Details

Autor(en) / Beteiligte

• van der Heide, T.
• Kullgren, J.
• Broqvist, P.
• Bačić, V.
• Frauenheim, T.
• Aradi, B.

Titel

Fortnet, a software package for training Behler-Parrinello neural networks

Ist Teil von

• Computer physics communications, 2023-03, Vol.284, p.108580, Article 108580

Ort / Verlag

Elsevier B.V

Erscheinungsjahr

2023

Link zum Volltext

Quelle

Elsevier ScienceDirect Journals Complete

Beschreibungen/Notizen

• A new, open source, parallel, stand-alone software package (Fortnet) has been developed, which implements Behler-Parrinello neural networks. It covers the entire workflow from feature generation to the evaluation of generated potentials, coupled with higher-level analysis such as the analytic calculation of atomic forces. The functionality of the software package is demonstrated by driving the training for the fitted correction functions of the density functional tight binding (DFTB) method, which are commonly used to compensate the inaccuracies resulting from the DFTB approximations to the Kohn-Sham Hamiltonian. The usual two-body form of those correction functions limits the transferability of the parametrizations between very different structural environments. The recently introduced DFTB+ANN approach strives to lift these limitations by combining DFTB with a near-sighted artificial neural network (ANN). After investigating various approaches, we have found the combination of DFTB with an ANN acting on-top of some baseline correction functions (delta learning) the most accurate one. It allowed to introduce many-body corrections on top of two-body parametrizations, while excellent transferability to chemical environments with deviating energetics could be demonstrated. Program title: Fortnet CPC Library link to program files:https://doi.org/10.17632/sjg3n9vr8p.1 Developer's repository link:https://github.com/vanderhe/fortnet Code Ocean capsule:https://codeocean.com/capsule/3992747 Licensing provisions: LGPL Programming language: Fortran, Python External routines/libraries: MPI, BLAS/LAPACK, HDF5, DFTB+ Supplementary material: See supplementary material for exemplary Human-friendly Structured Data (HSD) input listings, as well as the basic usage of the Fortformat Python layer for generating datasets and extracting results. Nature of problem: Semi-empirical quantum mechanical methods like density functional tight binding (DFTB) rely on fitting empirical energy correction terms, often represented by two-body potentials, to ab initio references. Hereby empirical, beyond-pairwise contributions are inevitably incorporated and therefore inadequately covered by a purely two-body description. Solution method: The new, open source, parallel, stand-alone software package Fortnet provides a powerful, yet accessible tool to construct many-body correction terms by resorting to high-dimensional neural networks of Behler-Parrinello type. Fortnet is characterized by its modern infrastructure, complementing the landscape of available implementations by a robust combination of Fortran and Python based code. Additional comments including restrictions and unusual features: Fortnet's core is supplemented by two additional projects that are BSD 2-clause licensed, namely fortnet-python [2], a collection of Python based tools for generating compatible datasets and extracting results, and fortnet-ase [3], an interface to the Atomic Simulation Environment (ASE) [1]. Both projects are available via the Python Package Index (PyPI). The interaction of all components is explained in cookbook-like recipes (see: https://fortnet.readthedocs.io/en/latest/), meant to guide new users, while learning about various basic and more advanced features by using comprehensible examples with physical reference. [1]A.H. Larsen, J.J. Mortensen, J. Blomqvist, I.E. Castelli, R. Christensen, M. Dułak, J. Friis, M.N. Groves, B. Hammer, C. Hargus, E.D. Hermes, P.C. Jennings, P.B. Jensen, J. Kermode, J.R. Kitchin, E.L. Kolsbjerg, J. Kubal, K. Kaasbjerg, S. Lysgaard, J.B. Maronsson, T. Maxson, T. Olsen, L. Pastewka, A. Peterson, C. Rostgaard, J. Schiøtz, O. Schütt, M. Strange, K.S. Thygesen, T. Vegge, L. Vilhelmsen, M. Walter, Z. Zeng, K.W. Jacobsen, J. Phys., Condens. Matter 29 (2017) 273002. doi:10.1088/1361-648x/aa680e.[2]T. van der Heide, fortnet-python, (2022), GitHub repository, https://github.com/vanderhe/fortnet-python.[3]T. van der Heide, fortnet-ase, (2022), GitHub repository, https://github.com/vanderhe/fortnet-ase.