EE635 - Control System Theory

Jitkomut Songsiri

2. Matrix exponential

definition and properties

calculation of eAt
Cayley-Hamilton theorem

2-1

Definition
let A be an n n matrix
for t 0, the matrix function eAt is defined by

eAt

X Ak tk
A2t2
Antn
= I + At +
+ +
+ =
2!
n!
k!
k=0

arises as the solution of a linear differential equation:

x(t)

= Ax(t)
the solution is given by
x(t) = eAtx(0), t 0

Matrix exponential

2-2

properties:
e0 = I
deAt

= AeAt = eAtA, t 0
dt
R t A
At
e = I + A 0 e d , t 0

eAteA = eA(t+ ) = eA eAt

(eAt)1 = eAt

(eAt is always nonsingular)

e(A+B)t = eAteBt only when AB = BA

L[eAt] = (sI A)1
(eAt) = e(A)t
det(eAt) = etr(A)t

Matrix exponential

2-3

Calculation of eAt
we can compute eAt via:
inverse Laplace transform
eAt = L1(sI A)1
(use when the inverse of (sI A) is easy to compute)
diagonalization
A = T DT 1

eAtT eDtT 1

(when A is diagonalizable)
Cayley-Hamilton theorem
Matrix exponential

2-4

Compute eAt via diagonalization

1 1 0
A = 0 2 1
0 0 0
the eigenvalues and eigenvectors of A are

1
1 = 1, v1 = 0 ,
0

1
2 = 2, v2 = 1 ,
0

1
3 = 0, v3 = 1
2

form T = v1 v2 v3 and compute eAt = T eDtT 1



eAt

1 1 1
et 0
= 0 1 1 0 e2t
0 0 2
0 0

Matrix exponential

t 2t
e e et
0 1 1 1

0 0 1 1/2 = 0
e2t
1 0 0 1/2
0
0

2t

e 2e +1
2

2t
e 1
2

2-5

Cayley-Hamilton theorem
the characteristic polynomial of a matrix A of size n n
X () = det(I A)
can be written as a polynomial of degree n:
X () = n + n1 n1 + + 1 + 0
Theorem: a square matrix satisfies its characteristic equation:
X (A) = An + n1An1 + + 1A + 0I = 0

result: for m n, Am is a linear combination of Ak , k = 0, 1, . . . , n 1.

Matrix exponential

2-6


1 2
example 1: A =
the characteristic equation of A is
0 3
X () = ( 1)( 3) = 2 4 + 3 = 0
the Cayley-Hamilton states that A satisfies its characteristic equation
X (A) = A2 4A + 3I = 0
use this equation to write matrix powers of A
A2 = 4A 3I
A3 = 4A2 3A

= 4(4A 3I) 3A

A4 = 13A2 12A
..

= 13(4A 3I) 12A = 40A 39I

..

= 13A 12I

powers of A can be written as a linear combination of I and A

Matrix exponential

2-7

example 2: with A in page 2-7, find the closed-form expression of Ak

for k 2, Ak is a linear combination of I and A, i.e.,
Ak = 0I + 1A
where 1, 0 are to be determined
multiply eigenvectors of A on both sides
Ak v1 = (0I + 1A)v1 k1 = 0 + 11
Ak v2 = (0I + 1A)v2 k2 = 0 + 12
substitute 1 = 1 and 2 = 3 and solve for 0, 1
  
 
3 3k
1
1 1 0
,
=
0 =
1 3 1
3k
2
k

3k 1
1 =
2


3 1
33
1 3 1
,
I+
A=
Ak =
k
0
3
2
2
Matrix exponential

k2

2-8

Function of square matrices

for any convergent power series f (A) =

k
k=0 ak A ,

we can write

f (A) = 0I + 1A + + n1An1
multiplying an eigenvector v of A on both sides and using v 6= 0, we get
f () = 0I + 1 + + n1n1
substitute with the n eigenvalues of A

n1
1

2n1

1 1
0
f (1)
f (2) 1 2
1
. = . .
..
.. ..

