Cardinal AMR

This repository houses a series of test cases for Adaptive Mesh Refinement (AMR) on unstructured mesh tallies in Cardinal-OpenMC. The models can be found in /models/*; each consists of the following files:

• A script which uses the OpenMC Python API to generate a Constructive Solid Geometry (CSG) model for a particular reactor type (make_openmc_model.py).
• A Cardinal input file to generate the initial unstructured mesh for fission power / flux tallies (mesh.i).
• A Cardinal input file which creates an unstructured mesh tally using the initial mesh and runs OpenMC (openmc.i).

The Python files include a command line arguement (-n) to specify the number of fuel axial subdivisions in the CSG geometry. This can be used to investigate the impact of AMR on temperature/density feedback that is applied to the underlying OpenMC geometry.

At present this repository contains the following models:

• Light Water Reactor (/models/lwr/*)

  • Single PWR pincell;
  • 3 pin x 3 pin PWR bundle;
  • 17 pin x 17 pin unrodded / rodded UO2 PWR assemblies;
  • 4 assembly x 4 assembly unrodded / rodded UO2/MOX PWR lattices;

• Sodium Cooled Fast Reactor (/models/sfr/*)

  • pincell model with two types of material composition ( inner and outer region of the reactor);
  • assembly model with two types of material composition ( inner and outer region of the reactor);

In addition to these models, this repository also contains a series of AMR algorithms - these can be found in /amr_strategies/*. To apply one of these AMR strategies to a test model in this repository, follow the steps below:

1. Add an !include STRATEGY_NAME to the beginning of the openmc.i file in one of the model sub directories, where STRATEGY_NAME is the name of a file in amr_strategies.
2. Run python make_openmc_model.py.
3. Run cardinal-opt -i mesh.i --mesh-only.
4. Run mpiexec -n NUM_PROC cardinal-opt openmc.i --n-threads=NUM_THREADS, replacing NUM_PROC with the number of MPI processors and NUM_THREADS with the number of OpenMP threads.

As an example, here we apply a relative error-aware refinement scheme to the LWR pincell case. First, we modify models/lwr/pincell/openmc.i to include the following on the first line:

!include ../../../amr_strategies/value_jump_rel_error.i

Then, we run the following:

cd ./models/lwr/pincell
python make_openmc_model.py
cardinal-opt -i mesh.i --mesh-only
mpiexec -n 2 cardinal-opt -i openmc.i --num-threads=2

Please note that some model/strategy combinations will require substantial computing power due to the size of the OpenMC model and the number of mesh tally bins.