M1 · Open baseline
Done 1 Oct 2026
Goal
Replace the toy physics with real open-source codes, wired into the coupling loop, and show that the result reproduces a published benchmark. After M1, a surrogate has something real to stand in for, and a trusted baseline to be compared against.
Acceptance criteria (fixed in the brief before work started)
- The benchmark matches the published profiles within an agreed tolerance.
- Energy and particle balance are conserved to tolerance.
- Gradients are finite and checked against finite differences.
- Human gate: physicist review.
The tolerance in (1) was set at the physicist review: the t = 10 s profiles within 2% of RAPTOR (densities 0.1%) and the time traces within 5%. Conservation in (2) must hold to 10-6 per step on converged steps (OQ-5). The physicist review passed on 1 Oct 2026, with the owner as reviewer of record (decisions, entry 13). The claims it judged are marked reviewed in the claims register.
Tasks
| task | state | log |
|---|---|---|
| coupling loop with Newton-Raphson, Picard kept as fallback | done | entry 1 |
| wrap TORAX; ITER hybrid reference runs with provenance | done | entry 2 |
| fixed-step loop driving TORAX’s own time step; time-convergence study | done | entry 3 |
| QLKNN behind the guard, inside TORAX | done | entry 4 |
| analytic Gaussian RF heating | done | entry 5 |
| wrap FreeGSNKE; MAST-U free-boundary equilibrium | done (standalone) | entry 6 |
| ASCOT5 sources | done: fusion-alpha heating from ASCOT5 into TORAX, checked against TORAX’s local model (71.8 MW vs 70.5 MWC-064, 6400 alphas); orbit tolerance 10-7 (0.348 at 10-6, 0.331 at 10-7C-058, electrons -0.16%, ions +0.08%C-066). The production run (OQ-13) finished on 2 Oct 2026 | entries 10, 12, 15 |
| reproduce a published benchmark | ITER L-mode case of the TORAX paper against RAPTOR and the paper’s TORAX run (OQ-12); done, within the tolerance set at the physicist review | entry 11 |
| wrapper check on the ITER hybrid scenarios | done | entry 3 |
| energy and particle balance on the benchmark | done; meets the agreed 10-6 per step (OQ-5) | entry 8 |
Against the criteria
| criterion | where it stands |
|---|---|
| benchmark within tolerance | published ITER L-mode case, NRMSD against RAPTOR at t = 10 s: T_i 1.23% (1.1%), T_e 0.28% (0.71%), n_e 0.03% (0.03%), psi 0.49% (0.41%), q 0.69% (0.70%)C-050; against the paper’s own TORAX run: T_i 0.13%, T_e 0.50%, n_e 0.01%, psi 0.08%, q 0.16%; Ohmic power density 3.6%C-051; time-step convergence of those numbers: at most 0.006 percentage pointsC-052. Loop vs TORAX’s own driver: better than 10-13C-049. Passes the tolerance set at the review |
| conservation | every step’s balance rebuilt from TORAX’s own equations: on the published benchmark energy 1.1 × 10-7, particles 3 × 10-13C-054; energy better than 10-6C-035, particles better than 10-10C-036 on the ITER hybrid ramp-up; the flat-top’s cheaper solver leaves about 0.8%C-037 of the energy unaccounted for. Conversion checks: stored energy 10−16C-009; heating power better than 10−8C-010 |
| gradients | through three coupled TORAX steps: 10−6C-008 against finite differences (linear solver); through three Newton steps of the published benchmark: 6 × 10-9C-055; forward mode equals reverse mode (2 × 10-15 (NaN before the fix)C-062) |
| physicist review | passed 1 Oct 2026 (entry 13) |
Deviations
- The benchmark is run through TORAX’s own time step, called from the Tokamak Toolkit’s fixed-step loop, rather than by re-implementing TORAX’s physics in the Tokamak Toolkit’s own solver. Same loop shape, far more physics. See coupling loop.
- The equilibrium is standalone. FreeGSNKE solves MAST-U equilibria, but they don’t yet feed the transport loop: that needs a transport model valid on a spherical tokamak, which QLKNN is not (M2, M3).
- The published benchmark is an L-mode case with constant transport (OQ-12), because it is the only TORAX benchmark published against an independent code. QLKNN and the pedestal are exercised on the ITER hybrid scenarios, which have no independent published profiles.
- The analytic RF model is TORAX’s, wrapped with a typed entry point and tests, rather than a second implementation.
Future extensions
Extending the benchmark to a STEP power-plant flat-top is planned as a future extension (OQ-7, deferred). For now the work stays with the ITER benchmark and MAST-U.