Coarse-Graining Auto-Encoders for Molecular Dynamics (1812.02706v2)

Abstract: Molecular dynamics simulations provide theoretical insight into the microscopic behavior of materials in condensed phase and, as a predictive tool, enable computational design of new compounds. However, because of the large temporal and spatial scales involved in thermodynamic and kinetic phenomena in materials, atomistic simulations are often computationally unfeasible. Coarse-graining methods allow simulating larger systems, by reducing the dimensionality of the simulation, and propagating longer timesteps, by averaging out fast motions. Coarse-graining involves two coupled learning problems; defining the mapping from an all-atom to a reduced representation, and the parametrization of a Hamiltonian over coarse-grained coordinates. Multiple statistical mechanics approaches have addressed the latter, but the former is generally a hand-tuned process based on chemical intuition. Here we present Autograin, an optimization framework based on auto-encoders to learn both tasks simultaneously. Autograin is trained to learn the optimal mapping between all-atom and reduced representation, using the reconstruction loss to facilitate the learning of coarse-grained variables. In addition, a force-matching method is applied to variationally determine the coarse-grained potential energy function. This procedure is tested on a number of model systems including single-molecule and bulk-phase periodic simulations.

• The paper introduces Autograin, a machine learning framework that automates coarse-graining via auto-encoders and force-matching, reducing reliance on manual tuning.
• It demonstrates effective mapping from all-atom to coarse-grained representations, validated on systems such as OTP, aniline, methane, and ethane.
• The results show enhanced simulation efficiency and structural consistency while identifying opportunities for further refinement in iterative learning.

Coarse-Graining Auto-Encoders for Molecular Dynamics

The paper "Coarse-Graining Auto-Encoders for Molecular Dynamics" introduces a method for enhancing molecular dynamics simulations via coarse-graining auto-encoders. The main contributions lie in optimizing coarse-grained representations of molecules while simultaneously refining the simulated interactions through auto-encoder-based frameworks and force-matching techniques.

Molecular dynamics simulations are recognized for their ability to predict and analyze the behavior of materials at microscopic levels. However, achieving computational feasibility is often a challenge when dealing with processes involving large temporal and spatial scales inherent to thermodynamic and kinetic phenomena. Coarse-graining methods address these challenges by reducing the simulation's dimensionality and enabling longer timesteps—thus, allowing atomic motions to be averaged out and simulated more efficiently.

Coarse-graining involves two interlinked learning tasks: establishing a mapping from an all-atom to a reduced representation, and parametrizing a Hamiltonian over coarse-grained coordinates. In existing methodologies, the all-atom to coarse-grained mapping is typically determined through manual tuning based on chemical intuition, which may not always be robust or scalable. This paper proposes Autograin, a machine learning-driven framework utilizing auto-encoders optimally to address both tasks. Autograin leverages reconstruction loss to identify coarse-grained variables and force-matching methods to variationally refine the coarse-grained potential energy function.

The architecture is applied to systems such as single-molecule simulations of ortho-terphenyl (OTP) and aniline, demonstrating effective coarse-graining through auto-encoders that could seamlessly identify relevant pseudo atoms without human intervention. The results indicate Autograin's success in establishing consistent coarse-grained dynamics and structural correlations. In liquid simulations of methane and ethane, Autograin adequately reproduced pair correlation functions and validated the learned coarse-grained force fields, albeit discrepancies revealing areas for further refinement, particularly in iterative learning and cross-correlation inclusion.

The implications of this research are both practical and theoretical. Practically, Autograin opens prospects for automating the coarse-graining process across a wide variety of molecular systems, thus reducing human bias and enhancing efficiency. Theoretically, it suggests future possibilities in bridging different scales in molecular simulations, advocating for advancements in predictive inference and probabilistic model integration.

Looking ahead, future directions could explore stochastic decoding for thermodynamic modeling and adaptive force field topologies, improving transferability among different thermodynamic states. The potential for machine learning methodologies to extend the capabilities of molecular dynamics simulations seems promising; ongoing developments may integrate learning frameworks that optimize time-series data for an enhanced understanding of non-equilibrium transport properties.

Overall, this paper lays the foundation for future research into automated and scalable coarse-graining approaches, signifying a meaningful stride in computational material science, aligning with the growing demands for efficiency and accuracy in simulating complex molecular systems.