Matrix Exponential. Fundamental Matrix Solution.

Objective: Solve
dx
dt
= Ax
with an n n constant coecient matrix A.
Here, the unknown is the vector function x(t) =

x
1
(t)
.
.
.
x
n
(t)

.
General Solution Formula in Matrix Exponential Form:
x(t) = e
tA

C = e
tA

C
1
.
.
.
C
n

,
where C
1
, , C
n
are arbitrary constants.
The solution of the initial value problem
dx
dt
= Ax, x(t
0
) = x
0
is given by
x(t) = e
(tt
0
)A
x
0
.
Denition (Matrix Exponential): For a square matrix A,
e
tA
=

k=0
t
k
k!
A
k
= I + tA +
t
2
2!
A
2
+
t
3
3!
A
3
+ .
1
Evaluation of Matrix Exponential in the Diagonalizable Case: Suppose that A is diago-
nalizable; that is, there are an invertible matrix P and a diagonal matrix D =

1
.
.
.

such that A = PDP

1
t
.
.
.
e
nt

P
1
.
EXAMPLE 1. Let A =

6 3 2
4 1 2
13 9 3

.
(a) Evaluate e
tA
.
(b) Find the general solutions of
dx
dt
= Ax.
(c) Solve the initial value probelm
dx
dt
= Ax, x(0) =

2
1
4

.
Solution: The given matrix A is diagonalized: A = PDP
1
with
P =

1 1 1/2
1 2 1/2
1 1 1

, D =

1 0 0
0 2 0
0 0 1

.
Part (a): We have
e
tA
= Pe
tD
P
1
=

1 1 1/2
1 2 1/2
1 1 1

e
t
0 0
0 e
2t
0
0 0 e
t

5 3 1
1 1 0
6 4 2

5e
t
e
2t
3e
t
3e
t
e
2t
2e
t
e
t
+ e
t
5e
t
+ 2e
2t
+ 3e
t
3e
t
+ 2e
2t
+ 2e
t
e
t
e
t
5e
t
+ e
2t
6e
t
3e
t
+ e
2t
4e
t
e
t
+ 2e
t

.
Part (b): The general solutions to the given system are
x(t) = e
tA

C
1
C
2
C
3

,
where C
1
, C
2
, C
3
are free parameters.
Part (c): The solution to the initial value problem is
x(t) = e
tA

2
1
4

11e
t
+ e
2t
+ 8e
t
11e
t
2e
2t
8e
t
11e
t
e
2t
+ 16e
t

.
2
Evaluation of Matrix Exponential Using Fundamental Matrix: In the case A is not
diagonalizable, one approach to obtain matrix exponential is to use Jordan forms. Here, we
use another approach. We have already learned how to solve the initial value problem
dx
dt
= Ax, x(0) = x
0
.
We shall compare the solution formula with x(t) = e
tA
x
0
to gure out what e
tA
is. We know
the general solutions of dx/dt = Ax are of the following structure:
x(t) = C
1
x
1
(t) + + C
n
x
n
(t),
where x
1
(t), , x
n
(t) are n linearly independent particular solutions. The formula can be
rewritten as
x(t) = [ x
1
(t) x
n
(t) ]

C with

C =

C
1
.
.
.
C
n

.
For the initial value problem,

C is determined by the initial condition
[ x
1
(0) x
n
(0) ]

C = x
0
=

C = [ x
1
(0) x
n
(0) ]
1
x
0
.
Thus, the solution of the initial value problem is given by
x(t) = [ x
1
(t) x
n
(t) ] [ x
1
(0) x
n
(0) ]
1
x
0
.
Comparing this with x(t) = e
tA
x
0
, we obtain
e
tA
= [ x
1
(t) x
n
(t) ] [ x
1
(0) x
n
(0) ]
1
.
In this method of evaluating e
tA
, the matrix M(t) = [ x
1
(t) x
n
(t) ] plays an essential
role. Indeed, e
tA
= M(t)M(0)
1
.
Denition (Fundamental Matrix Solution): If x
1
(t), , x
n
(t) are n linearly independent
solutions of the n dimensional homogeneous linear system dx/dt = Ax, we call
M(t) = [ x
1
(t) x
n
(t) ]
a fundamental matrix solution of the system.
(Remark 1: The matrix function M(t) satises the equation M

