A problem that violates one or more of the usual well-posedness criteria (existence, uniqueness, continuous dependence of solutions on data), so that solutions may not exist, may not be unique, or may change discontinuously under arbitrarily small perturbations of the input data.

Principle

Well-posedness is required for physical interpretability and stable numerical computation; ill-posedness indicates the inverse or direct mapping lacks continuity or injectivity/surjectivity in the chosen function spaces and typically necessitates reformulation or regularization.

Demonstration

Inverse heat conduction (backward heat equation) is classic: given final-time temperature, recovering earlier states is exponentially ill-posed—small high-frequency perturbations in data grow rapidly backward in time. Analytic continuation of a function from partial data is another ill-posed example.

Misapplication

Equating ill-posedness with mere computational difficulty or slow convergence, or failing to distinguish ill-posedness from ill-conditioning (a problem can be well-posed but ill-conditioned numerically), or applying regularization without verifying appropriate modelling assumptions.

Consequence

One must restrict data spaces, impose priors or regularization (Tikhonov, spectral cutoff, Bayesian priors), or reformulate as a well-posed problem (e.g. filtering, introducing physically motivated dissipation) to obtain stable, interpretable solutions.

Reversal

Well-posed problem: existence, uniqueness and continuous dependence hold in the chosen topology; this supports reliable inference and stable numerical approximation.

Boundary

Well- or ill-posedness is relative to the choice of function spaces, norms and topology; discretization may change the character (finite-dimensional truncation may render a problem numerically solvable), and stochastic formulations can recast ill-posedness as high uncertainty rather than impossibility.

Semantic Tension

Tension between ill-posed and ill-conditioned: the former is a mathematical failure of the inverse/direct operator properties, the latter a numerical amplification of perturbations; both affect computation but require different remedies.

Synthesis

An ill-posed problem lacks at least one of existence, uniqueness, or continuous dependence in the chosen mathematical setting; recognizing which criterion fails guides the choice of regularization, modeling restrictions or alternative formulations to restore useful predictive or reconstructive power.