GAP interatomic potential for C60

This is a <strong>Gaussian approximation potential</strong> (GAP [1]) for <strong>carbon</strong>. The potential can be used to model <strong>general</strong> carbon systems, but is particularly geared towards modeling systems containing<strong> C₆₀ molecules and graphitic carbons</strong>. It has been fitted with <strong>QUIP/GAP</strong> [1,2] by recomputing the a-C database of Deringer and Csányi [3] at the PBE level of theory [4] using the VASP code [5,6]. The a-C database from [3] has been augmented significantly to include other atomic structures, namely: Dimers at closer interatomic separation than in [3] Trimers Graphite and graphene (including bilayer graphene) throughout the entire exfoliation curve at different levels of biaxial strain Glassy carbon Isolated strained and distorted C₆₀ molecules, interacting C₆₀ dimers and colliding C₆₀ dimers Different surface reconstructions of diamond The main technical novelty introduced in this potential is <strong>van der Waals</strong> (vdW) <strong>corrections</strong> at different levels. It incorporates vdW corrections at the Tkatchenko-Scheffler (TS) level of theory [7] via a <strong>new machine learning based local parametrization of dispersion interactions</strong>. Environment independent ("fixed C₆") vdW corrections can also be used (mostly for banchmarking, since the new TS approach is superior). Check the comments in the gap_files/carbon.gap file for details. For the underlying PBE fit, this potential uses 2-body (distance_2b) and 3-body (angle_3b) descriptors [3] plus SOAP-type descriptors (soap_turbo) [8,9], as implemented in the <strong>TurboGAP</strong> code [10]. For the Hirshfeld volume fit, the potential reuses the soap_turbo descriptors for computational efficiency. The files can be used both with QUIP/GAP (compiled with the TurboGAP libraries) and TurboGAP. The authors acknowledge funding from the Academy of Finland (grants 321713, 329483 and 330488) and computational resources from the Finnish Center for Scientific Computing (CSC) and Aalto University's Science IT project. More details can be found in this paper: <strong>Machine learning force fields based on local parametrization of dispersion interactions: Application to the phase diagram of C₆₀</strong><br> Heikki Muhli, Xi Chen, Albert P. Bartók, Patricia Hernández-León, Gábor Csányi, Tapio Ala-Nissila, and Miguel A. Caro<br> Phys. Rev. B 104, 054106 (2021) <strong>References</strong> A.P. Bartók, M.C. Payne, R. Kondor, and G. Csányi. Phys. Rev. Lett. <strong>104</strong>, 136403 (2010). LibAtoms: https://libatoms.github.io V.L. Deringer and G. Csányi. Phys. Rev. B <strong>95</strong>, 094203 (2017). J.P. Perdew, K. Burke, and M. Ernzerhof. Phys Rev. Lett. <strong>77</strong>, 3865 (1996). VASP: http://vasp.at G. Kresse and J. Furthmüller. Phys. Rev. B <strong>54</strong>, 11169 (1996). A. Tkatchenko and M. Scheffler. Phys. Rev. Lett. <strong>102</strong>, 073005 (2009). A.P. Bartók, R. Kondor, and G. Csányi. Phys. Rev. B <strong>87</strong>, 184115 (2013). M.A. Caro. Phys. Rev. B <strong>100</strong>, 024112 (2019). TurboGAP: http://turbogap.fi <strong>Contact</strong> Miguel A. Caro: [e-mail] or [e-mail]