The RadCluster code

Introduction

RadCluster is a physics-based simulation suite for the evolution of radiation-induced defect populations in crystalline solids under neutron or ion irradiation: self-interstitial atom (SIA) dislocation loops, vacancy and helium-vacancy cavities, and free interstitial helium. The reference host material is the reduced-activation ferritic-martensitic steel EUROFER-97.

The cluster-dynamics master equation is not hand-written per material. A material is declared as a Reaction Admissibility Graph (RAG) whose edges are elementary reactions drawn from a fixed catalogue of ten abstract classes, and the right-hand side of the ODE system is assembled by walking that graph.

RadCluster contains the following modules:

• Abstract core (Layer 1)
• Material models (Layer 2)
• C++ solver
• Input, output and orchestration
• Verification and diagnostics

The source code, the releases and the issue tracker are on GitHub. User-oriented notes are kept on the wiki.

Installation

RadCluster has two layers: a Python package (model definition, orchestration, post-processing) and a C++17 solver built on SUNDIALS/CVODE. The Python layer alone is sufficient to build a material graph, inspect the model and run the reference graph walker. Production runs use the C++ solver.

git clone https://github.com/Ghoniem/RadCluster.git
cd RadCluster
python -m pip install -r requirements.txt

cd radcluster_code
cmake -S cpp_utils -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build --config Release

The C++ solver needs a C++17 compiler and CMake 3.15 or newer. The following compilers should do the job:

• GNU g++
• Clang (Apple Clang on macOS)
• Microsoft Visual C++

A published container image carries the compiled solver and the Python layer, so that nothing has to be built locally. See the dedicated Container image page.

Are you running RadCluster on the UCLA Hoffman2 cluster? Then make sure you read the dedicated Running on the Hoffman2 cluster page.

External libraries

The C++ solver requires the following external library:

• the SUNDIALS library (CVODE), version 7.1.1

Three further libraries are optional. Each is detected at configure time, and the solver degrades gracefully when one is absent:

• LAPACK/BLAS, used by the Woodbury preconditioner
• OpenMP, used by the parallel right-hand side of the active-window mode (see Solver modes and physics options)
• SuiteSparse/KLU, used by the sparse-direct linear solver

The Python layer requires numpy, scipy, pandas, matplotlib, openpyxl, jupyter and ipykernel. The package networkx is optional, and is used only by the graph-diagnostics export.

Other useful packages

This documentation is generated with Doxygen and Graphviz:

cd doxygen
doxygen Doxyfile

The pages are written to the folder doxygen/HTML, which is not tracked by git.

How to cite

N. M. Ghoniem, A Generalized Graph-Based Cluster Dynamics Framework for Irradiated Materials, Journal of Nuclear Materials (submitted, 2026).

Section and equation numbers in the code comments refer to that document. Its source is in the folder docs/Formulation/cluster_dynamics_framework of the repository.

License

RadCluster is released under the MIT License.