15.An introduction to H~ control theory
.pdf22/10/2004 |
15.3 Optimal model matching II. Nehari’s Theorem |
577 |
(iii)CR ◦ CR = CR;
(iv)OR ◦ OR = OR;
(v)OR is invertible.
Proof (i) Since −A is Hurwitz, by part (i) of Theorem 5.32 there is a unique symmetric matrix P so that (−At, P, bbt) is a Liapunov triple. What’s more, the proof of Theorem 5.32 gives P explicitly as
Z ∞
P = e−Atbbte−Att dt.
0
Now one sees trivially that (A, −P, −b, bt) is also a Liapunov triple. This part of the proposition now follows because CR = −P.
(ii) The proof here is exactly as for part (i).
(iii) This follows from the characterisations of CR and CR given in Proposition 15.24. (iv) This follows from the characterisations of OR and OR given in Proposition 15.24. (v) Since OR is square, injectivity is equivalent to invertibility. Suppose that OR is not
invertible. Then, since OR is positive-semidefinite, there exists x Rn so that xtORx = 0, or so that Z ∞
xte−Attccte−Atxdt.
0
578 |
|
|
15 An introduction to H∞ control theory |
22/10/2004 |
||
If u = |
CR |
R |
6 |
|
CR |
is |
O |
|
x then we claim that u = 0. Indeed, from part (v) of Proposition 15.24, |
|
injective, and from part (v) of Proposition 15.25, OR is injective. Thus u = 0 if and only if
x = 0. Thus we see that σ2 is an eigenvalue for ¯ |
¯ |
with eigenvalue u. |
|
R |
R |
|
Thus we have a characterisation of all nonzero singular values of the Hankel operator as eigenvalues of an n × n matrix. This is something of a coup. We shall suppose that the nonzero singular values are arranged in descending order σ1 ≥ σ2 ≥ · · · ≥ σk, so that σ1 denotes the largest of the singular values. We shall call σ1, . . . , σk the Hankel singular values for the Hankel operator R.
Now we wish to talk about the “size” of a Hankel operator R. Since R is a linear map between two inner product spaces—from RH+2 to RH−2 —we may simply define its norm in the same manner in which we defined the induced signal norms in Definition 5.19. Thus we
define |
|
|
|
|
k R(Q)k2 |
|
k |
|
Rk |
= |
sup |
. |
|
|
|
Q RH2+ |
kQk2 |
|||
|
|
|
|
Q not zero |
|
|
This is called the Hankel norm of the Hankel operator R. The following result follows easily from Theorem 15.26 if one knows just a little more operator theory than is really within the confines of this course. However, it is an essential result for us.
This means that cte−Atx = 0 for all t |
|
[0, |
∞ |
). Di erentiating successively with respect to |
|
15.27 Corollary If σ1 is the largest Hankel singular value, then |
k |
R |
k |
= σ1. |
|
|
||||||||||||||||||||||||||||||||||||
t at t = 0 gives |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|||||||||||||||||||||||||||||
|
ctx = 0, −ctAx = 0, . . . , (−1)n−1ctAn−1x = 0. |
|
|
|
|
|
Proof |
We can prove this using matrix norms. |
|
|
|
|
|
|
|
|
|
|
|
|||||||||||||||||||||||||||||
|
|
|
|
|
|
|
|
|
|
|
|
|
finish |
|
|
|
|
|
|
|
|
|
|
|
|
|||||||||||||||||||||||
This implies that (A, c) is not observable. It therefore follows that OR is indeed injective. |
|
Next, let us look a little more closely at eigenvectors induced by singular values. Thus |
||||||||||||||||||||||||||||||||||||||||||||||
