ASCOT5 second cluster run: a tighter orbit tolerance, and a production run lost to a time limit
Entry 10 computed the fusion-alpha heating with ASCOT5 and fed it back into TORAX (see the Primer). Two things were left open: whether the orbit-integration tolerance was tight enough, and how the heating is distributed in radius, which the first run’s sample of alphas was too small to resolve. The owner approved a second cluster job for both (OQ-11): ten times more alphas for the radial profile, and a smaller sample at a tolerance ten times tighter as the check.
What happened
The tolerance check worked. The production run did not. ASCOT5 stops any alpha that has used more than a set amount of computing time, as a guard against runaway markers. We set that limit from the average cost per alpha in the first job. ASCOT5, however, advances alphas in batches and charges every alpha the time of its whole batch, so the limit stopped 2823 of 6400 markers, 18% of the birth powerC-057. The heating they had not yet delivered is missing from the profiles, so the run is discarded. The error was ours, in reading what the limit measures. The Tokamak Toolkit now sets no such limit (the cluster’s job time limit bounds the run instead), records the undelivered share with every profile, and refuses to pass profiles to TORAX when that share exceeds one percent.
What the check showed
The tighter tolerance, in a sample of the same size as the first job’s, changes the share of the heating that goes to ions: 0.348 at 10-6, 0.331 at 10-7C-058. The share going to electrons does not change, and the energy accounting closes four times better. So the first job’s tolerance was not tight enough for ion heating, and the tighter one is now the working value. The first job’s analysis had put part of the energy-accounting error down to how ASCOT5’s output is binned; it was mostly the orbit tolerance, and the technical page now says so.
| ASCOT5 at the tighter tolerance vs TORAX’s own alpha model | |
|---|---|
| total alpha heating | 72.2 MW vs 70.5 MWC-059 |
| effect on TORAX after two seconds, heating held fixed | +0.03% in stored energy (central T_e +0.5%, T_e and T_i profiles 0.6%)C-060 |
| alphas stopped by the time limit in the discarded production run | 2823 of 6400 markers, 18% of the birth powerC-057 |
The total heating and its split between electrons and ions agree with TORAX’s simpler model to a few percent. Where in radius the heat lands is still not resolved: the heating’s centre moved from slightly inside TORAX’s to slightly outside it between the two samples, which is sampling noise. Resolving it needs the production run.
Where this stands: M1’s ASCOT5 task is done, with totals checked against TORAX. A production run that resolves the radial profile awaits a decision (OQ-13); the recommendation is not to hold the M1 physicist review for it.
Technical details → Backends: fusion-alpha heating
Technical details → Backends › ascot5 alpha
Decisions → OQ-11 · OQ-13