Numerical stability determines whether a finite element simulation produces physically meaningful results as discretization parameters change. The Lax equivalence theorem formulated by Peter Lax, Courant Institute of Mathematical Sciences New York University, formalizes that consistency together with stability yields convergence. In finite element methods this principle ties the choice of elements, the algebraic solvers, and time-integration schemes to the reliability of computed solutions. Stability is not an abstract property; it is the gatekeeper between a model that approximates reality and one that produces artefacts.
Stability in spatial discretization
Spatial stability depends on element shape, interpolation order, and the differential operator. Classic finite element analysis by Olek C. Zienkiewicz, Swansea University, and the error estimates developed by Gilbert Strang, Massachusetts Institute of Technology, show that badly shaped elements and extreme aspect ratios generate ill-conditioned stiffness matrices that amplify rounding errors. Conditioning of linear systems is central because iterative solvers propagate and sometimes magnify numerical errors. Gene H. Golub, Stanford University, established much of the modern theory of numerical linear algebra that practitioners use to assess conditioning. Practical consequences include spurious modes and locking phenomena in nearly incompressible elasticity where the discrete solution fails to represent the continuous field, producing misleading stress concentrations that can affect design or safety assessments.
Stability in time integration
Temporal stability controls the interaction between the time-stepping algorithm and spatial discretization. The Courant-Friedrichs-Lewy condition originally analyzed by Richard Courant and colleagues at the Courant Institute of Mathematical Sciences New York University remains the guiding constraint for explicit schemes: time steps must respect wave propagation speeds and mesh resolution. Implicit methods relax strict time-step limits but require robust linear solvers and preconditioners to prevent instability or nonphysical energy growth. Thomas J.R. Hughes, University of Texas at Austin, developed stabilized formulations and analyses showing how techniques such as variational multiscale and streamline upwind methods mitigate convective instabilities in advection-dominated problems. Choosing an integrator without regard for its stability region invites oscillatory errors or blow-up in transient simulations.
Causes of instability are often mixed: high contrast material coefficients, nonmatching boundary conditions, nonlinearities, and solver tolerances combine to produce degraded behavior. The environmental and territorial stakes can be real. In coastal flood modeling, unstable discretizations can misplace inundation extents, influencing evacuation planning and infrastructure siting. In seismic simulations, spurious high-frequency energy alters predicted ground motions that inform building codes. These are human consequences tied to numerical choices.
Remedies are well established in the literature. Mesh quality control, h and p refinement guided by error estimators, and stabilized finite element formulations address spatial instabilities; preconditioning and robust iterative methods grounded in the work of Gene H. Golub reduce solver-induced amplification; time integration must be chosen with its stability region in mind. Ultimately, numerical stability is a design criterion as important as the physical model, because it determines whether computational predictions can be trusted for engineering, environmental, and policy decisions.