[Paper Review] A Remark on Monotone I/O Systems

Eduardo D. Sontag|ArXiv.org|Mar 15, 2005

Nonlinear Dynamics and Pattern Formation10 references3 citations

TL;DR

This paper demonstrates that small-gain theorems for negative feedback loops in monotone input/output systems can be derived by embedding the original system into a larger, extended monotone system. By analyzing the extended system under positive feedback, the paper shows that stability conditions for the original negative feedback loop are equivalent to those for the monotone extended system, thereby unifying the treatment of positive and negative feedback in monotone I/O systems.

ABSTRACT

This note remarks that small-gain results for a negative feedback loop around a monotone system can be seen as consequences of results concerning an extended monotone system.

Motivation & Objective

• To unify the analysis of positive and negative feedback in monotone input/output systems by showing that negative feedback results can be reduced to positive feedback results on an extended system.
• To demonstrate that small-gain conditions for anti-monotone feedback loops are equivalent to stability conditions for a monotone extended system under positive feedback.
• To provide a conceptual framework that simplifies the derivation of small-gain theorems by leveraging the theory of monotone systems.
• To establish a connection between periodic behavior in the original system and multi-stability in the extended system.
• To explore the implications of the embedding approach for delay systems and the persistence of oscillatory behavior when the small-gain condition fails.

Proposed method

• Construct an extended 2n-dimensional system by cascading two copies of the original monotone I/O system, with state space $X \times X$.
• Define a non-standard partial order on the extended state space: $(x,z) \leq (x',z') \Leftrightarrow x \leq x' \text{ and } z \geq z'$, which ensures monotonicity of the extended system.
• Show that the extended system is input/output monotone under this order, even when the original system is anti-monotone in output mapping.
• Apply positive-feedback results (e.g., convergence to equilibria, spectral radius conditions) to the extended system under unity feedback $u = y$.
• Restrict the analysis to the diagonal subspace $x = z$ to recover the original small-gain theorems for the negative feedback loop.
• Use linear system analysis (e.g., Hurwitz matrix conditions, spectral radius $\rho(K) < 1$) to derive explicit stability conditions for linear monotone systems.

Experimental results

Research questions

• RQ1 Can small-gain theorems for negative feedback loops in anti-monotone I/O systems be derived from results on positive feedback in a monotone extended system?
• RQ2 What is the role of the extended system's monotonicity in unifying the analysis of positive and negative feedback in monotone I/O systems?
• RQ3 How does the embedding of the original system into a higher-dimensional monotone system preserve stability properties under feedback?
• RQ4 What is the relationship between periodic behavior in the original system and multi-stability in the extended system?
• RQ5 Under what conditions do periodic orbits persist in delay systems when the small-gain condition fails?

Key findings

• Small-gain results for negative feedback in anti-monotone I/O systems are equivalent to stability results for a positive-feedback loop in an extended monotone system.
• The extended system is monotone under a non-standard order on $X \times X$, where the order on the second component is reversed, ensuring monotonicity of the feedback loop.
• For linear monotone systems, the small-gain condition $\rho(K) < 1$ is equivalent to the Hurwitz stability of both $A - BC$ and $A + BC$, which ensures stability of the closed-loop system.
• The spectral radius condition $\rho(K) < 1$ is equivalent to the global convergence of the iteration $u^+ = k(u)$, which corresponds to the fixed-point convergence of the feedback loop.
• If the extended system has multiple equilibria, then the original system may exhibit periodic behavior, establishing a link between multi-stability in the extension and oscillations in the original.
• For large delays in delay differential equations, pseudo-oscillations can emerge when the small-gain condition fails, suggesting the potential existence of true periodic orbits for small $\varepsilon > 0$ in singularly perturbed systems.