2020
DOI: 10.26434/chemrxiv.12530468.v1
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

Creating Gaussian Process Regression Models for Molecular Simulations Using Adaptive Sampling

Abstract: <div>FFLUX is a new force field that combines the accuracy of quantum mechanics with the speed of force </div><div>fields, without any link to the architecture of classical force fields. This force field is atom‐focused and </div><div>adopts the parameter‐free topological atom from Quantum Chemical Topology (QCT). FFLUX uses </div><div>Gaussian Process Regression (GPR) (aka kriging) models to make predicƟons of atomic properties, which </div><div>… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
2
1

Citation Types

0
11
0

Year Published

2021
2021
2023
2023

Publication Types

Select...
4
2

Relationship

5
1

Authors

Journals

citations
Cited by 8 publications
(11 citation statements)
references
References 33 publications
0
11
0
Order By: Relevance
“…For two training strategies that achieve the same error, that which does so with fewer training points is more computationally efficient. Attempts to develop more computationally efficient training strategies have involved active learning or sequential design methods 19,29,59 , composite kernels 24 and new sampling methods 27,60 .…”
Section: A Gp Regressionmentioning
confidence: 99%
See 1 more Smart Citation
“…For two training strategies that achieve the same error, that which does so with fewer training points is more computationally efficient. Attempts to develop more computationally efficient training strategies have involved active learning or sequential design methods 19,29,59 , composite kernels 24 and new sampling methods 27,60 .…”
Section: A Gp Regressionmentioning
confidence: 99%
“…This approach proceeds by training a statistical technique on a relatively small set of data from ab initio calculations on the PES of interest, known as the training set. Many such techniques have been employed to predict the energy in these algorithms, including neural networks [9][10][11][12][13] , moment tensors [14][15][16] and Gaussian processes [17][18][19][20][21][22][23][24][25][26][27][28][29][30][31][32][33][34] (GPs).…”
Section: Introductionmentioning
confidence: 99%
“…Kriging makes predictions based on some training set containing train N training points. The training of kriging models is a topic unto itself 20 and is not covered here as it is not relevant. Predictions of a quantity of interest, ˆA Y , relating to (topological) atom A, are made according to and predicts based on this correlation.…”
Section: Krigingmentioning
confidence: 99%
“…DL_FFLUX is a force field that aims to perform accurate Molecular Dynamics (MD) calculations without sacrificing speed too much. The accuracy of the FFLUX methodology has been demonstrated at various levels of analysis from single molecules [18][19][20][21] to small clusters of molecules 22 and even ions 23 . However, before the advances presented in the current paper, the prohibitive computational cost of DL_FFLUX made 'bulk' simulations impossible.…”
Section: Introductionmentioning
confidence: 99%
“…The choice of a local frame is significant as it means that all multipole moments are invariant with respect to global rotations and translations. 20 and is not covered here as it is not relevant. Predictions of a quantity of interest, ˆA Y , relating to (topological) atom A, are made according to 42 resulting from these energies.…”
Section: Krigingmentioning
confidence: 99%