Accuracy and transferability of Gaussian approximation potential models for tungsten

By Wj Szlachta, A. P. Bart'ok, Gábor Csányi. 2014

247 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Materials Chemistry

Citations in the corpus

The MLIP package: moment tensor potentials with MPI and active learning (2020, methodology): “When used with the smooth overlap of atomic positions (SOAP) kernel [2,72], they can provably approximate an arbitrary local many-body interaction of atoms, in contrast to the neural network potentials employed on top of two- and three-body descriptors [59].”
A Performance and Cost Assessment of Machine Learning Interatomic Potentials. (2019, background): “1 + j2 k. In the SNAP formalism, the energy E SNAP , force F j SNAP , and stresses j SNAP are expressed as linear functions of the bispectrum components 15,16 as follows: E SNAP = 0 N + XN i=1 B i; (13) F j SNAP = XN i=1 @ B i @ r j ; (14) j SNAP = XN i=1 r j XN i=1 @ B i @ r j ; (15) where 0 and ”
Machine Learning a General-Purpose Interatomic Potential for Silicon (2018, methodology): “This representation was later simplified to the SOAP representation and kernel [7] and applied to tungsten [67], amorphous carbon [11,68], iron [69], graphene [12], and boron [70].”
Machine-learned potentials for next-generation matter simulations (2021, methodology): “Kernel-based methods—often used to model inorganic systems with one or a few different elements 68,75,86 —can achieve high accuracies when trained on only a few thousand data points.”
Accurate interatomic force fields via machine learning with covariant kernels (2016, background): “On the other hand, in many situations even the best available state-ofthe-art classical force fields may not guarantee accurate force prediction, as they may incur systematic errors [48,49], or may be hard to extend to complex chemical compositions [50], so that a technique that can indefinitely reu”
Active learning of linearly parametrized interatomic potentials (2016, methodology): “A very natural AL approach applicable to force ﬁelds based on Gaussian process regression [7, 17, 6, 27, 29, 30], which has not yet been implemented in practice, would be to use the Bayesian predictive variance, shown to correlate with the actual error, e.g., in [21].”
Moment Tensor Potentials: A Class of Systematically Improvable Interatomic Potentials (2015, methodology): “MTP2 is generated by limiting deg(Bα) + 8(#α) ≤ 52, additionally limiting (#α) ≤ 5, µ ≤ 3, and ν ≤ 5, and applying the `0 regularization algorithm with Ncap = 4, as described in Section 4.2.2, to extract 760 basis functions (so as to make the accuracy of GAP and MTP2 same, as discussed in the next p”