Autonomous molecular design by Monte-Carlo tree search and rapid evaluations using molecular dynamics simulations

Abstract

Functional materials, especially those that largely differ from known materials, are not easily discoverable because both human experts and supervised machine learning need prior knowledge and datasets. An autonomous system can evaluate various properties a priori, and thereby explore unknown extrapolation spaces in high-throughput simulations. However, high-throughput evaluations of molecular dynamics simulations are unrealistically demanding. Here, we show an autonomous search system for organic molecules implemented by a reinforcement learning algorithm, and apply it to molecular dynamics simulations of viscosity. The evaluation is dramatically accelerated (by three orders of magnitude) using a femto-second stress-tensor correlation, which underlies the glass-transition model. We experimentally examine one of 55,000 lubricant oil molecules found by the system. This study indicates that merging simulations and physical models can open a path for simulation-driven approaches to materials informatics.

Contributions of Molecular Dynamics Simulations to Elastohydrodynamic Lubrication

“…Despite significant recent progress, MD simulations of the high-pressure viscosity of lubricant-sized molecules (C 10 -C 40 [76]) are still a computationally demanding and time-consuming task. An interesting recent development by Kajita et al [83] has been to dramatically accelerate (by three orders of magnitude) the evaluation of viscosity using a femtosecond stress-tensor correlation. Using this method, they screened the viscosity-temperature behaviour of 55,000 lubricant-sized molecules.…”

“…This enabled a Monte-Carlo tree search algorithm to be used to design molecules with a highviscosity index. Kajita et al [83] justified the use of femtosecond stress-tensor correlations through the shoving model developed by Dyre et al [84]. Usually, nanosecond stresstensor correlations are required in the Green-Kubo method to obtain accurate viscosities [69].…”

Mechanochemistry of Phosphate Esters Confined between Sliding Iron Surfaces

“…It has recently been demonstrated that molecular dynamics (MD) simulations can be used to virtually screen and even autonomously design new lubricant molecules with a high viscosity index. 26 MD simulations have also been successfully applied to compare the thermal stability of different antiwear additives on steel surfaces. In particular, Ewen et al 27 have studied substituent effects on the thermal decomposition of phosphate esters on ferrous surfaces, reproducing the same order of reactivity as observed experimentally for ZDDP.…”

“…15 This study gives new insights into the rate-determining step for tribofilm formation (dissociative chemisorption) and its dependence on temperature and stress. It also represents an important step towards the virtual screening and autonomous molecular design 26 of antiwear additives with optimised molecular structures for tailored mechanochemical and tribological responses.…”

Discovery of superionic conductors by ensemble-scope descriptor

“…An ab initio calculation is used as a powerful evaluator of the static properties of materials in a restricted chemical space designed by researchers 3,4 . However, the number of ab initio evaluations critically drops to a few tens of evaluations when exploring non-equilibrium properties (e.g., ionic conductivity) because the computational cost is too large even with a massive multicore architecture 5,6 . A combination of ab initio calculations and machine learning is therefore employed to broaden the search space 5,[7][8][9] .…”