Uncorrected Proof: Steady-State Behaviors in Nonlinear Systems With An Application To Robust Disturbance Rejection

+ Models
JARAP 289 116
1 Annual Reviews in Control xxx (2008) xxxxxx

www.elsevier.com/locate/arcontrol
3 Steady-state behaviors in nonlinear systems with an application
F
4
to robust disturbance rejection
OO
5
A. Isidori a,*, C.I. Byrnes b
a
6 Department of Informatica e Sistemistica, University of Rome La Sapienza, Rome, Italy
7 b
Department of Electrical and Systems Engineering, Washington University, St. Louis, MO, USA
8 Received 1 November 2007; accepted 17 January 2008
9
10
PR
11
Abstract
12
13 Despite the widespread intuitive appeal of the concept of steady-state response and its use in shaping the asymptotic behavior of control
14 systems, this concept has only been rigorously defined for finite-dimensional, linear time invariant systems. In this paper, we investigate this
15 concept for nonlinear systems, following some classical developments in nonlinear dynamics. As an application, we show how the concept in
16 question plays a role of paramount importance in the design of control laws for asymptotic tracking and disturbance attenuation.
ED
17 # 2008 Elsevier Ltd. All rights reserved.
18
Keywords: Limit sets; Steady-state response; Behaviors; Internal model; Regulation; Nonlinear control
19
20 40
CT
1. The classical notion of steady state for instance the case in the classical set-point control problem, 41
21 in which the output of the controlled system is required to 42
22 One of the main concerns in the analysis and design of asymptotically converge to a fixed (but otherwise arbitrary or 43
23 control systems, is the ability to influence or shape the response undetermined) value. Another well-known instance is the case 44
of a given system to assigned external inputs. This can in which the output of a system is required to asymptotically
E
24 45
25 sometimes be achieved by finding the open-loop input which track (or reject) a prescribed sinusoidally varying trajectory (or 46
26 generates a prescribed trajectory. On the other hand, the use of disturbance) of fixed frequency (but otherwise arbitrary in 47
RR
27 closed-loop control is almost always the solution of choice in amplitude and phase). On the other hand, it is well known at 48
28 the presence of uncertainties affecting the control systems itself least in linear system theory that the ability to analyze and 49
29 as well as the external inputs to which the response has to be shape the steady-state response to sinusoidally varying inputs 50
30 shaped. Among various possible criteria by means of which also provides a powerful tool for the analysis and, to a some 51
31 responses can be analyzed and classified, a classical viewpoint extent, for the design of the transient behavior. 52
CO
32 dating back to the origins of control theory consists on the Given the central importance of the notion of steady-state 53
33 separation between steady state and transient responses. response, it is somewhat surprising that a rigorous investigation 54
34 There are several well-known strong arguments in support of and delineation of this concept has never been fully developed 55
35 the important role played by the idea of a steady-state response in the system and control literature, especially for nonlinear 56
36 in system analysis and design. On one hand, in a large number systems. The separation between steady and transient states 57
37 of cases it is actually required, as a design specification, that the presumes, of course, the ability to be able to discern whether or 58
UN
38 controlled system evolves, as time increases, toward a steady not a given system exhibits either one of these two kinds of 59
39 state in which one or more variables describing its behavior behavior. In this respect, a precise, but somewhat restricted, 60
40 exactly match one or more prescribed functions of time. This is definition of steady state is the one that can be found in the 61
classical textbook of Gardner and Barnes (1942): A dynamical 62

system is said to be in the steady state when the variables 63
A plenary address on this subject, and with this title, has been presented by describing its behavior are either invariant with time, or are 64
this author at the 17th IFAC Symposium on Automatic Control in Aerospace,
Toulouse, 2529 June 2007.
(sections of) periodic functions of time. A dynamical system is 65
* Corresponding author. said to be in the transient (or unsteady) state when it is not in 66
E-mail address: isidori@ese.wustl.edu (A. Isidori). steady state. In this distinction there is no predetermined 67
1367-5788/$ see front matter # 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.arcontrol.2008.01.001
Please cite this article in press as: Isidori, A., & Byrnes, C.I., Steady-state behaviors in nonlinear systems with an application to robust
disturbance rejection, Annu Rev Control (2008), doi:10.1016/j.arcontrol.2008.01.001
+ Models
JARAP 289 116
2 A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx

67 124
68 separation between inputs and outputs. Rather, the system is 2. Limit sets
69 only analyzed in terms of how the variables describing its 125
70 behavior depend on time. This viewpoint applies to general 2.1. The limit set of a point
71 dynamical systems and not necessarily to control systems with 126
72 input and output variables; it is a precursor (at least so long as The principle inspiring the study of the steady-state response 127
73 the notion of steady state is concerned) of the behavioral of a linear system to sinusoidally varying inputs, which had so 128
74 viewpoint proposed in much more recent times by Willems much influence in the analysis and design of linear control 129
75 (1991). Note also that, consistently with the behavioral systems, is indeed the same principle which is behind the 130
F
76 framework, this definition speaks of steady state and not of investigation of forced oscillations in nonlinear systems, a 131
77 steady-state response. However, if among the variables classical problem with its origin in celestial mechanics. In this 132
OO
78 describing the behavior of the system one or more can be respect it must be stressed that for a nonlinear system forced by 133
79 viewed as inputs, and if all such variables are either constant or a sinusoidally varying input, the situation is far more complex 134
80 periodically varying, then it makes sense to speak of a steady- than those outlined above, with the possibility of one, or 135
81 state response whenever all other variables describing the several, forced oscillations with varying stability characteristics 136
82 behavior of the system also exhibit the same periodic variation. occurring. In addition, the fundamental harmonic of these 137
83 This notion of steady state of course, is quite restrictive, as periodic responses may agree with the frequency of the forcing 138
PR
84 it only applies to cases in which all relevant variables which term (harmonic oscillations), or with integer multiples or 139
85 describe the behavior of a dynamical systems are periodic divisors of the forcing frequency (higher harmonic, or 140
86 (constant in particular) functions of time. It excludes a subharmonic, oscillations) or none of the above. Despite a 141
87 number of situations that one would under any circumstances vast literature on nonlinear oscillations, only for second order 142
88 consider as manifestations of a steady-state behavior. For systems is there much known about the existence and stability 143
89 instance, it excludes the simple case in which the variables EDof forced oscillation and, in particular, which of these kinds of 144
90 describing the behavior of the system can be expressed as periodic responses might be asymptotically stable. Essentially 145
91 linear combinations of sinusoidal functions (of time) with at most of methods for determining the existence and stability of 146
92 least two frequencies whose ratio is not a rational number. In forced periodic trajectories repose on H. Poincares classical 147
93 this case, the variables in question are not periodically idea of seeking the existence of fixed points for the map that 148
94 varying, but by all means it would be natural to say that the associates with any (initial) condition x0, the point xT 149
CT
95 system is still in steady state. reached after T units of time. In fact, fixed points of this map are 150
96 Motivated by this classical idea of a steady state (extended points from which a periodic trajectory of period T is generated. 151
97 to cover the case of irrationally related sinusoidal functions On the other hand, while an isolated periodic motion is truly 152
98 of time) and by the fact that, in a stable linear system, any nonlinear phenomenon, oscillations are not the only nontrivial 153
99 transient state asymptotically approaches a steady state, it is asymptotic behaviors for nonlinear systems, as the actual time 154
E
100 a common practice to regard a steady state as a kind of limit tends to 1 or, respectively, as the initial time tends to 1. 155
101 behavior. From this viewpoint, the steady state can be looked This motivates the need for a fresh approach to the whole 156
RR
102 at as either the limit behavior which is approached when the question of defining the steady state of a nonlinear system, 157
103 actual time t tends to 1 or, respectively, the limit behavior capable of covering the largest possible number of applications. 158
104 which is approached when the initial time t0 tends to 1. Since the idea of considering a system in steady state when the 159
105 The two alternatives are equivalent for a stable linear system. variables describing its behavior are periodically (or almost 160
106 From this viewpoint we note, for instance, that in the periodically) varying is too restrictive, it is reasonable to try to 161
107 classical book of James, Nichols, and Phillips (1947) it is look at the other classical characterizations outlined above, 162
CO
108 observed that the transient response of [a linear] filter is the associated to the limit behavior of the system. Fundamental 163
109 difference between the actual output of the filter for t > t0 and contributions, in the theory of dynamical systems, toward the 164
110 the asymptotic form that it approaches and that only when characterization of such behavior are the works of H. Poincare 165
111 a filter is stable it is possible to speak with full generality of and G.D. Birkhoff. In particular Birkhoff, in his classical 1927 166
112 its response to an input that starts indefinitely far in the past. essay (which was considered by J. Moser a continuation of 167
UN
113 In slightly more general terms, the book of Zadeh and Desoer Poincares profound and extensive work on celestial 168
114 (1963) defines a ground state [of the system], if it exists, mechanics), provided the appropriate definitions that made 169
115 [as] the limiting terminal state of [the system] when the zero him possible to claim that with an arbitrary dynamical system 170
116 input is applied, . . ., provided the limiting state g is the same . . . there is associated always a closed set of central motions 171
117 for all initial states and afterward define the steady-state which do possess this property of regional recurrence, toward 172
118 response [of the system] to an input ut0 ;t [as] the limit, if it which all other motions of the system in general tend 173
119 exists, of the ground-state response of [the system] to u as asymptotically (Birkhoff, 1927, p. 190). 174
120 t0 ! 1. Furthermore it is observed that usually [the The first of such definitions is the concept of v-limit (or a- 175
121 system] and u are such that, in [the expression of the limit) set of a given point, which is consists in what follows. 176
122 response], g can be replaced by an arbitrary initial state a Consider an autonomous ordinary differential equation
123 without affecting the limiting value of the response as
124 t0 ! 1. x f x (1) 178
177
+ Models
JARAP 289 116
A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx 3

