tutorial.md

Running Gyacomo Tutorial

This tutorial provides guidance on running Gyacomo in a new installation. It is designed to be used in conjunction with the new_prob.sh script, providing a refresher on the basics of Gyacomo commands. It is meant to be followed once sh new_prob.sh has been run in /gyacomo which creates an example problem directory /gyacomo/simulations/problem_01. The following commands are expected to be run from the example problem directory.

How to Run the Code Locally After Executing new_prob.sh in /gyacomo Directory?

• For a single-core run, execute:

  ./gyacomo.exe

• For a multi-core run, use:

  mpirun -np N ./gyacomo.exe np nky nz

  Here, N is the total number of processes, np is the number in the p direction (Hermite polynomials), nky is the number of poloidal wavenumbers, and nz is the number of parallel planes. The relation N = np x nky x nz must be satisfied. In both commands above, the input parameters are expected to be in a fort.90 file present in the directory where the code is run. You can append an additional integer X at the end of both single and multi-core runs, which points to an input file fort_XX.90 where XX is the two-digit version of the number X. This allows for clearer restarts by indexing each consecutive run.

Examples

• Single-core run with parameters located in fort_00.90:

  ./gyacomo.exe 0

• Multi-core run with 2 cores in Hermite, 4 in ky, 1 in z, and reading the file fort_00.90:

  mpirun -np 8 ./gyacomo.exe 2 4 1 0

• Same as above but redirecting the standard terminal output to the file out_00.txt and running in the background:

  mpirun -np 8 ./gyacomo.exe 2 4 1 0 > out_00.txt &

  You can monitor the simulation progress with:

  tail -f out_00.txt

How to Stop the Simulation?

Here are the stopping conditions of GYACOMO:

• A nan (overflow) is detected in any field, this can be caused by a too high dt (CFL condition) or an unsufficient resolution (Nkx, Nky, Nz)
• The number of steps exceeds the number provided in nrun.
• The simulation run time exceeds the time defined in maxruntime.
• The simulation reaches the maximal physical time tmax.
• A file named mystopis present in the current directory. Thus, you can smoothly stop your current simulation by creating an empty file named mystop in the directory where the code runs (here gyacomo/simulations/problem_01):

touch mystop

The code checks if such a file exists every 100 steps (this is set in the gyacomo/src/tesend.F90 module). If it finds it, it will run the ending procedure and remove the file from the directory.

Checking the Output

Once the simulation is finished, the Python script minimal_analysis.py provides a basic example of result analysis. If you followed the tutorial, you should be able to obtain plots by typing:

python minimal_analysis.py

Restarting a Simulation

In the parameter file, the parameter job2load defines from which previous outputs the code has to continue the simulation. If job2load=-1, a new start is made, and the output is located in the outputs_00.h5 file in the current directory.

If the outputs_XX file exists, you can continue the run by setting job2load=X (where XX is the double-digit version of the integer X). It's recommended to link all runs with an input file fort_XX.90. In any case, you can find the input file used in the simulations in outputs_XX.h5, located in outputs_XX.h5/files/STDIN.00.

You can use minimal_analysis.py X to analyze the restart. The script will here look for the outputs_XX.h5 file.

Example

This is a minimal example for a restart procedure

1. Copy the fort_00.90 to a new fort_01.90 input file.
2. Adapt the parameters of fort_01.90 (Tmax, grads, anything).
3. IMPORTANT: Ensure that fort_01.90 has job2load = 0 (otherwise, it will restart a simulation from 0).
4. Now you can run:

   mpirun -np 8 ./gyacomo.exe 2 4 1 1 > out_01.txt &

   Note the 1 input in the fourth position; the code reads fort_01.90. We also direct the std output to a new file out_01.txt
5. You can monitor the run with:

   tail -f out_01.txt

6. Once the simulation is done (end or mystop), you can analyze the new data using:

   python minimal_analysis.py 1

Running on Marconi

You can adapt and use the provided template submit_marconi.cmd.