Outline Preliminaries for Model Predictive Control course

Introduction Stability, equilibria, invariant sets Lyapunov function theory Disturbances and robust stability Basic MPC problem and nominal stability

James B. Rawlings
2 Department of Chemical and Biological Engineering University of WisconsinMadison 3

Graduate School in Systems, Optimization, Control and Networks (SOCN) K.U. Leuven Leuven, Belgium August 29September 5, 2013

SOCN 2013

MPC short course

1 / 45

SOCN 2013

MPC short course

2 / 45

Predictive control
Reconcile the past
11 00 00 11 1 0 1 0 1 0

State estimation
Forecast the future
00 11 11 00 00 11

Reconcile the past

11 00 00 11 1 0 1 0 1 0

Forecast the future

00 11 11 00 00 11

MH Estimate Measurement

Forecast MPC control

MH Estimate Measurement

Forecast MPC control

sensors y

1 0 1 0 0 1 1 0 0 1 1 0 0 1 0 1 1 0 00 0 1 11 0 1 0 1 0 1 0 0 1 0 01 1 0 0 1 01 1 0 1 1 0 1 1 0 1 0

1 0 1 0 1 0 0 1 0 1 0 1 1 0 1 0 0 1 0 1 10 0 1 0 1 0 1 0 1 00 1 1 1 0 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 1 0 1 0 01 1 0 1 00 1 0 1 0 1 0 1 1 0 0 1 0 1 1 0 1 0 1 0 0 1

1 0 0 1

1 0 0 1

1 0 0 0 0 0 0 0 0 0 0 0 0 01 1 01 1 01 1 01 1 01 1 01 1 01 1 01 1 01 1 01 1 01 1 0 1

1 0 0 1 1 0 1 0

1 0 0 1

1 0 0 1 1 00 1 1 0 0 1

sensors y

1 0 1 0 0 1 1 0 0 1 1 0 0 1 0 1 1 0 00 0 1 11 0 1 0 1 0 1 0 0 1 0 01 1 0 0 1 01 1 0 1 1 0 1 1 0 1 0

1 0 1 0 1 0 0 1 0 1 0 1 1 0 1 0 1 0 0 1 10 0 1 0 1 0 1 0 1 00 1 1 1 0 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 1 0 1 0 01 1 0 1 00 1 0 1 0 1 0 1 1 0 0 1 0 1 1 0 1 0 1 0 0 1

1 0 0 1

1 0 0 1

1 0 0 0 0 0 0 0 0 0 0 0 0 01 1 01 1 01 1 01 1 01 1 01 1 01 1 01 1 01 1 01 1 01 1 0 1

1 0 0 1 1 0 1 0

1 0 0 1

1 0 0 1 1 00 1 1 0 0 1

actuators u t time

actuators u t time

min
u (t ) 0

2 |ysp g (x , u )|2 Q + |usp u |R dt

min
x0 ,w (t ) T

|y g (x , u )|2 f (x , u )|2 R + |x Q dt

x = f (x , u ) x (0) = x0
SOCN 2013

x = f (x , u ) + w y = g (x , u ) + v
3 / 45 SOCN 2013

(process noise) (measurement noise)

MPC short course 4 / 45

(given)
MPC short course

y = g (x , u )

System model1
We consider systems of the form x + = f (x , u ) where the state x lies in Rn and the control (input) u lies in Rm ; In this formulation x and u denote, respectively, the current state and control, and x + the successor state. We assume in the sequel that the function f : Rn Rm Rn is continuous. Let (k ; x , u) denote the solution of = f (x , u ) at time k if the initial state is x (0) = x and the control sequence is u = {u (0), u (1), u (2), . . .}; The solution exists and is unique.
Most of this preliminary material is taken from Rawlings and Mayne (2009, Appendix B). Downloadable from www.che.wisc.edu/~jbraw/mpc.
SOCN 2013 MPC short course 5 / 45 1

Existence of solutions to model

If a state-feedback control law u = (x ) has been chosen, the closed-loop system is described by x + = f (x , (x )). Let (k ; x , ()) denote the solution of this dierence equation at time k if the initial state at time 0 is x (0) = x ; the solution exists and is unique (even if () is discontinuous). If () is not continuous, as may be the case when () is a model predictive control (MPC) law, then f ((), ()) may not be continuous. In this case we assume that f ((), ()) is locally bounded.

