FPE with simplified-SDC + Detonation

[zingale]

Build the detonation problem as:

make USE_SIMPLIFIED_SDC=TRUE -j 20

run as:

./Castro1d.gnu.SMPLSDC.ex inputs-det-x.simplified_sdc amrex.fpe_trap_invalid=1

and it will crash after 26 steps with an FPE

Curiously, if you run it via valgrind as:

valgrind ./Castro1d.gnu.SMPLSDC.ex inputs-det-x.simplified_sdc amrex.fpe_trap_invalid=1

then it runs for 450+ steps (I killed the job at that point).

[zingale]

after all the recent changes, I now run to 420 steps before getting the FPE

[zingale]

with PR #1607 we FPE after 132 steps

Clearly we are generating a bad state at some point in the VODE stepping

[zingale]

in particular, the FPE is happening in sneut5 where we are coming in with:

rho, T = 1e-12 1e+10

[zingale]

okay, so this is the problem of too much advection again. The issue is that Castro wants to try a burn of 1.277229732e-07 s, but VODE is internally trying a timestep of 0.00399574. So the advective terms evacuate the cell. We need to restrict VODE to not try a timestep that is too large.

[zingale]

this was fixed with the Microphysics PR (643) linked above. I was able to run the example problem for 2000 steps (evolving to 1.87e-4 s).