ASCOT5 fusion-alpha heating for the ITER flat-top: first cluster run
Fusion reactions in a deuterium-tritium plasma produce helium nuclei (alpha particles) at 3.5 MeV. They heat the plasma as they slow down. TORAX models this locally: the heating appears where the reaction happened, split between electrons and ions by a formula. ASCOT5, an open-source orbit-following code, follows each alpha along its orbit until it slows down or leaves the plasma. M1 asks for ASCOT5 sources in the coupled simulation (see the Primer). The owner approved fusion alphas first and one cluster job (OQ-9).
Method
We take the TORAX state at the end of the ITER hybrid flat-top run, where the alpha heating is highest. ASCOT5’s fusion-source module computes the alpha birth distribution from TORAX’s temperature and density profiles. Its birth rate agrees with TORAX’s to 1.002C-042 (ratio). ASCOT5 then follows a sample of alphas (markers) with Coulomb collisions until each one slows to twice the local ion temperature or crosses the separatrix. The heating to electrons and ions is taken from where the markers spent time and at what energy. The profiles are handed back to TORAX as a prescribed heating source.
The job ran 6400 markers at a guiding-centre orbit tolerance of 10-5, chosen because it was seven times faster than tighter settings on a short test. It also reran 640 markers at 10-5 and at 10-6 as a check.
Result
The check failed the production setting:
| 640 markers, same in both runs | result |
|---|---|
| markers lost across the separatrix | 18% at 10-5, 0.6% at 10-6C-043 |
| heating of ions | 13.2 MW at 10-5, 25.0 MW at 10-6 (TORAX: 24.8 MW)C-044 |
| total heating at 10-6, ASCOT5 vs TORAX’s local model | 73.6 MW vs 70.5 MWC-045 |
At 10-5 the integration error moves alphas outward. Many cross the separatrix before they have slowed down, and ions, which take most of their heating late in the slowing-down, receive half as much. The total heating changed much less than the ion heating, so the total alone would not have shown the error. We discarded the 6400-marker run and made 10-6 the default.
At 10-6 the two models give the same split between electrons and ions, ASCOT5’s heating is slightly more peaked, and its total is higher by the amount in the table. Most of that difference is numerical (alpha energy convention, the deposition calculation, interpolation onto TORAX’s cells), not orbit physics. With 640 markers the profile is noisy cell by cell, so this run compares totals, not shapes.
We then continued TORAX from that state three times, over the same interval. With ASCOT5’s profiles held fixed, stored energy differs by -0.005%C-046 from a run with TORAX’s own profiles from the same state held fixed. Holding the heating fixed at all, instead of letting TORAX update it, changes stored energy by -0.3% (central T_e -1.0%)C-047. That is the cost of updating ASCOT5 once per interval rather than every step.
Where this stands: M1 in progress. The ASCOT5 source works end to end. A second job with more markers and a 10-7 convergence check is awaiting a decision (OQ-11).
Technical details → Backends: ASCOT5
Technical details → Backends › ascot5
Decisions → OQ-9 · OQ-11