[Paper Review] Noncausal FIR Zames-Falb Multiplier Search for Exponential Convergence Rate

Jingfan Zhang, Peter Seiler+1 more|arXiv (Cornell University)|Feb 25, 2019

Stability and Control of Uncertain Systems23 references4 citations

TL;DR

This paper extends the search for noncausal finite impulse response (FIR) Zames-Falb multipliers to estimate exponential convergence rates in Lur’e systems, demonstrating that noncausal multipliers yield less-conservative stability bounds than causal or anticausal counterparts. It establishes equivalence between time-domain IQC and frequency-domain ρ-IQC frameworks under variable transformation and proposes a unified factorization method for causal, anticausal, and noncausal multipliers, validated through numerical examples showing improved convergence rate estimates.

ABSTRACT

In the existing literature, there are two approaches to estimate tighter bounds of the exponential convergence rate of stable Lure systems. On one hand, the classical integral quadratic constraint (IQC) framework can be applied under loop-transformation, so the stability of the new loop implies the convergence of the original loop. On the other hand, it is possible to modify the IQC framework, the so-called rho-IQC framework, in such a way that the convergence rate is directly obtained over the original loop. In this technical note, we extend the literature results from the search for a causal finite impulse response (FIR) Zames-Falb multiplier to the noncausal case. We show that the multipliers by the two approaches are equivalent by a change of variable. However, the factorisation of the Zames-Falb rho-IQC is restricted compared to the Zames-Falb IQC, so an unified factorisation is proposed. Finally, numerical examples illustrate that noncausal multipliers lead to less-conservative results.

Motivation & Objective

• To extend the use of noncausal FIR Zames-Falb multipliers beyond causal cases for tighter exponential convergence rate estimation in Lur’e systems.
• To establish equivalence between the time-domain IQC and frequency-domain ρ-IQC frameworks when using noncausal multipliers via variable transformation.
• To develop a unified factorization structure applicable to causal, anticausal, and noncausal FIR Zames-Falb multipliers in both IQC and ρ-IQC frameworks.
• To numerically validate that noncausal multipliers yield less-conservative convergence rate bounds compared to causal and anticausal alternatives.

Proposed method

• Extends the classical Zames-Falb multiplier framework to noncausal FIR forms, enabling tighter exponential convergence rate analysis.
• Applies variable transformation to show equivalence between the time-domain Zames-Falb IQC and frequency-domain ρ-IQC frameworks under noncausal multipliers.
• Proposes a unified lifting-based factorization method applicable to causal, anticausal, and noncausal FIR Zames-Falb multipliers in both IQC and ρ-IQC settings.
• Uses the Kalman-Yakubovich-Popov (KYP) lemma to convert frequency-domain conditions into computable linear matrix inequalities (LMIs).
• Implements numerical examples with varying system uncertainties and maximum slopes to compare convergence rate estimates across multiplier types.
• Employs MATLAB-based numerical computation with careful handling of noncausal matrix structures to avoid numerical instability.

Experimental results

Research questions

• RQ1 Can noncausal FIR Zames-Falb multipliers provide less-conservative estimates of the exponential convergence rate in Lur’e systems compared to causal and anticausal multipliers?
• RQ2 Are the time-domain Zames-Falb IQC and frequency-domain ρ-IQC frameworks equivalent when noncausal multipliers are used?
• RQ3 What unified factorization structure enables consistent analysis of causal, anticausal, and noncausal FIR Zames-Falb multipliers in both IQC and ρ-IQC frameworks?
• RQ4 How do different multiplier types (causal, anticausal, noncausal) perform in estimating convergence rates for systems with odd and general nonlinearities?
• RQ5 What is the impact of numerical challenges on the practical implementation of noncausal multipliers in MATLAB-based LMI solvers?

Key findings

• Noncausal FIR Zames-Falb multipliers yield less-conservative convergence rate estimates than causal and anticausal multipliers, particularly for systems near instability.
• The time-domain Zames-Falb IQC and frequency-domain ρ-IQC frameworks are equivalent under noncausal multipliers via a change of variable, with M(z) in (2) and M(ρz) in (3) being interchangeable.
• A unified lifting-based factorization is proposed that supports causal, anticausal, and noncausal FIR Zames-Falb multipliers in both IQC and ρ-IQC frameworks, resolving structural restrictions in existing factorizations.
• For systems violating the Kalman conjecture (e.g., Ex. 4), noncausal multipliers outperform anticausal ones, as seen in the case where anticausal multipliers fail at K=12 while noncausal ones remain valid.
• The results from multipliers M(z) and M(ρ,z) are nearly identical, confirming their equivalence, though M(ρ,z) incurs higher computational cost due to larger matrices.
• For odd nonlinearities, the optimal convergence rate is no greater than for general nonlinearities, reflecting stricter ℓ₁-norm constraints in the latter case.