{"contributor-id" "729049db-685a-43b1-97a8-617daa2586ba"
 "description" "A general purpose parallel NequIP equivariant graph neural network (GNN) interatomic potential for Si. The model is trained on the GAP Si PRX (Bartók et al., Phys. Rev. X, 8:041048, 2018) dataset consisting of 2475 configurations, inluding bulk diamond, beta-Sn, hexagonal, bcc, fcc, hcp, liquid, and amorphous silicon configurations, as well as diamond surfaces, vacancies, divacancy, and interstitial faults, and additional rare configurations includingsp2 and sp bonded Si. Given the wide variety of configurations this model was trained on, it is suitable for simulating diverse Si structures. The model has a cutoff radius of 4 angstroms, with the following hyperparameters: maximum order of spherical harmonics set to 1, size of intermediate features set to 64, and the number of graph convolutions set to 3. The model was trained until the error did not improve for 50 validation steps, and an adaptive learning rate < 10^-6 was achieved."
 "developer" ["729049db-685a-43b1-97a8-617daa2586ba"
              "360c0aed-48ce-45f6-ba13-337f12a531e8"
              "e1b97ecf-68df-423a-97de-e11a40dc4dde"]
 "doi" "10.25950/d7a965ba"
 "domain" "openkim.org"
 "executables" []
 "extended-id" "TorchML_NequIP_GuptaTadmorMartiniani_2024_Si__MO_196181738937_000"
 "kim-api-version" "2.3"
 "maintainer-id" "729049db-685a-43b1-97a8-617daa2586ba"
 "model-driver" "TorchML__MD_173118614730_000"
 "potential-type" "nequip"
 "publication-year" "2024"
 "source-citations" [{"author" "Bart{\\'o}k, Albert P and Kermode, James and Bernstein, Noam and Cs{\\'a}nyi, G{\\'a}bor"
                      "doi" "10.1103/PhysRevX.8.041048"
                      "issue" "4"
                      "journal" "Physical Review X"
                      "pages" "041048"
                      "publisher" "American Physical Society"
                      "recordkey" "MO_196181738937_000a"
                      "recordtype" "article"
                      "title" "Machine learning a general-purpose interatomic potential for silicon"
                      "volume" "8"
                      "year" "2018"}]
 "species" ["Si"]
 "title" "Parallel NequIP Equivariant GNN for Si developed by Gupta et al. (2024) v000"}