Ragnar [ᚱᛅᚴᚾᛅᚱᛦ]

Radiative post-processing code

Written in C++ with Kokkos portability and python API via pybind11. Supports loading data from PIC simulations for post-processing.

Usage

Download the code with:

git clone --recursive https://github.com/haykh/ragnar.git
cd ragnar

You can build/install the package with pip using the following command:

pip install . --config-settings="cmake.args=<FLAGS>"

All the CMake flags can be passed in the cmake.args as shown above, e.g., cmake.args=-DRAGNAR_USE_HDF5:BOOL=OFF;-DKokkos_ENABLE_OPENMP:BOOL=ON;....

Otherwise, you can also compile the code to produce the shared library directly with CMake:

cmake -B build [-D RAGNAR_USE_HDF5:BOOL=OFF] [-D Kokkos_ENABLE_CUDA:BOOL=ON] ...
cmake --build build -j

Compilation produces a shared library file .so (located in build/) which can be imported directly in python.

If Kokkos is installed externally, no special -DKokkos_*** flags are needed.

If you don't intend to use hdf5 for reading the simulation data, you may disable it by providing a flag -DRAGNAR_USE_HDF5:BOOL=OFF (it is set to ON by default).

Usage examples together with unit tests are located in src/examples and src/tests.

import ragnar as rg
rg.Initialize(); # always remember to initialize Kokkos

Generating/manipulating synthetic data

You can generate synthetic data using built-in functions:

dist_prtls = rg.TabulatedDistribution(
    rg.Logbins(1, 100, 200), rg.PlawGenerator(-2, 1, 100)
)

After that, you may use this data further; for instance, to generate a synchrotron signal from a given distribution of particles:

# first, generate the photon bins
bins_e_syn = rg.Logbins(0.01, 1e7, 200)
bins_e_syn.unit = rg.EnergyUnits.mec2

e_syn_2_f_syn = rg.SynchrotronSpectrumFromDist(dist_prtls, bins_e_syn, 1, 1)

# docstring can be accessed via `rg.SynchrotronSpectrumFromDist?`

Then you can plot the generated data:

plt.plot(bins_e_syn.as_array(), e_syn_2_f_syn.as_array())
plt.xscale("log")
plt.yscale("log")

Post-processing output data

All functions have docstrings which can be accessed via, e.g., rg.SynchrotronSpectrum_3D?.

While functions in rg can be accessed with any user-defined, interaction with the simulation data is done via the so-called plugins. Below is an example usage for Tristan v2 plugin:

plugin = rg.TristanV2_3D()
plugin.setPath("<PATH_TO_DATA>") # directory where the `output` is
plugin.setStep(55)

electrons = plugin.readParticles("e-", 1) # 1 -- is the species index (starting from 1)
positrons = plugin.readParticles("e+", 2)
protons = plugin.readParticles("p", 3)

.readParticles returns a Particles object, which is a special container to store all the read data (if compiled with GPU support, data is stored only on the GPU). With this object, one can, for instance, compute an energy distribution for the given species:

gbins = rg.Logbins(1e-2, 1e3, 200) # define gamma * beta bins
e_dist = electrons.energyDistribution(gbins)

# plot d N / d (gamma * beta)
plt.plot(e_dist.EnergyBins().as_array(), e_dist.F().as_array())
plt.xscale("log")
plt.yscale("log")

Unit tests

The code also has a set of unit tests that can be run after compilation using:

ctest --test-dir build

These tests are also ran using GitHub actions automatically on every push.

Dependencies

All the dependencies (except for pybind11) can be built in-tree (except for the HDF5), however, it is recommended to install them externally to speed up compilation.

• Kokkos
• HighFive : optional (used when -D RAGNAR_USE_HDF5=ON)
• pybind11 : no need to install

pybind11 is downloaded with the code when you clone with --recursive.

Using spack

You may install all of the dependencies using the spack package manager:

spack env create ragnar
spack env activate ragnar
spack install --add kokkos [+cuda] [+wrapper] [cuda_arch=...]
spack install --add highfive