Scribd
Upload
91 views4 pages

(2.2) Matrix Identities

(1) The document defines the exponential of a matrix A, denoted eA, as an infinite series and proves several properties about it, including that etA is the fundamental matrix solution to the linear system ẋ = Ax. (2) It proves five specific properties of the exponential of a matrix: (1) eA is invertible if A is, (2) eB-1eAB = eA if B is invertible, (3) eA+B = eAeB if AB = BA, (4) the derivative of etA is AetA, and (5) ||eA|| ≤ e||A||. (3) The document provides exercises

Uploaded by

audaciousribbon
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
91 views4 pages

(2.2) Matrix Identities

(1) The document defines the exponential of a matrix A, denoted eA, as an infinite series and proves several properties about it, including that etA is the fundamental matrix solution to the linear system ẋ = Ax. (2) It proves five specific properties of the exponential of a matrix: (1) eA is invertible if A is, (2) eB-1eAB = eA if B is invertible, (3) eA+B = eAeB if AB = BA, (4) the derivative of etA is AetA, and (5) ||eA|| ≤ e||A||. (3) The document provides exercises

Uploaded by

audaciousribbon
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

160 2.

Linear Systems and Stability of Nonlinear Systems

Proposition 2.32. Suppose that A, B ∈ L(E).

(0) If A ∈ L(Rn ), then eA ∈ L(Rn ).


−1
(1) If B is nonsingular, then B −1 eA B = eB AB
.

(2) If AB = BA, then eA+B = eA eB .

(3) e−A = (eA )−1 . In particular, the image of exp is in the general linear
group GL(E) consisting of the invertible elements of L(E).

(4) d tA
dt (e ) = Ae
tA
= etA A. In particular, t → etA is the principal fun-
damental matrix solution of the system (2.7) at t = 0.

(5) eA  ≤ eA .

Proof. The proof of (0) is obvious.


To prove (1), define

1 2 1 N
SN := I + A + A + ··· + A ,
2! N!

and note that if B is nonsingular, then B −1 An B = (B −1 AB)n . Thus, we


have that
1 −1 1
B −1 SN B = I + B −1 AB + (B AB)2 + · · · + (B −1 AB)N ,
2! N!
and, by the definition of the exponential map,
−1
lim B −1 SN B = eB AB
.
N →∞

Using the continuity of the linear map on L(E) defined by C → B −1 CB,


it follows that
lim B −1 SN B = B −1 eA B,
N →∞

as required.
While the proof of (4) given here has the advantage of being self con-
tained, there are conceptually simpler alternatives (see Exercises 2.33–
2.34). As the first step in the proof of (4), consider the following proposition:
If s, t ∈ R, then e(s+t)A = esA etA . To prove it, let us denote the partial
sums for the series representation of etA by

1 1
SN (t) := I + tA + (tA)2 + · · · + (tA)N
2! N!
1 1 N N
= I + tA + t2 A2 + · · · + t A .
2! N!
2.1 Homogeneous Linear Differential Equations 161

We claim that

2N
SN (s)SN (t) = SN (s + t) + Pk (s, t)Ak (2.8)
k=N +1

where Pk (s, t) is a homogeneous polynomial of degree k such that

(|s| + |t|)k
|Pk (s, t)| ≤ .
k!
To obtain this identity, note that the kth order term of the product, at
least for 0 ≤ k ≤ N , is given by

k
1 k−j j
 k 1
k
k!  1
s t A = sk−j tj Ak = (s + t)k Ak .
j=0
(k − j)!j! k! j=0 (k − j)!j! k!

Also, for N + 1 ≤ k ≤ 2N , the kth order term is essentially the same, only
some of the summands are missing. In fact, these terms all have the form

k
δ(j) 
sk−j tj Ak
j=0
(k − j)!j!

where δ(j) has value zero or one. Each such term is the product of Ak and
a homogeneous polynomial in two variables of degree k. Moreover, because
|δ(j)| ≤ 1, we obtain the required estimate for the polynomial. This proves
the claim.
Using equation (2.8), we have the following inequality

2N
SN (s)SN (t) − SN (s + t) ≤ |Pk (s, t)| Ak
k=N +1

2N
(|s| + |t|)k
≤ Ak .
k!
k=N +1

Also, because the series


∞

(|s| + |t|)k
Ak
k!
k=0

is convergent, it follows that its partial sums, denoted QN , form a Cauchy


sequence. In particular, if > 0 is given, then for sufficiently large N we
have
|Q2N − QN | < .
162 2. Linear Systems and Stability of Nonlinear Systems

