Homec4science
Diffusion GYACOMO (OBSOLETE: LATEST VERSION ON gitlab.epfl.ch) (half_ky_plane) Browse

gyacomo/README.md
663259d809f5half_ky_plane

Last Change
Raw File

README.md
View Options

# HeLaZ (Hermite-Laguerre Z-pinch solver, 2020)
To compile it check INSTALLATION.txt
How to run it
1. Be sure to have correct paths in local/dirs.inc for the different libraries
2. Compile from HeLaZ/ using make
3. To run the code, use HeLaZ/wk/local_run.m and run it to set up the parameters and the results folder
4. Then go to the results folder and launch HeLaZ using mpirun -np num_procs ./../../../bin/helaz num_p num_kr
5. You can obtain various plots and gifs using HeLaZ/wk/analysis_2D.m once the simulation is done. To select the correct output file, run parameters*.m with the corresponding simulation parameters and then run analysis_2D.m (everything with matlab from wk/)
// Comment : For some collision operators (Sugama and Full Coulomb) you have to run COSOlver from B.J.Frei first in order to generate the required matrices in HeLaZ/iCa folder.
# Logbook
(Current versions : 2.5)
0. Write MOLI matlab solver in Fortran using Monli1D as starting point
0.0 go from 1D space to 2D fourier and from Hermite basis to Hermite-Laguerre basis
0.1 implement linear Poisson equation in fourier space
0.2 implement moment hierarchy linear terms
0.3 RK4 time solver
0.4 Benchmark with MOLI matlab results for Z-pinch (cf. kz_linear script)
Note : benchmark_*.m compares MOLI and HeLaZ linear results
0.5 Load COSOlver matrices
0.6 Benchmarks now include Dougherty, Lenard-Bernstein and Full Coulomb collision operators
Note : for full Coulomb, one must store a precomputed matrix from COSOlver in the iCa folder
1. Implementation of the non linear Poisson brackets term
1.0 FFTW3 has been used to treat the convolution as a product and discrete fourier transform
1.1 Methods in fourier_mod.f90 have been validated by tests on Hasegawa Wakatani system
1.1 Qualitative test : find similar turbulences as Hasegawa Wakatani system with few moments
1.2 Zonal flows are observed in a similar way to Ricci Rogers 2006 with GS2
1.3 Linear analysis showed that a certain amount of PJ are recquired to trigger mode
1.3.1 The \eta_B = 0.5 case is easier since it converged better in linear analysis than \eta_B = 1.0
1.3.2 Collisionality helps
1.4 Quantitative study with stationary average particle flux \Gamma_\infty
1.4.1 Convergence study of \Gamma_\infty w.r.t. P and J
1.4.2 Direct comparison with GS2 results of Ricci,Rogers 2006
1.4.3 Code too expensive in sequential to reach PJ convergence
2. MPI parallel version (branch MPI)
2.1 First compilable parallel version (1D parallel along kr)
2.1.1 Benchmarks, profiling and portability of the code, max speed up is arround 10 for 12 proc. when using large enough resolution or polynomials degree. Should be inum_procs_kreased with other dimension distribution.
2.2 Allow restart with different P,J values
2.3 GK Dougherty operator
2.3.1 Convergence study of steady radial particle transport
2.3.1.1 Detailed analysis of numerical stability for P,J = 10,5.
(test were done with non linear s-sum truncated at 2, worked great.)
2.4 Data distribution along P
2.4.1 Parallel scaling and numerical stability
2.5 GK Sugama collision operator
2.7 GPU accelerated version
2.8 GK Full Coulomb collision operator
4. GK 3D version, kr,kz,kpar for linear device
5. DK 3D version, kr,kz,kpar for linear device
6. DK+GK 3D version, kr,kz,kpar for linear device
7. 3D version with curvature
c4science · Help