Validation
One measured check: a stored ChemDFT job compared with the same Quantum ESPRESSO input run directly. It is the only result the site compares with an independent run, and only within the limits below. Live platform checks of the hosted service are listed after the limits.
What was run
α-Cr2O3 (rhombohedral, 10 atoms), antiferromagnetic, Quantum ESPRESSO 7.6, pseudopotentials SSSP PBE Efficiency v2.0.
| Functional | PBE+U (Dudarev), U_eff(Cr) = 2.0 eV |
|---|---|
| Cutoffs | ecutwfc 45 Ry, ecutrho 540 Ry |
| k-points | Γ only |
| Smearing | Fermi-Dirac, 0.01 Ry |
| Platform | 2 vCPU AWS EC2 pilot machine |
| Container image | chemdft-pilot:runtime-qe-0046 (sha256:83d8d131…c77a) |
| Input file sha256 | 6d8b931c13e66180ba43199c0454ddc9d6468404ced9a3f93ac8d32191182432 |
What was compared
Job 44f6ed72, run id 65106558-cff4-40b7-8890-1e3f43b79369.
| Quantity | Through the live site | Same pw.x input, run directly in the same image |
|---|---|---|
| Total energy | -476.0906515824172 Ha | -952.18130316 Ry (= -476.09065158 Ha) |
| SCF iterations | 16 | 16 |
| Fermi energy | 12.95961 eV | 12.9596 eV |
| Löwdin Cr moments (μB) | -3.0028, -3.0028, +3.0028, +3.0028 | the same four values |
| Wall time of pw.x | 57.95 s | — |
An earlier stored job, 58a06173, with the same settings gave -476.0906515824173 Ha, 16 iterations and the same moments.
Limits
- The native run used ChemDFT's own input generator to write pw.in. This verifies execution, parsing and reproducibility. It does not verify that the generated input is the right physics.
- No comparison against another code or against the literature is claimed.
- This is one structure on one platform. A result from one platform is not evidence for another.
- The only result compared against an independent run of the engine is the one above. The live platform checks below are execution checks, not independent verification.
Live platform checks, 2026-11-14 (release 0.0.53)
Run by the owner on chemdft.com. Each result below passed the Result Firewall (CALCULATED and VALIDATED). Nothing here was independently verified, and each result belongs to this platform (AWS pilot, PySCF 2.14.0) only.
- H2, HF/STO-3G, through the AI-planned path (plan mode, approval card, approve): calculation d0c51937, PySCF 2.14.0, -1.11676 Eh (rounded to 6 digits).
- H2 geometry optimisation with PySCF and geomeTRIC, started manually: job 6408f6c2, converged in 4 evaluations, new structure "H2 - optimized V2".
Not done or open
- Hosted geometry optimisation is switched on by an owner decision. geomeTRIC 1.1.1 is still in the deployed compute image, and whether its no-AI-training clause (clause 4) applies is an open question for a lawyer.
- Not tested live: an AI-proposed geometry optimisation, an AI-proposed xTB calculation (plan mode offers single-point energy only), and any independent verification of these two results.
See Principles and the engine status in the Engines section.