A Δ-machine learning approach for force fields, illustrated by a CCSD(T) 4-body correction to the MB-pol water potential

Qu, Chen; Yu, Qi; Conte, Riccardo; Houston, Paul L.; Nandi, Apurba; Bowman, Joel M.

doi:10.1039/d2dd00057a

Cited by 16 publications

(22 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The inclusion of V PIP 4B in models 4b and 4c significantly improves the descriptions of 4-body energies when compared to the original MB-pol PEF, which does not include a PIP 4-body term (model a). This behavior is consistent with ref .…”

Section: Resultssupporting

confidence: 93%

MB-pol(2023): Sub-chemical Accuracy for Water Simulations from the Gas to the Liquid Phase

Zhu

Riera

Bull-Vulpe

et al. 2023

J. Chem. Theory Comput.

View full text Add to dashboard Cite

We use the MB-pol theoretical/computational framework to introduce a new family of data-driven many-body potential energy functions (PEFs) for water, named MB-pol(2023). By employing larger 2-body and 3-body training sets, including an explicit machine-learned representation of 4-body energies, and adopting more sophisticated machine-learned representations of 2-body and 3-body energies, we demonstrate that the MB-pol(2023) PEFs achieve sub-chemical accuracy in modeling the energetics of the hexamer isomers, outperforming both the original MB-pol and q-AQUA PEFs, which currently provide the most accurate description of water clusters in the gas phase. Importantly, the MB-pol(2023) PEFs provide remarkable agreement with the experimental results for various properties of liquid water, improving upon the original MB-pol PEF and effectively closing the gap with experimental measurements.

show abstract

Section: Resultssupporting

confidence: 93%

MB-pol(2023): Sub-chemical Accuracy for Water Simulations from the Gas to the Liquid Phase

Zhu

Riera

Bull-Vulpe

et al. 2023

J. Chem. Theory Comput.

View full text Add to dashboard Cite

show abstract

“…They all have two-body terms, such as a Lennard-Jones or exp-6 potential. This general form immediately suggests the following expression for a many-body Δ correction: .25em V normalΔ‐ML + MB‐FF = V MB‐FF + prefix∑ i > j N normalΔ V 2‐b ( i , j ) + prefix∑ i > j > k N normalΔ V 3‐b ( i , j , k ) + prefix∑ i > j > k > l N normalΔ V 4‐b ( i , j , k , l ) + ··· where V MB‑FF is the force field and the Δ V n ‑b are the many-body corrections to the MB-FF many-body terms. These are given by the difference between CCSD(T) and MB-FF n -body ( n -b) interaction energies.…”

Section: δ-Machine Learning Polarizable Force Fieldsmentioning

confidence: 99%

“…Using fourth-order 222111-symmetry PIPs, the fitting RMSE for the whole dataset is 9 cm –1 . For the Δ-ML 4-b correction potential, Δ V 4‑b , we employed the same PIP bases as described in q-AQUA 4-b and the Δ-ML 4-b to MB-pol using a dataset of 3692 tetramer energies computed at the CCSD(T)-F12/haTZ level of theory. The fitting RMSE for this correction potential is 6.3 cm –1 .…”

Section: δ-Machine Learning Polarizable Force Fieldsmentioning

confidence: 99%

Δ-Machine Learned Potential Energy Surfaces and Force Fields

Bowman

Qu²,

Conte

et al. 2022

J. Chem. Theory Comput.

Self Cite

View full text Add to dashboard Cite

There has been great progress in developing machinelearned potential energy surfaces (PESs) for molecules and clusters with more than 10 atoms. Unfortunately, this number of atoms generally limits the level of electronic structure theory to less than the "gold standard" CCSD(T) level. Indeed, for the well-known MD17 dataset for molecules with 9−20 atoms, all of the energies and forces were obtained with DFT calculations (PBE). This Perspective is focused on a Δmachine learning method that we recently proposed and applied to bring DFT-based PESs to close to CCSD(T) accuracy. This is demonstrated for hydronium, N-methylacetamide, acetyl acetone, and ethanol. For 15atom tropolone, it appears that special approaches (e.g., molecular tailoring, local CCSD(T)) are needed to obtain the CCSD(T) energies. A new aspect of this approach is the extension of Δ-machine learning to force fields. The approach is based on many-body corrections to polarizable force field potentials. This is examined in detail using the TTM2.1 water potential. The corrections make use of our recent CCSD(T) datasets for 2-b, 3-b, and 4-b interactions for water. These datasets were used to develop a new fully ab initio potential for water, termed q-AQUA.

show abstract

“…The fitting basis is the same as described in q-AQUA 4-b and the ∆-ML 4-b to MB-pol; 33 briefly, it consists of 200 "super PIPs" of Morse variables; these super PIPS are formed from sums of regular PIPs so as to ensure invariance with respect to permutation of monomers. 32 The fitting RMS error for the whole data set is 6.3 cm −1 .…”

Section: Methods and Computational Detailsmentioning

confidence: 99%

Interfacing q-AQUA with a Polarizable Force Field: The Best of Both Worlds

Bowman

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

Polarizable force fields are pervasive in the fields of computational chemistry and biochemistry; however, their empirical or semi-empirical nature gives them both weaknesses and strengths. Here, we have developed a hybrid water potential, named q-AQUA-pol, by combining our recent ab initio q-AQUA potential with the TTM3-F water potential. The new potential demonstrates unprecedented accuracy ranging from gas-phase clusters, e.g., the eight low-lying isomers of the water hexamer, to the condensed phase, e.g., radial distribution functions, the self-diffusion coefficient, triplet OOO distribution, and the temperature dependence of the density. This represents a significant advancement in the field of polarizable machine learning potential and computational modeling.

show abstract

A Δ-machine learning approach for force fields, illustrated by a CCSD(T) 4-body correction to the MB-pol water potential

Abstract: ∆-Machine Learning (∆-ML) has been shown to effectively and efficiently bring a low-level ML potential energy surface to CCSD(T) quality. Here we propose extending this approach to general force fields,...

Cited by 16 publications

References 39 publications

MB-pol(2023): Sub-chemical Accuracy for Water Simulations from the Gas to the Liquid Phase

MB-pol(2023): Sub-chemical Accuracy for Water Simulations from the Gas to the Liquid Phase

Δ-Machine Learned Potential Energy Surfaces and Force Fields

Interfacing q-AQUA with a Polarizable Force Field: The Best of Both Worlds

Contact Info

Product

Resources

About