Denition 1 (Locally bounded)

A function f : X X is locally bounded if, for any x X , there exists a neighborhood N of x such that f (N ) is a bounded set, i.e., if there exists a M > 0 such that |f (x )| M for all x N .

x+

SOCN 2013

MPC short course

6 / 45

Stability and equilibrium point

We would like to be sure that the controlled system is stable, i.e., that small perturbations of the initial state do not cause large variations in the subsequent behavior of the system, and that the state converges to a desired state or, if this is impossible due to disturbances, to a desired set of states. If convergence to a specied state, x say, is sought, it is desirable for this state to be an equilibrium point:

Positive invariant set

In other situations, for example when studying the stability properties of an oscillator, convergence to a specied closed set A Rn is sought. If convergence to a set A is sought, it is desirable for the set A to be positive invariant:

Denition 3 (Positive invariant set)

A set A is positive invariant for the system x + = f (x ) if x A implies f (x ) A. Clearly, any solution of x + = f (x ) with initial state in A, remains in A. The (closed) set A = {x } consisting of a (single) equilibrium point is a special case; x A (x = x ) implies f (x ) A (f (x ) = x ).

Denition 2 (Equilibrium point)

A point x is an equilibrium point of x + = f (x ) if x (0) = x implies x (k ) = (k ; x ) = x for all k 0. Hence x is an equilibrium point if it satises x = f (x )

SOCN 2013

MPC short course

7 / 45

SOCN 2013

MPC short course

8 / 45

Distance to a set; set addition

K, K , KL, and PD functions

Denition 4
Dene distance from point x to set A |x |A := inf |x z |
z A

A function : R0 R0 belongs to class K if it is continuous, zero at zero, and strictly increasing; : R0 R0 belongs to class K if it is a class K and unbounded ( (s ) as s ). A function : R0 I0 R0 belongs to class KL if it is continuous and if, for each t 0, (, t ) is a class K function and for each s 0, (s , ) is nonincreasing and satises limt (s , t ) = 0. A function : R R0 belongs to class PD (is positive denite) if it is continuous and positive everywhere except at the origin.

If A = {x }, then |x |A = |x x | which reduces to |x | when x = 0. Set addition: A B := {a + b | a A, b B }.

SOCN 2013

MPC short course

9 / 45

SOCN 2013

MPC short course

10 / 45

Some useful properties of K functions

Stability Denitions

Denition 5 (Various forms of stability (constrained))

The following useful properties of these functions are established in Khalil (2002, Lemma 4.2):
1 if 1 () and 2 () are K functions (K functions), then 1 () and (1 2 )() := 1 (2 ()) are K functions (K functions).

Suppose X Rn is positive invariant for x + = f (x ), that A is closed and positive invariant for x + = f (x ), and that A lies in the interior of X . Then A is
1

Moreover, if 1 () and 2 () are K functions and () is a KL function, then (r , s ) = 1 ( (2 (r ), s )) is a KL function.

locally stable in X if, for each > 0, there exists a = () > 0 such that x X (A B ), implies |(i ; x )|A < for all i I0 .a locally attractive in X if there exists a > 0 such that x X (A B ) implies |(i ; x )|A 0 as i . attractive in X if |(i ; x )|A 0 as i for all x X .
B denotes the unit ball in Rn .

3 a

SOCN 2013

MPC short course

11 / 45

SOCN 2013

MPC short course

12 / 45

Stability Denitions (cont.)

Asymptotic stability stronger denition

An alternative, stronger denition of asymptotic stability is becoming popular. Ill call it the KL version.

Denition 5 (Various forms of stability (constrained))

4

locally asymptotically stable in X if it is locally stable in X and locally attractive in X . asymptotically stable with a region of attraction X if it is locally stable in X and attractive in X . locally exponentially stable with a region of attraction X if there exist > 0, c > 0, and (0, 1) such that x X (A B ) implies |(i ; x )|A c |x |A i for all i I0 . exponentially stable with a region of attraction X if there exists a c > 0 and a (0, 1) such that |(i ; x )|A c |x |A i for all x X , all i I0 .

