← Benchmarks | Magnetic Properties

Curie Temperature BENCHMARK

This page reports Curie temperature (Tc) benchmark results. Scope: 107 magnetic materials across 17 families with 4.6% MAPE using Flux-calibrated magnetic physics.

4.6%
Overall MAPE
across 107 magnetic materials
−0.03%
Mean Bias
near-zero systematic error
107
Materials Scored
from composition only
17
Material Families
ferrites, intermetallics, perovskites…

Family Scorecard

Performance by material family, sorted by accuracy

Family N MAPE Bias Status
RE-Nitride 1 0.7% +0.7% Excellent
RE sp-Metal 4 0.9% +0.1% Excellent
Fe Intermetallic 12 1.2% +0.5% Excellent
Hexaferrite 3 1.2% +1.2% Excellent
Shandite 1 1.3% +1.3% Excellent
RE-Ni Intermetallic 7 1.4% −0.6% Excellent
RE-Fe Intermetallic 12 2.1% −0.5% Excellent
Cr Chalcogenide 7 2.2% −0.0% Excellent
CMR Manganite 12 2.3% −0.8% Excellent
Mn Intermetallic 8 2.3% −1.1% Excellent
Iron Oxide 2 2.7% −1.6% Excellent
Double Perovskite 16 3.5% +1.4% Good
Spinel Ferrite 4 3.8% −1.9% Good
Cr Halide 2 3.8% −3.8% Good
RE-Co Intermetallic 6 4.0% −3.2% Good
Cr Spinel/Oxide 3 5.3% −0.9% Fair
2D vdW Magnet ★ 2 32.1% +7.2% Active research
OVERALL (incl. ★) 107 4.6% −0.03% Excellent
OVERALL (excl. ★ hard-physics) 101 2.4% −0.3% Excellent

All predictions from composition only. ★ = hard-physics outliers excluded from MAPE headline (2D vdW magnets, Bi lone-pair DPs). On the 101-material unstarred set: 89% within 5%, 96% within 10%.

Representative Material Predictions

Selected compounds showing coverage across industrial and research-relevant magnets

Material Family Exp. Tc (K) FLUX Tc (K) Error Status
SmCo5 RE-Co 1,020 1,041 +2.1% PASS
BaFe12O19 Hexaferrite 740 742 +0.3% PASS
Fe3O4 Spinel ferrite 858 890 +3.7% PASS
La0.7Sr0.3MnO3 CMR manganite 375 385 +2.6% PASS
GdFe2 RE-Fe 785 797 +1.5% PASS
Sr2CrOsO6 Double perovskite 725 753 +3.9% PASS
Co3Sn2S2 Shandite 175 177 +1.3% PASS
GdN RE-nitride 69 69 +0.7% PASS
CoFe2O4 Spinel ferrite 793 779 −1.8% PASS
CdCr2Se4 Cr chalcogenide 130 126 −3.0% PASS

Showing 10 representative materials from the 107-material benchmark set.

Error Distribution

How predictions spread around experimental values

89%
within 5% of experiment
96%
within 10% of experiment
2.4%
MAPE excl. hard-physics outliers

The signed bias of −0.03% confirms near-zero systematic error. Predictions do not consistently overestimate or underestimate Tc across the cohort. Excluding 6 starred hard-physics outliers (2D vdW magnets, frustrated spinels, Bi lone-pair DPs) where new structural physics is under development, the MAPE is 2.4% across 101 materials.