(t) = AM(t). Moreover,

7 9 9
3 5 3
3 3 5

.
Solution: We rst solve dx/dt = Ax. We obtain
x(t) = C
1
e
t

3
1
1

+ C
2
e
2t

1
1
0

+ C
3
e
2t

1
0
1

.
This gives a fundamental matrix solution:
M(t) =

3e
t
e
2t
e
2t
e
t
e
2t
0
e
t
0 e
2t

.
The matrix exponential is
e
tA
= M(t)M(0)
1
=

3e
t
e
2t
e
2t
e
t
e
2t
0
e
t
0 e
2t

3 1 1
1 1 0
1 0 1

1
=

3e
t
2e
2t
3e
t
3e
2t
3e
t
+ 3e
2t
e
t
+ 2e
2t
e
t
+ 2e
2t
e
t
e
2t
e
t
e
2t
e
t
e
2t
e
t
+ 2e
2t

.
EXAMPLE 3. Evaluate e
tA
for A =

5 8 4
2 3 2
6 14 5

.
Solution: We rst solve dx/dt = Ax. We obtain
x(t) = C
1
e
t

3
1
1

+ C
2
e
3t

2
1
1

+ C
3
e
3t

1 2t
1/2 + t
t

.
This gives a fundamental matrix solution:
M(t) =

3e
t
2e
3t
e
3t
2te
3t
e
t
e
3t 1
2
e
3t
+ te
3t
e
t
e
3t
te
3t

.
The matrix exponential is
e
tA
= M(t)M(0)
1
=

3e
t
2e
3t
e
3t
2te
3t
e
t
e
3t 1
2
e
3t
+ te
3t
e
t
e
3t
te
3t

3 2 1
1 1 1/2
1 1 0

1
=

3e
t
2e
3t
8te
3t
6e
t
6e
3t
20te
3t
4te
3t
e
t
+ e
3t
+ 4te
3t
2e
t
+ 3e
3t
+ 10te
3t
2te
3t
e
t
e
3t
+ 4te
3t
2e
t
2e
3t
+ 10te
3t
e
3t
2te
3t

.
4
EXAMPLE 4. Evaluate e
tA
for A =

9 5
4 5

.
Solution 1: (Use Diagonalization)
Solving det(AI) = 0, we obtain the eigenvalues of A:
1
= 7 + 4i,
2
= 7 4i.
Eigenvectors for
1
= 7 + 4i: are obtained by solving [A(7 + 4i)I]v = 0:
v = v
2

1
2
+ i
1

. ()
Eigenvectors for
2
= 7 4i: are complex conjugate of the vectors in ().
The matrix A is now diagonalized: A = PDP
1
with
P =

1
2
+ i
1
2
i
1 1

, D =

7 + 4i 0
0 7 4i

.
We have
e
tA
= Pe
tD
P
1
=

1
2
+ i
1
2
i
1 1

e
(7+4i)t
0
0 e
(74i)t

1
2
i
1
2
+
1
4
i
1
2
i
1
2

1
4
i

(
1
2

1
4
i)e
(7+4i)t
+ (
1
2
+
1
4
i)e
(74i)t 5
8
ie
(7+4i)t

5
8
ie
(74i)t

1
2
ie
(7+4i)t
+
1
2
ie
(74i)t
(
1
2
+
1
4
i)e
(7+4i)t
+ (
1
2

1
4
i)e
(74i)t

.
Solution 2: (Use fundamental solutions and complex exp functions)
A fundamental matrix solution can be obtained from the eigenvalues and eigenvectors:
M(t) =