216
always defined for all negative (and positive) times, and is 217
moreover bounded, since the set vx0 is compact. Put in 218
different terms, the set vx0 is filled by motions of (1) which 219
are bounded backward and forward in time. The other 220
remarkable feature is that xt; x0 approaches vx0 as 221
t ! 1, in the sense that the distance of the point xt; x0 from 222
the set vx0 tends to 0 as t ! 1 (recall that the distance of a 223
point x of Rn from a set S of Rn , denoted distx; S, is the 224
F
nonnegative number inf y 2 S kx yk). Corresponding properties 225
hold, of course, for the a-limit set of a motion. 226
OO
We recognize in these properties some of the keywords 227
which have already occurred in the summary of the classical 228
notion of steady state: the existence of motions which are 229
defined backward and forward in time (as periodic motions are) 230
Fig. 1. The v-limit set of a point.
and the convergence of any actual motion to a set filled with 231
178
177 such special motions, as observed by Birkhoff. However, as 232
PR
179 with x 2 Rn , t 2 R. It is well known that, if f : Rn ! Rn is promising as this seems, there are some very limiting features 233
180 locally Lipschitz, for all x0 2 Rn the solution of (1) with initial about a definition of steady-state behavior based only on the 234
181 condition x0 x0 , denoted by xt; x0 , exists on some open concept of v-limit set of a point. Before proceeding further, we 235
182 interval of the point t 0 and is unique. Assume, in particular, illustrate this concept with a couple of simple examples. 236
183 that xt; x0 is defined for all t 0. A point x is said to be an v- 237
184 limit point of the motion xt; x0 if there exists a sequence of ED Example. Let the stable, one-dimensional, linear system
times ftk g, with lim k ! 1 tk 1, such that
lim xtk ; x0 x: y y u (2) 239
238
186
185 k!1
240
187 The v-limit set of a point x0 , denoted vx0 , is the union of all be forced by the input ut U sin vt f. According to the 241
188 v-limit points of the motion xt; x0 (see Fig. 1). classical definition given in the previous section, this system is 242
CT
189 Likewise, assume that xt; x0 is defined for all t 0. A point in steady state if yt is a periodic function of period T 2p=v 243
190 x is said to be an a-limit point of the motion xt; x0 if there and it is well known that this occur if y0 is chosen appro- 244
191 exists a sequence of times ftk g, with lim k ! 1 tk 1, such priately. To determine the value of y0 which makes this 245
that happen, i.e. to evaluate the steady-state response of the system 246
to the given ut, regard ut as one of the state variables of a
E
247
lim xtk ; x0 x: harmonic oscillator oscillating at angular frequency v, that is,
193
192 k!1
set
194 The a-limit set of a point x0 , denoted ax0 , is the union of all a-
RR
195 limit points of the motion xt; x0 . x1 t u U sin vt f;

196 It is obvious from this definition that an v-limit point i s not
197
1 249
248
necessarily a limit of xt; x0 as t ! 1, because the solution in x2 t ut U cos vt f
198 question may not admit any limit as t ! 1. It happens though, v
199 that if the motion xt; x0 is bounded, then xt; x0 asympto-
CO
in which case
200 tically approaches the set vx0 . This property is precisely
201

characterized in the following statement (Birkhoff, 1927, p. x1 0 v x1 250
251
252
202 198). : (3)
x2 v 0 x2 253
203
204
Set also y x3 , to represent the system, along with its
Lemma 1. Suppose there is a number M such that kxt; x0 k
UN
205
forcing input, as a three-dimensional autonomous linear system
M for all t 0. Then, vx0 is a nonempty compact connected
206 set, invariant under (1). Moreover, the distance of xt; x0 from 0 1 0 10 1 0 1
x1 0 v 0 x1 U sin f
207 vx0 tends to 0 as t ! 1. @ x2 A @ v 0 0 A@ x2 A; x0 @ U cos f A:
208 255
254
256
209 One of the remarkable features of vx0 , as indicated in x3 1 0 1 x3 y0
257
210 Lemma 1, is the fact that this set is invariant for (1). Invariance (4)
211 means that for all initial condition x0 2 vx0 the solution
212 xt; x0 of (1) exists for all t 2 1; 1 and that Integration of this system is a standard exercise. Change x3
213 xt; x0 2 vx0 for all such t. In particular, it is somewhat into 259
258
214 surprising to observe that, even in case the solution xt; x0 is
215 not defined for all negative times, the solution xt; x0 , i.e. the x
z x3 P 1
216 solution passing through a point x0 of the v-limit set of x0 , is x2
+ Models
JARAP 289 116

259
258
with P solution of the Sylvester equation

0 v
P P 1 0 :
261
260 v 0
This yields
263
262 z z
from which it is concluded that
F

U sin f
yt x3 t et y0 P
OO
U cos f

U sin vt f
P :
264
265
266 U cos vt f
267 It is seen from this that the value of y0 for which xt; x0 is Fig. 2. The phase portrait of Van der Pols oscillator (5).
268 periodic, i.e. for which the system is in steady state according to 298
PR
the classical definition, is From the phase portrait of this system (depicted in Fig. 2 for 299
e 1) it is seen that all motions except the trivial motion 300
U sin f
y0 P : occurring for x0 0 approach, as t ! 1, a limit cycle L. The 301
270
269 U cos f
system is in steady state, according to the classical definition, if 302
If this is the case, then and only if either x0 0 or x0 2 L. Note also that the limit 303
EDcycle L is the v-limit set of any point x0 6 0 while the point 304
U sin vt f x 0 is the v-limit set of the point x0 0. 305
yt P ; 306
272
271 U cos vt f
Since any motion xt; x0 which is bounded in positive time 307
273
and this is precisely what is commonly considered a steady- asymptotically approaches the v-limit set vx0 as t ! 1, one 308
274
275 state response of system (2) to the input ut U sin vt f. may be tempted to look, for a system (1) in which all motions 309
276 are bounded in positive time, at the union of the limit sets of all 310
CT
Viewing system (2) driven by the harmonic input ut as a
277
single autonomous system, such as (4), has a number of points x0 , that is, at the set
278
advantages. For instance, it simplifies the calculation of the [
279 V vx0
steady-state response to a given family of inputs: those obtained x 0 2 Rn 312
311
280
for different values of U and f. As a matter of fact, what we
E
281 and to say that the system is in steady state if x0 2 V. 313

need to consider in this case is no longer a specific response to a
282 In this respect, note that in system (4) the set V is the entire 314
forcing input, but rather the behavior of an autonomous system
283 plane P, while in system (5) the set V consists of the union of 315
as certain initial conditions (those of x1 and x2 in (4)) are
RR
284 the equilibrium point {0} and of the limit cycle L. Since all 316
allowed to vary. In this respect, the previous conclusion can be
285 motions of systems (4) and (5) are bounded in positive time and 317
rephrased by saying that system (4) is in steady state if and only
if the initial condition x0 is a point of the plane asymptotically approach the (respective) sets V as t increases, it 318
may seem that the set in question is the right object to look at for 319
x a definition of steady state. There is however a remarkable 320
P x 2 R : x3 P 1
3
:
CO
287
286 x2 321
difference between the two cases: the type of convergence.
288 In both cases, as Lemma 1 says, the distance of xt; x0 from 322
Revisiting this elementary analysis from the viewpoint of the
289 V tends to 0 as t ! 1, but while in the first example the 323
concept of v-limit set, it is readily observed that, for any
convergence is uniform in x0 , for all x0 within a set of finite 324
x0 2 R3 , the set vx0 is the ellipse defined as follows:
distance from V, in the second example it is not. To examine 325
8 9
UN
> > this difference in more detail, recall that to say that the distance 326
<x1 ; x2 ; x3 2 R3 : x21 x22 x21;0 >
> =
of xt; x0 from a set S tends to 0 as t ! 1 is to say that for every 327
vx0 :
> x2 ; x P x1
> >
> e there exists T such that
290
291 : 2;0 3 ;
292 x2
distxt; x0 ; S e for all t T: (6) 329
328
293
330
294 The number T in this expression obviously depends on e, but it
Example. As a second, nonlinear, example, consider now the 331
also generally depends on x0 ; as a matter of fact, the larger the
classical Van der Pol oscillator, written in state-space form as 332
distance of x0 from S is, the more one can expect to wait until
333
296
295 xt; x0 comes within an e-distance from S. However, one might
x y; y x e1 x2 y (5) 334
hope that if the initial distance of x0 from S is bounded by a
297 2 335
in which, as it is well known, the damping term e1 x y fixed number, then the time required to get within an e-distance
298 336
models the effect of a nonlinear resistor (see Khalil, 2002). from S would only depend on e and not on x0 . Formalizing this
+ Models
JARAP 289 116

