2 Chapter 1 Mathematical Notation and Review
Vectors
‘Two x1 vectors x and y are ealled linearly independent if no nontrivial linear
‘conibinaiion of « and y givesthe zero vector. Tis means that if ax + By = 0, then both
scalars @ and B are zero. Of course the definition extends to a linear combination of
any number of vectors. A set of n linearly independent n x 1 vectors forms a basis for
the vector space ofall» x1 vectors. The st ofall linear combinations of a specified set
‘of vectors is vector space called the span of the set of vectors. For example
span {x,y,z} is a 3-dimensional subspace of R". if x, y, and 2 are linearly
independent n x I vectors.
norm fot nx vectors, defined as
follows. Writing a vector and its sranspose in the form.
x
fn
5 !
.
Tet
Ikell = Wee (3 i «
Elementary inequalities relating the Euclidean norm of a vector to the absolute values of
entries are (max of course is short for maximum)
max. |.xj1 < [lal] 0
and Q20 are used to denote positive definiteness, and positive semidefiniteness,
respectively, Of course Q, 2 Q), simply means that Q,—Qy, is positive semidefinite.
All eigenvalues of a symmetric matrix must be real, It follows that positive
definiteness is equivalent to all eigenvalues positive, and positive semidefiniteness is
equivalent to all eigenvalues nonnegative. An important inequality for a symmetric
nxn matrix Q is the Rayleigh-Ritz inequality, which states that for any real nx 1
vector 1,
Aenin XTX S27 Ox S Ayy XT (4)
Where Ania and Aypay denote the smallest and largest eigenvalues of Q. See Exercise
1.10 for the spectral norm of Q. If we assume Q 20, then ||Q || = Aas and the trace is
bounded by
l@ii0, p=
Itis negative definite if and only if
(-p’ ea,
1.4 Theorem The symmetric matrix Q is positive semidefinite if and only if
ISip 1,, in which case the absolute value signs on the right side
can be erased.
The sum of the infinite series is called the matrix exponential and denoted as e'4 :
tA # ik
tay ea.
Ss
This series is absolutely convergent.3—
STATE EQUATION SOLUTION
‘state equations unencumbered by inputs and outputs. That is, we consider
A= A(X), 1) = 40 wm
The n xn matrix function A (1) is
assumed to be continuous and defined for all 1.
that satisfies (1) for all ¢, though at the
‘outset only solutions for +2f, are considered. Among other things this avoids
4) = OU, ty)Xy ay0, 92! + [A(o,) do, + J(01)f Aloe) doy doy
+ fave) | A002) J (os) dos do: doy + =» ay
3.1 Example For a scalar, time-invariant linear state equation, where we write
A(t) =, the approximating sequence in (2) generates,
and so on. The general term in the sequence is
a(t)= [: sae
and the limit of the sequence is the presumably familiar solution
sdyae (3)
‘Thus the transition matrix in this case is simply a scalar exponential.
3.5 Example For
Or
AW) = [° ‘| (20)
the Peano-Baker series (12) is
10], [0a j[oa]t [oe
mas E | * i[e 3] {fe ai [: 0
It is straightforward to verify that all terms in the series beyond the second are zero, and
thus
don doy + °*
(21)
Doge
dias [s G “alComplete Solution
The standard approach to considering existence and uniqueness of solutions of
H()= AWC) + BOUL, XG) (23)
with given f,, 2.) and continuous u(t), involves using properties of the transition matrix
48 Chapter3 State Equation Solution
that are discussed in Chapter 4. However the guess~:
successful, so in Exercise 3.1 the reader is invited to verify by direct differentiation that a
solution of (23) is
X(0= 00, ,)x, + [OG o)B((o)do. 121, (24)
A litile thought shows that this solution is unique since the difference >(r) between any
two solutions of (2) must satisfy (17). Thus z(t) must be identically zero,
Taking account of an output equation,
yO) = CW) + DOU) (25)
(24) leads to
Y= COG, f,)x, + [CHE OB (O)u(o)do + D(Nu(t) (26)
2,3.7 Example To compute the transition matrix for
10
so-[ S|
write the corresponding pair of scalar equations
FO =O), 116) = 20
¥2() = a(xr() +x). x2ty) = 20
From Example 3.1 we have
x=
“Xo
Then the second scalar equation can be written as a forced scalar state equation
(BQu(n =e x15)
a(t) =a(nra(r) + e's
tos X2() = X2y
The transition matrix for scalar a(t) is computed in Example 3.6, and applying (24)
gives
sy tfe™ ey, do
Repucking into matrix notation yields
x=], :
Jexplo-1,+Jaindddo ¢
from which we immediately ascertain ®,(r, 1,,).
ooo
Additional Examples
We illustrate aspects of the complete solution formula for linear state equations by
revisiting two examples from Chapter 2.4.1 Property If A(/) =A, ann xn constant matrix, then the transition matrix is
OE, )=e4-9
where the matrix exponential is defined by the power series
May ant @
i
that converges uniformly and absolutely on (~T, 7], where 7 > 0 is arbitrary.
5.7 Property If A and F aren x n matrices, then
eter — pA +Pr
for every ¢ if and only if AF = FA.
4.2 Property If for every ¢ and +,
A JA(@)do=[A@)doA() 8
Iams = 4 ‘
Oo, ) =e"
'
i lis ()do @)with no loss of generality. Therefore in the matrix exponential case we work with
M1, 0) =e" (2)
5.9 Example For the harmonic oscillator, where
o1
-10
eat *