Fitting potential energy surfaces with fundamental invariant neural network. II. Generating fundamental invariants for molecular systems with up to ten atoms
Problem
Neural-network fitting of permutationally invariant potential energy surfaces was limited to small molecules because generating fundamental invariants was computationally too slow.
Why it matters
Accurate potential energy surfaces are essential for simulating chemical reactions in larger molecules.
What the AI did
An improved scheme using computational invariant theory and parallel programming generates fundamental invariants fast enough to serve as neural-network inputs for high-dimensional PES fitting.
Results
Significantly reduces FI evaluation time, extending FI-NN PES fitting beyond five atoms (up to ten per title) with small fitting errors.
Impact
Enables accurate neural-network potential energy surfaces for larger molecular systems than previously practical.
Record
| Venue | The Journal of Chemical Physics |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Rongjun Chen, Kejie Shao, Bina Fu, Dong Hui Zhang |
| Citations | 103 (OpenAlex, at import time) |
| References | 46 |
| Model or method | fundamental invariant neural network (FI-NN) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1063/5.0010104 |
| Open access | closed |
| Keywords | neural networks, potential energy surface, fundamental invariants, permutation symmetry, molecular systems |
Importance rating
significant — thousands to millions affected
Accurate potential energy surfaces are essential for simulating chemical reactions; the method extends FI-NN PES fitting beyond five atoms, a genuinely new capability for a foundational chemistry field, though no application scale is given.