Moreover, since

2N
(|s| + |t|)k
Q2N − QN = Ak ,
k!
k=N +1

it follows that
lim SN (s)SN (t) − SN (s + t) = 0.
N →∞

Using this fact and passing to the limit as N → ∞ on both sides of the
inequality

esA etA − e(s+t)A  ≤ esA etA − SN (s)SN (t)


+ SN (s)SN (t) − SN (s + t)
+ SN (s + t) − e(s+t)A ,

we see that

esA etA = e(s+t)A , (2.9)

as required.
In view of the identity (2.9), the derivative of the function t → etA is
given by
d tA 1
e = lim (e(t+s)A − etA )
dt s→0 s
1 sA
= lim (e − I)etA
s→0 s
1 
= lim (esA − I) etA
s→0 s

= lim (A + R(s)) etA
s→0

where
∞ ∞
1
|s|k
|s|k−2
R(s) ≤ Ak ≤ |s| Ak .
|s| k! k!
k=2 k=2

Moreover, if |s| < 1, then R(s) ≤ |s|eA . In particular, R(s) → 0 as


s → 0 and as a result,
d tA
e = AetA .
dt
Since ASN (t) = SN (t)A, it follows that AetA = etA A. This proves the
first statement of part (4). In particular t → etA is a matrix solution of
the system (2.7). Clearly, e0 = I. Thus, the columns of e0 are linearly
independent. It follows that t → etA is the principal fundamental matrix
solution at t = 0, as required.
2.1 Homogeneous Linear Differential Equations 163

To prove (2), suppose that AB = BA and consider the function t →


et(A+B) . By (4), this function is a matrix solution of the initial value prob-
lem
ẋ = (A + B)x, x(0) = I.
The function t → etA etB is a solution of the same initial value problem. To
see this, use the product rule to compute the derivative
d tA tB
e e = AetA etB + etA BetB ,
dt
and use the identity AB = BA to show that etA B = BetA . The desired
result is obtained by inserting this last identity into the formula for the
derivative. By the uniqueness of the solution of the initial value problem,
the two solutions are identical.
To prove (3), we use (2) to obtain I = eA−A = eA e−A or, in other words,
(e ) = e−A .
A −1

The result (5) follows from the inequality


1 2 1 N 1 1
I + A + A +···+ A  ≤ I + A + A2 + · · · + AN . 2
2! N! 2! N!
We have defined the exponential of a matrix as an infinite series and
used this definition to prove that the homogeneous linear system ẋ = Ax
has a fundamental matrix solution, namely, t → etA . This is a strong result
because it does not use the existence theorem for differential equations.
Granted, the uniqueness theorem is used. But it is an easy corollary of
Gronwall’s inequality (see Exercise 2.5). An alternative approach to the
exponential map is to use the existence theorem and define the function
t → etA to be the principal fundamental matrix solution at t = 0. Proposi-
tion 2.32 can then be proved by using properties of the solutions of homo-
geneous linear differential equations.

Exercise 2.33. Show that the partial sums of the series representation of etA
converge uniformly on compact subsets of R. Use Theorem 1.248 to prove part (4)
of Proposition 2.32.
Exercise 2.34. (a) Show that exp : L(E) → L(E) is continuous. Hint: For r >
0, the sequence of partial sums of the series representation of exp(X) converges
uniformly on Br (0) := {X ∈ L(E) : X < r}. (b) By Exercise 2.30, matrices
with distinct eigenvalues are dense in L(E). Such matrices are diagonalizable
(over the complex numbers). Show that if A ∈ L(E) is diagonalizable, then
part (4) of Proposition 2.32 holds for A. (c) Use parts (a) and (b) to prove
part (4) of Proposition 2.32. (d) Prove that exp : L(E) → L(E) is differentiable
and compute D exp(I).
Exercise 2.35. Define exp(A) = Φ(1) where Φ(t) is the principal fundamental
matrix at t = 0 for the system ẋ = Ax. (a) Prove that exp(tA) = Φ(t). (b) Prove
that exp(−A) = (exp(A))−1 .

You might also like