336 387
337 concept, let Bd S denote the set of all points whose distance 2.2. The limit set of a set
from S does not exceed a given number d, that is 388
As it turns out, the features in question are in fact shared by 389
339
338 Bd S fx 2 Rn : distx; S dg: an object which extends the concept of limit set from the case of 390
340 a s ingle initial point to the case of a set of initial points. This 391
341 The distance of xt; x0 from S is said to tend to 0, as t ! 1, object, defined in Bhatia and Szego (1970) as p rolongational 392
342 uniformly in x0 on Bd S, if for every e there exists T, which limit set and defined in Hale, Magalhaes, and Oliva (2002) and 393
343 depends on e and d but not on x0 , such that (6) holds for all Sell and You (2002) as omega limit set of a given set B, 394
F
344 x0 2 Bd S. addresses in a convenient way the issue of uniform convergence 395
345 Well, it is readily seen that in the case of example (4) the (when initial conditions are picked in a bounded set) and also 396
convergence is uniform, while in the second example it is not.
OO
346 lends itself to the possibility of looking at a steady-states 397
347 To this end observe that in the first example, in which V behaviors as limits taken when the initial time tends to 1. 398
coincides with the plane P, it is always possible to find two The notion in question is defined as follows. 399
numbers c1 and c2 such that, for all x, Consider again system (1), let B be a subset of Rn and 400
349
348 suppose xt; x0 is defined for all t 0 and all x0 2 B. The v- 401
c1 distx; V z c2 distx; V: limit set of B, denoted vB, is the set of all points x for which 402
PR
there exists a sequence of pairs fxk ; tk g, with xk 2 B and 403
Thus, using the fact that z z, one obtains the estimate
lim k ! 1 tk 1 such that
351
350
c2
352 distxt; x0 ; V distx0 ; V et ; lim xtk ; xk x:
c1 k!1 404
405
406
353
It is clear from the definition that if B consists of only one 407
354 from which it immediately follows that xt; x0 converges to the ED single point x0 , all xk s in the definition above are necessarily 408
355 set V uniformly in x0 on Bd V, for any given d > 0.
equal to x0 and the definition in question reduces to the 409
356 Consider now the second example, in which the set V
definition of v-limit set of a point, given earlier. It is also clear 410
357 consists of the union of the equilibrium point f0g and of the
from this definition that, if for some x0 2 B the set vx0 is 411
358 limit cycle L. Observe that all x0 s inside the limit cycle L are
nonempty, all points of vx0 are points of vB. In fact, all 412
359 within a finite distance d from L. All x0 2 Bd V are such that
such points have the property indicated in the definition, if all 413
CT
360 distxt; x0 ; V ! 0 as t ! 1 (as a matter of fact, if x0 6 0, the
the xk s are taken equal to x0 . Thus, in particular, if all motions 414
361 motion xt; x0 asymptotically approaches L, while, if x0 0,
362 the motion xt; x0 trivially remains at 0). If the convergence with x0 2 B are bounded in positive time,
363 [
were uniform in x0 on Bd V, it would be possible, for any vx0 vB:
364 choice of e, to find a number T, only depending on e and not x0 , x0 2 B 416
415
E
365 such that (6) holds. This, however, is not the case. In fact,
417
366 observe that, if x0 6 0 is inside L, the motion xt; x0 is bounded However, the converse inclusion is not true in general.
418
367 in negative time and remains inside L for all t 0 (as a matter This can be immediately checked on the second of our
RR
419
368 of fact, it converges to 0 as t ! 1). Pick any x1 6 0 inside L earlier examples. In fact, consider again the van der Pol
420
369 such that distx1 ; L > e and let T 1 be the minimal time needed oscillator (5), and let the set B be, for instance, a closed disc of
421
370 to have distxt; x1 ; L e for all t T 1 . Now, go backwards sufficiently large radius, to include the limit cycle L in its
422
371 T 0 > 0 units of time, to the point x0 xT 0 ; x1 . Then, the interior. We already know that f0g and L, being v-limit sets of
423
372 minimal time T needed to have distxt; x0 ; V e for all t T points of B, are in vB. But it is also easy to see that any other
CO
424
373 is T T 0 T 1 and, since T 0 can be taken arbitrarily large point inside L is a point of vB. In fact, let x be any of such
425
374 (while keeping x0 inside the limit cycle, within distance d from points and pick any sequence ftk g such that lim k ! 1 tk 1. It
426
375 V), we see that the time T needed to have property (6) fulfilled is seen from the phase portrait that xt; x remains inside L (and
376 cannot be made independent of x0 . hence in B) for all negative values of t; hence
377 Now, recall that one of the main motivations for looking into xk : xtk ; x 428
427
UN
378 the concept of steady state is the aim to shape the steady-state
429
379 response of a system to a given (or to a given family of) forcing is a point in B for all k. Since the sequence fxk ; tk g is such that
430
380 input(s). But this motivation looses much of its meaning if the xtk ; xk x, the property required for x to be in vB is
431
381 time needed to get within an e-distance from the steady state trivially satisfied. This shows that vB includes not just f0g
432
382 may grow unbounded as the initial state changes (even when the and L, but also all points of the open region bounded by L. As a
433
383 latter is picked within a fixed bounded set). In other words, matter of fact, it is not difficult to prove that no other point can
434
384 uniform convergence to the steady state (which is automatically be a point of vB. This can be done either by direct arguments,
435
385 guaranteed in the case of linear systems) is an indispensable or more simply by appealing to the result indicated in the
436
386 feature to be required in a nonlinear version of this notion. next Lemma, which says that, in this example, all motions with
437
387 However, as shown above, the set consisting of the union of the initial conditions in vB have to be bounded backward in time.
438
v-limit sets of all points in the state space does not have this Now, it is observed from the phase portrait that for any point x
439
property of uniform convergence. which is not on or inside L, the motion xt; x is such that
+ Models
JARAP 289 116

439 494
440 kxt; xk ! 1 as t ! 1. Thus, any of such x cannot be a Vinograd (1957) (and thoroughly analyzed in Hahn (1967, 495
441 point of vB. pp. 191194)). 496
442 The relevant properties of the v-limit set of a set, which 497
443 extend those presented earlier in Lemma 1, can be
Example. Consider the nonlinear system
444 summarized as follows (see, for instance, Bhatia &

445 Szego, 1970; Hale, Magalhaes, & Oliva, 2002; Sell & x f x; y
446 You, 2002).
y gx; y 499
498
447
F
in which f 0; 0 g0; 0 0 and
448 Lemma 2. Let B be a nonempty bounded subset of Rn and 2
449 suppose there is a number M such that kxt; x0 k M for all f x; y 1 x y x y5
OO
2
450 t 0 and all x0 2 B. Then vB is a nonempty compact set, gx; y x2 y2 1 x2 y2 2 y y 2x
451 invariant under (1). Moreover, the distance of xt; x0 from (7) 501
500
452 vB tends to 0 as t ! 1, uniformly in x0 2 B. If B is connected,
502
453 so is vB. for x; y 6 0; 0. The phase portrait of this system is the one
454 503
depicted in Fig. 3. In particular, this system has only one
455 Thus, as it is the case for the v-limit set of a point, we equilibrium at x; y 0; 0 and any initial condition 504
PR
456 see that the v-limit set of a bounded set, being compact x0 ; y0 in the plane produces a motion that asymptotically 505
457 and invariant, is filled with motions which exist for all tends to this point. However, this equilibrium point is not stable 506
458 t 2 1; 1 and are bounded backward and forward in the sense of Lyapunov, because it is not possible to find, for 507
459 in time (the set of all such trajectories is a behavior, in the every e > 0, a number d > 0 such that every initial condition in a 508
460 sense of Willems, 1991). But, above all, we see that the set disc of radius d produces a motion which remains for all t 0 in 509
461 in question is u niformly approached by motions with EDa disc of radius e. As a matter of fact, as is seen from the phase 510
462 initial state x0 2 B, a property that the v-limit set of a portrait, there are points arbitrarily close to the origin from 511
463 point does not have. The property in question which the motion always travels a finite fixed distance away 512
464 makes this notion more suitable, as explained above, for from the origin. Thus, no matter how small d is chosen, it is 513
465 a definition of steady-state behavior of a nonlinear impossible to keep the motion within an e-distance from the 514
466 system. origin, if e is not large enough. 515
CT
467 We conclude the section with another property, that will be Note that, in this example, the point x; y 0; 0 is an v- 516
468 useful in the sequel. limit point of every point in the plane and thus the set V 517
469 518
introduced before, that is, the union of all v-limit sets of all
519
470 Lemma 3. If B is a compact set invariant for (1), then points in the plane, is simply the point 0; 0. However, if B is a
520
vB B. disc of sufficiently large radius, centered at the origin, the v-
E
471
472 521
limit set of B is the nontrivial set consisting of the larger figure
473 522
eight and of all points in its interior. As expected, the set vB
RR
523
474 Proof. As shown in the previous example, for any x 2 B it is in question is filled with motions of system (7), all bounded in
524
475 trivially always possible to find a sequence fxk ; tk g, with backward and forward time. Note, in particular, that the orbit of
525
476 xk 2 B and lim k ! 1 tk 1 such that xtk ; xk x, and this any of such motions is a motion in which the v-limit set and the
526
477 shows that B vB. To show that the reverse inclusion a-limit set coincide (a homoclinic orbit). 527
478 vB B also holds, pick any point x 2 vB, and observe It should be stressed that while, in general, convergence to 528
CO
479 that by definition for some sequence fxk g of points of B an equilibrium may not imply stability of the latter, this is no 529
480 there is a sequence of times ftk g such that
481 lim k ! 1 xtk ; xk x. Since B is by assumption invariant
482 for (1), xtk ; xk 2 B for all k. Thus x is the limit of a sequence
483
484 of points of B and, since B is also compact, necessarily
485
x 2 B. &
UN
486 3. Limit sets and stability

487
488 It is well known, in the classical theory of the stability of
489 motion, that (in a nonlinear system) an equilibrium point
490 which attracts all motions starting from initial conditions in
491 some (small) open neighborhood of this point is not
492 necessarily stable in the sense of Lyapunov. Examples of
493 systems in which convergence to an equilibrium does not
494 imply stability can even be built in dimension two, as seen
for instance, from the following classical example due to Fig. 3. The phase portrait of Vinograds system (7).
+ Models
JARAP 289 116