Denition 6 (Asymptotic stability (constrained KL version))

Suppose X Rn is positive invariant for x + = f (x ), that A is closed and positive invariant for x + = f (x ), and that A lies in the interior of X . The set A is asymptotically stable with a region of attraction X for x + = f (x ) if there exists a KL function () such that, for each x X |(i ; x )|A (|x |A , i ) i I0 (1)

See Teel and Zaccarian (2006) and the Notes on Recent MPC Literature link on: www.che.wisc.edu/~jbraw/mpc for further discussion of the dierences in the two denitions. If f () is continuous, the two denitions are equivalent.
SOCN 2013 MPC short course 14 / 45

SOCN 2013

MPC short course

13 / 45

Lyapunov function
Denition 7 (Lyapunov function)
A function V : Rn R0 is said to be a Lyapunov function for the system x + = f (x ) and set A if there exist functions i K , i = 1, 2 and 3 PD such that for any x Rn , V (x ) 1 (|x |A ) V (x ) 2 (|x |A ) V (f (x )) V (x ) 3 (|x |A ) (2) (3) (4)

Some ne print

Remark 8
It is shown in Jiang and Wang (2002), Lemma 2.8, that, under the assumption that f () is continuous, we can always assume that 3 () in (4) is a K function.

If V () satises (2)(4) for all x X where X A is a positive invariant set for x + = f (x ), then V () is said to be a Lyapunov function in X for the system x + = f (x ) and set A.

SOCN 2013

MPC short course

15 / 45

SOCN 2013

MPC short course

16 / 45

Lyapunov function Asymptotic stability

Converse Lyapunov theorem Asymptotic stability

Theorem 9 merely provides a sucient condition for global asymptotic stability that might be thought to be conservative. The next result, a converse stability theorem by Jiang and Wang (2002), establishes necessity under a stronger hypothesis, namely that f () is continuous rather than locally bounded.

The existence of a Lyapunov function is a sucient condition for global asymptotic stability as shown in the next result which we prove under the assumption, common in MPC, that 3 () is K function.

Theorem 9 (Lyapunov function and GAS)

Suppose V () is a Lyapunov function for x + = f (x ) and set A with 3 () a K function. Then A is globally asymptotically stable.

Theorem 10 (Converse theorem for asymptotic stability)

Let A be compact and f () continuous. Suppose that the set A is globally asymptotically stable for the system x + = f (x ). Then there exists a smooth Lyapunov function for the system x + = f (x ) and set A. A proof of this result is given in Jiang and Wang (2002), Theorem 1, part 3.

SOCN 2013

MPC short course

17 / 45

SOCN 2013

MPC short course

18 / 45

Asymptotic stability Constrained case

Exponential stability Constrained case

Theorem 12 (Lyapunov function for exponential stability) Theorem 11 (Lyapunov function for asymptotic stability (constrained case))
Suppose X Rn is positive invariant for x + = f (x ), that A is closed and positive invariant for x + = f (x ), and that A lies in the interior of X . If there exists a Lyapunov function in X for the system x + = f (x ) and set A with 3 () a K function, then A is asymptotically stable for x + = f (x ) with a region of attraction X . Suppose X Rn is positive invariant for x + = f (x ), that A is closed and positive invariant for x + = f (x ), and that A lies in the interior of X . If there exists V : Rn R0 satisfying the following properties for all x X
a1 |x | A V (x ) a2 |x |A

V (f (x )) V (x ) a3 |x | A in which a1 , a2 , a3 , > 0, then A is exponentially stable for x + = f (x ) with a region of attraction X .

SOCN 2013

MPC short course

19 / 45

SOCN 2013

MPC short course

20 / 45

Converse theorem for exponential stability

Exercise B.3: A converse theorem for exponential stability
a

Robust Stability
We turn now to stability conditions for systems subject to bounded disturbances (not vanishingly small) and described by x + = f (x , w ) where the disturbance w lies in the compact set W. This system may equivalently be described by the dierence inclusion x + F (x ) where the set F (x ) := {f (x , w ) | w W} Let S (x ) denote the set of all solutions of (5) or (6) with initial state x . (6) (5)

Assume that the origin is globally exponentially stable (GES) for the system x + = f (x ) in which f is Lipschitz continuous. Show that there exists a Lipschitz continuous Lyapunov function V () for the system satisfying for all x Rn a1 |x | V (x ) a2 |x | V (f (x )) V (x ) a3 |x | in which a1 , a2 , a3 , > 0. Hint: Consider summing the solution |(i ; x )| on i as a candidate Lyapunov function V (x ). Establish also that in the Lyapunov function dened above, any > 0 is valid, and the constant a3 can be chosen as large as one wishes.
SOCN 2013 MPC short course 21 / 45

SOCN 2013

MPC short course

22 / 45

Positive invariant sets

Stability and attraction

We require, in the sequel, that the set A is positive invariant for (5) (or for x + F (x )):

Denition 14 (Local stability (disturbances))

The closed positive invariant set A is locally stable for x + = f (x , w ), w W (or for x + F (x )) if, for all > 0, there exists a > 0 such that, for each x satisfying |x |A < , each solution S (x ) satises |(i )|A < for all i I0 .

