About

DXMag Computational HeuslerDB contains structures and properties for ternary, quaternary, and all-d Heusler compounds. The entries of ternary Heusler compounds were obtained using density functional theory (DFT) calculations. The entries of quaternary and all-d Heusler compounds were obtained using machine learning interatomic potential (MLIP) and machine learning regression model (MLRM).

Ternary Heusler Compounds

This database contains first-principles calculated data for ternary Heusler compounds. The data is generated using density functional theory (DFT) calculations. The database provides a wide range of physical and chemical properties for Heusler compounds, including electronic, magnetic, and structural properties.

• Number of compounds: 27,864 (regular, inverse, and half-Heusler compounds)
• More than 100,000 entries (including various magnetic configurations and structures)
• Properties:
  • Relaxed structures, energies, and magnetic moments: ~ 106,000
  • Electronic bands/DOS: ~ 50,000
  • Phonon: ~ 8,000
  • Curie/Néel temperatures: ~ 1,200
  • Anomalous Hall/Nernst conductivities: ~ 670

Download

• Unified file containing structures and properties: Download latest CSV (2025-09-09)
  • View previous versions
• Electronic structure
  • Full list of electronic DOS
  • Full list of band structures along high-symmetry lines
• Phonons
  • Full inputs for ALAMODE (will be soon available)
• Interactive web interface
  • Heuslerdb dash (accessible only from NIMS network)

Quaternary and all-d Heusler Compounds

These datasets aggregate quaternary and all-d Heusler compounds produced through a machine-learning–accelerated high-throughput workflow. The pipeline employs ML interatomic potentials (e.g., eSEN-30M-OAM) and transfer-learned regression models to perform structure optimization and property estimation. The precision of this ML-HTP workflow is demonstrated in our paper.

• Number of compounds: 131,544 (quaternary) and 105,763 (all-d)
• Ground-state structures and properties included
• Each entry contains properties:
  • Relaxed structures
  • Formation energy and energy above convex hull
  • Local magnetic moments
  • Minimum phonon frequency
  • Curie/Néel temperatures
  • Magnetic anisotropy energy

Download

• Unified file containing quaternary Heusler compounds structures and properties: Download latest CSV (2026-03-10)
• Unified file containing all-d Heusler compounds structures and properties: Download latest CSV (2026-03-10)

References

• E. Xiao and T. Tadano, “High-throughput computational screening of Heusler compounds with phonon considerations for enhanced material discovery”, Acta Mater. 297, 121312 (2025). [arXiv version]
• E. Xiao and T. Tadano, “Accurate Screening of Functional Materials with Machine-Learning Potential and Transfer-Learned Regressions: Heusler Alloy Benchmark”, npj Comput. Mater. (2026). [arXiv version]

License

• The database is released under the CC BY 4.0 license.

Computational details

• VASP: Structure optimization, band structure, DOS, phonon dispersion, Spin polarization

  • PREC = A; ENCUT = 520; ADDGRID = .True.; LASPH = .True.
  • ISMEAR = 2; SIGMA = 0.05 for structural opt., and ISMEAR = 2; SIGMA = 0.1 for phonon calculations, ISMEAR = -5 for DOS.
  • LREAL is set to Auto except for phonon calculations, where it is set to False.
  • EDIFF is set to 1.0e-7 for structure optimization and 1.0e-8 for phonon calculations.
  • EDIFFG = 1.0e-4 for structure optimization with ISIF = 3
  • KSPACING = 0.2 is used except for the body-centered tetragonal (bct) lattice; for the bct lattice, we used pymatgen.io.vasp.inputs.Kpoints.automatic_density_by_vol with a density of 450 per Å^-3.
  • GGA-PBE functional is used for all calculations.
  • No Spin-orbit coupling and no Hubbard U correction are considered.
  • Pseudopotentials: default set of PBE_54 by pymatgen. For the complete list, please refer to here.
• ALAMODE: Phonon band structure, phonon DOS
  • \(2\sqrt{2} \times 2\sqrt{2}\times 2\) conventional supercell is used for phonon calculations.
  • For the harmonic force constant calculations, the displacement magnitude is set to 0.01-0.02 Å.
• SPR-KKR: Curie Temperature, Anomalous Hall/Nernst conductivities
  • NKTAB = 500; NE = 48 for Self-consistent field (SCF) computation and Curie Temperature
  • NKTAB = 125000; NE = 20; NOSYMSIG for Anomalous Hall/Nernst conductivities.
  • TOL = 1e-08 for SCF computation.
  • Curie Temperatures are computed using ASA and FULLPOT
  • Anomalous Hall/Nernst conductivities are computed using ASA.
  • GGA-PBE functional is used for all calculations.
• ML interatomic potential: structure optimization, formation energy, and distance above convex hull
  • eSEN-30M-OAM interatomic potential is used.
  • Fast Inertial Relaxation Engine (FIRE) optimizer was employed.
  • Reference energies for formation energy and hull distance were also computed using eSEN-30M-OAM.
  • Approach implemented in the MLIP-HOT package.
• ML Regression model: Local magnetic moments, Minimum phonon frequency, Curie/Néel temperatures, Magnetic anisotropy energy
  • eSEN architecture.
  • Frozen transfer learned from eSEN-30M-OAM.
  • Trained using ternary Heusler compounds dataset.
  • Approach implemented in the MLIP-FTL package.

Funding and Computational resources

• MEXT Program: Data Creation and Utilization-Type Material Research and Development Project (Digital Transformation Initiative Center for Magnetic Materials) Grant Number JPMXP1122715503.

• MEXT Program: “Program for Promoting Researches on the Supercomputer Fugaku” (Data-Driven Research Methods Development and Materials Innovation Led by Computational Materials Science; Grant Number JPMXP1020230327) and used computational resources of (supercomputer Fugaku provided by the RIKEN Center for Computational Science ) (Project ID: hp230212)

• Super computer resources at the National Institute for Materials Science (NIMS)

• Super computer resources at the Institute for Solid State Physics, The University of Tokyo