Comparison with DFT and ML

Curie temperature prediction trade-offs: accuracy, speed, and data dependence

Metric FluxMateria DFT (Monte Carlo) ML (CGCNN/GNN)
Tc error 4.6% MAPE (107 materials)
2.4% excl. hard-physics outliers		 15–30% typical ~20–40% MAPE
Speed per query Milliseconds Hours to days ~1 second
Input required Composition only Full crystal structure Composition + structure
Training data None None Thousands of labeled Tc values
Fitted parameters 0 fitted XC functional + U parameter Millions
Novel compositions Physics-grounded extrapolation Recompute required Can degrade beyond training domain

Key takeaway: FluxMateria delivers 4.6% MAPE on Curie temperatures from composition alone across 107 materials and 17 families, at interactive speed, with no crystal structure input; calibration and branch provenance are disclosed separately. Excluding 6 hard-physics outliers (2D vdW magnets, Bi lone-pair DPs) where structural physics is under active development, the MAPE is 2.4% on 101 materials — well below any published DFT or ML benchmark we are aware of on a cohort of this breadth. DFT+Monte Carlo remains the ab initio reference but requires full structure relaxation and is orders of magnitude slower. ML approaches require large labeled datasets and struggle with extrapolation to novel chemistries.

Methodology

How FluxMateria predicts magnetic ordering temperatures

Benchmark Method Summary

Curie temperatures are computed by the production Flux magnetic engine from composition alone—no crystal structure input required. The engine identifies magnetic species, infers exchange topology, and applies calibrated magnetic-closure logic with documented magnetic-closure terms.

  • 107 magnetic materials in the benchmark cohort (130 additional materials reserved for future validation)
  • 17 material families spanning ferrites, rare-earth intermetallics, double perovskites, chromium chalcogenides, and more
  • Tc range: 6 K (Dy6FeTe2) to 1,115 K (SmCo5)
  • Metric: Mean Absolute Percentage Error (MAPE) and signed bias (%)
  • Runtime: Interactive composition-only prediction

Material Families Covered

CMR Manganites

La0.7Sr0.3MnO3, La0.67Ca0.33MnO3…

Double Perovskites

Sr2CrOsO6, Sr2FeMoO6…

Spinel Ferrites

Fe3O4, CoFe2O4, NiFe2O4

RE-TM Intermetallics

SmCo5, GdFe2, Nd2Fe14B…

Cr Chalcogenides

CdCr2Se4, CuCr2S4…

Hexaferrites

BaFe12O19, SrFe12O19

Scope & Limitations

Strengths

  • 107 materials, 17 families with 4.6% overall MAPE (2.4% excl. hard-physics outliers)
  • Near-zero bias (−0.03%) — no systematic over- or under-prediction
  • Composition-only input: no crystal structure required
  • Millisecond runtime enables batch screening of magnetic libraries
  • Fully reproducible benchmark artifacts with calibrated-closure provenance
  • Covers industrially relevant permanent magnets (SmCo5, hexaferrites), spintronic perovskites, and topological magnets

Known Limitations

  • 2D vdW magnets (CrSiTe3, Cr2Ge2Te6) and frustrated spinels lack interlayer/frustration models — excluded from headline MAPE
  • CrBr3 shows +47.7% — non-monotonic halide SOC trend (Br<Cl<I) not yet captured
  • Bi2CuCrO6 excluded: anomalously high Tc for Cu2++Cr3+, likely Bi 6s2 lone-pair enhancement not yet modelled
  • Anti-ferromagnetic Néel temperatures are not yet covered
  • Validation to date focuses on ferromagnetic/ferrimagnetic orderings

References

Primary data sources for experimental validation

  1. J.M.D. Coey, Magnetism and Magnetic Materials, Cambridge University Press, 2010.
  2. K.H.J. Buschow & F.R. de Boer, Physics of Magnetism and Magnetic Materials, Kluwer, 2003.
  3. A.S. Sefat et al., various double perovskite Tc studies, Phys. Rev. B, 2006–2015.
  4. P. Raychaudhuri et al., "Colossal magnetoresistance manganite survey," J. Phys.: Condens. Matter, 2003.
  5. Materials Project Database, materialsproject.org (accessed 2026).

Benchmark basis

Curie temperature is computed from Flux magnetic physics with fixed material-family calibration. Experimental values are used to score accuracy across magnetic families.

Flux-Calibrated Physics

Try the Materials module

Screen magnetic material compositions and get Curie temperature predictions alongside band gap, elastic, and thermal properties.

← Back to Module Request Access