ttnte

ttnte is a Python library for solving the discrete ordinates neutron transport equation (NTE) with a discontinuous isogeometric analysis (IGA) spatial discretization. This repository features IGA assembly through PyTorch sparse tensors and tensor trains (TTs). The application of TTs aims to exploit the multiscale structure commonly found in reactor applications. The IGA discretization offers higher continuity than traditional finite elements and benefits from working directly with CAD, cutting out the often expensive meshing step.

Requirements

• torch
• torchtt
• numpy
• igakit
• geomdl
• pandas
• matplotlib
• plotly
• h5py
• tqdm
• cotengra
• optuna
• cmaes
• cotengrust
• cytools
• kahypar
• loky
• networkx
• opt_einsum

Note

For best performance compile geomdl with Cython, PyTorch with CUDA, torchTT with its C++ extension, and ensure ttnte compiles by setting TTNTE_CPP_BACKEND=True environment variable. ttnte will compile the C++ backend by default but if it fails to compile it will fall back on the Python implementations.

Installation

Quick installation

pip install https://github.com/dalcinl/igakit/archive/refs/heads/master.zip
pip install git+https://github.com/ion-g-ion/torchTT.git
pip install git+https://github.com/myerspat/ttnte.git

From source

git clone https://github.com/myerspat/ttnte.git && cd ttnte
pip install https://github.com/dalcinl/igakit/archive/refs/heads/master.zip
pip install git+https://github.com/ion-g-ion/torchTT.git
pip install .

For developers

git clone git@github.com:myerspat/ttnte.git && cd ttnte
pip install https://github.com/dalcinl/igakit/archive/refs/heads/master.zip
pip install git+https://github.com/ion-g-ion/torchTT.git
pip install -e ".[dev]"
pre-commit install

Install Environment Variables

• TTNTE_CPP_BACKEND: Ensure the C++ backend compiles by setting this to True. You can also toggle the C++ backend at runtime.
• DEBUG: Compile the C++ backend in debug mode.
• NJOBS: Number of threads to use when compiling. This defaults to os.cpu_count().
• TTNTE_PROFILE: By default if DEBUG=True this environment variable defaults to True. This toggles compilation with -g and -fno-omit-frame-pointer flags.
• TTNTE_OPTIMIZED: If this is set to True then the C++ backend is compiled with -O3 and -march=native flags. This defaults to True unless DEBUG=True.
• USE_CUDA: Link PyTorch with CUDA support. This defaults to True.
• TORCH_INSTALL_PREFIX: Path to the base PyTorch directory. This defaults to os.path.abspath(os.path.dirname(torch.__file__)).
• _GLIBCXX_USE_CXX_ABI: This is either 0 or 1 and is used by PyTorch. This defaults to int(torch._C._GLIBCXX_USE_CXX11_ABI).

Classes and Methods

• ttnte.xs.Server: Class for handling multigroup cross section information.
• ttnte.cad.Patch: Patch class.
• ttnte.iga.IGAMesh: Meshing object for NURBS surfaces defined as igakit.nurbs.NURBS.
• ttnte.assemblers.MatrixAssembler: Assembler discretized system into ttnte.assemblers.operators.SparseOperators.
• ttnte.assemblers.TTAssembler: Assembler discretized system into torchtt.TTs.
• ttnte.linalg.gmres(): Method for solving the resulting discretized fixed source systems
• ttnte.linalg.power(): Method for solving the resulting discretized eigenvalue problem.
• ttnte.linalg.LinearSolverOptions: Options class for GMRES used in ttnte.linalg.power().

Modules

• ttnte.xs.benchmarks: XS data sets from common neutron transport benchmarks.
• ttnte.cad.curves: Methods for building NURBS curves used in the notebooks.
• ttnte.cad.surfaces: Methods for building NURBS surfaces used in the notebooks.
• ttnte.sources: Define fixed sources.

Notebooks

Fixed Source

• Homogeneous square
• Homogeneous circle
• Quarter circle with void
• Cruciform source with cylindrical wall

Eigenvalue

• Homogeneous square
• Homogeneous circle
• Homogeneous quarter circle
• C5G7 infinite pincell array
• Infinite array of cruciform four-lobe fuel with a
  • burnable absorder displacer
  • gas displacer