Machine learning force fields based on local parametrization of dispersion interactions: Application to the phase diagram of C60

Abstract

We present a comprehensive methodology to enable addition of van der Waals (vdW) corrections to machine learning (ML) atomistic force fields. Using a Gaussian approximation potential (GAP) [Bartók et al.., Phys. Rev. Lett. 104, 136403 (2010)] as baseline, we accurately machine learn a local model of atomic polarizabilities based on Hirshfeld volume partitioning of the charge density [Tkatchenko and Scheffler, Phys. Rev. Lett. 102, 073005 ( 2009)]. These environment-dependent polarizabilities are then used to parametrize a screened Londondispersion approximation to the vdW interactions. Our ML vdW model only needs to learn the charge density partitioning implicitly, by learning the reference Hirshfeld volumes from density functional theory (DFT). In practice, we can predict accurate Hirshfeld volumes from the knowledge of the local atomic environment (atomic positions) alone, making the model highly computationally efficient. For additional efficiency, our ML model of atomic polarizabilities reuses the same many-body atomic descriptors used for the underlying GAP learning of bonded interatomic interactions. We also show how the method enables straightforward computation of gradients of the observables, even when these remain challenging for the reference method (e.g., calculating gradients of the Hirshfeld volumes in DFT). Finally, we demonstrate the approach by studying the phase diagram of C 60 , where vdW effects are important. The need for a highly accurate vdW-inclusive reactive force field is highlighted by modeling the decomposition of the C 60 molecules taking place at high pressures and temperatures.

Neuroevolution machine learning potentials: Combining high accuracy and low cost in atomistic simulations and application to heat transport

“…The smoothing parameter is 2 Å. The size of the embedding net is (25,50,100) and the size of the fitting net is (240,240,240). The learning rate in the stochastic gradient descent algorithm decreases exponentially from 10 −3 to 10 −8 .…”

“…In this section, we present a few case studies to evaluate the performance of NEP implemented in GPUMD, as compared to the QUIP [15] package that implements the GAP-SOAP potential [7,30], the MLIP package [16] that implements the MTP potential [17], and the DeePMD-kit package [18] that implements the DP potential [19,20]. Because a good machine learning potential should be able to account for nearly all the phases of a given material, as demonstrated for elementary silicon [47], phosphorus [48], and carbon [49,50], we will consider fitting a general-purpose potential for silicon. In addition, we will consider fitting a specific potential for two-dimensional (2D) silicene and a specific potential for bulk PbTe.…”

Anisotropic and high thermal conductivity in monolayer quasi-hexagonal fullerene: A comparative study against bulk phase fullerene

“…The vdW interactions between the C 60 molecules are mostly captured by the radial descriptor components with a relatively long cutoff (7 Å) in our NEP model. In other MLPs, vdW interactions in carbon systems have been modelled by an explicit dispersion term with [38] or without [39,40] environment dependence.…”

Accurate Deep Learning-aided Density-free Strategy for Many-Body Dispersion-corrected Density Functional Theory

“…We note that a machine learning approach to model AIM polarizabilities has been recently proposed in modelling pairwise dispersion interactions in carbon-based materials. 36 Here we generalize the approach to model MBD interactions to a much broader class of systems thanks to the employed model's flexibility and broad data set.…”

“…Recently, Muhli and co-workers 36 have developed a pairwise dispersion corrected Gaussian approximation potential (GAP) force field for carbon on TS polarizability rescaling.…”