Numerical Modeling of Solid–Liquid Phase‐Change Materials: A Critical Review of Formulations, Validation, and Application‐Oriented Model Selection

ABSTRACT Numerical models are central to the design of phase‐change material (PCM) thermal systems, yet their credibility is more often asserted than demonstrated. Although reviews of PCM materials and applications are abundant, comparatively little attention has been given to how modeling assumptions, parameter choices, and validation practices determine predictive reliability. This review examines numerical models of solid–liquidphase change from three perspectives: benchmarking, parameter sensitivity, and application‐oriented model selection. Enthalpy‐based, enthalpy–porosity, effective heat‐capacity, simplified conduction, sharp‐interface, reduced‐order, and data‐driven approaches are assessed in terms of physical fidelity, computational cost, treatment of natural convection, interface representation, parameter sensitivity, and validation evidence. Particular attention is given to the mushy‐zone constant, which can alter melting time, flow structure, and phase‐front position while leaving temperature predictions apparently acceptable. Representative benchmarks involving cavity melting, cylindrical tubes and capsules, optical imaging, and X‐ray computed tomography are used to clarify adequate validation evidence and explain why temperature‐only comparison is insufficient. Application‐specific validation requirements and modeling risks are synthesized for building envelopes, battery and electronics thermal management, latent‐heat thermal‐energy‐storage units, solar systems, and reduced‐order or AI‐based frameworks. The analysis shows that no model is universally optimal: credible PCM simulation requires matching model complexity to the dominant physics, available validation observables, and intended engineering decision. Research priorities include standardized open benchmark data sets, systematic reporting of parameter sensitivity, uncertainty‐aware calibration, and hybrid physics–data‐driven models.