The tendency of solutions of a mathematical model or of an algorithm to remain bounded or to respond in a controlled way under small perturbations of input data, parameters, or initial conditions; in numerical analysis it often refers to bounded error propagation.

Principle

A stable system or method limits amplification of perturbations: small changes in data produce proportionally small changes in outcomes or in numerical errors, ensuring predictable behavior under uncertainty and discretization.

Demonstration

An explicit time-stepping scheme for a PDE is stable only under a Courant–Friedrichs–Lewy (CFL) condition; violating that condition produces growing numerical modes and blow-up, illustrating conditional numerical stability.

Misapplication

Confusing stability with accuracy (a stable method can be inaccurate) or assuming stability of a discretization without checking relevant norms and step-size conditions leads to misleading conclusions about reliability.

Consequence

Stability ensures that numerical computations and model predictions are robust to small input changes and round-off errors; combined with consistency it implies convergence of discretizations to the true solution.

Reversal

Instability describes unbounded growth or uncontrolled sensitivity to perturbations, where small errors amplify and render results meaningless without additional stabilization or reformulation.

Boundary

Applies to dynamical systems, inverse problems, and numerical algorithms; the meaning depends on the chosen norm, time horizon, and problem class (e.g., Lyapunov stability vs. numerical stability vs. conditional stability).

Semantic Tension

Tension exists between stability and responsiveness: highly stable controllers may respond sluggishly to legitimate inputs, while aggressive, responsive designs may be less stable under perturbations.

Synthesis

Stability is the property that bounds the effect of small perturbations on solutions or computations, providing a foundation for reliable prediction and ensuring that controlled growth of errors permits meaningful numerical and theoretical analysis.