Computational materials design: first-principles calculations, machine-learning interatomic potentials, interpretable descriptors, and LLM-agent workflows.
Founded and led by Jiahao Xie.
| Line | What we build |
|---|---|
| Interpretable descriptors | Physically constrained symbolic regression (SISSO++) with nested out-of-sample validation |
| ML potentials for hybrid perovskites | Fine-tuned interatomic potentials for molecule–perovskite interfaces and structure optimisation |
| Excited states in metal halides | Automated computational discovery of self-trapped-exciton emitters in zero-dimensional metal halides |
| Knowledge engineering | A literature-grounded knowledge base and ontology for materials design, with LLM-assisted screening |
| AI for physics | DerivationLab — an LLM-agent system for step-level physics derivations |
| Repository | What it does | Licence |
|---|---|---|
| elf-anisotropy-sisso | Command-line ELF anisotropy descriptor from a VASP ELFCAR, plus the SISSO++ candidate-search and nested leave-one-material-out validation workflow for 2D Sn/Pb iodide perovskites |
MIT |
| derivationlab | DerivationLab: hash-chained records of machine-checked physics derivations — the derivation runtime, service and web UI, plus a stdlib-only offline verifier and viewer with example records | Apache-2.0 |
| materials-discovery-skills | matdisc: an agent-drivable workflow for generative structure design, MLIP prescreening and DFT validation of inorganic semiconductors, packaged as seven agent skills (from arXiv:2606.10251) |
MIT |
| perovskite-adsorption-sampler | Adsorbate-placement samplers for Pb–I terminated perovskite surfaces on top of FAIRChem, with a resumable MLP-to-DFT screening workflow and a CPU-only example | MIT |
Jiahao Xie · Google Scholar · [email protected]