Input-to-State Stability Implications in Contraction Theory

Abstract: For nonlinear control systems on normed vector spaces, we characterize an incremental input-to-state stability (ISS) type property in which the overshoot constant multiplies both the initial-condition and the input terms. Working through the associated variational system, we show that two properties are equivalent: an ISS-type bound on the variational system, and the incremental ISS-type bound on the original system. We further establish the equivalence between an infinitesimal contraction condition, expressed through a Lyapunov-type function, and an incremental Lyapunov condition. Each of these equivalent conditions yields a necessary condition and a sufficient condition for the ISS-type bounds, differing only in the input Lipschitz constant of the vector field. When the overshoot constant equals one, the infinitesimal contraction condition reduces to the standard norm-based contraction conditions. We establish these implications under mere continuous differentiability of the vector field, and we illustrate the results through sensitivity matrices and Lyapunov characteristic exponents.

• The paper establishes a systematic equivalence chain linking infinitesimal contraction, incremental Lyapunov functions, variational ISS bounds, and incremental ISS under only continuous differentiability of the system vector field.
• The results show that the converse implication incurs only an overshoot-factor k loss in the input Lipschitz constant, while k = 1 yields full equivalence and standard logarithmic-norm contraction conditions.
• The paper proves global existence without separately assuming forward completeness and applies the theory to sensitivity bounds, non-exploding gradients in neural ODEs, and upper bounds on Lyapunov characteristic exponents.

This paper develops a systematic equivalence theory between contraction-type conditions and input-to-state stability (ISS) properties for nonlinear control systems $\dot x = f(t,x,u)$ on normed vector spaces, requiring only continuous differentiability of the vector field. The central result is a chain of necessary and sufficient conditions linking an infinitesimal contraction condition, an incremental Lyapunov condition, an ISS-type bound on the variational system, and an incremental ISS-type bound on the original system. These conditions differ, on either side of the equivalence, only in the admissible input Lipschitz constant of the vector field, with the overshoot factor $k$ mediating the gap.

Problem setting and variational machinery

The analysis concerns control systems on $\mathbb{R}^n$ with vector field $f$ satisfying only that $f$, $\mathcal{D}_x f$, and $\mathcal{D}_u f$ exist and are jointly continuous (Assumption 1). A key technical contribution is Theorem 1, which establishes that the solution $\delta x(t)$ of the variational system

$$\dot{\delta x} = \mathcal{D}_x f(t, \phi(t,t_0,x_0,u), u(t))\,\delta x + \mathcal{D}_u f(t, \phi(t,t_0,x_0,u), u(t))\,\delta u$$

coincides with the Gâteaux derivative of the flow with respect to simultaneous perturbations of the initial condition and input. Consequently, the state-transition matrix $\mathcal{D}_{x_0}\phi$ of the variational system exists and satisfies the usual semigroup identity under mere $k$0 regularity, relaxing the twice continuous differentiability assumed in much of the Finslerian literature. This relaxation is what permits the subsequent equivalences to hold under weak regularity.

A supporting lemma shows that an exponential growth bound $k$1 on the variational transition matrix, valid up to the forward escape time, implies global existence of solutions. This removes the need to assume forward completeness, an assumption commonly imposed in prior work on contraction equivalences.

Main equivalence theorem

The main result considers four properties parameterized by a growth rate $k$2, overshoot constant $k$3, and input Lipschitz constant $k$4:

1. an infinitesimal contraction condition: existence of a lower semicontinuous function $k$5 satisfying $k$6 with an upper Dini derivative bounded by $k$7 along the variational flow;
2. an incremental Lyapunov condition: an analogous function $k$8 bounding $k$9 along pairs of trajectories;
3. an ISS-type bound on the variational system: $\mathbb{R}^n$0;
4. an incremental ISS-type bound on the original system with the same structure relating $\mathbb{R}^n$1 to $\mathbb{R}^n$2 and $\mathbb{R}^n$3.

