You are on page 1of 4

Balakrishnan

Nov 13, 2003

EE695R Homework #6
Due: 12/02/03

1. How slowly can state trajectories grow?


In class, we derived LMI conditions for each of the four system types, which, given α,
guaranteed that for some K ∈ R, kx(t)k2 ≤ Kkx(0)k2e−αt for all t ≥ 0 and for all
x(0). In words, the trajectories decay “at least as fast as e−αt . In this problem, you
will explore the question of how slowly the trajectories can grow.
We will define the minimum growth rate of the trajectories of a system as
( )

For some positive K ∈ R, kx(t)k2 ≥ Kkx(0)k2 eβt for all
βmgr = max β
.
t ≥ 0, and for all x(0)

Thus, βmgr the smallest possible growth rate for the trajectories of the system: The
trajectories of the system grow “at least as fast as eβmgr t”.

(a) Consider the LTI system ẋ = Ax. Given some β0, use Lyapunov functions of the
form V (ψ) = ψ T P ψ, P > 0 to write down an LMI that guarantees that βmgr ≥ β0.
(b) For the LTI system ẋ = Ax, express βmgr in terms of the eigenvalues of A.
(c) Find the best lower bound on βmgr for the following system, using Lyapunov
functions of the form V (ψ) = ψ T P ψ, P > 0.
 
−1 −2 −1
d
x(t) =  0 −1 −1  x(t).
 
dt
1 2 −1

(d) Consider the polytopic system


ẋ = A(t)x, A(t) ∈ Co {A1 , . . . , AL } .
Given some β0 , use Lyapunov functions of the form V (ψ) = ψ T P ψ, P > 0 to
write down an LMI that guarantees that βmgr ≥ β0 .
(e) Find the best lower bound on βmgr for the following system, using Lyapunov
functions of the form V (ψ) = ψ T P ψ, P > 0.
 
−1 + a1(t) −2 −1 − a1(t)
d
x(t) = 

0 −1 + a2(t) −1  x(t),

dt
1 + a1(t) 2 −1 + a1(t)
where for all t ≥ 0, |ai(t)| ≤ 1, i = 1, . . . , 2.
(f) Consider the norm-bound system
ẋ = Ax + Bp p, q = Cq x + Dqp p, p(t)T p(t) ≤ q(t)T q(t) for all t ≥ 0.
Given some β0 , use Lyapunov functions of the form V (ψ) = ψ T P ψ, P > 0 to
write down an LMI that guarantees that βmgr ≥ β0 .

1
(g) Find the best lower bound on βmgr for the following system, using Lyapunov
functions of the form V (ψ) = ψ T P ψ, P > 0.
   
−1 −2 −1 0 1
d
x(t) =  0 −1 −1  x(t) +  1 0  p(t)
   
dt
1 2 −1 1 1
" #
−1 −1 1
q(t) = x(t)
0 0 2

with p(t)T p(t) ≤ q(t)T q(t) for all t.

2. LMIs for other well-posedness problems.


Consider the interconnection shown in Figure 1, where M ∈ Cn×n is a fixed matrix, and
∆ ∈ Cn×n belongs to some set ∆. We will refer to this as the “M –∆ interconnection”
in the sequel.
We say that the M –∆ interconnection is well-posed over ∆ if

det(I − M ∆) 6= 0 for all ∆ ∈ ∆, (1)

where I is the n × n identity matrix.

- M

∆ 

Figure 1: The M –∆ interconnection.

(a) Suppose that n o


∆pass = ∆ ∈ Cn×n | ∆ + ∆∗ > 0 .
Show that the M –∆ interconnection is well-posed over ∆pass if and only if

M + M ∗ ≤ 0.

(b) Suppose that


n o
∆pass,diag = ∆ ∈ Cn×n | ∆ + ∆∗ > 0, ∆ is diagonal .

Show that the M –∆ interconnection is well-posed over ∆pass,diag if

for some diagonal W ∈ Rn×n with W > 0, we have W M + M ∗ W ≤ 0. (2)

2
(c) Suppose that
n o
∆pass,diag,real = ∆ ∈ Rn×n | ∆ > 0, ∆ is diagonal .

