4:00 PM - 4:30 PM
[SIT21-07] Molecular Dynamics Simulation of Shock-Induced Structural Transformation of Silica Using Artificial Neural Network
★Invited Papers
Keywords:Molecular dynamics , Machine learning, Shock compression
Artificial neural network (ANN) potential, which is an interatomic potential constructed by machine-leaning, attracts attention as a promising method to achieve extra-large-scale molecular dynamics (MD) simulation with first-principles accuracy (Behler, 2007; Artrith, 2016). Application of this ANN-MD to far-from-equilibrium phenomena, such as fracture and pressure-induced transformation, is important to understand structural properties of materials under extreme conditions. For this purpose, we have tried to perform ANN-MD simulation of shock-induced structural transformation of silica.
To compute shock-compression behaviors within the framework of MD method, the multiscale-shock technique (Reed, 2003) was employed in our simulations. Potential energy of shock-compressed α-quartz obtained by first-principles MD (FPMD) method is used as a reference data for training of ANN-potential. As a result of ANN-MD simulation for elastic shock-wave region, elastic-deformation behaviors of α-quartz was successfully reproduced with high-accuracy (Misawa, 2020). On the other hand, for plastic shock-wave region, the ANN-potential has completely failed on prediction of both structure and energy, because of that irregular structure appears and rapid movement of atoms occurs during the elastic-to-plastic transition process.
In order to improve the predicting ability of ANN-potential, we introduce a more accurate training method that uses not only potential energy but also atomic force and pressure as reference data (Shimamura, 2020). Using this improved ANN-potential, it was succeeded that reproducing the elastic-to-plastic transition behavior and plastic deformation of α-quartz with an accuracy close to FPMD simulation. Thus, the ANN-potential trained with energy, force, and pressure is a potentially powerful tool to investigate wide range of far-from-equilibrium phenomena and will also provide useful information for solid earth physics field in future.
Reference
E. Reed et al., Phys. Rev. Lett. 90, 235503 (2003).
J. Behler and M. Parrinello, Phys. Rev. Lett. 98, 146401 (2007).
N. Artrith and A. Urban, Comput. Mater. Sci. 114, 135 (2016)
M. Misawa et al., J. Phys. Chem. Lett. 11, 4536 (2020).
K. Shimamura et al., J. Chem. Phys. 153, 234301 (2020).
To compute shock-compression behaviors within the framework of MD method, the multiscale-shock technique (Reed, 2003) was employed in our simulations. Potential energy of shock-compressed α-quartz obtained by first-principles MD (FPMD) method is used as a reference data for training of ANN-potential. As a result of ANN-MD simulation for elastic shock-wave region, elastic-deformation behaviors of α-quartz was successfully reproduced with high-accuracy (Misawa, 2020). On the other hand, for plastic shock-wave region, the ANN-potential has completely failed on prediction of both structure and energy, because of that irregular structure appears and rapid movement of atoms occurs during the elastic-to-plastic transition process.
In order to improve the predicting ability of ANN-potential, we introduce a more accurate training method that uses not only potential energy but also atomic force and pressure as reference data (Shimamura, 2020). Using this improved ANN-potential, it was succeeded that reproducing the elastic-to-plastic transition behavior and plastic deformation of α-quartz with an accuracy close to FPMD simulation. Thus, the ANN-potential trained with energy, force, and pressure is a potentially powerful tool to investigate wide range of far-from-equilibrium phenomena and will also provide useful information for solid earth physics field in future.
Reference
E. Reed et al., Phys. Rev. Lett. 90, 235503 (2003).
J. Behler and M. Parrinello, Phys. Rev. Lett. 98, 146401 (2007).
N. Artrith and A. Urban, Comput. Mater. Sci. 114, 135 (2016)
M. Misawa et al., J. Phys. Chem. Lett. 11, 4536 (2020).
K. Shimamura et al., J. Chem. Phys. 153, 234301 (2020).