(
1
2
+ i)e
(7+4i)t
(
1
2
i)e
(74i)t
e
(7+4i)t
e
(74i)t

.
The matrix exponential is
e
tA
= M(t)M(0)
1
=

(
1
2
+ i)e
(7+4i)t
(
1
2
i)e
(74i)t
e
(7+4i)t
e
(74i)t

1
2
+ i
1
2
i
1 1

1
=

(
1
2

1
4
i)e
(7+4i)t
+ (
1
2
+
1
4
i)e
(74i)t 5
8
ie
(7+4i)t

5
8
ie
(74i)t

1
2
ie
(7+4i)t
+
1
2
ie
(74i)t
(
1
2
+
1
4
i)e
(7+4i)t
+ (
1
2

1
4
i)e
(74i)t

.
Solution 3: (Use fundamental solutions and avoid complex exp functions)
A fundamental matrix solution can be obtained from the eigenvalues and eigenvectors:
M(t) =

e
7t

1
2
cos 4t sin 4t

e
7t

cos 4t +
1
2
sin 4t

e
7t
cos 4t e
7t
sin 4t

.
The matrix exponential is
e
tA
= M(t)M(0)
1
=

e
7t

1
2
cos 4t sin 4t

e
7t

cos 4t +
1
2
sin 4t

e
7t
cos 4t e
7t
sin 4t

1
2
1
1 0

1
=

e
7t
cos 4t +
1
2
e
7t
sin 4t
5
4
e
7t
sin 4t
e
7t
sin 4t e
7t
cos 4t
1
2
e
7t
sin 4t

.
5
APPENDIX: Common Mistakes
Since the years of freshmen calculus, we all loved the exponential function e
x
with scalar
variable x. There are tons of simple and beautiful formulas for the scalar function e
x
. The
matrix exponential is, however, a quite dierent beast. We need to be a little careful in handling
it.
Here are some common mistakes I have seen people make:
e
t(A+B)
= e
tA
e
tB
.
The solutions of
dx
dt
= A(t)x are x(t) = e
tA(t)

C.
The solutions of
dx
dt
= A(t)x are x(t) = e
R
t
0
A(s)ds

C.

e
B(t)

= B

(t)e
B(t)
.
All the above four statements are WRONG!
(1) It is true that e
t(A+B)
= e
tA
e
tB
if AB = BA. But in general, e
t(A+B)
= e
tA
e
tB
.
Example: For
A =

1 1
1 1

, B =

1 1
0 0

, A + B =

0 0
1 1

,
we have
e
tA
e
tB
=
1
2

1 + e
2t
1 e
2t
1 e
2t
1 + e
2t

e
t
1 e
t
0 1

=
1
2

e
t
+ e
t
2 e
t
e
t
e
t
e
t
2 e
t
+ e
t

,
but
e
t(A+B)
=

1 0
1 e
t
e
t

.
(2) We learned how to solve
dx
dt
= Ax where A is a constant matrix.
Unfortunately there is no general solution method for
()
dx
dt
= A(t)x where A(t) is nonconstant.
In particular,
x(t) = e
tA(t)

C.
does not solve (). This even fails in the scalar case.
Example: The solutions of the scalar equation
dx
dt
= (sin t)x is given by x(t) = e
cos t
C,
but not e
t sin t
C.
6
(3) Likewise,
x(t) = e
R
t
0
A(s)ds

C
is also a wrong solution formula for (). This formula is only valid for scalar equations,
i.e., when the space dimension is 1.
Example: Consider
dx
dt
=

1 0
2t 1

x, x(0) =

1
0

where A(t) =

1 0
2t 1

.
The solution of this initial value problem is
x(t) =

e
t
(t
1
2
)e
t
+
1
2
e
t

.
The function
exp

t
0
A(s)ds

1
0

does not give the solution. Indeed,

t
0
A(s)ds

1
0

= exp

t 0
t
2
t

1
0

e
t
0
1
2
te
t

1
2
te
t
e
t

1
0

e
t
1
2
te
t

1
2
te
t

.
(4) We do have:
(e
b(t)
)

= b

(t)e
b(t)
holds for scalar functions b(t), and
(e
tA
)

