Weighted Phase Volume Method in Stability Analysis: Integral Criteria and Ellipsoidal Reachable Sets

Technical Summary: Weighted Phase Volume Method in Stability Analysis

Problem Statement
The paper addresses the challenge of analyzing the stability of dynamical systems, particularly complex, nonlinear, nonautonomous, or discontinuous systems where constructing classical Lyapunov functions remains difficult. While existing geometric approaches based on Liouville's theorem and the Reynolds transport theorem link stability to the divergence of vector fields, they often lack constructive geometric estimates of reachable sets or are limited by specific assumptions about the vector field's structure. The authors propose a method to overcome these limitations by introducing a weighted phase volume and a time rescaling mechanism that allows for the manipulation of dissipativity properties without altering the system's topological phase portrait.

Methodology
The core of the proposed method involves transforming the original autonomous system $\dot{x} = F(x)$ into an auxiliary "scaled" system $\frac{dx}{d\tau} = G(x) = \rho(x)F(x)$. This transformation relies on two key components:

1. Matrix-Valued Scaling Function ($\rho(x)$): A smooth matrix function satisfying collinearity and positivity conditions ($F \wedge \rho F = 0$ and $\langle F, \rho F \rangle > 0$). This ensures that the vector field $G(x)$ is a positive scalar multiple of $F(x)$ at every point, preserving the trajectories and the topology of the phase portrait.
2. Time Rescaling: The transformation induces a time change $d\tau = \mu(x)dt$, where $\mu(x) = \langle F, \rho F \rangle / \|F\|^2$.

The authors introduce a weighting function $W(x, t)$ and analyze the evolution of the integrals of $W$ and $1/W$ over a moving domain $\Omega_t$ in the scaled time $\tau$. By applying the Reynolds transport theorem and Liouville's theorem to this scaled system, they derive integral identities that relate the evolution of these weighted volumes to the divergence of the scaled field and the derivative of the weighting function along trajectories.

Key Contributions and Results

• New Integral Identities and Dissipativity Conditions: The paper derives new integral identities (Theorem 3) that generalize Liouville's theorem and the Reynolds transport theorem. These identities depend on arbitrary weighting functions and time scaling. Based on these, the authors formulate "paired $\rho$-dissipativity conditions" for both $W$ and $1/W$ (Section 5). Unlike classical dissipativity which uses a fixed storage function, these conditions are parameterized by the free functions $\rho$ and $W$, significantly expanding the class of analyzable systems.
• New Stability Definitions: The authors introduce definitions for integral stability, integral asymptotic stability, and integral exponential stability with respect to the pair $(\mu, W)$ (Definitions 3–5). These definitions explicitly incorporate the scaling factor $\mu(x)$ and allow for two-sided estimates via $W$ and $1/W$.
• Ellipsoidal Reachable Sets: A major contribution is the development of a method to construct geometric estimates of reachable sets using ellipsoids.
  • Covering Ellipsoid: Under the first dissipativity condition, the authors derive an evolution equation for a covering ellipsoid that is guaranteed to contain the evolving domain $\Omega_t$ (Theorem 8).
  • Inner Ellipsoid: Under the second dissipativity condition, they derive an equation for an inner ellipsoid that is guaranteed to be free of points from $\Omega_t$ (Theorem 9).
  • The simultaneous use of these ellipsoids provides a complete geometric picture of the system's evolution as an "ellipsoidal annulus." Crucially, this approach reduces to integrating differential equations for the center and shape matrix, avoiding the need to solve optimization problems at each step, which is common in linear matrix inequality (LMI) approaches.
• Connection to Lyapunov Stability: The paper establishes a theoretical link between the proposed integral stability and classical Lyapunov stability (Theorem 10). It demonstrates that the existence of a contracting family of covering ellipsoids implies asymptotic (or exponential) Lyapunov stability. This generalizes previous results that only established such connections for specific system classes without geometric estimates.

Significance and Claims
The paper claims that the proposed method offers a constructive alternative to classical Lyapunov analysis, particularly for systems where finding a Lyapunov function is intractable. By allowing the user to choose the weighting function $W$ and the scaling matrix $\rho$, the method provides flexibility to "tune" the dissipativity properties of the system to satisfy stability criteria.

The authors highlight that their approach:

1. Preserves the topology of the phase portrait while altering the time parametrization, unlike methods that modify the vector field itself.
2. Provides analytical stability guarantees and constructive geometric estimates (ellipsoidal bounds) for reachable sets, which are absent in many divergence-based stability analyses.
3. Demonstrates effectiveness in numerical examples where standard divergence methods (with $\rho = E$) fail to establish stability, but a suitable choice of $\rho(x)$ succeeds.
4. Offers a framework for control synthesis and robustness analysis by allowing structural constraints (such as sparsity or passivity) to be imposed directly on the matrix $\rho$.

The work is presented as a generalization of existing divergence and density methods, unifying them under a weighted phase volume framework that yields new integral criteria and geometric approximations for dynamical systems.