Viscosity in water from first-principles and deep-neural-network simulations

Abstract

We report on an extensive study of the viscosity of liquid water at near-ambient conditions, performed within the Green-Kubo theory of linear response and equilibrium ab initio molecular dynamics (AIMD), based on density-functional theory (DFT). In order to cope with the long simulation times necessary to achieve an acceptable statistical accuracy, our ab initio approach is enhanced with deep-neural-network potentials (NNP). This approach is first validated against AIMD results, obtained by using the Perdew–Burke–Ernzerhof (PBE) exchange-correlation functional and paying careful attention to crucial, yet often overlooked, aspects of the statistical data analysis. Then, we train a second NNP to a dataset generated from the Strongly Constrained and Appropriately Normed (SCAN) functional. Once the error resulting from the imperfect prediction of the melting line is offset by referring the simulated temperature to the theoretical melting one, our SCAN predictions of the shear viscosity of water are in very good agreement with experiments.

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

“…DeePMD-kit 55 is an example of a soware package, which implements this architecture. While this type of approach has shown promise in various applications 52,[56][57][58][59][59][60][61][62][63] it can have large training data requirements, which limits its usefulness as the generation of sufficiently large training data is still computationally very demanding.…”

Size and Quality of Quantum Mechanical Data Set for Training Neural Network Force Fields for Liquid Water

“…On the other hand, more recently active learning procedures have allowed for the construction of a ML potential with fewer training data points . In the work of Malosso et al., for example, the ML potential for liquid water was trained with 4000 configurations, which is in agreement with the size of the training data set required for the convergence of the self-diffusion coefficient present in Figure . Thus, it is possible that by combining uncorrelated snapshots and active learning the size of the data set could be reduced even further.…”

Building robust machine learning models for small chemical science data: the case of shear viscosity of fluids

“…Atomistic molecular dynamics (MDs) simulations with ab initio or empirical force fields can be used to estimate the viscosity of any liquid, however complex, in silico [14][15][16][17][18][19][20][21][22]. While there exists many methods to estimate viscosity from MD simulations, they largely fall into two categories-equilibrium MD (EMD) [14,23] and non-equilibrium MD (NEMD) based methods [24][25][26].…”

“…Despite the progress in this area [14,15,19,23,[27][28][29][30][31][32][33][34], the state-of-the-art methods to estimate viscosity accurately from MD simulations require huge computing time especially for viscous fluids [23,[35][36][37], as it is a collective quantity. This drawback precludes the use of MD simulations in viscosity-based high throughput screening processes in the industry [12].…”