Matthew Francis Dixon
Adaptive finite elements refine meshes or increase polynomial degree, but coherent non-polynomial structure may remain expensive in either coordinate. We introduce variational enrichment, in which compact problem-informed functions complement ordinary h- and p-refinement. The framework applies to Galerkin and minimum-residual formulations whenever admissible functions and their residual value can be evaluated; it is not tied to one PDE class. Quotient-Schur analysis removes redundant additions and measures their variational value. For symmetric coercive problems, the present theory also gives an exact energy-error reduction and a reference-independent upper bound. Across heterogeneous media, stochastic coefficients, irregular domains, singularities, interfaces, transport, and oscillatory tests, hc helps when a compact function captures unresolved structure aligned with the PDE operator; it offers little benefit when that structure is redundant or mismatched. On SPE10 Models 1 and 2, coefficient-adapted functions improve polynomial enrichment in all eleven prescribed cases at equal dimension, by up to $41.6\%$. The wider positive and negative evidence supports the same conclusion: hc selects useful operator-compatible structure and otherwise falls back to ordinary hp refinement.