--- title: Fixed-Time Stable Systems url: https://www.emergentmind.com/topics/fixed-time-stable-system type: topic --- # Fixed-Time Stable Systems A fixed-time stable system is a nonlinear, linear, or stochastic dynamical system whose solutions are rigorously guaranteed to converge to an equilibrium or invariant set within a time bound that is uniform over all admissible initial conditions. The defining feature is that the convergence (settling) time is independent of the initial state, and this property is certified by Lyapunov-based inequalities involving multiple power terms in the state or a related Lyapunov function. This concept underpins the synthesis of controllers, observers, algorithms, and learning dynamics that guarantee deterministic or probabilistic convergence to target sets in a time explicitly specified by parameters, even in the presence of exogenous disturbances or stochastic noise. ## 1. Formal Definition and Core Concepts A system $\dot{x}=f(x)$ (continuous-time) or $y(k+1)=F(y(k))$ (discrete-time) is **fixed-time stable (FxTS)** if it is Lyapunov stable and there exists a uniform upper bound $T^*<\infty$ such that all trajectories starting from admissible initial conditions reach the equilibrium within $T^*$ and remain there for all future times. For stochastic systems, fixed-time stability in probability requires an analogous bound holding almost surely or in expectation, independent of the initial condition [2202.11225, 2403.20258]. Specifically, for discrete-time autonomous systems, the fixed-time stability of the equilibrium $y=0$ is characterized by the existence of a settling-time function $K(y_0)$ and $K_{\max}$ such that for all $y_0$ in the domain, $y(k)=0$ for all $k\ge K(y_0)$ and $K(y_0)\le K_{\max}$; $K_{\max}$ does not depend on $y_0$ [2202.11225]. In continuous time, the standard Lyapunov condition asserts that, for a positive definite, radially unbounded $V$, $$ \dot V(x) \leq -aV(x)^p - bV(x)^q \quad\text{with }0 < p < 1 < q,\;a,b>0, $$ implies fixed-time stability with $$ T(x_0) \leq \frac{1}{a(1-p)} + \frac{1}{b(q-1)} =: T_{\max}. $$ This bound is uniform and independent of initial $x_0$ [2411.09118]. ## 2. Lyapunov-Based Synthesis and Certification Lyapunov conditions play a central role in certifying fixed-time stability. For deterministic discrete-time systems, the fixed-time condition is: $$ \Delta V(y(k+1)) := V(y(k+1)) - V(y(k)) \leq -\alpha \min\left\{\frac{V(y(k))}{\alpha}, \max\left\{V(y(k))^{r_1}, V(y(k))^{r_2}\right\}\right\} $$ for suitable $0<\alpha<1$, $00 $$ is enforced almost surely or in expectation; under suitable martingale and observer error control, this yields a uniform fixed-time bound $T^* = 1/(a(1-\alpha)) + 1/(b(\beta-1))$ for the probability that the process enters the target set within $T^*$ [2403.20258]. ## 3. Robustness to Perturbations and Stochasticity Fixed-time stability can be extended to systems with deterministic perturbations and stochastic noise. For perturbed deterministic systems $y(k+1) = F(y(k)) + g(k, y(k))$, with $|g(k, y)| \leq \delta_0$, the Lyapunov descent is modified, and fixed-time *attractiveness* into an invariant ball is guaranteed, with explicit bounds both on the ball radius and the maximal time to enter this ball [2202.11225]. Under suitable parameter restrictions, solutions enter a computed invariant set within a number of steps independent of the initial condition. For stochastic discrete-time systems $y(k+1) = f(y(k)) + g(y(k))\nu(k)$ with i.i.d. zero-mean noise, a similar Lyapunov function $V$ can be employed with a generator/expectation-based condition: $$ \mathbb E[V(F(y, \nu))] \leq \gamma(V(y)), $$ where $\gamma(\cdot)$ has a Polyakov-type structure. This ensures that the mean hitting time (expected settling time) is uniformly bounded [2202.11225]; recent advances further provide high-probability (risk-aware) bounds with explicit probability control [2403.20258]. ## 4. Optimality, Prescribed-Time and Design Flexibility Traditionally, the settling-time bound for fixed-time systems depended only on system parameters and not on initial state, but was often highly conservative. Recent work derives exact least upper bounds for settling time by completing the integral or recursion calculations, avoiding unnecessary conservatism [1809.07012]. Moreover, *predefined-time* or *prescribed-time* stability strengthens fixed-time stability by allowing the designer to set the uniform convergence time $T_c$ arbitrarily via gain rescaling of the system, achieving $\sup_{x_0}T(x_0)\leq T_c$ [1809.07012, 1901.02782, 1910.14604]. Most standard fixed-time systems can be algorithmically transformed to this strong form by explicit gain design or time-scale modifications, including bounded time-varying gains [2001.06707]. The general framework is now unified: arbitrary classes of continuous- or discrete-time systems can be