Learning consistent molecular mechanics force fields from first principles
grappa-fullFF learns bonded and nonbonded force-field parameters together from ab initio data.
The paper says the model removes the need for externally assigned nonbonded parameters in practical simulations. It uses electrostatic-potential supervision and a charge-equilibration-friendly architecture to recover electric response properties. The authors report state-of-the-art accuracy on geometry optimization benchmarks and conformational sampling in line with classical and existing machine-learned force fields. Accepted to the ML4Molecules Workshop at NeurIPS 2026. ArXiv · AI/CL/LG's note
The paper says the model removes the need for externally assigned nonbonded parameters in practical simulations. It uses electrostatic-potential supervision and a charge-equilibration-friendly architecture to recover electric response properties. The authors report state-of-the-art accuracy on geometry optimization benchmarks and conformational sampling in line with classical and existing machine-learned force fields. Accepted to the ML4Molecules Workshop at NeurIPS 2026. ArXiv · AI/CL/LG's note
score 5