Meijie Wang
AI4S Algorithm Intern, Deep Principle
Ph.D. student in Condensed Matter Physics at Xiamen University, currently working on AI for Science research at Deep Principle. My recent work focuses on the materials property-prediction foundation model (MPA), where I develop large-scale training and evaluation infrastructure, design the mid-/post-training pipelines, and support distributed model training, benchmarking, and iteration.
My background combines scientific computing and machine learning for materials science. Previously, I worked on structure–electronic-structure–activity relationships in single-atom and dual-atom catalytic systems using DFT and ML methods (Appl. Surf. Sci. 2024, J. Mater. Chem. A 2024, J. Phys. Chem. Lett. 2026, ACS Catal. 2026).
Education
Experience
Deep Principle
AI4S Algorithm Intern
Jan 2026 - Present
- MPA materials foundation model: brought LLM-style multi-phase training (pre-/mid-/post-training) to experimental property prediction — SOTA on 35/40 tasks and 14.6% lower MAE than direct fine-tuning under scaffold (OOD) splits, surpassing Uni-Mol2, Suiren, and ChemProp.
- Implemented the full mid-training (physics-guided alignment on large-scale first-principles data) and post-training pipelines, including a Hybrid Readout head (attention-pooling + atom-additive); built the training/evaluation infrastructure and ran all large-scale training.
Xiamen University
Ph.D. Candidate
Sep 2022 - Present
- Dual-atom catalyst design & mechanism: designed Si-based dual-atom catalysts for CO₂ reduction (Appl. Surf. Sci. 2024); uncovered the p–d orbital-coupling mechanism and screened 360+ candidates with a GBR pipeline (J. Mater. Chem. A 2024).
- Curvature-driven catalysis: established curvature as an independent activity knob (inverted-volcano relation, interpretable descriptor; J. Phys. Chem. Lett. 2026) and generalized it into a unified geometric-electronic principle (ACS Catal. 2026).
Selected Publications
Materials Property Axiom: Scaling Foundation Models to Experimental Property Generalists via Multi-phase Training
Deep Principle Team
Technical Report · Deep Principle (2026) · DOI
A geometric-electronic principle for curvature-driven catalysis
Meijie Wang, Yuxing Lin, Zhulin Huang, Yang Sun, Zi-zhong Zhu, Shunqing Wu, Xinrui Cao
ACS Catal., ASAP (2026) · DOI
Curvature Engineering of SiFe Dual-Atom Catalysts for Enhanced CO₂ Electroreduction
Meijie Wang, Yuxing Lin, Yaowei Xiang, Yang Sun, Zi-zhong Zhu, Shunqing Wu, Xinrui Cao
J. Phys. Chem. Lett., 17, 1227-1234 (2026) · DOI
p-d Orbital coupling in silicon-based dual-atom catalysts for enhanced CO₂ reduction
Meijie Wang, Yaowei Xiang, Yuxing Lin, Yang Sun, Zi-zhong Zhu, Shunqing Wu, Xinrui Cao
J. Mater. Chem. A, 12, 31902-31913 (2024) · DOI
SiFeN₆-graphene: A Promising Dual-Atom Catalyst for Enhanced CO₂-to-CH₄ Conversion
Meijie Wang, Yaowei Xiang, Wengeng Chen, Shunqing Wu, Zi-Zhong Zhu, Xinrui Cao
Appl. Surf. Sci., 643, 158724 (2024) · DOI
Handbook
A structured handbook for new group members, covering the most-used parts of Linux, research tooling, and DFT workflows.