Jay Arcities | Searcharxiv

arXiv2026

CodeJeNN: A simple C++ neural network generator for physics applications

Machine learning has shown speedups for numerical methods in physics applications, but integrating Python-based libraries into high-performance C++ solvers creates performance bottlenecks. We present CodeJeNN, which bridges this gap by auto-generating self-contained C++ code from trained Keras models for inference. This eliminates external dependencies through minimal inlined functions, allowing seamless integration into existing frameworks. We describe the Keras-to-C++ workflow, supported architectures, and limitations. CodeJeNN is demonstrated through inference benchmarks against Keras in eager and JIT modes and a CFD test case modeling viscosity in a hydrogen-air mixing layer, showing speedups without sacrificing accuracy.

physics.comp-ph↗

arXiv2025

ChemGen: Code Generation for Multispecies Chemically Reacting Flow Simulations

This paper introduces ChemGen, a software package that uses code generation to integrate multispecies thermodynamics and chemical kinetics into C+-based computational physics codes. ChemGen aims to make chemical kinetics more accessible in existing simulation frameworks and help bridge the gap between combustion modeling and computational physics. The package employs the concept of decorators which enable flexible C++ code generation to target established software ecosystems. ChemGen generates code to evaluate thermodynamic properties, chemical source terms, and their analytical derivatives for Jacobian calculations. Also included are a variety of implicit time integration schemes, linear solvers, and preconditioners. The various components of Chemgen are verified by demonstrating agreement with Cantera and/or theoretical convergence rates. Finally, we integrate ChemGen into OpenFOAM and achieve a speedup over its native chemistry solver by approximately four times. ChemGen is an ongoing project released under the NRL Open License, a source-available license provided by the U.S. Naval Research Laboratory.

physics.comp-ph↗