FluxMateria and PBE: measured comparison
Band-gap MAPE 10.3% versus 37.5% for the specified PBE/GPAW setup on the same 10 materials with positive experimental gaps. FluxMateria has 3.66× lower MAPE in this comparison.
Band-gap MAPE 10.3% versus 37.5% for the specified PBE/GPAW setup on the same 10 materials with positive experimental gaps. FluxMateria has 3.66× lower MAPE in this comparison.
| Quantity | FluxMateria measurement | Cohort and comparison |
|---|---|---|
| Band gap MAPE | 10.3% | 10 positive-gap materials; PBE/GPAW 37.5% on the same experimental references. |
| Band gap MAE | 0.651 eV | All 15 materials, including zero-gap references. |
| Lattice constant | Median absolute percentage error 0.64%; full-cohort MAPE 6.6% | 15 materials. MAPE 0.61% on the 13 non-layered cells; layered cells require structural input. |
| Magnetic moment | MAPE 3.64% | Fe and Ni; two materials. This small panel retains its own scope. |
FluxMateria calculates more than 40 material properties from a chemical formula, covering electronic, structural and mechanical quantities. Each property has its own input requirements and benchmark evidence.
For this comparison, the FluxMateria inputs are composition on its declared property path. PBE/GPAW uses an explicit crystal structure, plane-wave cutoff 200 eV and 6³ k-points. The comparison evaluates that specified screening setup; it does not cover every density functional or electronic-structure method.
The FluxMateria measurements use the evaluated production configuration of 23 October 2026. The panel has 15 materials; percentage gap errors use its 10 positive-gap references. All denominators are retained. The PBE and experimental references are fixed across the comparison.
Choose the properties, input conditions and scoring criteria for your application. A frozen system and partner-selected cases provide a direct product evaluation.
Acceleration depends on the calculation being compared
The benchmark timing compares a single FluxMateria composition call (approximately 3 ms after initialization) with a seven-point GPAW-PBE equation-of-state scan (11–294 seconds per material; median 50 seconds). The per-material acceleration is of order 104. Across all 15 materials, the recorded totals are 1.4 seconds for FluxMateria and 22.1 minutes for GPAW-PBE, approximately 950× with initialization included.
These timings compare the stated prediction and SCF workflows. They do not measure all DFT settings or replace a timing of a complete application report; the accuracy replay and the timing benchmark are separate measurements.