Prediction-sealed risk routing limits negative transfer in cross-workflow translation of two-dimensional material properties
Computational databases can assign different electronic properties to equivalent two-dimensional materials because their density-functional workflows differ. These inconsistencies can change screening decisions and can make flexible cross-database corrections fail after transfer. We present FidelitySwitch-2D, a prediction-sealed framework that combines structure matching, material-group-disjoint development, candidate-specific risk estimation, adaptive routing and group-level conformal triage. Development used 4,540 directed pairs from 1,056 material groups across C2DB, JARVIS-DFT and 2DMatPedia. In sealed external audits, the nonlinear Residual Bridge transferred negatively; on 490 independent MC2D groups, the frozen switch reduced Residual Bridge error by 0.0557 eV. In a target-sealed hybrid-perovskite audit, RiskArgminRouter achieved the lowest observed band-gap and conduction-band-minimum errors and consistently enriched low-error subsets across band gap, CBM and VBM. These results support FidelitySwitch-2D as an auditable reliability layer for workflow-dependent materials labels, while broader validation across properties, routes and experimental cohorts remains necessary.
Authors
- Amira Samy Talaat (ORCID: https://orcid.org/0000-0002-9413-3079)
Institutions
- Electronics Research Institute (EG)
Publication Details
- Journal
- Scientific Reports
- Published
- 2026-11-18
- DOI
- https://doi.org/10.1038/s41598-026-75789-2
- Primary Topic
- Machine Learning in Materials Science
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Science and Technology Development Fund