.png){kind=icon}
.png){kind=icon}
Return
Multibaric sampling for machine learning potential construction
.png){kind=icon}
PRE
AI
DOI：10.1103/PhysRevB.103.144106.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The on-the-fly machine learning potential (MLP) generation scheme is combined with the multibaric-isothermal (MUBA) ensemble simulation to construct the MLP applicable for the prediction of the phase stability under various pressures. The MUBA simulation performs a random walk in the volume space and provides an efficient way to sample physically relevant configurations over a wide range of volume. In the present MUBA formulation, the explicit construction of a bias potential to realize a random walk is not required. The sample structures for training are dynamically collected by the simulation using the MLP itself, in which the simultaneous error estimation is utilized to judge whether an updated structure should be added to the sample data set or not. The utility of the method is demonstrated for aluminum nitride. The MUBA ensemble is sampled using the hybrid Monte Carlo (HMC) method at 300 K. Starting from the metastable zinc-blende structure, the simulation correctly reproduces experimentally observed two phases, the wurtzite and rocksalt structures. During 600 000 configuration updates in the HMC sweeps, the total number of density-functional theory (DFT) calculations required is only 107. The constructed MLP shows high accuracy comparable to the DFT calculations.
Keywords:
EQUATION-OF-STATE
ROCK-SALT PHASE
MOLECULAR-DYNAMICS
ALUMINUM NITRIDE
SUPERCONDUCTIVITY
SIMULATIONS
ENSEMBLE
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.