arXiv:2009.00727v2 [eess.SY] 11 Sep 2020
Performance Analysis and Non-Quadratic Lyapunov Functions for
Linear Time-Varying Systems
Matthew Abate, Corbin Klett, Samuel Coogan, and Eric Feron
Abstract— Performance analysis for linear time-invariant
(LTI) systems has been cl osely tied to quadratic Lyapunov
functions ever since it was shown that LTI system stability is
equivalent to the existence of such a Lyapunov function. Some
metrics for LTI systems, however, have resisted treatment via
means of quadratic Lyapunov functions. Among these, point-
wise-in-time metrics, such as peak norms, are not captured
accurately using these techniques, and this shortcoming has
prevented the development of tools to analyze system behavior
by means other than e.g. time-domain simulations. This work
demonstrates how the more general class of homogeneous poly-
nomial Lyapunov functions can be used to approximate point-
wise-in-time behavior for LTI systems with greater accuracy,
and we extend this to the case of linear time-varying (LTV)
systems as well. Our findings rely on th e recent observation that
the search for homogeneous polynomial Lyapunov functions for
LTV systems can be recast as a search for quadratic Lyapunov
functions for a related hierarchy of time-varying Lyapunov
differential equations; thus, performance guarantees for LTV
systems are attainable without heavy computation. Numerous
examples are provided to demonstrate the findings of this work.
I. INTRODUCTION
Beginner’s courses on linear systems quickly introduce the
Lyapunov function as a natur al means to expre ss system
stability in terms of energy lo ss. O ne essential result of
Lyapunov states that the stability of a linear time invariant
(LTI) system is equ ivalent to the existence of a quadratic
energy function that decays along system trajector ie s [1],
and since then qua dratic stability theory has been greatly
extended to develop metrics and indicators of performance
such as passivity [2, Chapte r 14], [3] and robustness [4].
These metrics generally leverage the ubiq uitous presence of
the quadratic Lyapunov functions that are n a turally embed-
ded in stable LTI systems [5].
Some metric s for LTI systems, however, have resisted
treatment via means of quadra tic Lyapunov functions.
Among these, poin t-wise-in-time metrics, such as peak
norms, are not captured accurately [6], and this shortcoming
has prevented the development of tools to analyze system
behavior by me a ns oth er than time-domain simulations.
This work was supported by the KAUST baseline budget.
M. Abate is with the School of Mechanical Engineering and the School
of Electrical and Computer Engineering, Georgia Institute of Technology,
Atlanta, 30332, USA: Matt.Abate@GaTech.edu.
C. Klett is with the School of Aerospace Engineering, Georgia Institute
S. Coogan is with the School of Electrical and Computer Engineering
and the School of Civil and Environmental Engineering, Georgia Institute
E. Feron is with the Department of Electrical Engineering, King Ab-
dullah University of Science and Technology, Thuwal, Saudi Arabia:
Eric.Feron@Kaust.edu.sa.
When extending to the case of linear time-varying (LTV)
systems, new cha llenges emerge: for instance, it is known
that not a ll stable LTV systems c an be certified v ia quadratic
Lyapunov functions [7], and the time-varying na ture of these
systems reduces the ease of simulation. Fur ther, analytical
considerations in simulation are often steere d by subjective
criteria: for example, the stopping-time of a simulation is, in
practice, generally chosen by either analysing the poles of
the system or the relative distance to the steady-state output
(See, e.g. [8, impulse.m]). For these reasons, it is useful to
have means other than simulation for extracting time domain
properties f or LTV systems.
The topic addressed in this paper relies on the re cent ob-
servation in [9] that the search for homogene ous polynomial
Lyapunov functions for LTV systems can be recast as the
search for quadratic Lyapunov fun ctions for a related hierar-
chy of Lyapunov differential equa tions. Indeed, every stable
LTV system induces a h omogeneous polynomial Lyapunov
function [10], [11], and the search for such a Lyapunov
function is easily expr essed as sum-of-squares and foun d
by solving a convex, semi-definite feasibility program. Our
contribution is to show that the aforementioned hierarchy
of LTI systems defines a p owerful framework for extracting
time-dom ain properties of LTV systems, a nd we particularly
show how one can compute bounds on the impulse an d step
response of LTV systems using homoge neous polynomial
Lyapunov functions.
This paper is structured as follows. We in troduce our
notation in Section II. In Section III we recall a procedure
for computing norm bounds on the impulse response of LTI
systems, and this procedure relies on the use of quad ratic
Lyapunov functions. In the same section, we introduce a
hierarchy of LTI systems that can be used to compute homo-
geneous polynomial Lyapunov functions for LTI systems. We
show how th e afor ementioned hierarchy is used to compute
bounds on the impulse responses of LTI systems in Section
IV and bo unds on the step responses of LTI systems in
Section V. Similar bounds are co mputed for LTV systems
in Section VI, and we additionally present a procedure for
computing convergence envelopes on the impulse response of
LTV systems. We demonstrate our findings through numer-
ous examples that appear througho ut the work and through
a case study presented in Section VII.
II. NOTATION
We denote by S
n
++
⊂ R
n×n
the set of symmetric positive
definite n ×n matr ices. We denote by I
n
the n ×n identity
matrix, and we denote by 0
n
∈ R
n
the zero vector in R
n
.