0% found this document useful (0 votes)
163 views4 pages

8 KYP Lemma: N M N N M

The document summarizes key theorems known as the KYP Lemma related to dissipation inequalities. Specifically: 1) The KYP Lemma refers to theorems about dissipation inequalities where the system is linear and the supply rate is quadratic. 2) It establishes that if a dissipation inequality has a solution, it has a quadratic solution. 3) Frequency domain versions of the KYP Lemma relate positive definiteness of the supply rate over frequency subspaces to the existence of solutions to the dissipation inequality.
Copyright
© Attribution Non-Commercial (BY-NC)
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
0% found this document useful (0 votes)
163 views4 pages

8 KYP Lemma: N M N N M

The document summarizes key theorems known as the KYP Lemma related to dissipation inequalities. Specifically: 1) The KYP Lemma refers to theorems about dissipation inequalities where the system is linear and the supply rate is quadratic. 2) It establishes that if a dissipation inequality has a solution, it has a quadratic solution. 3) Frequency domain versions of the KYP Lemma relate positive definiteness of the supply rate over frequency subspaces to the existence of solutions to the dissipation inequality.
Copyright
© Attribution Non-Commercial (BY-NC)
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

MIT EECS 6.241 (FALL 2009) LECTURE NOTES BY A.

MEGRETSKI

8 KYP Lemma
The dissipation inequality associated with a continuous time state space model
ẋ(t) = f (x(t), u(t)) (8.1)
defined by a function f : Rn × Rm 7→ Rn , and a supply rate σ (a function σ : Rn × Rm 7→
R), the associated dissipation inequality is the condition
V̇ (x)f (x, u) ≤ σ(x, u) ∀ x ∈ Rn , u ∈ Rm , (8.2)
where V : Rn → R (the associated storage function) is a continuously differentiable
function, and V̇ (x) means the gradient of V at the point x ∈ Rn . In a typical application,
f (·, ·) and σ(·, ·) are given, and (8.2) is to be solved with respect to V (·). While having
a dissipation inequality established for a given state space model is always beneficial
(unless (8.2) is trivial, as in the case when V ≡ 0 and σ ≥ 0 is non-negative), dissipation
inequalities with non-negative storage functions are especially useful, as they establish
that the integrals
Z T Z T
σ(x(t), u(t))dt ≥ dV (x(t)) = V (x(T )) − V (x(0)) ≥ −V (x(0))
0 0

are bounded from below as T → ∞.


When f : X × U 7→ X defines a discrete time state space model
x(t + 1) = f (x(t), u(t)), (8.3)
the similar dissipation inequality with supply rate σ : X × U 7→ R and storage function
V : X 7→ R has the form
V (f (x, u)) − V (x) ≤ σ(x, u) ∀ x ∈ X, u ∈ U. (8.4)
The notion of ”KYP Lemma” refers to a loosely defined set of frequently used theorems
related to dissipation inequalities in which the function is linear and the supply rate σ is
a quadratic form, i.e.
 0  
x x
f (x, u) = Ax + Bu, σ(x, u) = Q ∀ x ∈ X = Rn , u ∈ U = Rm , (8.5)
u u
where Q = Q0 , A, B are fixed real matrices. (For example, the case
σ(x, u) = γ 2 |u|2 − |Cx + Du|2
is relevant to establishing upper bounds for the L2 gain from u to y = Cx + Du subject
(8.1) or (8.3).) This section presents some of these theorems.

39
8.1 Storage Functions: Quadratic vs. Non-Quadratic
The simplest part of the KYP Lemma states that dissipation inequalities (8.2) or (8.4)
with linear f and quadratic σ are satisfied for some continuous V if and only if they are
satisfied for some quadratic storage function V (x) = x0 P x (where P = P 0 ), in which case
(8.2) takes the form of the quadratic form inequality

2x0 P (Ax + Bu) ≤ σ(x, u) ∀ x, u, (8.6)

or, equivalently, the matrix inequality

P A + A0 P P B
 
≤ Q, (8.7)
B0P 0

and (8.4) takes the form of the quadratic form inequality

(Ax + Bu)0 P (Ax + Bu) − x0 P x ≤ σ(x, u) ∀ x, u, (8.8)

or, equivalently, the matrix inequality


 0
A P A − P A0 P B

≤ Q. (8.9)
B0P A B0P B

Theorem 8.1 is the statement for the continuous time case, without a positive defi-
niteness restriction on V .

Theorem 8.1 If functions f : Rn × Rm 7→ Rn and σ : Rn × Rm 7→ R defined by


(8.5) satisfy (8.2) for some continuously differentiable function V : Rn →
7 R then (8.6) is
satisfied for some n-by-n real symmetric matrix P = P 0 .

Theorem 8.2 is the similar statement for the continuous time case, with a positive
definiteness restriction on V .

Theorem 8.2 If functions f : Rn × Rm 7→ Rn and σ : Rn × Rm 7→ R defined by (8.5)


satisfy (8.2) for some continuously differentiable function V : Rn 7→ [0, ∞) then (8.6) is
satisfied for some n-by-n real symmetric positive semidefinite matrix P = P 0 ≥ 0.

