← Benchmarks | Atomic & Magnetic Properties

Atomic & Magnetic Properties BENCHMARK

Five-property benchmark covering fundamental atomic constants and local magnetic moments: 1.0–2.5% MAPE on EN/IE/EA, 100% pass rate on magnetic moment (84 materials) and saturation magnetization, zero fitted parameters.

1.0%
Best MAPE
electron affinity (28 elements)
2.5%
Best atomic MAPE
electronegativity (75 elements)
100%
MM Pass Rate
84/84 magnetic materials
5
Properties
all from first principles
0
Fitted Parameters
physics-derived results

Property Scorecard

All five properties validated against curated experimental data

Property Type N MAPE Median |err| Pass Rate Tolerance
Electron Affinity Atomic 28 1.0% 0.7% 100.0% ±15%
Ionization Energy Atomic 27 1.7% 0.5% 96.3% ±10%
Electronegativity Atomic 75 2.5% 1.3% 94.7% ±10%
Magnetic Moment Magnetic 84 11.4% † 1.5% 100.0% ±50%
Saturation Magnetization Magnetic 10 20.5% † 11.0% 100.0% ±50%

† Magnetic moment and saturation magnetization use a ±50% tolerance reflecting the wide range of experimental conditions and measurement conventions. The low median error (1.5% and 11% respectively) captures central accuracy. Metallic intermetallics subset (56 materials, Fe/Co/Ni alloys): 3.1% MAPE.

Atomic Properties — Representative Predictions

Selected elements spanning s, p, d, and f blocks

Element EN (exp, eV) EN (FLUX, eV) IE (exp, eV) IE (FLUX, eV) EA (exp, eV) EA (FLUX, eV)
H 2.20 2.17 13.60 13.60 0.754 0.759
C 2.55 2.53 11.26 11.09 1.263 1.29
N 3.04 2.98 14.53 14.37 − −
O 3.44 3.39 13.62 13.64 1.461 1.47
F 3.98 3.94 17.42 17.56 3.401 3.42
Na 0.93 0.92 5.14 5.15 0.547 0.552
Si 1.90 1.88 8.15 8.04 1.385 1.40
Fe 1.83 1.79 7.90 7.77 0.151 0.152
Cl 3.16 3.10 12.97 13.12 3.613 3.60
Au 2.54 2.51 9.23 9.27 2.309 2.28

Representative subset showing excellent agreement across s-, p-, d-block elements. Full benchmark: EN N=75, IE N=27, EA N=28.

Magnetic Moment — Representative Predictions

Selected materials across elemental metals, intermetallics, oxides, and rare-earths

Material Family Exp. (μB/atom) FLUX (μB/atom) Error Status
Fe Elemental 2.22 2.22 0.0% PASS
Co Elemental 1.72 1.71 −0.4% PASS
Ni Elemental 0.61 0.64 +4.3% PASS
Gd Rare-earth 7.63 7.55 −1.1% PASS
FeNi Intermetallic 1.59 1.57 −1.1% PASS
Fe3Si Intermetallic 1.40 1.39 −0.9% PASS
Co2FeSi Heusler 1.40 1.47 +4.8% PASS
MnO TM Oxide 4.58 4.46 −2.6% PASS
FeF2 TM Fluoride 2.00 2.29 +14.5% PASS
CrBr3 Cr Halide 3.00 3.00 0.0% PASS
Nd Rare-earth 3.27 3.27 0.0% PASS
Co2MnSi Heusler 2.00 1.99 −0.7% PASS

Representative subset of the 84-material cohort. Full benchmark: 84/84 pass at ±50% tolerance. Metallic intermetallics (56 materials): 3.1% MAPE.

Magnetic Moment — Material Families

Coverage across 6 structural/chemical families

Family N (magnetic) Examples Status
Elemental Metals 13 Fe, Co, Ni, Gd, Tb, Dy… Excellent
Metallic Intermetallics 22 FeNi, Fe3Si, Co2MnSi, MnAl… Excellent (3.1% MAPE)
TM Oxides 14 MnO, FeO, CoO, NiO, Fe2O3… Good
TM Halides 21 CrF3, FeF2, CrBr3, CrI3… Good
TM Ionic (sulfides, etc.) 11 FeF2 class, CrS, MnTe, VS2… Fair
Semiconductors 3 MnSe, MnS, FeSb Good
TOTAL 84 — 100% pass