529 583
530 longer the case if the convergence to the equilibrium is uniform subset X 0 of X, there is a time T > 0 such that xt; x0 2 B for all 584
531 (in the sense made precise in the previous section). As a matter t T and all x0 2 X 0 . In other words, if the motions of the 585
532 of fact, uniform convergence to an equilibrium does imply system are ultimately bounded, every motion eventually enters 586
533 stability in the sense of Lyapunov. This is a consequence of the and remains in the bounded set B. 587
534 fact that xt; x0 depends continuously on x0 (see for example Note that, since by hypothesis X is positively invariant, there 588
535 Hahn (1967, p. 181)). Of course, in the previous example the is no loss of generality in assuming B X in the definition 589
536 convergence to the equilibrium is not uniform: no matter how above. Hence, there exists a number M such that kxt; x0 k M 590
537 large T is taken it is always possible to pick initial states for all t 0 and all x0 2 B and, from Lemma 2, it is concluded 591
F
538 arbitrarily close to the origin (for instance on any of the that the set vB is nonempty (and has all the properties 592
539 homoclinic orbits) from which the motion needs a time larger indicated in that Lemma). It is worth stressing that, for a system 593
OO
540 than T to enter a disc of small radius e. whose motions are ultimately bounded, the set vB is a unique 594
541 In this respect, the v-limit set of a (bounded) set B of initial well-defined set, regardless of how B is taken. 595
542 conditions, which carries with it the property of uniform 596
543 convergence, appears to be a more satisfactory object to look at
Lemma 5. Let the motions of (1) be bounded and let B0 be any 597
544 in the quest for a set (other than a simple equilibrium) having
other bounded subset of X with the property that, for every 598
545 both the properties of attractivity and stability in the sense of
compact subset X 0 of X, there is a time T > 0 such that
PR
599
546 Lyapunov. For motions converging to a closed invariant set A,
xt; x0 2 B0 for all t T and all x0 2 X 0 . Then, vB vB0 . 600
547 the notion of asymptotic stability, a straightforward extension 601
548 of the notion of asymptotic stability of an equilibrium, is 602
549 defined as follows. Let A Rn be a closed set invariant for (1). 0
Proof. Pick x be a point of B . By hypothesis, there exists a 603
The set A is asymptotically stable if the following hold:
sequence fxk ; tk g, with xk 2 B0 and lim k ! 1 tk 1 such that 604
552
551
550
ED xtk ; xk converges to x as k ! 1. As all such xk s are in a 605
552 (i) for every e > 0, there exists d > 0 such that,
compact subset of X, by definition of B there exist a time T > 0 606
distx0 ; A d implies distxt; x0 ; A e such that all points xk xT; xk are points of B. Set tk tk T 607
553
554 and consider the sequence fxk ; tk g. Trivially xtk ; xk , being 608
555 for all t 0:
555
554
553 equal to xtk ; xk , converges to x as k ! 1. Thus, x is a point of 609
(ii) there exists a number d > 0 such that B also. We have shown in this way that vB0 vB. Rever- 610
CT
sing the role of the two sets shows that vB vB0 , that is, 611
556
557 distx0 ; A d implies lim distxt; x0 ; A 0: that the two sets in question are identical. & 612
557
556 t!1
559
558 613
560 For systems whose motions are ultimately bounded, the 614
561 As in the case of equilibria, for a closed invariant set A notion of steady state can be defined as follows. 615
E
562 which is asymptotically stable for (1), the domain of attraction 616
563 is the set of all x0 for which xt; x0 is defined for all t 0 and
Definition. Suppose the motions of system (1), with initial 617
distxt; x0 ; A ! 0 as t ! 1.
RR
564
conditions in a closed and positively invariant set X, are 618
565 It is not difficult to show (see for example Celani, 2003;
ultimately bounded. A steady-state motion is any motion with 619
566 Sontag & Wang, 1995) that if the set A in Rn is also bounded
initial condition in x0 2 vB. The set vB is the steady-state 620
567 and hence compact, and the convergence in (ii) is uniform in x0 ,
locus of (1) and the restriction of (1) to vB is the steady-state 621
568 then property (ii) implies property (i). Using this fact and the
behavior of (1). 622
569 result of Lemma 2, this yields the following important property. 623
CO
The notion thus introduced recaptures the classical notion of 624

570
n steady state for linear systems and provides a new powerful tool 625
571 Lemma 4. Let B be a nonempty bounded subset of R and
to deal with similar issues in the case of nonlinear systems. To 626
572 suppose there is a number M such that kxt; x0 k M for all
see how this notion includes the classical viewpoint, consider 627
573 t 0 and all x0 2 B. Then vB is a nonempty compact set,
an n-dimensional, single-input, linear system
574 invariant under (1). Suppose also that vB is contained in the
UN
575 interior of B. Then, vB is asymptotically stable, with a z Fz Gu (8) 629

628
576
domain of attraction that contains B.
forced by the harmonic input ut U sin vt f. As shown 630
in the first example, a simple method to determine the periodic 631
577 4. The steady-state behavior of a nonlinear system motion of (8) consists in viewing the forcing input ut as 632
578 provided by an autonomous signal generator of the form (3) and 633
579 Consider now again system (1), with initial conditions in a in analyzing the state state behavior of the associated augmen- 634
580 closed subset X Rn . Suppose the set X is p ositively invariant, ted system
581 which means that for all initial conditions x0 2 X, the solution 0 1
0 v
582 xt; x0 exists for all t 0 and xt; x0 2 X for all t 0. The w 0 w
@ v 0 A (9)
583 motions of this system are said to be ultimately bounded if there z z
is a bounded subset B with the property that, for every compact G 1 0 F 636
635
+ Models
JARAP 289 116

636
635 684
637 As a matter of fact, let P be the unique solution of the Sylvester set, invariant for the upper subsystem of (11). As in the previous 685
equation example, the v-limit set of W under the motion of the upper 686
subsystem of (11) is the subset W itself. Moreover, since the 687
0 v lower subsystem of (11) is a linear asymptotically stable system
P FP G 1 0 688
639
638 v 0 driven by the bounded input ut qwt, it is easy to check 689
and observe that the graph of the linear map that the motions of system (11), with initial conditions taken in 690
W Rn , are ultimately bounded. As a matter of fact, so long as 691
p : R2 ! Rn w0 2 W, the input qwt to the lower subsystem of (11) is 692
F
641
640 w 7! Pw bounded by some fixed number U and standard arguments can 693
642 is an invariant subspace for the system (9). Since all trajectories be invoked to show that
OO
643 of (9) approach this subspace as t ! 1, the steady-state kztk K elt kz0k LU 695
694
644 behavior of (9) is determined by the restriction of its motion
645 to this invariant subspace. 696
for all t 0, in which K; l and L are appropriate positive
646 Revisiting this analysis from the viewpoint of the more numbers. Thus, any bounded set B of the form
647 general notion of steady state introduced above, let W R2 be a 698
697
set of the form B fw; z 2 Rr Rn : w 2 W; kzk 2LUg
PR
699
649
648 W fw 2 R2 : kwk cg (10) has the property indicated in the definition of ultimate bound- 700
edness. 701
650 in which c is a fixed number, and suppose the set of initial Moreover, it is possible to show that, regardless of how B is
651 conditions for (9) is W Rn . This is in fact the case when the taken, vB is the graph of the map
652 problem of evaluating the periodic response of (8) to harmonic
653 inputs whose amplitude does not exceed a fixed number c is ED p : W ! Rn 703
702
654 addressed. The set W is compact and invariant for the upper w 7! pw;
655 subsystem of (9). Therefore, as shown before, the v-limit set of
656 W under the motion of the upper subsystem of (9) is the subset defined by
657 W itself. Z 0
704
705
706
658 The set W Rn is closed and positively invariant for the full pw eFt Gqwt; w dt: (12)
CT
659 system (9) and, moreover, since the lower subsystem of (9) is a 1 707
660 linear asymptotically stable system driven by a bounded input, 708
The explanation of this fact reposes on the following 709
661 it is immediate to check that the motions of system (9), with arguments. First of all, observe that since qwt; w is by 710
662 initial conditions taken in W Rn , are ultimately bounded. As hypothesis a bounded function of t and all eigenvalues of F 711
E
a matter of fact, any bounded set B of the form have negative real part the improper integral on the right-hand 712
664
663 B fw; z 2 R2 Rn : w 2 W; kz Pwk dg side of (12) exists. Then, a simple calculation shows that the 713
graph of the map p is invariant for (11). To see why this is the
RR
665 714
in which d is any positive number, has the property indicated in case, pick any initial condition w0 ; z0 on the graph of p, i.e.
666 715
the definition of ultimate boundedness. Note also that any such with z0 pw0 and compute the solution zt of the lower
B satisfies B W Rn . It is easy to check that equation of (11) by means of the classical variation of constants
formula, to obtain
668
667 vB fw; z 2 R2 Rn : w 2 W; z Pwg;
Z 0
CO
669
that is, vB is the graph of the restriction of the map p to the set zt eFt eFt Gqwt; w0 dt
670
W. Note that vB is independent of the choice of B (so long as 1
671
B is a set having the properties indicated in the definition of Z t 717
716
672 Ftt
ultimate boundedness). The restriction of (9) to the invariant set e Gqwt; w0 dt 718
673 0
vB characterizes the steady-state behavior of (8) under the 719
UN
674
family of all harmonic inputs of fixed angular frequency v, and From this, an easy manipulation yields zt pwt; w0 , 720
675
amplitude not exceeding c. proving the invariance of the graph of p for (11). Then, it is 721
676
A similar result, that is, the fact that the steady-state locus is immediately concluded that any point of the graph of p is 722
677
the graph of a map, can be reached if the signal generator is any necessarily a point of vB. To complete the proof of the claim 723
678
nonlinear system, with initial conditions chosen in a compact it remains to show that no other point of W Rn can be a point 724
679
invariant set W. More precisely, consider an augmented system of vB. But this is a direct consequence of the fact that F has 725
680
of the form (we retain, throughout, the assumption that both eigenvalues with negative real part. In fact, this assumption 726
sw and qw are locally Lipschitz functions) implies that all motions of (11) whose initial condition is not on 727
682
681 the graph of p are unbounded in backward time and therefore 728
w sw; z Fz Gqw; (11)
cannot be contained in vB, which we know is a bounded set. 729
683
in which w 2 W Rr , z 2 Rn , and assume that: (i) the eigen- There are various ways in which this result can be
684
values of F have negative real part, (ii) the set W is a compact generalized. For instance, it can be extended to describe the
+ Models
JARAP 289 116

