LECTURE 32

SYSTEMS OF DIFFERENTIAL EQUATIONS

Systems of differential equations y0 = Ay may be easily solved by implementing the

eigenvalues and eigenvectors of A.
If A is a 3 3 matrix with linearly independent eigenvectors v1 , v2 and v3 , and associated
eigenvalues 1 , 2 and 3 , then the general solution to y0 = Ay takes the form
y = c1 v1 e1 t + c2 v2 e2 t + c3 v3 e3 t
More complicated systems may be simplified through the transformation y = P z where
P is the usual matrix of eigenvectors of A.
We start this lecture by proving the validity of the algorithms used to analyse quadratic
forms in the previous lecture.
T
x
x

y
Consider the quadratic form
A y = 1 where A is a symmetric matrix. Since
z
z
A is symmetrix it admits a full set of orthogonal eigenvectors. Let P be the matrix of unit
eigenvectors of A. The columns of P are orthogonal and also of unit length implying that
P is an
matrix. Via the usual process of diagonalisation P 1 AP = D where
orthogonal
1 0 0

0 2 0 is the diagonal matrix of eigenvalues. But since P is orthogonal we

D=
0 0 3
T
have P AP = D.

We now implement the orthogonal transformation (a rotation in space)

x
X
y = P Y .
Z
z
The quadratic form becomes
T

X
X
P Y AP Y = 1 implying that
Z
Z

X
X
Y P T AP Y = 1 and hence we have
Z
Z

X
X
X
1 0 0
X
Y D Y = 1 Y 0 2 0 Y = 1
Z
Z
Z
0 0 3
Z

As promised in the last lecture this yields the simplified form (without mixed terms) with
respect to the principal axes:

1 X 2 + 2 Y 2 + 3 Z 2 = 1
We turn now to our second major application of eigenvectors, systems of differential
equations.

SYSTEMS OF DIFFERENTIAL EQUATIONS

Example 1 Solve the system of differential equations.
y10 = 2y1 + y2
y20 = y1 + y3
y30 = y1 + y2 + y3
where y1 (0) = 6, y2 (0) = 5, and y3 (0) = 7.
We begin by noting that the system may be
0
2
y1
y20 = 1
y30
1

written in matrix form as

1 0
y1
0 1 y2
1 1
y3

which is expressed as y0 = Ay. The usual eigenanalysis yields eigenvalues 0,1, and 2 with

1
1
1

2 ,
1
0 .
associated eigenvectors
, and
1
0
1
The solution may now be simply written down as

1
1
1
y = c1 2 e0t + c2 1 e1t + c3 0 e2t
1
0
1
Reading across the rows we have a general solution:
y1 = c1 c2 et + c3 e2t
y2 = 2c1 + c2 et
y3 = c1 + c3 e2t

Lets prove that this all works:

Claim: If A is a 3 3 matrix with linearly independent eigenvectors v1 , v2 and v3 , and
associated eigenvalues 1 , 2 and 3 , then the general solution to y0 = Ay takes the form
y = c1 v1 e1 t + c2 v2 e2 t + c3 v3 e3 t
where c1 , c2 and c3 are arbitrary constants.
Proof:
Method 1: Assume a solution to y0 = Ay of the form y = vet where v is a vector and
is a number.

Method 2: Make the substitution y = P z where P is the matrix of eigenvectors of A.

It is clear from the above that once we have the eigenvectors and eigenvalues of A the
solution to the system y0 = Ay is just one step away!
3

The i.c.s are implemented at the last stage to evaluate the three arbitrary constants.
Recall that
y1 = c1 c2 et + c3 e2t
y2 = 2c1 + c2 et
y3 = c1 + c3 e2t
and that y1 (0) = 6, y2 (0) = 5, and y3 (0) = 7.
So

So we have c1 = 3,

c2 = 1,

c3 = 4.

Hence the final solution is

y1 = 3 et + 4e2t
y2 = 6 + et
y3 = 3 + 4e2t
These three functions satisfy both the system of differential equations and the i.c.s. Lets
check that the last equation y30 = y1 + y2 + y3 is satisfied:
LHS = y30 =

RHS = y1 + y2 + y3 =

F y1 = 3 et + 4e2t , y2 = 6 + et , y3 = 3 + 4e2t

In more complicated examples y0 = Ay + b our approach is to actually implement the

substitution y = P z to yield P z0 = AP z + b. We then have z0 = P 1 AP z + P 1 b
implying z0 = Dz + P 1 b. Since the diagonal matrix D has so little structure this final
system when separated out, is trivial to solve using our standard first order linear theory.
32

You can now do Q 93 and 94