Open zingale opened 2 months ago
closes #729
some Strang data using burn_cell
. Build as:
make NETWORK_DIR=subch_simple
then run the old way:
./main3d.gnu.ex inputs_subch_simple integrator.jacobian=1 unit_test.nsteps=1 integrator.correct_jacobian_for_const_e=0
and it take 8084 steps.
Running with the new correction, integrator.correct_jacobian_for_const_e=1
and it takes
6788 steps
curiously, with aprox21
, it goes the other way, and we are slower
for He-C-Fe-group, using the correction terms speeds things up by 4x
./main3d.gnu.ex inputs_subch_simple integrator.jacobian=1 unit_test.nsteps=1 integrator.correct_jacobian_for_const_e=1 unit_test.tmax=1.e-2
104499 without the corrections 24735 with the corrections
Here's some test_react
runs:
aprox13
with new terms:
Run time = 2.016436636
min number of rhs calls: 6
avg number of rhs calls: 96
max number of rhs calls: 1171
min number of steps: 3
avg number of steps: 75
max number of steps: 878
without new terms:
Run time = 2.397303023
min number of rhs calls: 6
avg number of rhs calls: 116
max number of rhs calls: 3932
min number of steps: 3
avg number of steps: 94
max number of steps: 3767
subch_simple
(using inputs_aprox13
):
with new terms:
Run time = 101.0353978
min number of rhs calls: 6
avg number of rhs calls: 3962
max number of rhs calls: 147079
min number of steps: 3
avg number of steps: 2219
max number of steps: 79369
without new terms:
Run time = 100.0271792
min number of rhs calls: 6
avg number of rhs calls: 3975
max number of rhs calls: 157375
min number of steps: 3
avg number of steps: 2284
max number of steps: 86380
here's a Castro run with these changes: http://groot.astro.sunysb.edu/Castro/test-suite/gfortran/2024-07-12-002/index.html
this is the dT/dX |_e term that appears in all species it is now controlled by integrator.correct_jacobian_for_const_e