The established implications are:

• (infinitesimal contraction with $\mathbb{R}^n$4) $\mathbb{R}^n$5 (variational ISS bound) $\mathbb{R}^n$6 (incremental ISS bound) $\mathbb{R}^n$7 (infinitesimal contraction with $\mathbb{R}^n$8);
• infinitesimal contraction $\mathbb{R}^n$9 incremental Lyapunov condition.

The converse direction recovers the contraction condition only with the input gain inflated by the overshoot factor $f$0; this lossy step is the sole asymmetry in the chain. The authors state that this converse implication — incremental ISS implying an infinitesimal contraction condition with a quantified input gain bound — was not previously available in the literature. When $f$1, all four properties become fully equivalent, and property 1 reduces to the standard norm-based contraction condition $f$2 in terms of the logarithmic norm. For $f$3, the ISS-type bounds imply both incremental ISS and incremental integral ISS via standard $f$4 and $f$5 bounds on the convolution kernel; for $f$6 they quantify finite-time divergence rates. A byproduct is that any of the four properties implies global existence and uniqueness of solutions, and that boundedness of $f$7 is not required.

In the unforced case ($f$8), the four properties collapse to full equivalences for arbitrary norms, the same $f$9 and $f$0, and arbitrary overshoot constants — generalizing prior results that required $f$1, specific norms ($f$2, or $f$3/$f$4 for cooperative systems), time invariance, $f$5 regularity, or bounded Jacobians. The paper also derives, as a self-contained corollary, that a time-invariant system satisfying the exponential incremental estimate with $f$6 possesses a unique globally exponentially stable equilibrium, extending a known result from $f$7 to general $f$8.

Two caveats qualify the constructions. First, the Lyapunov functions $f$9 and $\mathcal{D}_x f$0 constructed in the converse direction are lower semicontinuous but not necessarily continuous; continuity can be restored only by restricting inputs to a compact set and accepting a slightly conservative rate $\mathcal{D}_x f$1 in the Dini inequality. Second, the equivalence holds for the specific ISS-type bound shape with the overshoot constant multiplying both the initial-condition and input terms; the paper does not claim equivalence for arbitrary ISS Lyapunov characterizations in the sense of Angeli's classical framework.

Applications

Two illustrations connect the equivalences to classical dynamical-systems notions. For the sensitivity matrix $\mathcal{D}_x f$2, the equivalence with $\mathcal{D}_x f$3 yields the uniform bound $\mathcal{D}_x f$4 whenever the logarithmic norm of $\mathcal{D}_x f$5 is bounded above by $\mathcal{D}_x f$6 — a result closely related to non-exploding gradient analyses in contractive neural ODEs. For Lyapunov characteristic exponents, the same bound gives $\mathcal{D}_x f$7 and, uniformly, $\mathcal{D}_x f$8, so that a finite logarithmic-norm supremum certifies non-positive maximal Lyapunov exponents. These applications are direct consequences of the equivalence theory rather than independent analyses, which is precisely their value: they show that sensitivity and exponent bounds can be certified by an infinitesimal condition on the vector field.

Limitations and open questions

The paper is explicit about the boundaries of its results. The converse implication from incremental ISS to contraction incurs the factor $\mathcal{D}_x f$9 on the input gain, and the paper does not determine whether this gap is an artifact of the proof technique or inherent. The constructed Lyapunov certificates are generally discontinuous, and continuity requires compact input ranges and rate slack. All results are stated on normed vector spaces; extensions to Riemannian or Finslerian geodesic metrics with the same weak regularity are not treated. Whether the equivalence chain extends to broader ISS Lyapunov characterizations (e.g., with nonlinear gain functions rather than linear convolution bounds) remains open.

Conclusion

The paper provides a complete, regularity-minimal characterization of incremental ISS-type behavior through contraction analysis: under mere $\mathcal{D}_u f$0 regularity, variational ISS bounds and incremental ISS bounds are equivalent, and both are characterized by infinitesimal contraction conditions up to a factor-$\mathcal{D}_u f$1 loss in the input gain. The equivalence for $\mathcal{D}_u f$2 recovers classical norm contraction, while the general-$\mathcal{D}_u f$3 results unify and strictly generalize a fragmented set of prior equivalences that each imposed stronger smoothness, norm-specific, or completeness assumptions.