Denition 13 (Positive invariance with disturbances)

The set A is positive invariant for x + = f (x , w ), w W if x A implies f (x , w ) A for all w W; it is positive invariant for x + F (x ) if x A implies F (x ) A. Clearly the two denitions are equivalent; A is positive invariant for x + = f (x , w ), w W, if and only if it is positive invariant for x + F (x ). In Denitions 14-16, we use positive invariant to denote positive invariant for x + = f (x , w ), w W or for x + F (x ).

Denition 15 (Global attraction (disturbances))

The closed positive invariant set A is globally attractive for the system x + = f (x , w ), w W (or for x + F (x )) if, for each x Rn , each solution () S (x ) satises |(i )|A 0 as i .

SOCN 2013

MPC short course

23 / 45

SOCN 2013

MPC short course

24 / 45

Asymptotic stability

Lyapunov function
To cope with disturbances we require a modied denition of a Lyapunov function.

Denition 16 (GAS (disturbances))

The closed positive invariant set A is globally asymptotically stable for x + = f (x , w ), w W (or for x + F (x )) if it is locally stable and globally attractive. An alternative denition of global asymptotic stability of A for x + = f (x , w ), w W, if A is compact, is the existence of a KL function () such that for each x Rn , each S (x ) satises |(i )|A (|x |A , i ) for all i I0 .

Denition 17 (Lyapunov function (disturbances))

A function V : Rn R0 is said to be a Lyapunov function for the system x + = f (x , w ), w W (or for x + F (x )) and set A if there exist functions i K , i = 1, 2 and 3 PD such that for any x Rn , V (x ) 1 (|x |A ) V (x ) 2 (|x |A ) sup V (z ) V (x ) 3 (|x |A )
z F (x )

(7) (8) (9)

SOCN 2013

MPC short course

25 / 45

SOCN 2013

MPC short course

26 / 45

GAS (disturbances)

Input-to-State Stability
In input-to-state stability (Sontag and Wang, 1995; Jiang and Wang, 2001) we seek a bound on the state in terms of a uniform bound on the disturbance sequence w := {w (0), w (1), . . .}. Let denote the usual norm for sequences, i.e., w := supk 0 |w (k )|.

Inequality 9 ensures V (f (x , w )) V (x ) 3 (|x |A ) for all w W. The existence of a Lyapunov function for the system x + F (x ) and set A is a sucient condition for A to be globally asymptotically stable for x + F (x ) as shown in the next result.

Denition 19 (Input-to-state stable (ISS))

The system x + = f (x , w ) is (globally) input-to-state stable (ISS) if there exists a KL function () and a K function () such that, for each x Rn , and each disturbance sequence w = {w (0), w (1), . . .} in |(i ; x , wi )| (|x | , i ) + ( wi ) for all i I0 , where (i ; x , wi ) is the solution, at time i , if the initial state is x at time 0 and the input sequence is wi := {w (0), w (1), . . . , w (i 1)}.

Theorem 18 (Lyapunov function for GAS (disturbances))

Suppose V () is a Lyapunov function for x + = f (x , w ), w W (or for x + F (x )) and set A with 3 () a K function. Then A is globally asymptotically stable for x + = f (x , w ), w W (or for x + F (x )).

SOCN 2013

MPC short course

27 / 45

SOCN 2013

MPC short course

28 / 45

The basic nonlinear, constrained MPC problem

Feasible sets
The set of feasible initial states and associated control sequences

The system model is x

+

= f (x , u )

(10) ZN = {(x , u) | u (k ) U, (k ; x , u) X for all k I0:N 1 , and (N ; x , u) Xf } and Xf X is the feasible terminal set. The set of feasible initial states is XN = {x Rn | u UN such that (x , u) ZN } (11)

Both state and input are subject to constraints x (k ) X , u (k ) U for all k I0

Given an integer N (referred to as the nite horizon), and an input sequence u of length N , u = {u (0), u (1), . . . , u (N 1)}, let (k ; x , u) denote the solution of (10) at time k for a given initial state x (0) = x . Terminal constraint (and penalty) (N ; x , u) Xf X

