Systems of Differential Equations
Matrix Methods
Characteristic Equation
Cayley-Hamilton
Cayley-Hamilton Theorem
An Example
The Cayley-Hamilton Method for u0 = Au
A Working Rule for Solving u0 = Au
Solving 2 2 u0 = Au
Finding ~c1 and ~c2
A Matrix Method for Finding ~c1 and ~c2
Other Representations of the Solution u
Another General Solution of u0 = Au
Change of Basis Equation
Characteristic Equation
Definition 1 (Characteristic Equation)
Given a square matrix A, the characteristic equation of A is the polynomial equation
det(A rI) = 0.
The determinant det(A rI) is formed by subtracting r from the diagonal of A.
The polynomial p(r) = det(A rI) is called the characteristic polynomial.
If A is 2 2, then p(r) is a quadratic.
If A is 3 3, then p(r) is a cubic.
The determinant is expanded by the cofactor rule, in order to preserve factorizations.
Characteristic Equation Examples
Create det(A rI) by subtracting r from the diagonal of A.
Evaluate by the cofactor rule.
A=
2 3
0 4
2 3 4
A = 0 5 6 ,
0 0 7
2r
3
p(r) =
0
4r
= (2 r)(4 r)
2r
3
4
5r
6
p(r) = 0
0
0
7r
= (2r)(5r)(7r)
Cayley-Hamilton
Theorem 1 (Cayley-Hamilton)
A square matrix A satisfies its own characteristic equation.
If p(r) = (r)n + an1 (r)n1 + a0 , then the result is the equation
(A)n + an1An1 + + a1A + a0I = 0,
where I is the n n identity matrix and 0 is the n n zero matrix.
Cayley-Hamilton Example
Assume
2 3 4
A=0 5 6
0 0 7
Then
2r
3
4
5r
6
p(r) = 0
0
0
7r
= (2 r)(5 r)(7 r)
and the Cayley-Hamilton Theorem says that
0 0 0
(2I A)(5I A)(7I A) = 0 0 0 .
0 0 0
Cayley-Hamilton Method
Theorem 2 (Cayley-Hamilton Method for u0 = Au)
A component function uk (t) of the vector solution u(t) for u0 (t) = Au(t) is
a solution of the nth order linear homogeneous constant-coefficient differential
equation whose characteristic equation is det(A rI) = 0.
Let atom1 , . . . , atomn denote the atoms constructed from the characteristic equation det(A rI) = 0 by Eulers Theorem. Then constant vectors c1 , . . . , cn
exist, uniquely determined by A and u(0), such that
u(t) = (atom1)~c1 + + (atomn)~cn
A Working Rule for Solving u0 = Au
The Theorem says that u0 = Au can be solved from the formula
u(t) = (atom1)~c1 + + (atomn)~cn
The problem of solving u0 = Au is reduced to finding the vectors ~c1, . . . , ~cn.
The vectors ~c1, . . . , ~cn are not arbitrary, but instead uniquely determined by A and
u(0)!
A 2 2 Illustration
Let us solve ~
u0 = A~
u when A is the non-triangular matrix
A=
1 2
2 1
.
The characteristic polynomial is
1r
2
2
1r
= (1 r)2 4 = (r + 1)(r 3).
Eulers theorem implies solution atoms et , e3t .
Then ~
u is a vector linear combination of the solution atoms,
~
u = et~c1 + e3t~c2.
Finding ~c1 and ~c2
To solve for c1 and c2 , differentiate the above relation. Replace ~
u0 by A~
u, then set t = 0
and ~
u(0) = ~
u0 in the two formulas to obtain the relations
~
u0 = e0~c1 + e0~c2
A~
u0 = e0~c1 + 3e0~c2
Adding the equations gives ~
u0 + A~
u0 = 4~c2 and then
~c1 =
1
u0 Au0,
4
4
~c2 =
1
u0 + Au0.
