Predictive qualification of capacitive deionization models: A framework for transferability and engineering deployment

Predictive modeling is increasingly important for the design, optimization, and scale-up of capacitive deionization (CDI), including membrane capacitive deionization (MCDI), yet model performance is commonly assessed from agreement with calibrated experiments rather than from demonstrated prediction across relevant operating, physical, or system boundaries. This review develops a predictive-qualification framework for evaluating CDI models according to the engineering claim being made. Physical representations and computational architectures are examined in terms of their governing assumptions, validation evidence, transfer limitations, and mechanisms of predictive failure. A hierarchical Q 0 –Q 3 framework distinguishes calibrated reproduction from prediction within a characterized system, transfer across physical or operational domains, and transfer across architectures or scales. The analysis shows that predictive reliability is governed not by model complexity alone, but by whether the required physical closures, parameterization, and computational resolution remain valid across the intended predictive boundary. On this basis, a predictive-gap-driven development strategy is proposed in which additional physics is introduced selectively in response to identified transfer failures and subsequently subjected to independent validation. The resulting framework connects model qualification with engineering design, cross-scale prediction, reduced-order deployment, and digital-twin development, providing a systematic pathway toward more transferable and decision-relevant CDI modeling.