This, then, is interesting as it a ords us the possibility of characterising the singular |
|
we let σ2 be a nonzero singular value for ¯ ¯ |
with eigenvector u |
1 |
|
L¯ |
2 |
[0, |
∞ |
). Now define |
||||||||||||||||||||||||||||||||||||||
|
¯ |
(−∞, 0] by u2 = |
1 ¯ |
|
|
R |
R |
|
|
|
|
|
|
|
|
|||||||||||||||||||||||||||||||||
values of the Hankel operator in terms of the eigenvalues of an n |
× |
n matrix. This is |
|
u2 L2 |
σ R(u1). Then one readily computes |
|
|
|
|
|
|
|
|
|
|
|||||||||||||||||||||||||||||||||
summarised in the following result. |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
¯ |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||||||
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||||||
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
R(u1) = σu2 |
|
|
|
|
|
|
|
|
|
|
|
|
|||||
15.26 Theorem The nonzero eigenvalues of the following three operators, |
|
|
|
|
|
|
¯ (u |
) = σu |
. |
|
|
|
|
|
|
|
|
|
|
|||||||||||||||||||||||||||||
|
|
|
|
|
|
R |
2 |
|
1 |
|
|
|
|
|
|
|
|
|
|
|
|
|||||||||||||||||||||||||||
(i) |
|
◦ |
|
R |
, |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
When an operator and its adjoint possess the same eigenvalue in this manner, the resulting |
||||||||||||||||||||||
|
R |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||||||||||||||||||
(ii) |
¯ |
◦ |
¯ |
R |
, and |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
eigenvectors (u1, u2) are called a σ-Schmidt pair for the operator. Of course, if Rj is the |
|||||||||||||||||||||||
|
R |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||||||||||||||||||
(iii) CROR, |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Laplace transform of uj, j = 1, 2, then we have |
|
|
|
|
|
|
|
|
|
|
|
|
|
|||||||||||||
agree. |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
R(R1) = σR2 |
|
|
|
|
|
|
|
|
|
|
|
|||||||
Proof That the eigenvalues for |
R and ¯ |
|
¯R agree is a simple consequence of Proposi- |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||||||||||||||||||||||||
|
|
|
|
|
(R |
) = σR |
|
, |
|
|
|
|
|
|
|
|
|
|
||||||||||||||||||||||||||||||
|
|
|
|
|
|
|
|
|
|
R |
|
|
|
|
R |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||
tion 15.23: the Laplace transform or its inverse will deliver eigenvalues and eigenvectors for |
|
|
|
|
R |
2 |
|
1 |
|
|
|
|
|
|
|
|
|
|
|
|||||||||||||||||||||||||||||
either of |
|
|
R |
or ¯ |
¯ |
given eigenvalues and eigenvectors for the other. |
|
|
|
|
+ |
− |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||||||||||||
|
|
|
R |
|
|
R |
R |
|
|
|
|
¯ |
|
|
|
|
|
|
L¯ |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|||
Now let σ2 > 0 be an eigenvalue for ¯ |
R |
with eigenvector u |
|
2 |
( |
, 0]. By part (i) |
|
so that (R1, R2) RH2 × RH2 |
are a σ-Schmidt pair for R. The matter of finding Schmidt |
|||||||||||||||||||||||||||||||||||||||
of Proposition 15.24 this means that |
|
R |
|
|
|
|
|
|
−∞ |
|
|
|
pairs for Hankel operators is a simple enough proposition as one may use Theorem 15.26. |
|||||||||||||||||||||||||||||||||||
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Indeed, suppose that σ2 |
