Brief communication
Input–output behavior for stable linear systems

https://doi.org/10.1016/S0016-0032(00)00085-5Get rights and content

Abstract

In this paper, a controllable, observable, asymptotically stable, finite-dimensional linear system is considered. The input–output problem considered is whether the input in a fairly general class of Hilbert spaces will produce the output in the same class. The problem in this generality appears to be very difficult and in this paper a large class of Hilbert spaces is determined for which the result is true and a series of counter examples are given to the more obvious conjectures.

Introduction

In this paper, we consider a controllable, observable, asymptotically stable, single-input–single-output finite-dimensional linear system. The problem we want to consider is motivated by a classical result on BIBO stability. It is well known that a controllable and observable linear system of BIBO is stable if and only if it is internally exponentially stable. It means that for such a system, a “bounded” input, namely, a function in the Lp space, where p∈[1,∞], produces a “bounded” output in the same Lp space. However, this result is limited due to the fact that this assumes that the input is either bounded or “small” for large values of t. We are often interested in the response of a system to other inputs — for example, ramp inputs. In this paper, we only consider the case p=2 where the input space is also Hilbert. We will show that this classical result can be extended to other classes of inputs. We consider the case, that the input is in a fairly general class of Hilbert spaces and ask if this implies that the output is in the same class. The problem in this generality appears to be very difficult, but we will determine a large class of Hilbert spaces for which the result is true and give a series of counterexamples to the more obvious conjectures.

We begin by stating the well-known result in the L2 case (see, for example, [1], [2], [3], [4]).

Theorem 1.1

The linear, controllable and observable system ẋ=Ax+bu,y=cx is asymptotically stable if and only if for every u(t)∈L2[0,∞) the output y(t)∈L2[0,∞).

We will begin by proving the following theorem which includes many of the inputs of interest.

Theorem 1.2

The linear finite-dimensional controllable observable system ẋ=Ax+bu,y=cx has its spectrum in the closed left half-plane if and only if for every λ>0 and for every control u(t)∈L2([0,∞),e−λtdt) y(t)∈L2([0,∞),e−λtdt).

It is tempting to believe that the following would be true.

Conjecture 1.1

Let H be a L2([0,∞,dμ(t)) space where dμ(t) is an arbitrary positive measure. Let ẋ=Ax+bu,y=cx be an asymptotically stable finite-dimensional linear system, then uH implies that 0tceA(t−s)bu(s)ds∈H.

We will show that this conjecture is false and that it is quite difficult to find a completely suitable substitute.

The main thrust of this paper will be devoted to attempting to find an answer to the following problem.

Problem 1.1

Let f have the following properties:

  • 1.

    f is locally Lebesgue integrable on the positive real numbers.

  • 2.

    f(t)⩾0 for t⩾0.

  • 3.

    f is piecewise continuous.

What are necessary and sufficient conditions on the function f which ensure the following statement:

Letẋ=Ax+Bu,y=Cx,where xRn, yRp and uRm, be a controllable, observable and exponentially stable system. If0|u(t)|2f(t)dt<∞then0|y(t)|2f(t)dt<∞.

In Section 2, we present two theorems with their proofs concerning Problem 1.1. In Section 3, we present a series of counterexamples.

Section snippets

Main results

In this section, we begin by proving Theorem 1.2. Theorem 1.2 generalizes the classical result of Theorem 1.1 to include many examples of interest including inputs with polynomial growth. We then prove two quite general results to provide a partial answer to Problem 1.1.

Proof of Theorem 1.2

For any u(t)∈L2([0,∞),e−λtdt), we can formally writey(t)e−(λ/2)t=e−(λ/2)t0tceA(t−s)bu(s)ds=0tce(A−(λ/2)I)(t−s)be−(λ/2)su(s)ds.Obviously, if u(t)∈L2([0,∞),e−λtdt), then e−(λ/2)tu(t)∈L2[0,∞), and conversely. Therefore, y(t)e−(λ

Counterexamples

In this section, we construct a series of examples of functions f which do not satisfy the input–output stability condition stated in Problem 1.1. These examples along with the theorems give strong indications as to what the nature of the functions must be that satisfy the conditions imposed in Problem 1.1.

The first example shows that the zero set of f must be small in some sense.

Example 3.1

Let f be defined as follows:f(t)=0,t∈[2n,2n+1)n=0,1,…,1otherwise.

Here it is clear that if we take a control that

References (4)

  • T. Kailath

    Linear Systems

    (1980)
  • C.T. Chen, Linear System Theory and Design, HRW,...
There are more references available in the full text version of this article.

Cited by (0)

1

The work was supported in Part by the Swedish TFR Grant TFR-98-154.

2

The work of Martin was supported in Part by NASA Grants NAG2-902 and NAG2-899, NSF Grant DMS 9628558 and NFS Grants ECS 9705312 and 9720357. This work was done while Martin was visiting KTH.

View full text