cell	kmesh	method	band_window	contour_points	mae_mev_per_fu	bands_kept_per_fu
FePt 2-atom	8x8x6	TB2J second order, 5.1 eV band cut (production before v61)	5.1 eV cut	100	3.1588	25
FePt 2-atom	8x8x6	TB2J second order, all bands (production from v61)	all	100	2.8598	92
FePt 2-atom	8x8x6	TB2J second order, 5.1 eV band cut	5.1 eV cut	200	3.1482	25
FePt 2-atom	8x8x6	TB2J second order, all bands	all	200	2.8491	92
FePt 2-atom	8x8x6	Direct diagonalisation with SOC along each axis (TB2J smearing)	all		3.548
FePt 2x2x2 (16 atoms)	4x4x3	TB2J second order, 5.1 eV band cut (production before v61)	5.1 eV cut	100	2.8999	24
FePt 2x2x2 (16 atoms)	4x4x3	TB2J second order, all bands (production from v61)	all	100	2.8598	92
FePt 2x2x2 (16 atoms)	4x4x3	TB2J second order, 5.1 eV band cut	5.1 eV cut	200	2.8896	24
FePt 2x2x2 (16 atoms)	4x4x3	TB2J second order, all bands	all	200	2.8491	92
FePt 2x2x2 (16 atoms)	4x4x3	Direct diagonalisation with SOC along each axis (TB2J smearing)	all		3.548
FePt 2-atom	8x8x6	Self-consistent SOC total energy, E(100) - E(001) (reference)	n/a		2.8647
FePt 2-atom	8x8x6	Deployed route (v61 code, service.run_property)	all	100	2.8704

Sign in to see the full dataset

Sep 29

FePt MAE accuracy: TB2J band window vs self-consistent SOC

Data Views Docs Logs

Overview

44

Attribution

Connections

1

Referenced by

Ouro DFT's MAE route: an idle A100, a slow TB2J loop, and a band cut that made results depend on the cell
MAE jobs now run on CPUs, 2–4× faster and 6–12× cheaper. A TB2J band cutoff that biased FePt by 10% and made the MAE depend on the cell is fixed; results now match self-consistent SOC to within 5 µeV. MAE also respects ferrimagnetic seeds and +U.

Attribution

CC-BY-4.0 license

L1₀ FePt magnetocrystalline anisotropy (meV/f.u.) from Ouro DFT (ABACUS LCAO, PBE, DZP, 100 Ry). The 2-atom cell (8×8×6) and a 2×2×2 supercell (4×4×3, an exactly equivalent k-mesh) are compared under different TB2J band windows and contour densities, a direct-diagonalisation variant, and a self-consistent SOC total-energy reference. TB2J's former 5.1 eV band cut made the result depend on the cell; keeping all bands matches the reference to within 5 µeV. Unrelaxed fixture lattice (a = 2.73 Å, c = 3.71 Å).

cell	kmesh	method	band_window	contour_points	mae_mev_per_fu	bands_kept_per_fu
FePt 2-atom	8x8x6	TB2J second order, 5.1 eV band cut (production before v61)	5.1 eV cut	100	3.1588	25
FePt 2-atom	8x8x6	TB2J second order, all bands (production from v61)	all	100	2.8598	92
FePt 2-atom	8x8x6	TB2J second order, 5.1 eV band cut	5.1 eV cut	200	3.1482	25
FePt 2-atom	8x8x6	TB2J second order, all bands	all	200	2.8491	92
FePt 2-atom	8x8x6	Direct diagonalisation with SOC along each axis (TB2J smearing)	all		3.548
FePt 2x2x2 (16 atoms)	4x4x3	TB2J second order, 5.1 eV band cut (production before v61)	5.1 eV cut	100	2.8999	24
FePt 2x2x2 (16 atoms)	4x4x3	TB2J second order, all bands (production from v61)	all	100	2.8598	92
FePt 2x2x2 (16 atoms)	4x4x3	TB2J second order, 5.1 eV band cut	5.1 eV cut	200	2.8896	24
FePt 2x2x2 (16 atoms)	4x4x3	TB2J second order, all bands	all	200	2.8491	92
FePt 2x2x2 (16 atoms)	4x4x3	Direct diagonalisation with SOC along each axis (TB2J smearing)	all		3.548
FePt 2-atom	8x8x6	Self-consistent SOC total energy, E(100) - E(001) (reference)	n/a		2.8647
FePt 2-atom	8x8x6	Deployed route (v61 code, service.run_property)	all	100	2.8704

Sign in to see the full dataset

4

Stats

Rows: 12
Columns: 8
Size: 32 KB
Downloads: 0

Join to comment

Connections

1

Referenced by

Ouro DFT's MAE route: an idle A100, a slow TB2J loop, and a band cut that made results depend on the cell
MAE jobs now run on CPUs, 2–4× faster and 6–12× cheaper. A TB2J band cutoff that biased FePt by 10% and made the MAE depend on the cell is fixed; results now match self-consistent SOC to within 5 µeV. MAE also respects ferrimagnetic seeds and +U.