= Ae
tA
= e
tA
A for constant matrices A.
But, in general, for nonconstant matrix function B(t),
(e
B(t)
)

is neither B

(t)e
B(t)
nor e
B(t)
B

(t).
Example: Let B(t) =

t 0
t
2
t

. We have e
B(t)
=

e
t
0
1
2
te
t

1
2
te
t
e
t

, and hence

e
B(t)

e
t
0
t+1
2
e
t
+
t1
2
e
t
e
t

,
B

(t)e
B(t)
=

1 0
2t 1

e
t
0
1
2
te
t

1
2
te
t
e
t

e
t
0
3
2
te
t
+
1
2
te
t
e
t

,
e
B(t)
B

(t) =

e
t
0
1
2
te
t

1
2
te
t
e
t

1 0
2t 1

e
t
0
1
2
te
t
+
3
2
te
t
e
t

.
All three matrix functions (e
B(t)
)

, B

(t)e
B(t)
and e
B(t)
B

(t) are dierent from each other.

5 3
1 1

.
[2] (a) Evaluate e
tA
for A =

4 12
3 8

.
(b) Solve
dx
dt
= Ax, x(0) =

5
1

.
[3] Evaluate e
tA
for A =

1 4 2
3 4 0
3 1 3

.
[4] Evaluate e
tA
for A =

5 4 2
12 9 4
12 8 3

.
[5] (a) Evaluate e
tA
for A =

9 7 3
16 12 5
8 5 2

.
(b) Solve
dx
dt
= Ax, x(0) =

1
1
1

.
[6] Evaluate e
tA
for A =

5 4
2 1

.
[7] Evaluate e
tA
for A =

1 1 0
2 3 2
0 2 1

.
See next page for answers
8
Answers:
[1] e
tA
=

3
2
e
4t

1
2
e
2t

3
2
e
4t
+
3
2
e
2t
1
2
e
4t

1
2
e
2t

1
2
e
4t
+
3
2
e
2t

[2] (a) e
tA
=

e
2t
6te
2t
12te
2t
3te
2t
e
2t
+ 6te
2t

(b) x(t) = e
tA

5
1

5e
2t
42te
2t
e
2t
21te
2t

[3] e
tA
=

3e
t
2e
2t
5e
t
+ 6e
2t
e
3t
3e
t
4e
2t
+ e
3t
3e
t
3e
2t
5e
t
+ 9e
2t
3e
3t
3e
t
6e
2t
+ 3e
3t
3e
t
3e
2t
5e
t
+ 9e
2t
4e
3t
3e
t
6e
2t
+ 4e
3t

[4] e
tA
=

3e
t
2e
t
2e
t
2e
t
e
t
+ e
t
6e
t
+ 6e
t
4e
t
+ 5e
t
2e
t
2e
t
6e
t
+ 6e
t
4e
t
+ 4e
t
2e
t
e
t

[5] (a) e
tA
=

3e
t
2e
t
+ 4te
t
2e
t
2e
t
+ 3te
t
e
t
+ e
t
te
t
6e
t
+ 6e
t
4te
t
4e
t
+ 5e
t
3te
t
2e
t
2e
t
+ te
t
6e
t
+ 6e
t
+ 4te
t
4e
t
+ 4e
t
+ 3te
t
2e
t
e
t
te
t

(b) x(t) = e
tA

1
1
1

4e
t
3e
t
+ 6te
t
8e
t
+ 9e
t
6te
t
8e
t
+ 9e
t
+ 6te
t

[6] e
tA
= e
3t

cos 2t + sin 2t 2 sin 2t

,
or, equivalently,
e
tA
=
1
2

(1 i)e
(3+2i)t
+ (1 + i)e
(32i)t
2ie
(3+2i)t
2ie
(32i)t
ie
(3+2i)t
+ ie
(32i)t
(1 + i)e
(3+2i)t
+ (1 i)e
(32i)t

[7] e
tA
=
1
5

2e
3t
+ 3 cos t + sin t 2e
3t
+ 2 cos t sin t e
3t
cos t + 3 sin t
4e
3t
+ 4 cos t 2 sin t 4e
3t
+ cos t 3 sin t 2e
3t
+ 2 cos t + 4 sin t
2e
3t
+ 2 cos t 6 sin t 2e
3t
2 cos t 4 sin t e
3t
+ 6 cos t + 2 sin t