729 777
steady-state response of a nonlinear system Ramsey (2003)), it is possible to show that, for each 778
w0 2 W, there is one and only one value z0 2 R from which 779
731
730 z f z; u (13)
the motion of
732 in the neighborhood of a locally exponentially stable equili-
733 brium point. To this end, suppose that f 0; 0 0 and that the w1 w2 ; w2 w1 ; z z3 w1 781
780
matrix 782
is a periodic motion. The set of all such pairs identifies a map 783
@f p : W ! R, whose graph coincides with the steady-state locus
F 0; 0 784
F
735
734 @z vB of the system. The map in question, depicted in Fig. 4, is 785
736 has all eigenvalues with negative real part. Then, it is well continuously differentiable at any nonzero w but only contin- 786
uous at w 0, as shown in Byrnes et al. (2003). 787
OO
737 known (see for example Hahn (1967, p. 275)) that it is always
788
738 possible to find a compact subset Z Rn , which contains z 0 Note that the motions of the autonomous system generator 789
739 in its interior and a number s > 0 such that, if z0 2 Z and (14) that drives system (13) are not supposed to be periodic 790
740 kutk s for all t 0, the solution of (13) with initial con- motions. For instance, the system in question could be a stable 791
741 dition z0 z0 satisfies zt 2 Z for all t 0. Suppose that the Van der Pol oscillator, with W defined as the set of all points 792
742 input u to (13) is produced, as before, by a signal generator of inside and on the boundary of the limit cycle. In this case, our
PR
793
the form approach makes it possible to define the steady-state response 794
w sw; u qw (14) of (13) not just to the (single) periodic motion generated by (14) 795
744
743
when the initial condition is taken on the boundary of W, but 796
745 with initial conditions chosen in a compact invariant set W and, also to all (non-periodic) motions generated by (14) when the 797
746 moreover, suppose that, kqwk s for all w 2 W. If this is the initial condition is taken in the interior of W. We consider this as 798
case, the set Z W is positively invariant for ED an advantage of the proposed approach. 799
w sw; z f z; qw; (15) A common feature of the examples discussed above is the 800
748
747
fact that the set vB can be expressed as the graph of a map 801
749 and the motions of the latter are ultimately bounded, with z pw. This means that, so long as this is the case, a system 802
750 B Z W. The set vB may have a complicated structure but of the form (13) has a unique well defined steady-state response 803
751 it is possible to show, using the Center Manifold theorem, that if to the input ut qwt. As a matter of fact, the response in
CT
804
s0 0 and the matrix question is precisely zt pwt. Of course, in general, this 805
may not be the case and the global structure of the steady-state
@s 806
S 0 locus can be very complicated. In particular, the set vB may 807
753
752 @w
fail to be the graph of a map z pw and multiple steady-state 808
754 has all eigenvalues on the imaginary axis and if Z and B are responses to a given input may occur. This is the counterpart
E
809
755 small enough, the set in question can still be expressed as the in the context of forced motions of the fact that, in general, a 810
756 graph of a map z pw. Specifically, the graph in question is nonlinear system may possess multiple equilibria. In these 811
RR
757 precisely the center manifold of (15) at 0; 0. cases, the steady-state response is determined not only by the 812
758 Of course, the possibility of expressing the steady-state forcing input, but also by the initial state of the system to which 813
759 locus of a system of the form (15) as the graph of a map the input is applied. 814
760 z pw is not necessarily tied to the assumption that the Even though, in general, uniqueness of the steady-state 815
761 equilibrium point w; z 0; 0 of (13) be locally exponen- response of system (13) to inputs generated by a system of the 816
CO
762 tially stable. This is shown for instance in the following simple form (14) cannot be guaranteed, it is useful to stress that, if the
763 example. It should be stressed, though, that if the equilibrium
764 w; z 0; 0 of (13) is not locally exponentially stable, the
765 map p may fail to be differentiable at the point w 0.
766
UN
Example. Consider the system
768
767 z z3 u (16)
769 forced by an input u w1 provided by the autonomous signal
770 generator (3), in which we assume, for simplicity v 1 and W
771 as in (10). The set W R is positively invariant and, by means
772 of simple arguments, it is easy to see that the motions with
773 initial conditions in W R are ultimately bounded. Now, by
774 means of the classical method of Poincare for the study of
775 periodic solutions of a nonlinear differential equation (whose
776 details cannot be included here for reasons of space, but which
777 can be found for instance in Byrnes, Gilliam, Isidori, and Fig. 4. The steady-state locus of Example (16).
+ Models
JARAP 289 116

816 872
817 set W is compact and invariant (as assumed above), for each particular Francis and Wonham (1976) and Francis (1977) 873
818 w 2 W there is always at least one z 2 Z such that the pair w; z provide a very elegant analysis, that was later taken as a 874
820
819 produces a steady-state response. paradigm for the study of the nonlinear version of the problem. 875
One of the contributions of Francis and Wonham (1976) was a 876
821 Lemma 6. Consider a system of the form (15) with
precise characterization of what the author called (and since 877
822 w; z 2 W Z. Suppose its motions are ultimately bounded.
then has become known as) the internal model principle. In a 878
823 If W is a compact set invariant for w sw, the steady-state
suitable framework, they proved that the property of perfect 879
824 locus of (15) is the graph of a (possibly set-valued) map defined
tracking is insensitive to plant parameter variations only if the 880
825 on W.
F
827
826 controller utilizes feedback of the regulated variable, and 881
incorporates in the feedback path a suitably reduplicated model 882
828 Proof. Recall that, as shown above, the limit set of W under the
OO
of the dynamic structure of the exogenous signals which the 883
829 flow of w sw coincides with W itself, that is vW W. As
regulator is required to process. The converse of this property, 884
830 a consequence, for all w 2 W there is a sequence fwk ; tk g with
also shown in Francis (1977), is that if the controller embeds an 885
831 wk in W for all k such that w lim k ! 1 wtk ; wk : Set p
internal model of the exogenous signals, stable perfect tracking 886
832 colw; z and let ft; p0 denote the integral curve of (15)
can be achieved, regardless of plant parameter variations (so 887
833 passing through p0 at time t 0. Pick any point z0 2 Z and
long as the stability of the closed loop is preserved). 888
let pk colwk ; z0 . If the motions of (15) are ultimately
PR
834
A nonlinear enhancement of this theory, which uses a 889
835 bounded, there is a bounded set B and a time T > 0 such that
combination of geometry and nonlinear dynamical systems 890
836 ft; pk 2 B for all t T and all k > 0. Pick any integer h such
theory, was initiated by pioneering works of Isidori and Byrnes 891
837 that th T, set pk fth ; pk and tk tk th , for k h, and
(1990), Huang and Rugh (1990) and Huang and Lin (1994) who 892
838 observe that, by construction, ftk ; pk ftk ; pk . The
showed how to design a controller that provides a local solution 893
839 sequence fftk ; pk g is bounded. Hence, there exists a sub-
near an equilibrium point, in the presence of exogenous signals 894
840 sequence fftk ; pk g converging to a point p colw; x, EDwhich were produced by a Poisson stable system. In particular, 895
841 which is a point of vB because all pk s are in B. Since
Isidori and Byrnes (1990) showed how the use center manifold 896
842 system (15) is upper triangular, necessarily w w. This shows
theory determines also in the case of nonlinear systems the 897
843 that, for any point w 2 W, there is a point z 2 Z such that
844 necessity of the existence of an internal model and Khalil 898
845 w; Z 2 vB, as claimed. &
(1994) showed how issues of global convergence to the required 899
CT
steady state could be addressed. Since these early contributions, 900
5. Asymptotic tracking and disturbance rejection the theory has experienced a tremendous growth, culminating 901
846
in the recent development of design methods able to handle the 902
5.1. Background case of parametric uncertainties affecting the autonomous 903
847
(linear) system which generates the exogenous signals (such as 904
E
848
One of the main motivations for the importance of an in Delli Priscoli, Marconi, & Isidori, 2006; Serrani, Isidori, & 905
849
appropriate notion of steady state is the need to address control Marconi, 2001), the case of nonlinear exogenous systems (such 906
850
RR
problems in which the output of a system is required to as in Byrnes & Isidori, 2004), or a combination thereof (as in 907
851 908
asymptotically track prescribed trajectories and/or to asymp- Marconi, Praly, & Isidori, 2006).
852
totically reject prescribed disturbances. Problems of this kind
853 5.2. The generalized tracking problem
are commonly known as g eneralized tracking problems, as
854 909
generalized servomechanism problems, or more often
855 The generalized tracking problem is cast in the following 910
output regulation problems. In any realistic scenario, this
CO
856 terms. The controlled plant is a finite-dimensional, time- 911

control goal has to be achieved in spite of a good number of
857 invariant, nonlinear system modelled by equations of the form
phenomena which would cause a system to behave differently
858
than expected. These phenomena could be endogenous, for x f w; x; u; e hw; x; y kw; x; (17) 913
912
859
instance parameter variations, or exogenous, such as additional
860
undesired inputs affecting the behavior of the plant. in which x 2 Rn is a vector of state variables, u 2 Rm is a vector 914
of inputs used for control purposes, w 2 Rs is a vector of inputs
UN
861 915
In most cases of practical interest, the trajectories to be tracked
862 which cannot be controlled and include exogenous commands, 916
(or the disturbance to be rejected) are not available for
863
measurement. Rather, it is only known that these trajectories exogenous disturbances and model uncertainties, e 2 R p is a 917
864 vector of regulated outputs which include tracking errors and 918
are simply (undefined) members of a set of functions, for instance
865
the set of all possible solutions of an ordinary differential any other variable that needs to be steered to 0, y 2 Rq is a 919
866 vector of outputs that are available for measurement and hence 920
equation. Theses cases include the classical problem of the set-
867 used to feed the device that supplies the control action. The 921
point control, the problem of active suppression of harmonic
868 problem is to design a controller, which receives yt as input 922
disturbances of unknown amplitude, phase and even frequency,
869 and produces ut as output, to the purpose that, in the resulting 923
the synchronization of nonlinear oscillations, and similar others.
870 closed-loop system, xt remains bounded and
For linear multivariable systems, the generalized servome-
871
chanism problem has been successfully addressed by various lim et 0; (18)
872 925
924
authors (Davison, 1976; Francis, 1977; Francis & Wonham). In t!1
+ Models
JARAP 289 116