For each x XN , the corresponding set of feasible input sequences is UN (x ) = {u | (x , u) ZN }

SOCN 2013

MPC short course

29 / 45

SOCN 2013

MPC short course

30 / 45

Cost function and control problem

Control law and closed-loop system

For any state x Rn and input sequence u UN , we dene

N 1

The control law is N (x ) = u 0 (0; x ) The optimum may not be unique; then N () is a point-to-set map Closed-loop system x + = f (x , N (x )) x + f (x , N (x )) dierence equation dierence inclusion

VN (x , u) =
k =0

((k ; x , u), u (k )) + Vf ((N ; x , u))

(x , u ) is the stage cost; Vf (x (N )) is the terminal cost Consider the nite horizon optimal control problem PN (x ) :
uUN

min VN (x , u)

Nominal closed-loop stability question; is the origin stable? If yes, what is the region of attraction? All of XN ?

SOCN 2013

MPC short course

31 / 45

SOCN 2013

MPC short course

32 / 45

Inherent robustness of the nominal controller

Basic MPC assumptions

Consider a process disturbance d , x + = f (x , (x )) + d A measurement disturbance xm = x + e Nominal controller with disturbance x + f (x , N (xm )) + d x + f (x , N (x + e )) + d x + F (x , w ) w = (d , e )

Assumption 20 (Continuity of system and cost)

The functions f : Rn Rm Rn , : Rn Rm R0 and Vf : Rn R0 are continuous, f (0, 0) = 0, (0, 0) = 0, and Vf (0) = 0.

Assumption 21 (Properties of constraint sets)

The set U is compact and contains the origin. The sets X and Xf are closed and contain the origin in their interiors, Xf X.

Robust stability; is the system x + F (x , w ) input-to-state stable considering w = (d , e ) as the input.

SOCN 2013

MPC short course

33 / 45

SOCN 2013

MPC short course

34 / 45

Basic MPC assumptions

Assumption 22 (Basic stability assumption)
For any x Xf there exists u := f (x ) U such that f (x , u ) Xf and Vf (f (x , u )) + (x , u ) Vf (x ). Note: understanding this requirement created a big research challenge for the development of nonlinear MPC. Credit the celebrated quasi-innite horizon work of Chen and Allg ower (1998) for cracking this problem.

Optimal MPC cost function as Lyapunov function

0 () is a Lyapunov function for the We show that the optimal cost VN closed-loop system. We require three properties. Lower bound. 0 VN (x ) 1 (|x |)

for all

x XN

Assumption 23 (Bounds on stage and terminal costs)

The stage cost () and the terminal cost Vf () satisfy (x , u ) 1 (|x |) Vf (x ) 2 (|x |) x XN , u U x Xf

Given the denition of VN (x , u) as a sum of stage costs, we have using Assumption 23 VN (x , u) (x , u (0; x )) 1 (|x |) so the rst property is established. for all x XN , u UN

in which 1 () and 2 () are K functions

SOCN 2013 MPC short course 35 / 45 SOCN 2013 MPC short course 36 / 45

MPC cost function as Lyapunov function cost decrease

Next we require the cost decrease
0 0 VN (f (x , N (x )) VN (x ) 3 (|x |)

Cost decrease (cont.)

But by Assumption 22 Vf (f (x , f (x ))) + (x , f (x )) Vf (x ) so we have that
0 ) VN VN (x + , u (x ) (x , u (0; x ))

for all

x XN

At state x XN , consider the optimal sequence u0 (x ) = {u (0; x ), u (1; x ), . . . , u (N 1; x )}, and generate a candidate sequence for the successor state, x + := f (x , N (x )) = {u (1; x ), u (2; x ), . . . , u (N 1; x ), f (x (N ))} u with x (N ) := (N ; x , u). This candidate is feasible for x + because Xf is control invariant under control law f () (Assumption 22). The cost is
0 ) = VN VN (x + , u (x ) (x , u (0; x ))

for all

x Xf

The optimal cost is certainly no worse, giving

0 + 0 VN (x ) VN (x ) (x , u (0; x )) 0 + 0 VN (x ) VN (x ) 1 (|x |)

for all

x XN

which is the desired cost decrease with the choice 3 () = 1 ().