> 0 is an eigenvalue for CROR with eigenvector x. Then a σ- |
|||||||||||||||||||||||||||||||
|
|
|
|
|
|
|
|
|
|
CRORORCR(u) = σ2u |
|
|
|
|
|
|
|
|
|
|
|
¯ |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|||||||
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Schmidt pair for R is readily verified to be given by (u1, u2) where |
|
|
|
|
|
|
||||||||||||||||||||||
|
|
|
|
|
|
|
|
|
|
= CRCRORORCR(u) = σ2CR(u). |
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|||||||||
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
|
|
|
|
|
2 |
|
|
|
|
|
u1 = |
|
|
OR(ORx) |
|
|
|
|
|
|
|
|
|
|
||||
If x = CR(u) then x 6= 0 since otherwise it would follow that σ |
= 0. This shows that σ |
is |
|
|
|
|
|
σ |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||||||||||||||||||||
|
|
|
|
|
|
u2 = OR(x). |
|
|
|
|
|
|
|
|
|
|
|
|
||||||||||||||||||||||||||||||
an eigenvalue of CROR with eigenvector x. |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||||||||
Now suppose that σ2 6= 0 is an eigenvalue for CROR with eigenvector x. Thus |
|
|
|
15.3.4 Nehari’s Theorem |
In this section we state and prove a famous theorem of Ne- |
|||||||||||||||||||||||||||||||||||||||||||
|
|
|
|
|
|
|
|
|
|
|
CRORx = σ2x |
|
|
|
|
|
|
|
|
|
|
|||||||||||||||||||||||||||
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
hari [1957]. This theorem is just one in a sweeping research e ort in “Hankel norm approxi- |
|||||||||||||||||||||||||||
|
|
|
|
|
|
|
|
|
|
= |
CRORCROR = σ2CRORx. |
|
|
|
|
|
|
|
|
|
|
mation,” with key contributions being made in a sequence of papers by Adamjan, Arov, and |
22/10/2004 |
15.3 Optimal model matching II. Nehari’s Theorem |
579 |
Krein [1968a, 1968b, 1971]. Our interest in this section is in a special version of this rather general work, as we are only interested in rational functions, whereas Nehari was interested in general H∞ functions.
15.28 Theorem Let R0 RH−∞, let σ1 > 0 be the largest Hankel singular value for R0, and let (R1, R2) RH+2 × RH−2 be a σ1-Schmidt pair. Then
|
|
|
|
|
inf |
kR0 − Rk∞ = σ1, |
|||
|
|
|
|
R |
RH+ |
||||
|
|
|
|
|
|
|
∞ |
|
|
and if |
R RH∞+ satisfies R1(R0 − R) = σ1R2 then kR0 − Rk∞ = σ1. |
||||||||
Proof |
First let us show that σ1 is a lower bound for kR0 − Rk∞. For any R RH2+ we |
||||||||
compute, using part (i) of Theorem 5.21, |
|
||||||||
|
k |
R |
0 − |
R |
k∞ |
= |
sup |
k(R0 − R)Qk2 |
|
|
kQk2 |
||||||||
|
|
|
|
Q RH2+ |
|||||
|
|
|
|
|
|
|
|
Q not zero |
|
|
|
|
|
|
|
|
≥ |
sup |
kΠ−(R0 − R)Qk2 |
|
|
|
|
|
|
|
kQk2 |
||
|
|
|
|
|
|
|
Q RH2+ |
||
|
|
|
|
|
|
|
|
Q not zero |
|
|
|
|
|
|
|
|
= |
sup |
kΠ−(R0)Qk2 |
|
|
|
|
|
|
|
kQk2 |
||
|
|
|
|
|
|
|
|
Q RH2+ |
|
|
|
|
|
|
|
|
|
Q not zero |
|
= k Rk .
Now let (R1, R2) be a σ1-Schmidt pair and write Q = (R0 − R)R1 for R RH+∞. Since R1 RH+2 , R0 (R1) RH+2 . Since RR1 RH+2 , Π−(Q) = Π−(R0R1) = R0 (R1). Therefore,
we compute
0≤ kQ − R0 (R1)k22
=kQk22 + h R0 (R1), R0 (R1)i2 − 2 hQ, R0 (R1)i2
=kQk22 + h R0 (R1), R0 (R1)i2 − 2 Π−(Q), R0 (R1) 2
=kQk22 − h R0 (R1), R0 (R1)i2
=kQk22 − R1, R0 R0 (R1) 2
=kQk22 − σ12 hR1, R1i2
=kQk22 − σ12 kR1k22
≤ kR0 − Rk2∞ kR1k22 − σ12 kR1k22 = kR0 − Rk2∞ − σ12 kR1k22
≥ 0.