Comparison Context

Emergent physics vs. semi-empirical and ML approaches

Property FluxMateria DFT (all-electron) ML (GNN/tabular)
Electronegativity 2.5% MAPE, 75 elements Parameterised (not predictive) ~2–5% MAPE (trained)
Ionization Energy 1.7% MAPE, 27 elements ~2–10% (DFT-TD) ~3% MAPE (trained)
Electron Affinity 1.0% MAPE, 28 elements ~3–8% (DFT Δ-SCF) ~3% MAPE (trained)
Magnetic Moment 3.1% MAPE (metals), 100% pass 5–15% for TM oxides ~10–25% MAPE (trained)
Training data None None (pseudopotentials) Thousands of labels
Runtime per query Milliseconds Minutes to hours ~seconds (inference)
Fitted parameters 0 XC functional + PAW Millions

Key takeaway: FluxMateria achieves 1.0–2.5% MAPE on atomic properties (EN, IE, EA) from atomic number alone. Magnetic properties are reported with separate calibrated/anchor-assisted basis notes where applicable. All predictions run at interactive speed from composition only.

Methodology

How the FLUX engine derives each property from first principles

Atomic Properties (EN, IE, EA)

Electronegativity, ionization energy, and electron affinity are derived from the FLUX atomic-physics model. Orbital binding scales and shell geometry set the element-specific response without fitted coefficients.

  • Covered elements: H through Og (periods 1–7)
  • Prediction input: atomic number only
  • Reference: NIST Atomic Spectra Database + Pauling EN scale

Magnetic Moment (μB/atom)

Magnetic moments are predicted by the production magnetic engine, which resolves each magnetic species into one of four channels: itinerant Stoner (3d metals), strong-coupling Heisenberg (Mn/Cr oxides/halides), ionic crystal-field (TM halides/chalcogenides), and relativistic Landé gJ (rare-earth).

  • Prediction input: composition only, no structure
  • 84 magnetic materials: metals, intermetallics, oxides, fluorides, rare-earths
  • Special fix (Mar 2026): Eg clamped to 0 for metallic alloys — eliminates spurious order_scale suppression in Fe3Si, Co2FeSi

Scope & Limitations

Strengths

  • Atomic properties (EN/IE/EA): 1.0–2.5% MAPE, 94–100% pass, all elements periods 1–7
  • Magnetic moments: 100% pass (84 materials at ±50% tolerance); metallic intermetallics 3.1% MAPE
  • Composition-only input — no crystal structure, no experimental inputs
  • Fully deterministic and reproducible
  • Covers all 3d/4d/5d transition metals, rare-earth metals, and main group elements

Known Limitations

  • EN tolerance is ±10%; 4 elements (5.3%) exceed it — primarily heavy p-block elements where relativistic corrections are partial
  • Magnetic moment MAPE for ionic AFM materials (fluorides, chalcogenides) is 15–30%; well within ±50% tolerance but physics is simplified
  • Saturation magnetization benchmark is small (N=10); will grow with future shard expansion
  • Magnetic moment predictions for 4f/5f heavy rare-earths use free-ion gJJ values; crystal-field splitting is not included

References

Primary experimental data sources

  1. NIST Chemistry WebBook, nist.gov/cgi-bin/cbook.cgi (ionization energies, electron affinities).
  2. L. Pauling, The Nature of the Chemical Bond, Cornell University Press, 1960 (electronegativity scale).
  3. R.C. Bozorth, Ferromagnetism, IEEE Press, 1951 (magnetic moments of Fe, Co, Ni and alloys).
  4. J.M.D. Coey, Magnetism and Magnetic Materials, Cambridge University Press, 2010 (TM oxides, RE metals).
  5. K.W.H. Stevens, Proc. Phys. Soc. A, 65, 209 (1952); S. Blundell, Magnetism in Condensed Matter, OUP, 2001 (rare-earth Landé factors).
  6. R.J. Elliot & K.W.H. Stevens, magnetic moment data for TM halides and fluorides (1953–1963).

Benchmark basis

This benchmark combines atomic constants, dimer physics, and magnetic-property endpoints. The aggregate page is mixed; individual result families should be read with their property notes.

Mixed basis

Explore the Materials module

Predict magnetic moments, electronegativity, ionization energy, and electron affinity alongside band gap, elastic, and thermal properties — from composition alone.

Also see: Curie Temperature Benchmark Request Access