Vf (x (N )) + (x (N ), f (x (N ))) + Vf (f (x (N ), f (x (N ))))
SOCN 2013 MPC short course 37 / 45 SOCN 2013 MPC short course 38 / 45

Upper bound
Finally we require the upper bound.
0 VN (x ) 2 (|x |)

Dynamic programming recursion

Next consider N = 2, and optimal control law 2 () for all x XN
0 0 V2 (x ) = min{ (x , u ) + V1 (f (x , u )) | f (x , u ) X1 } u U 0 = (x , 2 (x )) + V1 (f (x , 2 (x )))

x X2

Surprisingly, this one turns out to be the most involved. First, we have the bound from Assumption 23 Vf (x ) 2 (|x |) Next we show that Consider N = 1,
0 (x ) VN

x X2 x X1 x X1

(x , 1 (x )) + = Therefore
0 V1 (x )

0 V1 (f (x , 1 (x )))

for all

x Xf

(x , 1 (x )) + Vf (f (x , 1 (x ))) x X1

Vf (x ) for x Xf , N 1.

0 V2 (x ) Vf (x ) x Xf u U

0 (x ) = min{ (x , u ) + Vf (f (x , u )) | f (x , u ) Xf } V1

Continuing this recursion gives for all N 1

0 VN (x ) Vf (x )

Vf (x )

x Xf

x Xf

SOCN 2013

MPC short course

39 / 45

SOCN 2013

MPC short course

40 / 45

Extending the upper bound from Xf to XN

Question: When can we extend this bound from Xf to the (possibly 0 () is not necessarily continuous. unbounded!) set XN ? Recall that VN Answer: A function can be upper bounded by a K function if and only if it is locally bounded.2 We know from continuity of f () (Assumption 20) that VN (x , u) is a continuous function, hence locally bounded, and therefore so is 0 (x ). VN Therefore, there exists () K such that
0 VN (x ) (|x |)

Asymptotic stability of constrained nonlinear MPC

Why you want a Lyapunov function
0 () is a Lyapunov We have established that the optimal cost VN function on XN for the closed-loop system.

Therefore, the origin is asymptotically stable (KL version) with region of attraction XN . We can also establish robust stability, but lets do that later. If we strengthen the properties of (), we can strengthen the conclusion to exponential stability.
0 () plays in the stability analysis of Notice the essential role that VN MPC.

for all

x XN

Be aware that the MPC literature has been confused about the requirements for this last result.
See Proposition 10 of Notes on Recent MPC Literature link on: www.che.wisc.edu/~jbraw/mpc. Thanks also to Andy Teel.
SOCN 2013 MPC short course 41 / 45 2

In economic MPC we lose this Lyapunov function and have to work to bring it back.

SOCN 2013

MPC short course

42 / 45

Recommended exercises

Further Reading I
H. Chen and F. Allg ower. A quasi-innite horizon nonlinear model predictive control scheme with guaranteed stability. Automatica, 34(10): 12051217, 1998. Z.-P. Jiang and Y. Wang. Input-to-state stability for discrete-time nonlinear systems. Automatica, 37:857869, 2001. Z.-P. Jiang and Y. Wang. A converse Lyapunov theorem for discrete-time systems with disturbances. Sys. Cont. Let., 45:4958, 2002. H. K. Khalil. Nonlinear Systems. Prentice-Hall, Upper Saddle River, NJ, third edition, 2002. J. B. Rawlings and D. Q. Mayne. Model Predictive Control: Theory and Design. Nob Hill Publishing, Madison, WI, 2009. 576 pages, ISBN 978-0-9759377-0-9.

Stability denitions. Example 2.3 Lyapunov functions. Exercise B.2B.4.4 Dynamic programming. Exercise C.1C.2.4 MPC stability results. Theorem 7 and Example 1.3 Exercises 2.11, 2.14, 2.154

Notes on Recent MPC Literature link on: www.che.wisc.edu/~jbraw/mpc. Rawlings and Mayne (2009, Appendices B and C). Downloadable from www.che.wisc.edu/~jbraw/mpc.
4 SOCN 2013 MPC short course 43 / 45

E. D. Sontag and Y. Wang. On the characterization of the input to state stability property. Sys. Cont. Let., 24:351359, 1995.
SOCN 2013 MPC short course 44 / 45

Further Reading II

A. R. Teel and L. Zaccarian. On uniformity in denitions of global asymptotic stability for time-varying nonlinear systems. Automatica, 42: 22192222, 2006.

SOCN 2013

MPC short course

45 / 45