Resolving the Mechanisms of Ca and Mg Carbonate Ion-Pair Formation with Multi-Level Molecular Dynamics/Quantum Mechanics Simulations

J.-N. Boyn and E. A. Carter, submitted to J. Phys. Chem. B (2023)

Abstract: coming soon!

 

A Neural Network Water Model based on the MB-pol Many-Body Potential

Muniz, Maria, Car, Roberto, Panagiotopoulos, Athanassios, accepted, J. Phys. Chem. B (2023)

Abstract
The MB-pol many-body potential accurately predicts many properties of water, including cluster, liquid phase and vapor-liquid equilibrium properties, but its high computational cost can make applying it in large-scale simulations quite challenging. In order to address this limitation, we develop a “Deep Potential” neural network (DPMD) model based on the MB-pol potential for water. We find that a DPMD model trained on mostly liquid configurations yields a good description of the bulk liquid phase, but severely underpredicts vapor-liquid coexistence densities. By contrast, adding cluster configurations to the neural network training set leads to good agreement for vapor coexistence densities. Liquid phase densities at supercooled conditions are also represented well, even though they were not included in the training set. These results confirm that neural network models can combine accuracy and transferability, if sufficient attention is given to the construction of a representative training set for the target system.

Critical behavior in a chiral molecular model

Pablo M. Piaggi, Roberto Car, Frank H. Stillinger, Pablo G. Debenedetti, submitted to J Chem Phys 2023

Abstract
Understanding the condensed-phase behavior of chiral molecules is important in biology, as well as in a range of technological applications, such as the manufacture of pharmaceuticals. Here, we use molecular dynamics simulations to study a chiral four-site molecular model that exhibits a second-order symmetry-breaking phase transition from a supercritical racemic liquid, into subcritical D-rich and L-rich liquids. We determine the infinite-size critical temperature using the fourth-order Binder cumulant, and we show that the finite-size scaling behavior of the order parameter is compatible with the 3D Ising universality class. We also study the spontaneous D-rich to L-rich transition at a slightly subcritical temperature T≈0.985Tc and our findings indicate that the free energy barrier for this transformation increases with system size as N2/3 where N is the number of molecules, consistent with a surface-dominated phenomenon. The critical behavior observed herein suggests a mechanism for chirality selection in which a liquid of chiral molecules spontaneously forms a phase enriched in one of the two enantiomers as the temperature is lowered below the critical point. Furthermore, the increasing free energy barrier with system size indicates that fluctuations between the L-rich and D-rich phases are suppressed as the size of the system increases, trapping it in one of the two enantiomerically-enriched phases. Such a process could provide the basis for an alternative explanation for the origin of biological homochirality. We also conjecture the possibility of observing nucleation at subcritical temperatures under the action of a suitable chiral external field.

Ab initio Generalized Langevin Equation

Pinchen Xie, Roberto Car, Weinan E., submitted to Proc. Nat. Acad. Sci. 2023, arXiv preprint arXiv:2211.06558

Abstract
We propose an approach for learning accurately the dynamics of slow collective variables from atomistic data obtained from ab-initio quantum mechanical theory, using generalized Langevin equations (GLE). The force fields, memory kernel, and noise generator are constructed within the Mori-Zwanzig formalism under the constraint imposed by the fluctuation-dissipation theorem. Combined with Deep Potential Molecular Dynamics (DeePMD) and density functional theory, this GLE approach opens the door to carrying out first-principles multi-scale modeling for a variety of systems. Here, we demonstrate this capability with a study of the dynamics of twin domain walls in ferroelectric lead titanate. The importance of memory effects is illustrated by the fact that while ab-initio GLE agrees well with molecular dynamics at near-equilibrium conditions, Markovian Langevin dynamics underestimates the rate of rare events by several orders of magnitude.

 

 

Effect of Phonons and Impurities on the Quantum Transport in XXZ Spin-Chains

Amartya Bose, preprint https://arxiv.org/abs/2206.11156 (2022)

Abstract
Numerical and analytic results have been used to characterize quantum transport in spin chains, showing the existence of both ballistic and diffusive motion. Experiments have shown that heat transfer is surprisingly always diffusive. The scattering from phonons and impurities have been postulated to be the two factors critical in causing the diffusive transport. In this work, we evaluate the transport process by incorporating a bath of phonons and impurities in order to understand the role played by each of the factors. While methods like time-dependent density matrix renormalization group (tDMRG) can be used to simulate isolated spin chains, the coupling with phonons make simulations significantly more challenging. The recently developed multisite tensor network path integral (MS-TNPI) method builds a framework for simulating the dynamics in extended open quantum systems by combining ideas from tDMRG and Feynman-Vernon influence functional. This MS-TNPI is used to characterize dynamics in open, extended quantum systems. Simulations are done with the commonly used sub-Ohmic, Ohmic and super-Ohmic spectral densities describing the phononic bath. We show that while the transport in presence of impurities eventually becomes diffusive, the exact details are dependent on the specifics of the interactions and amount of impurities. In contrast, the presence of a bath makes the transport diffusive irrespective of the parameters characterizing the bath.

