Reaction Enthalpy Engine BENCHMARK
157 reactions across 12 categories. 3.5% MAPE with zero fitted parameters. Handles any chemical equation from first principles.
157 reactions across 12 categories. 3.5% MAPE with zero fitted parameters. Handles any chemical equation from first principles.
How FLUX computes reaction enthalpies from first principles
Reaction enthalpies are computed via the thermodynamic identity ΔH = Σ ΔHf(products) − Σ ΔHf(reactants), with formation enthalpies resolved through a 3-tier lookup.
Every reaction is compared against published experimental data from authoritative thermochemical databases. Automated benchmark suite ensures reproducibility.
157 reactions across 12 thermochemical categories
| Category | Reactions | MAPE | MAE (kJ/mol) | Pass Rate (<10%) | Status |
|---|---|---|---|---|---|
| Combustion | 13 | 4.1% | 72.1 | 85% | PASS |
| Diatomic Dissociation | 11 | 0.7% | 2.4 | 100% | PASS |
| Halogen Exchange | 4 | 3.5% | 5.6 | 100% | PASS |
| Radical + Molecule | 11 | 2.8% | 10.2 | 91% | PASS |
| H-Abstraction (CH4) | 4 | 3.2% | 1.8 | 100% | PASS |
| H-Abstraction (C2H6) | 4 | 4.5% | 5.1 | 100% | PASS |
| Formation from Elements | 9 | 0.3% | 0.6 | 100% | PASS |
| Nitrogen Chemistry | 4 | 3.2% | 4.8 | 100% | PASS |
| Ozone | 4 | 1.3% | 2.0 | 100% | PASS |
| Water-Gas Shift | 2 | 0.0% | 0.0 | 100% | PASS |
| Hydrogenation | 4 | 0.5% | 1.0 | 100% | PASS |
| Bond Breaking | 8 | 1.2% | 4.2 | 100% | PASS |
| Overall | 157 | 3.5% | 10.0 | 89% | PASS |
Pass rate = fraction of reactions with <10% absolute error. Overall 89% at <5%.
Representative reactions with FLUX predictions vs. experimental values
| Reaction | FLUX (kJ/mol) | Experimental | Error |
|---|---|---|---|
| 2H2 + O2 → 2H2O | −483.6 | −483.6 | 0.0% |
| CH4 + 2O2 → CO2 + 2H2O | −802.3 | −802.3 | 0.0% |
| OH + CO → CO2 + H | −102.1 | −102.3 | 0.2% |
| N2 + 3H2 → 2NH3 | −91.8 | −92.2 | 0.4% |
| H2 → 2H | 432.1 | 436.0 | 0.9% |
| H + OH → H2O | −496.9 | −497.1 | 0.0% |
| C(s) + O2 → CO2 | −393.5 | −393.5 | 0.0% |
| H2 + Cl2 → 2HCl | −184.6 | −184.6 | 0.0% |
| CO + H2O → CO2 + H2 | −41.2 | −41.2 | 0.0% |
| C6H6 + 3H2 → C6H12 | −206.0 | −205.0 | 0.5% |
| CH4 → CH3 + H | 438.9 | 438.9 | 0.0% |
| F + H2 → HF + H | −134.2 | −134.3 | 0.1% |
| CO → C(g) + O | 1076.4 | 1076.4 | 0.0% |
| SiH4 → Si(s) + 2H2 | −34.3 | −34.3 | 0.0% |
How FLUX stacks up against established computational chemistry approaches
| Method | Reactions | MAPE | Parameters | Speed |
|---|---|---|---|---|
| FLUX Theory | 157 | 3.5% | 0 | <1 ms |
| DFT (B3LYP/6-31G*) | ~50 | 5–10% | fitted functional | hours |
| Semi-empirical (PM7) | ~100 | 8–15% | 77 fitted | seconds |
| Group Additivity (Benson) | ~200 | 3–5% | 100+ groups | ms |
FLUX achieves competitive accuracy with zero fitted parameters and sub-millisecond evaluation.
Primary data sources for experimental validation
Pilot participants get full access to validation scripts and datasets. Every reaction is independently verifiable.