = P 1 X so that X = P Y . Since P is all constants we have
dX
dY
dt = P dt . Then we work with the equation above:
dX
dt
dY
P
dt
dY
dt
dY
dt

= MX + Q
= P DP 1 X + Q
= DP 1 X + P 1 Q
= DY + P 1 Q

1
Q then we can find X = P Y . In practice because D
In theory if we can solve dY
dt = DY + P
is diagonal it turns out that solving this is often not hard at all (famous last words.)

6. May I See an Example? Consider for example:

dx1
= 0.1x1 + 1.2x2 + et
dt
dx2
= 0.4x1 + 0.3x2 + 7
dt

We have M =


0.1
0.4


1.2
.
0.3

The eigenstuff gives us D =

 t
e
.
We have Q =
7

0.9
0




3 2
0
so P 1 =
and P =
2 1
0.5

1
7

1
2


2
.
3

We will therefore solve

dY
= DY + P 1 Q
dt
 



dY
1 1 2 et
0.9
0
Y +
=
0 0.5
dt
7 2 3 7
 dy1  
   1 t

y1
0.9
0
dt
7e + 2
+
=
dy2
0 0.5 y2
27 et + 3
dt

 dy1  
  1 t
e +2
0.9y1
dt
7
+
dy2 =
0.5y2
27 et + 3
dt
 dy1  

0.9y1 + 71 et + 2
dt
=
dy2
0.5y2 27 et + 3
dt

which corresponds to the system

1
dy1
= 0.9y1 + et + 2
dt
7
dy2
2
= 0.5y2 et + 3
dt
7
These are actually first-order linear if we view them as:
1
dy1
0.9y1 = et + 2
dt
7
2
dy2
+ 0.5y2 = et + 3
dt
7

Remember: First order linear means of the following form where weve used t in place of x:
dy
+ P (t)y = Q(t)
dt

The first has s(t) = 0.9t and hence solution


Z 
1 t
y1 = e0.9t
e + 2 e0.9t dt
7
Z
1
e0.1t + 2e0.9t dt
= e0.9t
7


20 0.9t
0.9t 10 0.1t
e
e
+ C1
=e
7
9
10 t 20
=
e
+ C1 e0.9t
7
9
The second has s(t) = 0.5t and hence solution

Z 
2 t
0.5t
e + 3 e0.5t dt
y2 = e
7

Z 
2 1.5t
0.5t
0.5t
e
dt
=e
+ 3e
7


4 1.5t
0.5t
0.5t
e
=e
+ 6e
+ C2
21
4
= et + 6 + C2 e0.5t
21
So then in closing
 
x1
=X
x2
 
x1
= PY
x2

  

10 t
20
0.9t
x1
3 2 7 e 9 + C1 e

=
4 t
x2
2 1
21
e + 6 + C2 e0.5t

 
20
4 t
t
0.9t
0.5t
3 10
e

+
C
e

e
+
6
+
C
e
x1
1
2
9
21
= 7




20
4 t
t
0.9t
0.5t
x2
2 10
e

+
C
e
+

e
+
6
+
C
e
1
2
7
9
21

 
56
14 t
0.9t
0.5t
e
+
3C
e

2C
e

x1
1
2
3
=3
8 t
x2
e + 2C e0.9t + C e0.5t + 14
3

So when all is said and done:

14 t
56
e + 3C1 e0.9t 2C2 e0.5t
3
3
8 t
14
0.9t
0.5t
x2 = e + 2C1 e
+ C2 e
+
3
9
x1 =

7. Initial Values! Suppose in addition we know that x1 (0) = 100 and x2 (0) = 200 we can plug
these in and solve for both C1 and C2 , giving us the specific solution.
14
56
+ 3C1 2C2
3
3
14
8
200 = + 2C1 + C2 +
3
9
100 =

which yields
C1 =

650
9

and C2 =

154
3

And so finally




154 0.5t 56
650 0.9t
14 t
e
2
e

e +3
3
9
3
3




154 0.5t 14
8 t
650 0.9t
e
+
e
+
x2 = e + 2
3
9
3
9

x1 =

8. Holy Macaroni! Okay, this was pretty rough but mostly because of the Q business. Without
that its actually really easy, for example if q1 = q2 = 0 its simple and if theyre constants its
still not so bad.
9. Bigger! Whats more, all of this is true with 3 3 matrices. What fun!
10. Summary: In summary the method is as follows:
(a) Let M be the matrix which corresponds to the coefficients of x1 and x2 in the system.
(b) Find the eigenvalues and eigenvectors and create the matrices D and P . Then find P 1 .
(c) Solve the two equations forming
first-order linear.

dY
dt

= DY + P 1 Q, these are (for us) separable or

(d) Use X = P Y to find X, thereby finding x1 and x2 .

11. Here is another easier example so you can see that this isnt always so bad!
Suppose
dx1
= 0.6x1 + 0.8x2
dt
dx2
= 0.9x1
dt

We have M =


0.6
0.9


0.8
.
0

The eigenstuff (work omitted) gives us D =



1 3 2
18
.
3 4
 
0
We have Q =
.
0


1.2
0

0
0.6

and P =


4 2
so P 1 =
3 3

We will therefore solve

dY
= DY + P 1 Q
dt



 
1 3 2 0
1.2
0
=
Y
0 0.6
0
18 3 4
 

y1
1.2
0
=
0 0.6 y2


1.2y1
=
0.6y2

which corresponds to the system

dy1
= 1.2y1
dt
dy2
= 0.6y2
dt
These are separable. In fact these are of the form
y = Cekt . Thus we have solutions

dy
dt

= ky and we know the solution is

y1 = C1 e1.2t
y2 = C2 e0.6t

So then

4 2
X = PY =
3 3




 
4C1 e1.2t + 2C2 e0.6t
C1 e1.2t
=
3C1 e1.2t 3C2 e0.6t
C2 e0.6t

So then the final solution is

x1 = 4C1 e1.2t + 2C2 e0.6t
x2 = 3C1 e1.2t 3C2 e0.6t