. . .
n1
f (n)
1 n nn1
Fact: if all k s are distinct, the system is solvable and has a unique sol.
Matrix exponential

2-9

Vandermonde matrix
a Vandermonde matrix has the form
n1
1
n1
2

1 1
1 2
V =
..
..
.. ..
1 n nn1
(with a geometric progression in each row)
one can show that the determinant of V can be expressed as
det(V ) =

(j i)

1i<jn

hence, V is invertible as long as is are distinct

Matrix exponential

2-10

example: compute f (A) = etA given

1 1 0
A = 0 2 1
0 0 0
the eigenvalues of A are = 1, 2, 0 (all are distinct)
form a system of equations:
f (i) = 0(t) + 1(t)i + 2(t)2i
or equivlently,

e
1 1 1
0(t)
e2t = 1 2 22 1(t)
e0t
1 0 0
2(t)
which has the solution
0(t) = 1,

Matrix exponential

1(t) = 2et e2t/2 3/2,

2(t) = et + e2t/2 + 1/2

2-11

substituting 0(t), 1(t), 2(t) in

eAt = 0(t)I + 1(t)A + 2(t)A2
gives

eAt

1 3 1
1 1 0
1 0 0
= 0 0 1 0 + 1 0 2 1 + 2 0 4 2
0 0 0
0 0 0
0 0 1

0(t) + 1(t) + 2(t)

1(t) + 32(t)
2(t)
0
0(t) + 21(t) + 42(t) 1(t) + 22(t)
=
0
0
0(t)
t 2t

t
2t
t
e e e (e 2e + 1)/2
e2t
(e2t 1)/2
= 0
0
0
1

(agree with the result in page 2-5)

Matrix exponential

2-12

Repeated eigenvalues
A has repeated eigenvalues, i.e., i = j for some i, j

goal: compute f (A) using C-H theorem

however, we can no longer apply the result in page 2-9 because

the number of independent equations on page 2-9 is less than n
the Vandermonde matrix (page 2-10) is not invertible
cannot form a linear system to solve for the n coefficients, 0, . . . , n1

Matrix exponential

2-13

solution: for the repeated root with multiplicity r

get r 1 independent equations by taking derivatives on f () w.r.t

f () = 0 + 1 + + n1n1
df ()
= 1 + 22 + + (n 1)n1n2
d
.. = ..
dr1f ()
n1r
=
(r

1)!
+

+
(n

r)

(n

2)(n

1)

r1
n1
dr1

Matrix exponential

2-14

example: find eAt given

0 0 2
A = 0 1 0
1 0 3
the eigenvalues are 2, 1, 1

repeated roots

substitute 1 = 2 and 2 = 1 in to the equation

et = 0(t) + 1(t) + 2(t)2

(1)

and we obtain
et = 0(t) + 1(t) + 2(t)
e2t = 0(t) + 21(t) + 42(t)

Matrix exponential

2-15

for another repeated root, we differentiate (1) w.r.t

tet = 1(t) + 22(t)
substituting = 1 gives
tet = 1(t) + 22(t)
arrange the three equations into a matrix form:

e
1 1 1
0(t)
tet = 0 1 2 1(t)
1 2 4
2(t)
e2t
solving these equations gives
0(t) = 2tet + e2t,

1(t) = (2 + 3t)et 2e2t,

2(t) = (t + 1)et + e2t

and substitute them in eAt = 0(t)I + 1(t)A + 2(t)A2

Matrix exponential

2-16

Summary
by C-H theorem, any polynomial function of A Rnn can be
recursively written as a linear combination of I, A, A2, . . . , An1
by definition, eAt is a polynomial of A, so it can be also expressed as a
linear combination of I, A, . . . , An1, where the coefficients are
functions of t
if all eigenvalues of A are distinct, the coefficients in the linear
combination can be solved uniquely via the inverse of a Vandermonde
matrix
if A has repeated roots, differentiate f () (or et) as many times as
needed, to obtain additional independent equations

Matrix exponential

2-17