1082 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 45, NO. 6, JUNE 2000
A Characterization of Integral Input-to-State Stability
David Angeli, Eduardo D. Sontag, and Yuan Wang
Abstract—The notion of input-to-state stability (ISS) is now
recognized as a central concept in nonlinear systems analysis. It
provides a nonlinear generalization of finite gains with respect to
supremum norms and also of finite gains. It plays a central
role in recursive design, coprime factorizations, controllers for
nonminimum phase systems, and many other areas. In this paper,
a newer notion, that of integral input-to-state stability (iISS), is
studied. The notion of iISS generalizes the concept of finite gain
when using an integral norm on inputs but supremum norms of
states, in that sense generalizing the linear “ ” theory. It allows
one to quantify sensitivity even in the presence of certain forms of
nonlinear resonance. We obtain here several necessary and suf-
ficient characterizations of the iISS property, expressed in terms
of dissipation inequalities and other alternative and nontrivial
characterizations. These characterizations serve to show that
integral input-to-state stability is a most natural concept, one that
might eventually play a role at least comparable to, if not even
more important than, ISS.
Index Terms—Dissipation inequalities, finite gain, input-to-state
stability, nonlinear systems, tracking.
I. INTRODUCTION
O
NE OF the main issues in control design concerns the
study of closed-loop sensitivity to disturbances, and, more
generally, of the dependence of state trajectories on actuator
and measurement errors, magnitudes of tracking signals, and
the like. In linear systems theory, classical frequency-domain
measures of performance such as root loci and gain-phase char-
acteristics have led to the modern theories of “ ” control and
its variants.
During the last ten years or so, the notion of input-to-state
stability (ISS) was formulated (in [23]), and quickly became a
foundational concept upon which much of modern nonlinear
feedback analysis and design rest. As an illustration, let us point
out Kokotovic’s recent survey paper [10], intended as a sum-
mary of current work and future directions on nonlinear design,
in which the notion of ISS plays a central unifying concept. Sev-
eral current textbooks and monographs, including [12]–[14] and
[21], make use of the ISS notion and results, sometimes in an
essential manner.
Applications of input-to-state stability are now widespread.
Besides the many applications to recursive design in the above-
Manuscript received July 6, 1998; revised September 15, 1999. Recom-
mended by Associate Editor, L.-S. Wang. This work was supported in part by
the U.S. Air Force under Grant F49620-98-1-0242 and in part by the National
Science Foundation under Grant DMS-9457826.
D. Angeli is with the Dipartimento Sistemi e Informatica, University of Flo-
rence, Firenze 50139 Italy (e-mail: angeli@dsi.unifi.it).
E. D. Sontag is with the Department of Mathematics, Rutgers—The State
University, New Brunswick, NJ 08903 USA (e-mail: sontag@hilbert.rut-
gers.edu).
Y. Wang is with the Department of Mathematics, Florida Atlantic University,
Boca Raton, FL 33431 USA (e-mail: ywang@math.fau.edu).
Publisher Item Identifier S 0018-9286(00)05275-2.
mentioned books, let us merely cite a few additional references:
singular perturbation analysis [3], powerful global small-gain
theorems [11], foundations of tracking design [19], supervi-
sory/switching adaptive control [6], observers [8], almost-dis-
turbance decoupling for non-minimum-phase systems [9], and
feedback stabilization with bounded controllers [29]. Moreover,
this concept has many equivalent versions, which indicates that
it is mathematically natural: there are characterizations in terms
of dissipation, robustness margins, and classical Lyapunov-like
functions; see, e.g., [25] and [26].
As remarked in [24], input-to-state stability is a nonlinear
generalization both of finite gain with respect to supremum
norms and of finite gain (“nonlinear ”); this property
takes account of initial states in a manner fully compatible
with classical Lyapunov stability and replaces finite linear
gains, which represent far too strong a requirement for general
nonlinear operators, with “nonlinear gains.”
A system that is ISS exhibits low overshoot and low total
energy response when excited by uniformly bounded or en-
ergy-bounded signals, respectively. These are highly desirable
qualitative characteritics. However, it is sometimes the case that
feedback design does not render ISS behavior, or that only a
weaker property than ISS is verified in a step in recursive de-
sign.
One such weaker, but still very meaningful, property was
given the name of integral input-to-state stability (iISS) in a
recent paper [24]. This property reflects the qualitative property
of small overshoot when disturbances have finite energy and
provides a qualitative analog of “finite norm” for linear
systems. This is a property with obvious physical significance
and relevance. The paper [24] showed that iISS is, in general,
strictly weaker than ISS, and provided a very conservative
Lyapunov-type sufficient condition. This paper provides sev-
eral foundational results, showing that the iISS property is a
most natural one to be expected for well-behaved nonlinear
systems, being equivalent to the combination of well-known
dissipation and detectability properties, and admitting elegant
Lyapunov-theoretic characterizations. We are confident that
once the results in this paper become more widely known,
iISS will play a role at least as prominent as the one that ISS
currently has.
In fact, the notion of iISS, and the results in this paper,
which were previously announced in electronic preprint form,
have already played a role in several recent control works. For
example, the iISS property appears in the latest approaches to
supervisory design in adaptive control. In [7], Hespanha and
Morse—citing preprints of this work—studied the closed-loop
system obtained when a high-level supervisor directs the
switching among a family of candidate controllers for an
uncertain plant. Their convergence analysis was based on the
0018–9286/00$10.00 © 2000 IEEE