A MACE interatomic potential for two dimensional and curved sp2 carbon: family level transfer benchmarks and a SOAP based structural novelty map
COMPUTATIONAL MATERIALS SCIENCE, cilt.276, ss.927-929, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 276
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.commatsci.2026.115176
- Dergi Adı: COMPUTATIONAL MATERIALS SCIENCE
- Derginin Tarandığı İndeksler: Applied Science & Technology Source, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO), Scopus, Science Citation Index Expanded (SCI-EXPANDED), Chimica, Compendex, INSPEC
- Sayfa Sayıları: ss.927-929
- İstanbul Ticaret Üniversitesi Adresli: Evet
Özet
Graphene-derived allotropes, nanotubes, fullerenes and triply periodic schwarzites are costly to screen with density functional theory when their cells are large or have low symmetry. We trained a carbon only MACE potential on a consistent GPAW/PBE dataset comprising 21,161 configurations with 1.16 million atoms of planar, defective, amorphous, and curved sp carbon structures. On the held out test set, the model achieves mean absolute errors of 2.83 meV/atom for energies and 33.9 meV/Å for force components. It reproduces the near equilibrium lattice constants of 50 two dimensional allotropes with mean absolute errors of 2.5 and 2.7 mÅ along the two lattice axes. Transfer tests on sp 2 families absent from the training database yield energy mean absolute errors of 5.4 meV/atom for egg-tray graphenes and 3.9 meV/atom for azulenoid kekulene allotropes. Of the 39 schwarzites absent from training, 31 pass the post-relaxation structural screening; their median maximum lattice parameter error is 1.1% relative to the VASP/PBE + D3(BJ) reference cells. The nanotube bending modulus (1.525 eV) is close to the PBE reference of approximately 1.49 eV, whereas the excess energies of strongly curved small fullerenes are overestimated (by approximately 30–44% for C₆₀), which marks the high curvature limit of the model. The nearest neighbor distance in structure averaged SOAP space provides a low cost, first pass indicator of structural novelty, but it should not be interpreted as a calibrated uncertainty measure or formal applicability boundary.