925
924 969
971
970
926 regardless of what the exogenous input wt actually is. (i) the positive orbit of W X J is bounded, i.e. there exists 971
972
927 The specific characteristic of the problem at issue is that the a bounded subset S of Rs Rn Rn such that, for any 972
973
928 exogenous input wt is assumed to be a (undefined) member of w0 ; x0 ; j0 2 W X J, the integral curve 973
974
929 a fixed family of functions of time, the family of all solutions wt; xt; jt of (21) passing through w0 ; x0 ; j0 at 974
975
930 obtained when the initial condition w0 of a fixed ordinary time t 0 remains in S for all t 0. 975
977
976
differential equation of the form (ii) lim t ! 1 et 0, uniformly in the initial condition, i.e. for 976
978
every e > 0 there exists a time t, depending only on e and not 977
979
932
931 w sw (19) on w0 ; x0 ; j0 such that the integral curve wt; xt; jt 978
980
F
933 is allowed to vary on a prescribed set W. This system is usually of (21) passing through w0 ; x0 ; j0 at time t 0 yields 979
981
934 referred to as the exosystem. This approach can be viewed as ketk e for all t t. 980
982
OO
935 intermediate between one extreme case in which wt is totally 981
983
936 unknown, and the opposite extreme case in which wt is Condition (i) replaces, and actually extends to a nonlinear 984
937 available for measurement. As observed earlier, there is abun- setting, the classical requirement of linear system theory 985
938 dance of design problems in which parameter uncertainties, that the un-driven closed-loop system be an asymptotically 986
939 reference commands and/or exogenous disturbances can be stable system. Condition (ii) expresses the property of 987
940 modelled in this way: the case of set point control, in which the asymptotic regulation (or tracking). The property that 988
PR
941 exosystem has a trivial dynamics, namely w 0, the case in convergence of the regulated variable et to zero be uniform, 989
942 which wt is any combination sinusoidal signals of fixed which is granted in the case of a linear system, needs now to be 990
943 frequency but unspecified amplitude and phase, in which case explicitly requested, since in the case of nonlinear systems (as 991
944 the exosystem is characterized by a bench of harmonic oscil- shown earlier) this may no longer be the case even if the initial 992
lators conditions are taken in a compact set. Form a practical 993
ED viewpoint, in fact, the only meaningful case is the one in which 994
w1 0 v w1 there is a guaranteed rate of decay of the regulated variable 995
;
946
945 w2 v 0 w2 to zero. 996
947
but even the case in which the frequencies of such sinusoidal 5.3. Steady-state analysis
948
signals are undetermined, in which case the exosystem is 997
CT
characterized by a bench of (nonlinear) equations of the form The notion of steady state introduced earlier is instrumental 998
0 1 0 10 1 to prove the following, elementary but fundamental result, 999
w1 0 w3 0 w1
@ w2 A @ w3 0 0 A@ w2 A: which is a nonlinear enhancement of a Lemma of Francis 1000
950
949
951 (1977) on which all the theory of output regulation for linear 1001
w3 0 0 0 w3
952 systems is based (see Byrnes and Isidori (2003) for a proof). 1002
E
The control law for (17) is to be provided by a system 1003

modelled by equations of the form Lemma 7. Suppose the positive orbit of W X J is 1004
RR
j jj; y bounded. Then

954
953 (20)
u gj; y lim et 0
955 1006
1005
t!1
n
956 with state j 2 R . The initial conditions x0 of the plant (17),
957 w0 of the exosystem (19) and j0 of the controller (20) are if and only if
CO
958 allowed to range over a fixed compact sets X Rn , W Rs and, 1007

1009
1008
959 respectively J Rn. All maps characterizing the model of the vW X J fw; x; j : hw; x 0g: (22) 1010
960 1011
controlled plant, of the exosystem and of the controller are
961 assumed to be sufficiently differentiable. This result is simply a characterization of the generalized 1012
962 The problem to be addressed, is to design a feedback tracking problem in geometric terms. The problem in question, 1013
963 controller of the form (20) so as to obtain a closed-loop system in fact, can be cast as a problem of shaping the steady-state 1014
UN
964 in which all trajectories are bounded and the regulated output locus of the closed-loop system, in such a way that the latter 1015
et asymptotically decays to 0 as t ! 1. More precisely, it is becomes a subset of the set of all points at which the regulated 1016
required that variable is 0. 1017

To proceed with the analysis in a more concrete fashion, we 1018
w sw; x f w; x; gj; kw; x; restrict our discussion to a very special case, on which 1019
966
965 (21) nevertheless most of the relevant features of the theory can be 1020
j jj; kw; x; illustrated. This is the case of a controlled plant modelled by
equations of the form
968
967 viewed as an autonomous system with output
e hw; x; z f 0 w; z f 1 w; z; ee;
(23) 1022
1021
970
969
be such that: e q0 w; z q1 w; z; ee u; y e:
+ Models
JARAP 289 116

1022
1021 1076
1023 This system is very special because the relative degree (Isidori, (19). Then, there exists a (possibly set-valued) map, defined on 1077
1024 1995, p. 137) between the control input u and the regulated the whole of W, whose graph Ass is invariant for the 1078
1025 output e is equal to one, the coefficient of u in the second autonomous system (26). Moreover, for each w0 ; z0 2 Ass 1079
1026 equation (known, with an abuse of terminology, as high- there is a point j0 2 Rn such that the integral curve of (26) issued 1080
1027 frequency gain) is unitary, and, moreover, the regulated from w0 ; z0 and the integral curve of
1028 variable e is assumed to coincide with the measured variable
1029 y. It must be observed, though, that there is not much loss of j jj; 0 1082
1081
1030 generality in considering a system having this simple because
F
1031 structure, as shown for instance in Delli Priscoli et al. (2006) issued from j0 satisfy
1032 and Marconi et al. (2006), the case of a more general system 1084
1083
OO
1033 (having arbitrary relative degree, but with the dynamics of z still uss wt; zt gjt; 0 8 t 2 R: 1085
1034 driven only by e, and having non-unitary high-frequency gain) 1086
1035 can be handled in a very similar manner, after suitable pre- This is a nonlinear version of the celebrated internal model
1036 liminary manipulations. The initial conditions of (23) are principle of Francis and Wonham (1976).
1037 assumed to range on a set Z E, in which Z is a fixed compact 1087
1038 subset of Rn1 and E fe 2 R : jej cg, with c a fixed 6. Regulator design
PR
1039 number. 1088
1040 Since the problem in question is a problem concerning how 6.1. Controller structure 1089
1041 the closed-loop system behaves in steady state, there is no 1090
1042 special interest in considering exosystems that are not in The steady-state analysis presented above has identified 1091
1043 steady state. Thus without loss of generality we assume certain features that any controller must have to be able to solve 1092
1044 that the set W is invariant for (19), and hence by Lemma 3 EDthe problem at issue. As a matter of fact, this controller must 1093
1045 that W vW. include a subsystem that behaves as a generator of all inputs 1094
1046 Suppose that a controller of the form (20) solves the problem of the form uss wt; zt in which wt; zt is a trajectory of 1095
1047 of output regulation. Then Lemma 7 applies and, the following (26), issued at any point w0 ; z0 of the compact invariant set 1096
conclusions immediately come true: Ass . As such, the characterization seems to be independent of 1097
1048
1050
1049 the controller, because the system (26) and the map uss w; z 1098
CT
1050
1051 The steady-state locus vW Z E J of the closed-loop only depend on the plant. However, this is not strictly speaking 1099
1051
1053
1052 system is a subset of the set W Rn1 f0g Rn . true, because the invariant set Ass may depend on how the 1100
1052
1054 The restriction of the closed-loop system to its steady-state controller is chosen. In view of the property thus established, it 1101
1053 locus vW Z E J reduces to appears that a first, fundamental, step in the design of a 1102
1054 controller that solves the problem of output regulation is to find 1103
E
1057
1056
1055 w sw; z f 0 w; z; j jj; 0: (24) a suitable candidate for the invariant set Ass and to build a 1104
For each w; z; 0; j 2 vW Z E J device able to generate as outputs all inputs of the form 1105
1055
1056 uss wt; zt. Driving the controlled plant by means of a
RR
1057 0 q0 w; z gj; 0: (25) 1106

1059
1058
1060 device of this kind yields a system possessing a (compact) 1107
1061 invariant set on which the regulated variable is identically zero. 1108
1059
1058
1062 With this in mind we observe that, by Lemma 6, if the This is a necessary prerequisite for the solution of the problem 1109
1063 positive orbit of W Z E J under the flow of (21) is in question, but clearly not sufficient yet, because the 1110
1064 bounded, then vW Z E J is the graph of a (possibly convergence to this invariant set still has to be secured. To 1111
CO
1065 set-valued) map defined on the whole of W. As a consequence, this purpose, an additional assumption (which in general is not 1112
the set necessary, though) is useful.
1113
Ass fw; z : w; z; 0; j 2 vW Z E J 1114
Assumption (A). There exists a bounded subset B W 1115
1067
1066 for some j 2 Rn g Rn1 which contains the positive orbit of the set W Z under
UN
1116
1117
1068 is the graph of a (possibly set-valued) map defined on the whole the flow of (26) and the resulting omega-limit set A : vW
1118
of W, and is invariant for the dynamics of Z is locally exponentially stable.
1119
w sw; z f 0 w; z: (26) While in the analysis of the necessity we have only identified 1120
1070
1069
the existence of a compact set (actually, the graph of a map 1121
Define the map defined on W) which is invariant for (26), Assumption (A) 1122
implies, in its first part, the existence of a compact set A (still 1123
uss : Ass ! R
the graph of a map defined on W) which is not only invariant but 1124
1071
1072
1073 w; z 7! q0 w; z:
also uniformly attractive of all trajectories of (26) issued from 1125
1074 The conclusions reached above can be rephrased in the points of W Z. The second part of the Assumption, in turn, 1126
1075 following terms. Suppose that a controller of the form (20) strengthens this property by also requiring the set A to be 1127
1076 solves the problem of output regulation for (23) with exosystem locally exponentially stable.
+ Models
JARAP 289 116

