Direct discovery · Materials science · Chemistry · Physics
SparksMatter proposes CaMg₂Si₂ thermoelectric candidates through an autonomous in-silico loop
Six CaMg₂Si₂ crystal candidates that passed surrogate stability screening, with three polymorphs receiving post hoc lattice-dynamics analysis.
Summary
A peer-reviewed npj Computational Materials paper reports that SparksMatter combined specialized language-model agents with Materials Project retrieval, MatterGen, MatterSim, and CGCNN to propose and screen CaMg₂Si₂ as an earth-abundant thermoelectric candidate. Six structures passed surrogate stability screening, and follow-up lattice-dynamics calculations found three polymorphs dynamically stable at 0 K and estimated room-temperature lattice thermal conductivity near 6.2 W m-1 K-1.
AI role
Coordinated agents to form a materials hypothesis, generate Ca-Mg-Si structures with MatterGen, execute stability and property screening, critique results, and plan validation.
Narrative role
This adds a concrete example of an agentic system organizing multiple materials-AI tools into an adaptive in-silico discovery loop and producing inspectable candidate structures rather than only retrieving known compounds or answering a benchmark.
Caveat
No candidate was synthesized or experimentally tested. Stability, electronic properties, thermal transport, synthesizability, and thermoelectric performance remain computational predictions, and the paper's model comparison used an LLM evaluator rather than independent domain experts.