Turbulence driving broken - Decaying turbulence

[saisandeepdammati]

Howdy!

I am using athena++ from the master brach with latest git commit ca8ba57 and trying to run a simple decaying turbulence problem (using src/pgen/turb.cpp) in a periodic cube of size 0.1125 cm on a 128^3 uniform grid on my local macbook on single core with one meshblock. I see that the code runs for 38 cycles and the then time-step drops drastically to very small value (1.e-22) and the run breaks. The only changes I made in turb.cpp is that I read the value of density and pressure from the of the athinput file as shown below in MeshBlock::ProblemGenerator(ParameterInput *pin) function. In history output, I can see that after cycle 38 suddenly total energy in the domain blows up. I am not sure what is the source of the error.

The configuration script and athena configuration for the run are shown below. I am also attaching the job file, the history output file and stdout as .txt files.

Any input is highly appreciated.

------------------ CHANGE IN TURB.CPP ----------------------------------
phydro->u(IDN,k,j,i) = pin->GetReal("problem","d");
if (NON_BAROTROPIC_EOS) { phydro->u(IEN,k,j,i) = pin->GetReal("problem","pr")/(peos->GetGamma()-1);

------------------ CONFIGURATION SCRIPT ----------------------------------
python3 configure.py --prob=turb -mpi --eos=adiabatic --flux=hllc --nghost=2 -hdf5 -fft --fftw_path=/usr/local --hdf5_path=/usr/local/hdf5-install

------------------ ATHENA CONFIGURATION ----------------------------------

Your Athena++ distribution has now been configured with the following options:
Problem generator: turb
Coordinate system: cartesian
Equation of state: adiabatic
Riemann solver: hllc
Magnetic fields: OFF
Number of scalars: 0
Special relativity: OFF
General relativity: OFF
Frame transformations: OFF
Self-Gravity: OFF
Super-Time-Stepping: OFF
Debug flags: OFF
Code coverage flags: OFF
Linker flags: -L/usr/local/lib -L/usr/local/hdf5-install/lib -lfftw3 -lhdf5
Floating-point precision: double
Number of ghost cells: 2
MPI parallelism: ON
OpenMP parallelism: OFF
FFT: ON
HDF5 output: ON
HDF5 precision: single
Compiler: g++
Compilation command: mpicxx -I/usr/local/include -I/usr/local/hdf5-install/include -O3 -std=c++11

hit.decaying.job.txt
HIT.n128.Decaying.vl2.hst.txt
HIT.n128.Decaying.vl2.out.txt

[c-white]

I see the following big changes from the standard problem:

• The CFL number has been raised from 0.3 to 0.45.
• The adiabatic index is about 1.2 rather than 5/3.
• Instead of density and pressure both being 1, they are now about 1e-3 and 1e6, so the temperature has changed from 1 to 9e8. If this is trying to make the code use CGS internally, then I'll warn that this is usually a bad idea. There are hard-coded small and large values used in traps for bad numerical behavior all throughout the code, so it's best to scale the problem to have order-unity numbers being evolved.
• The problem is 128^3 instead of 32^3.
• nlow, nhigh, and expo are no longer specified. Someone more familiar with the turbulence module can comment, but I can only imagine this results in driving all the way down to the grid scale.

[saisandeepdammati]

Hi Chris,

Thanks for your input. I figured that it is an issue of numerical stability. I repeated the same run with a CFL of 0.3 and it works fine.