prisae / libEMM

A C + CUDA library for 3D CSEM modelling

libEMM

A C + CUDA library for 3D CSEM modelling

Author: Pengliang Yang, Harbin Institute of Technology, China

Email: [e-mail]

Programming language: C, CUDA, Fortran, Shell

Operating System: Linux

Software dependencies: MPI, FFTW, CUDA (optional)

Nature of problem: Land/Marine Controlled-source electromagnetics (CSEM)

Solution method: High-order finite-dfference time-domain (FDTD) on staggered non-uniform grid by fictious wave domain transformation

Governing equation: 1st order diffusive Maxwell equation

Code structure:

• src: the source code in .c and .cu/.cuh.

• include: the header files in .h

• doc: the mannual for libEMM

• bin: the folderto store executable after compilation

• run_1d: Application example in 3D model of 1D layered sturture

• run_bathy_2d: Appliation example in 3D model of 2D structure with seafloor bathymetry

Instructions to run

1. Compile the code:

   cd src;

   make (to compile with mpicc)

   or

   make GPU=1 (to comipile with mpicc and CUDA)

2. Running reproducible examples

   Example 1: run_1d

   cd run_1d;

   bash run.sh

   To plot the result, use: python3 plot_emdata.py

   Example 2: run_bathy_2d

   cd run_bathy_2d

   bash run.sh

   To plot the result，use:

   python3 plot_cmp_libEMM_mare2dem.py

   python3 plot_emdata.py

NB: the input files may be generated prior to running. The resistivity files and nonuniform grid in binary format will be generated in src_nugrid:

cd src_nugrid;

make

./main

One may need to copy rho11, rho22, rho33 and x3nu into /run_1d and /run_bathy_2d.

Acknowledgement:

The initiative to start this ficititious wave domain modelling project begins when I was a scientist working in Electromagnetic Geoservices ASA (EMGS). I developed some modeling codes in my free time, but they never work correctly. After I left EMGS in 2020, I restarted everything from scratch using all things I learned from Madagascar open software development. It took me more than one year to make it work correctly: the solution now matches the semi-analytic one. During the development, I benefit from the discussion with Rune Mittet, in order to understand his method.

Please give a credit to the following publication if any component of libEMM is used in your research:

@Article{Yang_2023_libEMM, author = {Pengliang Yang}, title = {{libEMM: A fictious wave domain 3D CSEM modelling library bridging sequential and parallel GPU implementation}}, journal = {Computer Physics Communications}, volume = {288}, pages = {108745}, year = {2023}, issn = {0010-4655}, doi = {https://doi.org/10.1016/j.cpc.2023.108745}, }