AI for Science · DIGA@SMU
We believe the next phase of AI will be decided not in the service sector but in industry and manufacturing — and that AI for science is the pilot. We build agents and models that read the literature, design alloys, and run simulations on their own, and we are now making every number they produce carry a machine-checked proof.
Our projects cover one loop of scientific work: gather what is known, predict and design, simulate, validate in the lab — and, finally, prove that each step of the computation is right.
NewOur latest progress · Sep 2026
Rewriting DFT, molecular dynamics and finite elements with formal methods.
Materials design, drug screening and structural safety assessment rely more and more on numbers produced by simulation. Today every step from equation to floating-point number is vouched for mainly by testing and experience. We want every step to carry a proof anyone can check — Lean 4 and Mathlib for the mathematics, TLA+ and model checking for the parallel software, and a refinement relation between the two.
“A result is only as trustworthy as the weakest link.”
Agents and machine learning frameworks that turn design targets into new alloys — with simulations in the loop and experiments as the final judge.
From ideation to experimental validation without hand-curated datasets: large language models, automated CALPHAD simulations, residual-learning correction, and AI-guided optimization.
A curated alloy database, multi-task property predictors with physics-informed inductive biases, and constraint-aware inverse design, joined by an iterative AI–experiment feedback loop.
LLM reasoning in every stage of a DFT calculation: a strategic planner, a just-in-time step planner, and a monitor–recover–reflect cycle that repairs failures and revises the plan.
An automated knowledge-driven framework for materials research.
MATAI Platform
An all-in-one platform for AI-driven materials design: a holistic alloy database, a foundational
property predictor, a generalist designer, and MATAI Chat.
MatVerse Paper Collection
A starting point for researchers new to AI for (materials) science: a regularly updated paper
collection, with the machine learning details of each paper.
We are always open to collaborations with materials scientists, physicists, and formal methods researchers, and to students who want to work on AI for science.