Uncertainty estimation for molecular dynamics and sampling

Imbalzano, Giulio; Zhuang, Yongbin; Kapil, Venkat; Rossi, Kevin; Engel, Edgar A.; Grasselli, Federico; Ceriotti, Michele

doi:10.48550/arxiv.2011.08828

Cited by 1 publication

(5 citation statements)

References 71 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We also demonstrate how the committee model can be used to estimate the error both static-lattice quantities and finitetemperature thermodynamic averages. [45] In order to provide compelling arguments for the quality of the potential, we have tested the predictions across many different scenarios. Some, such as the calculation of static lattice properties, are directly comparable to the DFT predictions.…”

Section: Discussionmentioning

confidence: 99%

“…During molecular dynamics simulations, the uncertainty computed in this way allows to identify configurations that are poorly predicted, suggesting that the simulation is moving in previously unexplored regions of the phase space. It has recently been shown that this kind of uncertainty estimation can also be used to compute the error on thermodynamical averages, that results from the ML approximation of the reference potential [45]. We show an early application of this approach to the evaluation of the ML error for the pair distribution functions.…”

Section: Uncertainty Estimationmentioning

confidence: 99%

“…If we use this approach, we have to account for the limited dataset and implicit correlation of the models with a correction parameter α [39,45] that ensures that the predicted variance matches the expected variance over an unrelated validation set, as…”

Section: Uncertainty Estimationmentioning

confidence: 99%

“…In these figures we also provide a comparison between the thermodynamic uncertainty obtained by reweighting the trajectories for each potential in the committee (here called σ V following the same notation as Ref. 45) and the statistical uncertainty due to the finite time of the simulations (indicated as σ M D ). In the case of liquid Ga, our potential is able to reproduce with striking accuracy the g(r) both at low and high temperature, similarly to the results of other ab initio studies run with GGA [93,96] or LDA potentials [29].…”

Section: B Liquid Propertiesmentioning

confidence: 99%

“…17. The electronic density of states of liquid and solid Ga, As, and GaAs as predicted by a committee of 16 built following Ref 45…”

mentioning

confidence: 99%

See 4 more Smart Citations

Modeling the Ga/As binary system across temperaturesand compositions from first principles

imbalzano,

Ceriotti

2021

Preprint

Self Cite

View full text Add to dashboard Cite

Materials composed of elements from the third and fifth columns of the periodic table display a very rich behavior, with the phase diagram usually containing a metallic liquid phase and a polar semiconducting solid. As a consequence, it is very hard to achieve transferable empirical models of interactions between the atoms that can reliably predict their behavior across the temperature and composition range that is relevant to the study of the synthesis and properties of III/V nanostructures and devices. We present a machine-learning potential trained on density functional theory reference data that provides a general-purpose model for the GaxAs1−x system. We provide a series of stringent tests that showcase the accuracy of the potential, and its applicability across the whole binary phase space, computing with ab initio accuracy a large number of finite-temperature properties as well as the location of phase boundaries. We also show how a committe model can be used to reliably determine the uncertainty induced by the limitations of the ML model on its predictions, to identify regions of phase space that are predicted with insufficient accuracy, and to iteratively refine the training set to achieve consistent, reliable modeling.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Uncertainty Estimationmentioning

confidence: 99%

Section: Uncertainty Estimationmentioning

confidence: 99%

Section: B Liquid Propertiesmentioning

confidence: 99%

“…17. The electronic density of states of liquid and solid Ga, As, and GaAs as predicted by a committee of 16 built following Ref 45…”

mentioning

confidence: 99%

See 3 more Smart Citations

Modeling the Ga/As binary system across temperaturesand compositions from first principles

imbalzano,

Ceriotti

2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

Uncertainty estimation for molecular dynamics and sampling

Cited by 1 publication

References 71 publications

Modeling the Ga/As binary system across temperaturesand compositions from first principles

Modeling the Ga/As binary system across temperaturesand compositions from first principles

Contact Info

Product

Resources

About