Python tools for reading and manipulating plasma fluid simulations

Utilises MP parallelisation, lots of vectorisation through the numpy library, JIT compiling for I/O through numba, and joblib for any trivial parallelisation. There are a number of FFTs the library can use, depending upon what is available. The fastest (by a factor of a few) is pyfftw. Currently I am working on MPI implmentations for both FFT and derivative class, which is invaluable for large mem. jobs with more than (1k^3) cells.

Currently the reading of FLASH data is handled by the read.py code, which has classes for particles and fields, and can read FLASH, RAMSES and BHAC simulation data.

The post-processing functions are contained within aux_funcs/derived_var_funcs.py, aux_funcs/spectral_var_funcs.py, aux_funcs/shell_trans_funcs.py for derived variable functions, spectral variable functions and transfer functions respectively. The derivative class is in derivative.py.

All scalar/vector operations can be used in 1D, 2D or 3D.

At the moment, the post-processing functions can be used as stand-alone functions that are directly applied to data, but for quite a few of them (look at read.py) they can be called as a method for the data object, which adds new derived fields directly to the data object.

The functions are:

Read I/O:

• BHAC (tested in 2.5D)
• FLASH (any D)
• RAMSES (only tested in 3D)
• AthenaK coming soon.

Derivatives:

• a range of derivative stencils for first-order, central difference spatial derivatives, from two to eight-point with periodic, Neumann, and Dirichlet BCs.
• limited stencils for second-order, central spatial derivatives.
• all derivatives are vectorised, but currently working towards MPI implementation for large mem. problems.

Scalar operations:

• scalar gradient
• scalar laplacian

Vector operations:

• vector cross product
• vector dot product
• vector field magnitude
• vectof field RMS
• vector curl
• vector divergence
• construction of TNB basis ( Frenet-Serret coordinates of the vector field )
• coordinate transformation into TNB basis
• field line curvature
• magnetic helicity ( in Coloumb gauge \partial_ia_i = 0; a . b )
• current helicity ( j.b )
• kinetic helicity ( vort . v )

Tensor operations:

• tensor outer product
• tensor contraction
• vector dot tensor
• gradient tensor
• eigen values of Hermitian tensors ( for, e.g., symmetric stretching tensor; used for dynamo growth rate modelling )

Decompositions:

• orthogonal gradient tensor decomposition ( symmetric, antisymmetric and trace tensor )
• helmholtz decomposition (incompressible and compressible modes) of a vector field
• vorticity decomposition (stretching, compression, baroclinicity)
• decomposition into left and right helical eigen modes of a vector field
• decomposition into vorticity sources ( compressive, stretching, baroclinic, tension )

Spectral operations:

• vector potential ( via fft in Coloumb gauge )
• 3D vector sclar power spectrum
• 3D vector field power spectrum
• 3D tensor field power spectrum
• spherical shell binning
• cylindrical shell binning
• k space filtering through isotropic k shells ( for transfer function analysis )
• k space filtering through cylindrical k shells ( for transfer function analysis )

Generating synthetic stochastic fields:

• generating chiral random field ( e.g. net j.b on system scale )
• generating power-law random field

Energy flux transfer functions:

• kinetic energy transfer functions
• magnetic energy transfer functions
• kinetic - magnetic energy interaction transfer functions
• helmholtz decomposed transfer functions

Transfer functions (all use parallelisation over shells, but are not good for larger mem. jobs, e.g., >= 2k^3 cells):

• 2D transfer functions for Newtonian kinetic (momentum) and magnetic (induction) field, specifically for casacade transfers (u' to u'' or b' to b''). Scalable to 10k^2 grids.
• 3D transfer functions for Newtonian ideal MHD equations.

Plotting functions:

• vectorised implementation of line integral convolution algorithm. Works very fast, even up to (10k)^2 grids.

Critial point analysis

• o and x point detector based on vector potential / stream function. Only works in 2D.

Functionality coming to the repo.

• faster, mpi parallelised ffts for large mem. jobs, specifically important for transfer functions.