1126 1163
1127 For convenience, rewrite the augmented system (19)(23) Suppose now that the control v is chosen as v ke, as 1164
as indicated earlier. This yields a closed-loop system that can be 1165
regarded as pure feedback interconnection of
1129
1128 z f0 z f1 z; ee e q0 z q1 z; ee u (27)
p F p P p; ee (34) 1167
1166
1130 having set z w; z. Consistently let Z : W Z denote the
1131 compact set where the initial condition z0 is supposed to viewed as a system with input e and state p, and
1132 range, and set uss z q0 z. In these notations, Assumption 1169
1168
(A) expresses the property that, in the autonomous system e Q p R p; ee ke (35)
F
1170
1134
1133 z f0 z; (28) viewed as a system with input p and state e. To obtain bounded 1171
trajectories, and to steer et to zero, one might invoke the so-
OO
1135 1172
the compact invariant set A is asymptotically and locally called small gain theorem, actually an enhanced version of it 1173
1136
exponentially stable, with a domain of attraction that contains (see e.g. Teel & Praly, 1995). As a matter of fact, suppose that
1137
the set Z. the following properties hold: 1174
1176
1175
1138
With a view to the internal model principle, we choose the 1176
following candidate controller (P1) the dynamics
PR
1178
1177
1140
1139 j jj Gv; u gj v; v ke: (29) p F p (36)
1177
1178
1179
1141
In fact, when the regulated variable e is identically zero (as it possesses a compact invariant set S which is asymptoti- 1179
1180
1142
should occur in steady state), this controller reduces to the cally (and locally exponentially) stable, with a domain of 1180
1181
autonomous system with output attraction that contains the set P : Z X of all admissible 1181
1183
1182
initial conditions, and 1182
1184
1144
1143 ED
j jj u gj; (30) (P2) the function Q p vanishes on this invariant set. 1183
1185
1145
which is supposed to be a generator capable as the internal 1186
1146 Then, a version of the small-gain theorem (enhanced to
model principle dictates to produce inputs of the form 1187
1147 allow the case in which one of the two pieces of the
uss zt, with zt a trajectory of (28) issued at any point z0 1188
1148 interconnection, in this case system (34), possesses an
of A. 1189
1149
CT
asymptotically stable compact invariant set) can be invoked 1190
Controlling system (27) by means of (29) yields a closed-
to conclude that, if k is large enough (the lower limit of k being 1191
loop system
dependent on the actual choice of the set P E of admissible 1192
z f0 z f1 z; ee; e q0 z q1 z; ee gj v; initial conditions), all trajectories of the interconnected system 1193
(33) are bounded and do converge, as t ! 1, to the invariant set 1194
E
j jj Gv S f0g. Thus, in particular, et ! 0 as t ! 1 and the problem 1195

1151
1150
(31) of output regulator is solved. 1196
1152 Motivated by this observation, in what follows we will seek
RR
1197
1153 which, regarded as a system with input v and output e, has properties such as (P1) and (P2).
relative degree 1. It can be put in normal form by changing
variables as 6.2. The internal model property
1155
1154 1198
x j Ge 1199
Properties (P1) and (P2) are only determined by the
CO
which yields properties of the autonomous system (28) and of the function
1201
1200
z f0 z f1 z; ee; u q0 z (37)
1202
x jx Ge Ggx Ge Gq0 z Gq1 z; ee; (32) which, in the composite system (36), can be viewed as an output
1157
1156
of (28) driving a system of the form
UN
1158 e q0 z q1 z; ee gx Ge v: 1204
1203
x jx Ggx u: (38) 1205
Setting p z; x, this system can be further rewritten in the
1206
1160
1159 form For convenience, we will say that triplet fjx; G; gxg is an
1207
asymptotic internal model of the pair (28)(37) if properties
p F p P p; ee e Q p R p; ee v (33) (P1) and (P2) are satisfied. In this terminology, we can sum-
1208
1209
in which marize the conclusion of the previous subsection as follows.
1210
f0 z 1211
1162
1161 F p Proposition 1. Pick compact sets Z, E and J for the initial
jx Ggx q0 z 1212
1163 Q p gx q0 z conditions of the closed-loop system (19), (23), (29). Suppose
1213
1164 that Assumption A holds and that the triplet fjx; G; gxg is
1214
and P p; e, R p; e are suitable continuous functions. an asymptotic internal model for (28)(37). Then there exists
+ Models
JARAP 289 116

1213 1258
$ $
1214 k > 0such that for all k k the controller (29) with v ke which the map tz that characterizes the steady-state locus of 1259
1215 solves the generalized tracking problem. (36) is singled-valued and C 1 . To this end, we stress that the 1260
1216
properties in question are quite similar to properties that are 1261
1217 The notion of steady state provides a useful interpretation of
usually sought in the design of state observers. As a matter of 1262
1218 the properties in question. In fact, recall that, by Assumption A,
fact it is seen from (41)(42) that, for each z0 2 A, the function 1263
1219 all trajectories of system (28) with initial conditions in Z
of time
1220 asymptotically converge to the compact invariant set A, and the
1221 latter is also locally exponentially stable. If property (P1) holds, xt tzt; z0 1265
1264
all trajectories of the composite system
F
which is defined (and bounded) for all t 2 R satisfies
z f0 z dxt
OO
1223
1222 x jx Ggx q0 z jxt (43)
dt 1267
1266
1224 with initial conditions in Z X asymptotically converge to the and, moreover,
1225 limit set vZ X. Since (36) is a triangular system, its is
1226 readily seen (see also Lemma 6), that the set vZ X is the gxt q0 zt; z0 : 1268
1269
1270
1227 graph of a set-valued map defined on A, i.e. that there exists a In view of the latter, system (38) can be rewritten in the form
PR
map
x jx Ggx gx (44) 1272
1271
1229
1228 t : z 2 A 7! tz Rn ;
and interpreted as a copy of the dynamics (43) of x corrected by 1273
such that an innovation term gx gx weighted by an output 1274
injection gain G. This is the classical structure on an observer 1275
1230
1231
1232 vZ X fz; x : z 2 A; x 2 tzg : grt: 1276
EDand the requirement in (P1) precisely expresses the property
1233 The set grt is the steady-state locus of (36) and the that the difference xt xt (the observation error, in our 1277
1234 restriction of the latter to this invariant set characterizes its interpretation) should asymptotically decay to zero (with ulti- 1278
1235 steady-state behavior. Property (P2), on the other hand, mate exponential decay). 1279
expresses the property that at each point of z; x 2 grt This interpretation is at the basis of a number of major recent 1280
advances in the design of regulators. In fact, in a number of 1281
1237
1236 q0 z gx: (39)
CT
recent papers, this interpretation has been pursued and, taking 1282
1238 into consideration various approaches to the design of nonlinear 1283
Thus, looking again at system (36), it is realized that grt is in
fact invariant for the simpler system observers, has lead to effective design methods. In Byrnes and 1284
Isidori (2004), the approach of BornardGauthierHammouri 1285
1240
1239
1241 z f0 z x jx: (40) Kupca to the design of high-gain observers (as described, e.g. 1286
E
1242 by Gauthier and Kupka (2001)) has been followed. This design 1287
Note that, if the map tz is single-valued and C 1 , its
requires the extra assumption that the set of all functions of the 1288
invariance for (40) is expressed by the property that
RR
form ut : uss zt obtained as z0 ranges over A be a subset 1289

@tz of the set of solutions of a (nonlinear) differential equation
1244
1243 f z jtz 8 z 2 A; (41)
@z 0
1245 ud fud1 ; . . . ; u1 ; u 0: (45) 1291
1290
while the fact that (39) holds at each point of z; x 2 grt is
The controller obtained in this way cannot handle cases in 1292
expressed by the property that
1293
CO
1246
1247
which the exosystem contains uncertain constant parameters
1248 q0 z gtz 8 z 2 A: (42) 1294
(as it is in the case of harmonic oscillator of unknown fre-
1249 1295
Properties (41) and (42) have been usually referred to, in the quency). To handle this case, an adaptive nonlinear observer is
1250 1296
literature (see, e.g. Isidori, 1995), as properties of immersion of needed, as that of Bastin and Gevers (1988), which in turn
though requires stronger hypotheses, such as the possibility of 1297
system
expressing the state-space version of (45) in a form which can 1298
UN
z f0 z be made linear by means of output injection and diffeomorph- 1299

1252
1251 1300
u q0 z isms. The design of a regulator on the basis of the theory of
adaptive observers was pursued by Serrani et al. (2001) and by 1301
into system Delli Priscoli et al. (2006). Finally, in Marconi et al. (2006), the 1302
more recent advances in the theory of nonlinear observers 1303
1253
1254 x jx
1255 obtained by Andrieu and Praly (2006) have been exploited, 1304
u gx:
to show that a triplet fjj; G; gjg having internal model 1305
property controller does in fact exist always, and no assumption 1306
1256 6.3. Nonlinear observers as internal models like the existence of an equation of the form (45) is actually 1307
1257 required. In the triplet in question, jj has the following form
1258 In this section, we discuss how the properties (P1) and (P2)
can be enforced. To simply matters, we refer to the case in jj Fj Ggj 1309
1308
+ Models
JARAP 289 116