4
4
A Matrix Method for Finding ~c1 and ~c2
The Cayley-Hamilton Method produces a unique solution for c1 , c2 because the coefficient
matrix
e0 e0
e0 3e0
is exactly the Wronskian W of the basis of atoms evaluated at t = 0. This same fact
applies no matter the number of coefficients ~c1 , ~c2 , . . . to be determined.
The answer for ~c1 and ~c2 can be written in matrix form in terms of the transpose W T of
the Wronskian matrix as
aug(~c1,~c2) = aug(~
u0, A~
u0)(W T )1.
Solving a 2 2 Initial Value Problem
1
1 2
, A=
.
2
2 1
1
1 2
1
3
Then ~
u0 =
, A~
u0 =
=
and
2
2 1
2
0
T !1
1 3
1 1
3/2 1/2
aug(~c1,~c2) =
=
.
2 0
1 3
3/2 1/2
~
u0 = A~
u,
~
u(0) =
The solution of the initial value problem is
~
u(t) = et
3/2
3/2
+ e3t
1/2
1/2
=
23 et + 21 e3t
3 t
e + 21 e3t
2
.
Other Representations of the Solution u
Let y1 (t), . . . , yn (t) be a solution basis for the nth order linear homogeneous constantcoefficient differential equation whose characteristic equation is det(A rI) = 0.
Consider the solution basis atom1 , atom2 , . . . , atomn . Each atom is a linear combination of y1 , . . . , yn . Replacing the atoms in the formula
u(t) = (atom1)c1 + + (atomn)cn
by these linear combinations implies there are constant vectors d1 , . . . , dn such that
u(t) = y1(t)~
d1 + + yn(t)~
dn
Another General Solution of u0 = Au
Theorem 3 (General Solution)
The unique solution of u0 = Au, u(0) = u0 is
u(t) = 1(t)u0 + 2(t)Au0 + + n(t)An1u0
where 1 , . . . , n are linear combinations of atoms constructed from roots of the
characteristic equation det(A rI) = 0, such that
Wronskian(1 (t), . . . , n (t))|t=0 = I.
Proof of the theorem
Proof: Details will be given for n = 3. The details for arbitrary matrix dimension n is an easy modification
of this proof. The Wronskian condition implies 1 , 2 , 3 are independent. Then each atom constructed from
the characteristic equation is a linear combination of 1 , 2 , 3 . It follows that the unique solution u can be
written for some vectors d1 , d2 , d3 as
u(t) = 1 (t)~
d1 + 2 (t)~
d2 + 3 (t)~
d3 .
Differentiate this equation twice and then set t = 0 in all 3 equations. The relations u0 = Au and u00 = Au0 =
AAu imply the 3 equations
u0
= 1 (0)d1 + 2 (0)d2 + 3 (0)d3
Au0 = 01 (0)d1 + 02 (0)d2 + 03 (0)d3
00
00
A2 u0 = 00
1 (0)d1 + 2 (0)d2 + 3 (0)d3
Because the Wronskian is the identity matrix I , then these equations reduce to
u0
= 1d1 + 0d2 + 0d3
Au0 = 0d1 + 1d2 + 0d3
A2 u0 = 0d1 + 0d2 + 1d3
which implies d1 = u0 , d2 = Au0 , d3 = A2 u0 .
The claimed formula for u(t) is established and the proof is complete.
Change of Basis Equation
Illustrated here is the change of basis formula for n = 3. The formula for general n is
similar.
Let 1 (t), 2 (t), 3 (t) denote the linear combinations of atoms obtained from the vector
formula
1(t), 2(t), 3(t) = atom1(t), atom2(t), atom3(t) C 1
where
C = Wronskian(atom1, atom2, atom3)(0).
The solutions 1 (t), 2 (t), 3 (t) are called the principal solutions of the linear homogeneous constant-coefficient differential equation constructed from the characteristic
equation det(A rI) = 0. They satisfy the initial conditions
Wronskian(1 , 2 , 3 )(0) = I.