Predicting properties of periodic systems from cluster data: A case study of liquid water

Abstract

The accuracy of the training data limits the accuracy of bulk properties from machine-learned potentials. For example, hybrid functionals or wave-function-based quantum chemical methods are readily available for cluster data, but effectively out-of-scope for periodic structures. We show that local, atom-centered descriptors for machine-learned potentials enable the prediction of bulk properties from cluster model training data, agreeing reasonably well with predictions from bulk training data. We demonstrate such transferability by studying structural and dynamical properties of bulk liquid water with density functional theory and have found an excellent agreement with experimental as well as theoretical counterparts.

High-dimensional neural network potential for liquid electrolyte simulations

“…Therefore, it is of interest to compare the accuracy of predicted bulk properties of MLFFs that are trained to higher rungs of the "Jacob's ladder" of density functional approximations 66 . In the context of MLFFs for liquids, the current state of the art achieves DFT-level accuracy on prototypical systems like water 27,45 . Moreover, most MLFFs that are developed for bulk property prediction are trained to a single chemical system, which lacks generalizability 27 .…”

“…Moreover, most MLFFs that are developed for bulk property prediction are trained to a single chemical system, which lacks generalizability 27 . Recently, a HDNNP was developed for liquid water simulations and trained to hybrid DFT cluster data 45 . Overall, this model predicted accurate density and self-diffusivity 45 .…”

“…Recently, a HDNNP was developed for liquid water simulations and trained to hybrid DFT cluster data 45 . Overall, this model predicted accurate density and self-diffusivity 45 . In contrast, our model covers a broader chemical space, including common multi-component liquid electrolytes, and is validated more rigorously against multiple experimental of pure solvents and electrolyte mixtures.…”

“…The model was trained using an active learning approach [41][42][43][44] . In addition, the QRNN was trained exclusively to non-periodic DFT cluster data without any parametrization against experimental data (Scheme 1), a strategy that has been proven to be successful for liquid water 45 . With the addition of LiPF6, the common carbonates that we selected for this study were EC, propylene carbonate (PC), vinylene carbonate (VC), fluoroethylene carbonate (FEC), dimethyl carbonate (DMC), diethyl carbonate (DEC), and ethyl methyl carbonate (EMC).…”

Towards predictive design of electrolyte solutions by accelerating ab initio simulation with neural networks.

“…95 Alternatively, training a full NNP purely on data from small cluster calculations, which can be performed at higher levels, may also be feasible. 96 Another potential solution is employing a substantially higher level of theory such as the random phase approximation (RPA), second order Møller-Plesset (MP2) or double hybrids DFT functionals, which are now becoming feasible for periodic systems. [97][98][99] .…”