Theorem 8.3 is the statement for the discrete time case, without a positive definiteness
restriction on V .

Theorem 8.3 If functions f : Rn × Rm 7→ Rn and σ : Rn × Rm 7→ R defined by (8.5)


satisfy (8.4) for some continuous function V : Rn 7→ R then (8.8) is satisfied for some
n-by-n real symmetric matrix P = P 0 .

40
Theorem 8.4 is the similar statement for the discrete time case, with a positive defi-
niteness restriction on V .
Theorem 8.4 If functions f : Rn × Rm 7→ Rn and σ : Rn × Rm 7→ R defined by (8.5)
satisfy (8.4) for some continuous function V : Rn 7→ [0, ∞) then (8.8) is satisfied for some
n-by-n real symmetric positive semidefinite matrix P = P 0 ≥ 0.
Example 8.1 For example, according to Theorem 8.1, the dissipation inequality
V̇ (y)(ay + u) ≤ yu ∀ y, u ∈ R (8.10)
(where a ∈ R is a parameter) has a continuously differentiable solution V : R 7→ R if and only
if it has a solution V (y) = P y 2 , in which case (8.10) has the form
2P y(ay + u) ≤ yu ∀ y, u ∈ R,
which is equivalent to P = 0.5, P a ≤ 0, i.e. allows one to conclude that (8.10) is feasible iff
a ≤ 0.

8.2 Frequency Domain Inequalities


While originally given for real vector arguments x, u, the quadratic form definition from
(8.5) can be naturally extended to a Hermitian form definition
 0  
x x
σ(x, u) = Q ∀ x ∈ X = Cn , u ∈ U = Cm . (8.11)
u u
Substituting a pair x, u of complex vectors satisfying the equality
jωx = Ax + Bu (ω ∈ R), or x = 0 (ω = ∞) (8.12)
into (8.6) yields σ(x, u) ≥ 0. Similarly, substituting a pair x, u of complex vectors satis-
fying the equality
zx = Ax + Bu (z ∈ C, |z| = 1) (8.13)
into (8.8) yields σ(x, u) ≥ 0 as well.
The ”frequency domain inequality” versions of the KYP Lemma establish the inverse
implications, stating that positive definiteness of σ subject to (8.12) implies the existence
of P = P 0 satisfying (8.6), while positive definiteness of σ subject to (8.13) implies the
existence of P = P 0 satisfying (8.6).
Theorem 8.5 is the continuous time version with strict inequalities.
Theorem 8.5 If real matrices Q = Q0 , A, B are such that the Hermitian form (8.11) is
positive definite on the subspaces defined by (8.12) for all ω ∈ R ∪ {∞} then there exists
real matrix P = P 0 such that the quadratic form
σ(x, u) − 2x0 P (Ax + Bu)
is positive definite.

41
Theorem 8.6 is the discrete time version with strict inequalities.

Theorem 8.6 If real matrices Q = Q0 , A, B are such that the Hermitian form (8.11) is
positive definite on the subspaces defined by (8.13) for all z ∈ C such that |z| = 1 then
there exists real matrix P = P 0 such that the quadratic form

σ(x, u) + x0 P x − (Ax + Bu)0 P (Ax + Bu)

is positive definite.

The standard non-strict frequency domain inequality versions of the KYP Lemma
require an additional assumption of reachability (or controllability) of the pair (A, B),
which means that the matrix

Mc (A, B) = [B AB A2 B . . . An−1 B]

must be right invertible. Otherwise, the implication is not valid, as shown by the following
example.

Example 8.2 For n = m = 1, A = −1, B = 0, σ(x, u) = Re(x0 u) we have σ(x, u) = 0 ≥ 0


whenever jωx = Ax + Bu or x = 0, but the quadratic form

σ(x, u) − 2x0 P (Ax + Bu) = xu + 2P x2

is not positive semidefinite for all P .

Theorem 8.7 is the continuous time version with non-strict inequalities.

Theorem 8.7 If real matrices Q = Q0 , A, B are such that the pair (A, B) is reachable,
and the Hermitian form (8.11) is positive semidefinite on the subspaces defined by (8.12)
for all ω ∈ R then there exists real matrix P = P 0 satifying (8.6).

Theorem 8.6 is the discrete time version with strict inequalities.

Theorem 8.8 If real matrices Q = Q0 , A, B are such that the pair (A, B) is reachable,
and the Hermitian form (8.11) is positive semidefinite on the subspaces defined by (8.13)
for all z ∈ C such that |z| = 1 then there exists real matrix P = P 0 satisfying (8.8).

Example 8.3 According to Theorem 8.7, the dissipation inequality from Example 8.1 has a
quadratic form solution V (y) = P y 2 if and only if the Hermitian form σ(y, u) = Re(y 0 u) is
positive semidefinite subject to jωy = ay + u for all ω ∈ R. Since, for jωy = ay + u,

Re(y 0 u) = Re(jω − a)|y|2 ) = −a|y|2 ,

the condition holds iff a ≤ 0.

42

You might also like