Feature/thrift vmec restart

[lazersos]

This pulls request does the following:

• Merges changes from ORNL-fusion/PARVMEC
• Fixes VMEC restart feature to work when more than one processor is requested
• VMEC restarting is implemented in THRIFT.

[ajchcoelho]

When running this branch with the input files in BENCHMARKS/THRIFT_TEST (profiles_analytical.h5 and input.analytical), the code crashes with a segmentation fault error at t=12.5s, substep=12.

I tracked it down to line 38 of file thrift_run_equil.f90. The phi variable suddenly is not allocated. I don't understand how it runs for more than 12 time iterations and then this error pops up.

[lazersos]

I've added a feature to allow the user to control use of the VMEC 'reset' feature -vmec_reset.

[lazersos]

The issue you're experiencing is because the VMEC equilibria is failing to converge. At which point ier_paraexe=-1, which we don't handle. From that point on the code basically stops calculating VMEC equilibria which eventually leads to a crash. A check needs to be implemneted so that once VMEC fails to converge, the code exits out with the proper message.

[lazersos]

The code has been updated so that if VMEC fails to converge in reset mode then it'll attempt a second run in the normal mode of operation.

[lazersos]

@ajchcoelho I've updated the code so that the wout_reset_file.nc files are only written after THRIFT successfully converges on a total current. This seems to fix the issue I had where at t=2.5 s in the BENCHMARKS case the code would seemling fail to converge the equilibria. The idea being here that wout_reset_file.nc should always be the previous timestep not the previous iteration. I also added some initializations which VMEC wasn't doing.

[ajchcoelho]

Looks good. I've made a comparison for the analytical case, with restart and w/o restart at different times:

In terms of timing, as already expected there is no time improvement: THRIFT spends 25 minutes running the equilibrium code (VMEC) in both situations (with and without restart mode)