Show that the M –∆ interconnection is well-posed over ∆pass,diag if

for some diagonal W ∈ Cn×n with W + W ∗ > 0, we have W M + M ∗ W ∗ ≤ 0.


(3)

Let
 
−0.4326 + 0.1746j 0.2877 − 0.5883j 1.1892 + 0.1139j
M =  −1.6656 − 0.1867j −1.1465 + 2.1832j −0.0376 + 1.0668j  . (4)
 

0.1253 + 0.7258j 1.1909 − 0.1364j 0.3273 + 0.0593j

This matrix can also be loaded into matlab by executing the .m file
http://ece.www.ecn.purdue.edu/~ragu/695R/for-web/hw6M.m

(d) Is the M –∆ interconnection is well-posed over ∆pass?


(e) Solve the LMI feasibility problem (2) with M as given by (4). What can you
conclude about the well-posedness of the M –∆ interconnection over ∆pass,diag?
(f) Solve the LMI feasibility problem (3) with M as given by (4). What can you
conclude about the well-posedness of the M –∆ interconnection over ∆pass,diag,real?
Comments and hints: Note that the LMIs in (2) and in (3) are complex LMIs.
While solving these LMI feasibility problems, for numerical conditioning, add the
following additional constraints, wherever applicable:
1
10, 000I ≥ W + W ∗ ≥ I, 10, 000I ≥ (W − W ∗ ) ≥ −10, 000I.
2j

3. Suppose A ∈ Rn×n. Find conditions on A such for every P = P T > 0, we have


AT P + P A < 0. In other words, for what finite-dimensional linear systems does every
quadratic function serve as a Lyapunov function?

4. Consider the following system models:


LTI system:
   
−1 −2 −1 1 −1
d
x(t) =  0 −1 −1  x(t) +  0 1 
 w(t)
  
dt
1 2 −1 1 2
" #
1 1 −1
z(t) = x(t)
1 −1 2

Polytopic system:

3
   
−1 + a1(t) −2 −1 − a1(t) b1 (t) −1
d
x(t) =  0 −1 + a2(t) −1  x(t) +  0 0  w(t)
   
dt
1 + a1(t) 2 −1 + a1(t) 1 2
" #
1 1 −1
z(t) = x(t)
1 −2 2
where for all t ≥ 0, |ai (t)| ≤ α, i = 1, . . . , 2, |b1(t)| ≤ 1. Unless otherwise specified,
α = 0.1.

Norm-bound system:
     
−1 −2 −1 0 1 1 −1
d
x(t) =  0 −1 −1  x(t) +  1 0  p(t) +  0 1 
 w(t)
    
dt
1 2 −1 1 1 1 2
" #
−1 −1 1
q(t) = x(t)
0 0 2
" #
1 1 −1
z(t) = x(t)
1 −1 2

with p(t)T p(t) ≤ (0.1)2 q(t)T q(t) for all t.

Diagonal norm-bound system: Same data as with the above norm-bound system, but
here
|pi (t)| ≤ 0.1|qi (t)|, i = 1, 2.

For each of the systems:


(a) Find the smallest ellipsoidal approximation containing the set of states reachable
from the origin with unit-energy inputs w.
(b) Find the smallest ellipsoidal approximation containing the set of states reachable
from the origin with componentwise unit-energy inputs w.
(c) Find the best upper bound, using quadratic Lyapunov functions, on the energy
of the output z, for the initial condition [−1 1 − 1]T .
(d) Find the best upper bound, using quadratic Lyapunov functions, on the peak
value of the output z, for the initial condition [−1 1 − 1]T .
(e) Find the best upper bound, using quadratic Lyapunov functions, on the L2 gain
of the system from w to z.
5. With Bu = [1 0 1]T , design a (constant) state-feedback matrix K using the techniques
discussed in class such that u = Kx is “optimal”, whenever applicable, for items (a),
(c) and (e) in Problem 1, with the additional constraint that ku(t)k2 ≤ 10, for all t ≥ 0
when the system is initialized with x(0) = [−1 1 − 1]T .

You might also like