芝 隼人

Recent developments in machine learning have enabled accurate predictions of the dynamics of slow structural relaxation in glass-forming systems. However, existing machine learning models for these tasks are mostly designed such that they learn a single dynamic quantity and relate it to the structural features of glassy liquids. In this study, we propose a graph neural network model, “BOnd TArgeting Network,” that learns relative motion between neighboring pairs of particles, in addition to the self-motion of particles. By relating the structural features to these two different dynamical variables, the model autonomously acquires the ability to discern how the self motion of particles undergoing slow relaxation is affected by different dynamical processes, strain fluctuations and particle rearrangements, and thus can predict with high precision how slow structural relaxation develops in space and time.

Abstract Recently, a two-dimensional liquid cooled toward the glass transition was found to exhibit a t⁻¹ long-time tail in the velocity autocorrelation function (VACF) owing to the presence of long-wavelength fluctuations. To directly observe this power-law behaviour, it is necessary to simulate a large system with millions of particles, which is a challenging task from the computational viewpoint. In this study, to address this difficulty, I first show that this power-law tail can be reproduced by differentiating the finite-time diffusivity with respect to time. In addition, the feasibility of another direction, a direct on-the-fly computation of the VACFs utilizing GPGPUs, wherein VACFs are evaluated as the simulation runs, is also demonstrated. A performance benchmark was executed on Wisteria/BDEC-01 (Aquarius subsystem) supercomputer using a simulation code developed by the author, which enabled the direct computation of the VACF of 4 million particlesx for as long as the 10⁸ simulation steps within 10 days.

<p>The out-of-equilibrium phase diagrams obtained by the oscillatory shear in a 1-dodecyl-3-methylimidazolium hexafluorophosphate ([C₁₂mim][PF₆]) ionic liquid crystal on molecular-dynamics simulations.</p>