Table 1 Details of the training dataset for α-Fe

Datasets	Dataset	N_str	N_atom	N_force	Energy (meV/atom)	Force (meV/Å)
DE	Perfect crystal	1776	54	95936	2.20	69.38
	Vacancy	500	53	26500	1.33	63.88
	SIA	214	129	27606	1.86	36.43
	Total	2490		150042	2.03	63.60
RANDSPGs	RANDSPG	691	3–10	4078	16.28	88.49
	VOLMIN	1273	3–10	7918	12.72	77.98
	CELLMIN	1836	3–10	11882	11.14	83.20
	INTMIN	2015	3–10	13236	11.30	54.00
	TRIAX	5187	3–10	35806	16.15	62.70
	SHEAR	3209	3–10	22134	21.66	83.85
	RATTLE	3249	3–10	22092	9.10	114.45
	Total	17460		117146	15.14	81.94
All	Total	19950		267188	14.11	71.88

The training dataset for α-Fe consists of the domain expertize (DE) dataset, which is a training dataset for reproducing the basic properties and lattice defects of α-Fe, and the RANDSPG dataset, which is a training dataset for reproducing general grain boundaries. \({N}_{{\rm{str}}}\) and \({N}_{{\rm{force}}}\) are the number of atomic structures (number of total energies) and atomic forces, respectively, in the training dataset. \({N}_{{\rm{atom}}}\) is the number of atoms in each atomic structure in the training dataset. The root mean squared errors (RMSEs) of the energies and forces for each training dataset are also shown.

Quick links

Search