Yifei Mo

Yifei Mo

Professor of Materials Science and Engineering, University of Maryland.

FPBench: A New Benchmark for Evaluating Foundation Potentials

less than 1 minute read

Published:

We introduce FPBench, an application-oriented benchmark for evaluating machine-learned foundation potentials (FPs) — models that promise to replace expensive first-principles calculations across atomistic simulations.

While FPs often report near-DFT accuracy on average energy and force errors, we found that these averages frequently fail to predict how a model actually performs on the tasks that matter in practice: force prediction for atomistic simulations, energy ranking of substitutional and vacancy orderings, and ion/vacancy migration.

FPBench introduces error-decomposition metrics that resolve force errors among highly accurate, large-error, and far-from-equilibrium atoms; relative-energy errors among competing orderings, phases, and compositions; and endpoint and along-path errors in ion migration — pinpointing exactly where and why a model breaks down, and providing targeted guidance for improving foundation potentials.

We have released an open benchmark, evaluation code, and a public leaderboard for the community to assess and compare foundation potentials.

Read the paper on arXiv, view the code, or explore the leaderboard.

You May Also Enjoy

New Oxyhalide Electrolyte Breaks Barriers for Solid-State Battery Performance

1 minute read

Published:

A team led by Professor Yifei Mo, working with the University of Western Ontario and Oak Ridge National Laboratory, has developed a new class of oxyhalide solid electrolytes that combine fast lithium-ion transport with strong chemical stability — two properties that have historically been difficult to achieve in the same material.

Advanced Solid Electrolytes Break World Record for Ionic Conductivity

less than 1 minute read

Published:

A team led by Professor Yifei Mo, together with the Georgia Institute of Technology and Oak Ridge National Laboratory, has pushed the ionic conductivity of halide solid electrolytes to a new record by strategically tuning how anions move within the crystal lattice.

New, Superfast Method for Ceramic Manufacturing Could Open Door to AI-Driven Materials Discovery

1 minute read

Published:

Conventional ceramic sintering — the firing step that hardens ceramic materials — typically takes hours, a bottleneck for developing new solid-electrolyte and other functional ceramics. A UMD team led by Professor Liangbing Hu reinvented the process into an ultrafast high-temperature sintering method, sandwiching a pressed ceramic pellet between carbon strips that heat it by radiation and conduction to temperatures up to 3,000°C in under 10 seconds — over 1,000 times faster than a traditional furnace.

The Battery Revolution

less than 1 minute read

Published:

Fires in phones, laptops, and even a jumbo jet share a common cause: the flammable liquid electrolyte inside conventional lithium-ion batteries. Alongside their fire risk, these batteries also fall short on charging speed and energy density — driving researchers to look for a safer, better alternative.