Verification-Driven MLIP and DFT Workflows for Energy-Materials Diffusion

Overview: Seeking grant collaborators for a proposed research program on reliable machine-learning-assisted diffusion and defect-migration modelling in energy materials. No grant award or specific funding call is being claimed; the target program, budget, and timeline will be agreed with eligible partners. Existing Research Foundation: NEBWalk is an open-source, ASE-compatible NEB/CI-NEB framework with calculator-agnostic workflows, MLIP-assisted selection of barrier-sensitive images, and interfaces supporting Quantum ESPRESSO refinement. Its public research evidence includes 13 artifact-audited benchmark cases and matched-input consistency comparisons with ASE. These demonstrate defined workflow checks, not universal accuracy or industrial qualification. Scientific Challenge: MLIP-predicted migration barriers can be unreliable when model chemistry, configurations, or transition states fall outside validated domains. A low-cost pathway estimate is insufficient for materials qualification without independent reference calculations and documented numerical checks. Proposed Research:

1. Select representative diffusion and defect-migration problems relevant to batteries and related energy materials.
2. Benchmark MLIP-assisted NEB pathways against targeted DFT calculations using specified convergence criteria and controlled reference settings.
3. Evaluate endpoint ordering, barrier errors, model transferability, out-of-distribution indicators, and failure modes.
4. Develop transparent verification criteria, reproducible datasets, software tests, and decision-oriented reports. Partnership Opportunities: Seeking eligible academic or industrial co-applicants for joint grant development. Collaborators offering experimental expertise, reference data, or complementary computational capabilities are particularly welcome. Letters of support and subcontract arrangements may be considered where eligibility and institutional requirements permit. Research Contribution: I bring NEBWalk software and its documented public benchmark foundation, DFT/ASE/Quantum ESPRESSO experience, scientific Python engineering, and a validation-focused research approach. Expected Deliverables: Proposed outcomes include open-source workflow improvements (subject to partner IP terms), curated benchmark cases, numerical and uncertainty assessments, documented verification procedures, and scientific validation reports. The research scope, funding requirements, and implementation timeline will be defined jointly with interested collaborators. Public Research Evidence: NEBWalk: https://github.com/Rifat19R/NEBwalk ChemRxiv preprint: https://doi.org/10.26434/chemrxiv.15010372/v1