This shows that Q = R0 (R1), or, equivalently,
(R0 − R)R1 = R0 (R1) = σ1R2,
as claimed. |
|
580 |
15 An introduction to H∞ control theory |
22/10/2004 |
15.3.5 |
Relationship to the model matching problem The previous buildup has been |
significant, and it is perhaps not transparent how Hankel operators and Nehari’s Theorem relate in any way to the model matching problem. The relationship is, in fact, quite simple, and in this section we give a simple algorithm for obtaining a solution to the model matching problem using the tools of this section. However, as with Nevanlinna-Pick theory, there is a drawback in that on occasion a hack will have to be employed. Nonetheless, the process is systematic enough.
Let us come right out and state the algorithm.
15.29Model matching by Hankel norm approximation Given T1, T2 RH+∞.
15.4A robust performance example
15.5 Other problems involving H∞ methods
It turns out that the robust performance problem is only one of a number of problems falling under the umbrella of H∞ control. In this section we briefly indicate some other problems whose solution can be reduced to a model matching problem, and thus whose solution can be obtained by the methods in this chapter.
Exercises for Chapter 15 |
581 |
582 |
15 An introduction to H∞ control theory |
22/10/2004 |
Exercises
E15.1 Exercise in the ∞-norm giving a Banach algebra.
E15.2 Exercise on existence of solutions to the model matching problem.
E15.3 Verify that the following algorithm for reducing the modified robust performance problem for additive uncertainty actually works.
15.30 Algorithm for obtaining model matching problem for additive uncertainty Given
¯ |
|
|
|
|
|
RP , Wu, and Wp as in Problem 15.2. |
|
||||
1. |
Define |
|
WpWp WuWu |
|
|
|
|
|
|
||
|
U3 = |
|
. |
|
|
|
WpWp + WuWu |
|
|||
2. |
If kU3k∞ ≥ 21 , then Problem 15.2 has no solution. |
¯ |
|||
3. |
Let (P1, P2) be a coprime fractional representative for RP . |
||||
4. |
Let (ρ1, ρ2) be a coprime factorisation for P1 and P2: |
|
|||
|
|
ρ1P1 + ρ2P2 = 1. |
|
||
5. |
Define |
|
|
|
|
|
R1 = Wpρ2P2, |
|
S1 = Wuρ1P1, |
||
|
R2 = WpP1P2, |
|
S2 = − WuP1P2. |
6.Define Q = [R2R2 + S2S2 ]+.
7.Let V be an inner function with the property that
|
|
|
|
|
R1R2 + S1S2 |
V |
|||
|
|
|
|
|
|
Q |
|||
has no poles in |
|
+. |
|
|
|
|
|
|
|
C |
|
|
|
|
|
|
|||
8. Define |
R |
+ S |
S |
||||||
|
|
|
R |
||||||
|
|
U1 = |
1 |
2 |
1 |
2 |
V, U2 = QV. |
||
|
|
|
|
Q |
|
||||
|
|
|
|
|
|
|
|
9.Define U4 = [12 − U3]+.
10.Define
T1 |
= |
U1 |
, T2 |
= |
U2 |
. |
|
|
|||||
|
|
U4 |
|
U4 |
11.Let θ be a solution to Problem 15.3.
12.If kT1 − θT2k∞ ≥ 1 then Problem 15.2 has no solution.
13.The controller
|
RC = |
ρ1 + θP2 |
, |
|
ρ2 − θP1 |
||
|
is a solution to Problem 15.2. |
|
|
E15.4 |
M¨obius functions |
|
Finish |
E15.5 |
Show that RH2− and RH2+ are orthogonal with respect to the inner product on RL2 |
||
|
defined in equation (15.5). |
|
|