constructed to be fixed-time or predefined-time stable by leveraging time-scale transformations, composite Lyapunov inequalities, or operator-theoretic splitting, directly controlling and optimizing convergence rate and maximum allowable time [2407.08139, 1910.14604]. ## 5. Extensions: Discrete-Time, Composite, and Large-Scale Systems The fixed-time stability paradigm is applicable across a diverse set of contexts: - Discrete-time nonlinear and stochastic systems with explicit Lyapunov-based and comparison-lemma characterizations [2202.11225]. - Composite and interconnected systems, including singularly perturbed multi-timescale dynamics, where composite Lyapunov functions and quadratic forms are used to certify global fixed-time stability provided subsystem interaction terms are controlled; the overall settling time is then expressed in terms of subsystem and interaction coefficients [2408.16905]. - Distributed, learning, and optimization schemes, e.g., forward-backward splitting for monotone inclusions, and proximal-type algorithms for variational inequalities and generalized equations, where fixed-time stability translates into a uniform iteration bound in discrete-time schemes [2407.08139, 2601.08700]. The fixed-time property is also preserved under suitable explicit Euler time-discretization, as long as the step size is sufficiently small and Lipschitz constants are controlled [2507.00531, 2407.08139, 2409.11713]. ## 6. Control, Estimation, Learning, and Safety Applications Fixed-time stability has become a key paradigm in modern control, estimation, and learning, enabling rigorous guarantees on convergence deadlines: - Static and dynamic feedback controllers for nonlinear and linear plants that drive the state to equilibrium exactly within user-specified time even in the presence of input delays, exogenous perturbations, and measurement noise [2307.01621, 2202.07717]. - Adaptive control laws and parameter estimators for safety-critical control under parametric uncertainty, guaranteeing convergence of estimates in uniform fixed-time and integration within robust control barrier function (RaCBF) frameworks [2011.12362]. - Neural ODEs and deep learning architectures endowed with fixed-time convergence to target predictions, providing robustness against input perturbations and precise control over learning dynamics [2411.09118]. - Safe control via nonovershooting fixed-time safety filters enforcing state or output constraints with fixed restraint times, applicable to chain-of-integrator, nonlinear, or input-output linearizable systems [2202.07717]. - Large-scale problems in convex optimization, traffic assignment, and mixed variational inequalities, where fixed-time stable primal-dual or proximal-type algorithms yield practical iteration bounds independent of initialization [2503.04590, 2601.08700, 1908.03517]. ## 7. Representative Theorems and Settling-Time Benchmarks The following table synthesizes key fixed-time results and their explicit uniform bounds: | System (Setting) | Lyapunov / Drift Inequality | Uniform Bound Expression | Reference | |---------------------------------------------|--------------------------------------|--------------------------------------------|-----------------------| | CT ODE: $\dot{x}=f(x)$ | $\dot{V}\le -aV^p - bV^q$ | $T_{\max} = 1/(a(1-p)) + 1/(b(q-1))$ | [2411.09118] | | DT: $y(k+1)=F(y(k))$ | $\Delta V \le -\alpha\min\{\cdots\}$ | $K_{\max}$ explicit in $(\alpha, r_1,r_2)$ | [2202.11225] | | Stoch. DT: $y(k+1)=f(y)+g(y)\nu(k)$ | $\mathbb{E}[V(\cdot)] \le \gamma(V)$ | $\mathbb{E}[K]\le K_{\max}$ | [2202.11225, 2403.20258]| | Prescribed-time CT: | see [1910.14604], [2001.06707] | Arbitrary $T_c$ through gain design | [1809.07012, 1910.14604]| | Forward-Backward Splitting (FBF) / Proximal | $\dot{V}\le -p_1V^{a_1}-p_2V^{a_2}$ | $T^* = 1/(p_1(1-a_1))+1/(p_2(a_2-1))$ | [2407.08139] | All explicit claims regarding Lyapunov drift, settling time, and extension to stochastic or perturbed systems are directly reflected in the theorems and constructions found in these sources. --- **References** - "Deterministic and Stochastic Fixed-time Stability of Discrete-time Autonomous Systems" [2202.11225] - "Enhancing the settling time estimation of a class of fixed-time stable systems" [1809.07012] - "Risk-Aware Fixed-Time Stabilization of Stochastic Systems under Measurement Uncertainty" [2403.20258] - "FxTS-Net: Fixed-Time Stable Learning Framework for Neural ODEs" [2411.09118] - "Fixed-time Stabilization with a Prescribed Constant Settling Time by Static Feedback for Delay-Free and Input Delay Systems" [2307.01621] - "Consistent discretization of finite/fixed-time controllers" [2207.03235] - "A Lyapunov-like Characterization of Predefined-Time Stability" [1910.14604] - Additional references as cited throughout the article. Source: https://www.emergentmind.com/topics/fixed-time-stable-system