1309
1308 1366
1310 in which F; G is a controllable pair, with F a d d Hurwitz Francis, B. A. (1977). The linear multivariable regulator problem. SIAM 1367
Journal Control and Optimization, 14, 486505. 1368
1311 matrix, and sufficiently large d. However, no closed form is
Francis, B. A., & Wonham, W. M. (1976). The internal model principle of 1369
1312 immediately available for gj and this function is only guar- control theory. Automatica, 12, 457465. 1370
1313 anteed to be continuous. Gardner, M. F., & Barnes, J. L. (1942). Transients in linear systems. Wiley. 1371
Gauthier, J. P., & Kupka, I. (2001). Deterministic observation theory and 1372
applications. Cambridge University Press. 1373
7. Conclusions
1314 Hahn, W. (1967). Stability of motions. Springer-Verlag. 1374
Hale, J. K., Magalhaes, L. T., & Oliva, W. M. (2002). Dynamics in infinite 1375
1315 Classically, engineers have developed two intuitive dimensions. New York, NY: Springer-Verlag. 1376
F
1316 approaches to formale the concept of steady-state response. Huang, J., & Lin, C. F. (1994). On a robust nonlinear multivariable servome- 1377
1317 The first is that such a motion should be the response of the chanism problem. IEEE Transactions on Automatic Control, AC-39, 1510 1378
OO
1318 1513. 1379
system as the initial time tends to 1, independent of the
Huang, J., & Rugh, W. J. (1990). On a nonlinear multivariable servomechanism 1380
1319 initial condition. The second is based on the idea that every problem. Automatica, 26, 963972. 1381
1320 motion can be decomposed into the superposition of a transient Isidori, A. (1995). Nonlinear control systems (3rd ed.). Springer-Verlag. 1382
1321 response and a steady-state response. Any rigorous interpreta- Isidori, A., & Byrnes, C. I. (1990). Output regulation of nonlinear systems. 1383
1322 tion of the first approach would require the trajectories to be IEEE Transactions on Automatic Control, AC-25, 131140. 1384
1323 James, H. M., Nichols, N. B., & Phillips, R. S. (1947). Theory of servomechan- 1385
bounded backward in time, while any effective use of the
PR
1324 isms. McGraw-Hill. 1386
second interpretation for analysis and design would require Khalil, H. (1994). Robust servomechanism output feedback controllers for 1387
1325 some form of uniformity: that is, for any given tolerance and feedback linearizable systems. Automatica, 30, 5871599. 1388
1326 any given (compact) set of initial conditions there should exist a Khalil, H. (2002). Nonlinear systems (3rd ed.). Prentice-Hall. 1389
1327 time after which the magnitude of the transient response is less Marconi, L., Praly, L., & Isidori, A. (2006). Output stabilization via nonlinear 1390
1328 luenberger observers. SIAM Journal on Control and Optimization, 45, 1391
than this given tolerance regardless of the choice of initial
1329 22772298. 1392
condition. Sell, G. R., & You, Y. (2002). Dynamics of evolutionary equations. New York,
ED 1393
1330 In this paper, we outlined concepts and tools necessary to NY: Springer-Verlag. 1394
1331 motivate, for general nonlinear systems, a simple definition of Serrani, A., Isidori, A., & Marconi, L. (2001). Semiglobal nonlinear output 1395
1332 steady-state behavior which captures both of the classical regulation with adaptive internal model. IEEE Transactions on Automatic 1396
1333 Control, AC-46, 11781194. 1397
intuitive approaches: boundendess of steady-state trajectories
Sontag, E. D., & Wang, Y. (1995). On the characterizations of input-to-state 1398
1334 backward and forward in time and uniform attractivity of the stability with respect to compact sets. IEEE Conference on Decision and 1399
CT
1335 steady-state behavior. A rigorous technical treatment is Control, 226231. 1400
1336 provided as well as an indication of the recent applications Teel, A. R., & Praly, L. (1995). Tools for semiglobal stabilization by partial state 1401
1337 of these ideas and tools to the generalized tracking problem for and output feedback. SIAM Journal on Control and Optimization, 33, 1443 1402
1338 1485. 1403
nonlinear control systems.
Vinograd, R. E. (1957). The inadequacy of the method of characteristic 1404
E
exponents for the study of nonlinear differential equations. Mathematics 1405

References of the USSR-Sbornik, 41, 431438 (in russian). 1406
1339 Willems, J. C. (1991). Paradigms and puzzles in the theory of dynamical 1407
1340
RR
Andrieu, V., & Praly, L. (2006). On the existence of a KazantisKravaris/ systems. IEEE Transactions on Automatic Control, AC-36, 259294. 1408
1341 Zadeh, L. A., & Desoer, C. A. (1963). Linear system theory. McGraw-Hill. 1409
1342 Luenberger observer. SIAM Journal on Control and Optimization, 45, 432
456. 1410
1343 1411
1344 Bhatia, N. P., & Szego, G. P. (1970). Stability theory of dynamical systems. Alberto Isidori was educated in electrical engineering at the University of
Berlin: Springer-Verlag. 1412
1345 Rome La Sapienza, where he also became a professor of Automatic
Bastin, G., & Gevers, M. R. (1988). Stable adaptive observers for non-linear 1413
1346 Control in 1975. Since 1989 he has also been affiliated with the Department
CO
time varying systems. IEEE Transactions Automatic Control, AC-33, 650 1414
1347 of Electrical and Systems Engineering at Washington University in St.
657. 1415
1348 Louis. His research activity has been mostly addressed to problems of
Birkhoff, G. D. (1927). Dynamical systems. American Mathematical Society. 1416
1349 analysis and design of nonlinear control systems. He is author of an
Byrnes, C. I., & Isidori, A. (2003). Limit sets, zero dynamics and internal 1417
1350 influential book and has received numerous major awards, which include
models in the problem of nonlinear output regulation. IEEE Transactions on 1418
1351 the Giorgio Quazza Medal from IFAC in 1996, the H. Bode Lecture
Automatic Control, AC-48, 17121723. 1419
1352 Prize from IEEE in 2001, the Ktesibios award from the Mediterranean
Byrnes, C. I., & Isidori, A. (2004). Nonlinear internal models for output 1420
UN
1353 Control Association in 2000. He received twice the G. Axelby Award for
regulation. IEEE Transactions on Automatic Control, AC-49, 22442247. 1421
1354 papers published in the IEEE Transactions on Automatic Control, in 1981
Byrnes, C. I., Gilliam, D. S., Isidori, A., & Ramsey, J. (2003). On the steady- 1422
1355 and 1990, and twice the Automatica Award for papers published in Auto-
state behavior of forced nonlinear systems. In W. Kang, M. Xiao, & C. 1423
1356 matica, in 1991 and 2004. He is a Fellow of IEEE (1986) and IFAC (2005).
Borges (Eds.), New trends in nonlinear dynamics and control, and their 1424
1357 He has been President of the European Community Control Association in
applications (pp. 119144). London: Springer-Verlag. 1425
1358 19951997 and is currently President Elect of IFAC in the triennium 2005
Celani, F. (2003). Omega-limit sets of nonlinear systems that are semiglobally 1426
1359 2008.
1360 practically stabilized. Doctoral Dissertation. Washington University in St. 1427
1361 Louis. 1428
Davison, E. J. (1976). The robust control of a servomechanism problem for Christopher I. Byrnes holds the E. H. and F. G. Skinner Professorship of 1429
1362
1363 linear time-invariant multivariable systems. IEEE Transactions on Auto- Systems Science and Mathematics in the Department of Electrical and 1430
1364 matic Control, AC-21, 2534. Systems Engineering at Washington University. Prior to coming to Washing- 1431
Delli Priscoli, F., Marconi, L., & Isidori, A. (2006). A new approach to adaptive ton University in 1989, he was on the faculty at the University of Utah, 1432
1365
1366 nonlinear regulation. SIAM Journal on Control and Optimization, 45, 829 Harvard University and Arizona State University and has held visiting 1433
855. positions in Austria, France, Germany, Italy, Japan, the Netherlands, Swe-
+ Models
JARAP 289 116

1438
1433
1434 den, the US and the Academy of Sciences of the former USSR. A Fellow Mathematics from SIAM for his contributions to Control Theory and 1439
1435 of the IEEE, he received the IEEE George Axelby Award in 1991 and Differential Equations. He holds an Honorary Doctorate of Technology 1440
1436 in 2003, and was named the recipient of the IEEE 2008 H. Bode from the Royal Institute of Technology in Stockholm and, in 2002, was 1441
1437 Prize Lecture Award. He received the 1993 IFAC for the Best Paper named a Foreign Member of the Royal Swedish Academy of Engineering 1442
1438 appearing in the journal Automatica and in 2005, and the Reid Prize in Sciences. 1443
F
OO
PR
ED
E CT
RR
CO
UN

Uncorrected Proof: Steady-State Behaviors in Nonlinear Systems With An Application To Robust Disturbance Rejection

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Uncorrected Proof: Steady-State Behaviors in Nonlinear Systems With An Application To Robust Disturbance Rejection

Uploaded by

Copyright:

Available Formats

+ Models

JARAP 289 116

1 Annual Reviews in Control xxx (2008) xxxxxx

3 Steady-state behaviors in nonlinear systems with an application

2 A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx

A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx 3

195 limit points of the motion xt; x0 . x1 t u U sin vt f;

4 A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx

281 and to say that the system is in steady state if x0 2 V. 313

A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx 5

6 A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx

486 3. Limit sets and stability

A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx 7

The notion thus introduced recaptures the classical notion of 624

575 interior of B. Then, vB is asymptotically stable, with a z Fz Gu (8) 629

8 A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx

A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx 9

Example. Consider the system

10 A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx

856 terms. The controlled plant is a finite-dimensional, time- 911

A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx 11

The control law for (17) is to be provided by a system 1003

j jj; y bounded. Then

958 allowed to range over a fixed compact sets X  Rn , W  Rs and, 1007

required that variable is 0. 1017

12 A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx

1057 0 q0 w; z gj; 0: (25) 1106

A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx 13

j jj Gv S  f0g. Thus, in particular, et ! 0 as t ! 1 and the problem 1195

14 A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx

form ut : uss zt obtained as z0 ranges over A be a subset 1289

z f0 z be made linear by means of output injection and diffeomorph- 1299

A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx 15

exponents for the study of nonlinear differential equations. Mathematics 1405

16 A. Isidori, C.I. Byrnes / Annual Reviews in Control xxx (2008) xxxxxx

You might also like

958 allowed to range over a fixed compact sets X Rn , W Rs and, 1007

j jj Gv S f0g. Thus, in particular, et ! 0 as t ! 1 and the problem 1195