Molecular dynamics simulations are powerful tools to extract the microscopic mechanisms characterizing the properties of soft materials. We recently introduced machine learning surrogates for molecular dynamics simulations of soft materials and demonstrated that artificial neural network based regression models can successfully predict the relationships between the input material attributes and the simulation outputs. Here, we show that statistical uncertainties associated with the outputs of molecular dynamics simulations can be utilized to train artificial neural networks and design machine learning surrogates with higher accuracy and generalizability. We design soft labels for the simulation outputs by incorporating the uncertainties in the estimated average output quantities, and introduce a modified loss function that leverages these soft labels during training to significantly reduce the surrogate prediction error for input systems in the unseen test data. The approach is illustrated with the design of a surrogate for molecular dynamics simulations of confined electrolytes to predict the complex relationship between the input electrolyte attributes and the output ionic structure. The surrogate predictions for the ionic density profiles show excellent agreement with the ground truth results produced using molecular dynamics simulations. The high accuracy and small inference times associated with the surrogate predictions provide quick access to quantities derived using the number density profiles and facilitate rapid sensitivity analysis.

The convergence of HPC and data intensive methodologies provide a promising approach to major performance improvements. This paper provides a general description of the interaction between traditional HPC and ML approaches and motivates the "Learning Everywhere" paradigm for HPC. We introduce the concept of "effective performance" that one can achieve by combining learning methodologies with simulation based approaches, and distinguish between traditional performance as measured by benchmark scores. To support the promise of integrating HPC and learning methods, this paper examines specific examples and opportunities across a series of domains. It concludes with a series of open computer science and cyberinfrastructure questions and challenges that the Learning Everywhere paradigm presents.• HPCforML: Using HPC to execute and enhance ML performance, or using HPC simulations to train ML algorithms (theory guided machine learning), which are then used to understand experimental data or simulations.

No abstract

In elastohydrodynamic lubrication (EHL), the lubricant experiences pressures in excess of 500 MPa and strain rates larger than $$10^5$$ 10 5  $$\text{s}^{-1}$$ s - 1 . The high pressures lead to a dramatic rise in the Newtonian viscosity and the high rates lead to large shear stresses and pronounced shear thinning. The extraction of accurate rheological properties using non-equilibrium molecular dynamics simulations (NEMD) has played a key role in improving our understanding of lubricant flow in EHL conditions. However, the high dimensionality of the output data generated by NEMD simulations often makes a deeper interrogation of the link between molecular-scale features and rheological properties challenging. In this paper, we explore the use of machine learning to analyze and visualize the high-dimensional output data generated in typical NEMD simulations. We show that dimension reduction of NEMD simulation data describing the shear flow of squalane enables a clear visualization of the transition from Newtonian to non-Newtonian shear thinning with increasing shear rate and provides a reliable assessment of the correlation between shear thinning and the evolution in molecular alignment. The end-to-end atom pairs dominate the largest variations in pair orientation tensor components for low-pressure systems (0.1, 100 MPa) and provide the clearest separation of the orientation tensors with rate. On the other hand, the side atom pairs dominate the largest variation in the tensor components for the high-pressure systems ($$P\ge 400$$ P ≥ 400 MPa) which exhibit an overall limited evolution in orientation tensors as a function of rate. Dimension reduction using all the six components of the orientation tensors of all 435 pairs associated with a squalane molecule shows that the decrease in viscosity with rate for low pressures is strongly correlated with changes in molecular alignment. However, for high pressures, shear thinning occurs at saturated orientational order.

Classical molecular dynamics simulations are based on solving Newton’s equations of motion. Using a small timestep, numerical integrators such as Verlet generate trajectories of particles as solutions to Newton’s equations. We introduce operators derived using recurrent neural networks that accurately solve Newton’s equations utilizing sequences of past trajectory data, and produce energy-conserving dynamics of particles using timesteps up to 4000 times larger compared to the Verlet timestep. We demonstrate significant speedup in many example problems including 3D systems of up to 16 particles.