 

Zero-Cost Corrections to Influence Functional Coefficients from Bath Response Functions

Amartya Bose, preprint https://arxiv.org/abs/2205.15072 (2022)

Abstract
 Recent work has shown that it is possible to circumvent the calculation of the spectral density and directly calculate the coefficients of the discretized influence functionals using data from classical trajectory simulations. However, the accuracy of this procedure depends on the validity of the high temperature approximation. In this work, an alternative derivation based on the Kubo formalism is provided. This enables the calculation of additional correction terms that increases the range of applicability of the procedure to lower temperatures. Because it is based on the Kubo-transformed correlation function, this approach enables the direct use of correlation functions obtained from methods like ring-polymer molecular dynamics and centroid molecular dynamics in determining the influence functional coefficients for subsequent system-solvent simulations. The accuracy of the original procedure and the corrected procedure is investigated across a range of parameters. It is interesting that the correction term comes at zero additional cost. Furthermore, it is possible to improve upon the correction using zero-cost physical intuition and heuristics making the method even more accurate.

Improving the Generality of Neural Network Potentials with Constant Input Scaling

Mark DelloStritto and Michael Klein, submitted to Journal of Chemical Physics

Abstract

The use of neural network potentials (NNPs) to fit ab-initio potential energy surfaces is one of the most accurate and versatile methods to expand the utility of high quality, small-scale quantum calculations, allowing for ab-initio accuracy in simulations of large systems with an increase of efficiency by several orders of magnitude. As with all neural network applications however, NNPs present challenges with respect to over-fitting and generalization. In particular, choosing and normalizing the training data can be a complex task, as biases or gaps in the data can negatively impact the NNP performance. Here, it is shown that normalizing the inputs to the NNP by a constant related to the interaction cutoff volume leads to a significant improvement in the generalization of NNPs trained for Ar, Si, and NaCl compared to the standard approach, where each input is normalized by subtracting the average and scaling by the standard deviation. Specifically, NNPs trained using inputs scaled by a constant yield more accurate energies for defect structures not in the training set, and a NNP trained on isolated Ar clusters yields an accurate potential for solid and liquid Ar. The standard approach to NNP input scaling generally yields worse defect formation energies and cannot be transferred from isolated cluster calculations to periodic systems. We note that, when using volume scaling, the first layer of the neural network effectively renormalizes the inputs, reversing trends in the input data and narrowing the distribution of input values. Seemingly, the first layer is learning the distribution of inputs and renormalizing it for processing by subsequent layers, thereby leading to improved performance by removing a priori assumptions on the input distributions.

Using differentiable programming to obtain an energy and density-optimized exchange-correlation functional

Sebastian dick and Marivi Fernandez-Serra, submitted to Phys. Rev. Lett. (2021)

Abstract
Using an end-to-end differentiable implementation of the Kohn-Sham self-consistent field equations, we obtain an accurate neural network-based exchange and correlation (XC) functional of the electronic density. The functional is optimized using information on both energy and density while exact constraints are enforced through an appropriate neural network architecture. We evaluate our model against different families of XC approximations and show that, for non-empirical functionals, there is a strong linear correlation between energy and density errors. We use this correlation to define a novel XC functional quality metric that includes both energy and density errors, leading to a new, improved way to rank different approximations. Judged by this metric, our machine-learned functional significantly outperforms those within the same rung of approximations.

Automatic machine-learning potential generation scheme and simulation protocol for the LiGePS-type superionic conductors

Jianxing Huang, Linfeng Zhang, Han Wang, Jinbao Zhao, Jun Cheng, Weinan
J. Phys. Chem. Lett.
arXiv: 2006.03320.

Abstract
It has been a challenge to accurately simulate Li-ion diffusion processes in battery materials at room temperature using {\it ab initio} molecular dynamics (AIMD) due to its high computational cost. This situation has changed drastically in recent years due to the advances in machine learning-based interatomic potentials. Here we implement the Deep Potential Generator scheme to \textit{automatically} generate interatomic potentials for LiGePS-type solid-state electrolyte materials. This increases our ability to simulate such materials by several orders of magnitude without sacrificing {\it ab initio} accuracy. Important technical aspects like the statistical error and size effects are carefully investigated. We further establish a reliable protocol for accurate computation of Li-ion diffusion processes at experimental conditions, by investigating important technical aspects like the statistical error and size effects. Such a protocol and the automated workflow allow us to screen materials for their relevant properties with much-improved efficiency. By using the protocol and automated workflow developed here, we obtain the diffusivity data and activation energies of Li-ion diffusion that agree well with the experiment. Our work paves the way for future investigation of Li-ion diffusion mechanisms and optimization of Li-ion conductivity of solid-state electrolyte materials.