A lightweight, high-performance chemical equilibrium solver written in C and compiled to WebAssembly.
This project computes the equilibrium composition of ideal gas mixtures using direct Gibbs free energy minimization (the RAND method). It utilizes a damped Newton-Raphson approach to solve the Karush-Kuhn-Tucker (KKT) conditions and relies on NASA-9 polynomials for highly accurate thermodynamic properties.
This is a C/WASM port of my original Python implementation available at marciorvneto/chem-eq.
- Bare-Metal C / WASM: Zero external dependencies. Everything here is tailor-made from scratch: it uses a custom arena allocator and my own
stb-style linear algebra single-header library, tinyla.h, for maximum performance in the browser. - Direct Gibbs Minimization: Avoids the need to specify independent reaction sets.
- NASA-9 Thermodynamics: Accurate specific heat, enthalpy, and entropy calculations across wide temperature ranges.
- Warm-Start Capable: State is preserved between runs, allowing temperature or composition sweeps to converge in just 1–2 iterations.
- Gas-Phase Combustion: Currently, the solver only supports ideal gas phase calculations. It is tailored specifically for combustion applications, relying on NASA's 9-coefficient thermodynamic datasets. By the way, if you'd like to take a look at the complete NASA-9 Thermodynamics database, be sure to refer to NASA's amazing CEA repository.
Because this project relies on tinyla as a submodule, make sure to initialize it when cloning:
# Clone the repo with submodules
git clone --recursive [https://github.com/marciorvneto/tiny-gibbs.git](https://github.com/marciorvneto/tiny-gibbs.git)
cd tiny-gibbs
# Or, if you already cloned it without submodules, run:
# git submodule update --init
To compile the project, simply run make. The Makefile relies on standard gcc/clang for the native build and emcc (Emscripten) for the WebAssembly build.
make
This will generate:
out/main: The native C executable for local debugging.docs/main.js&docs/main.wasm: The compiled WebAssembly bundle (with the NASA database embedded) ready to be served alongside your HTML.
For a complete derivation of the KKT conditions and the resulting Jacobian block matrix formulated using Fréchet differentials, please refer to the included derivation.pdf.
Acknowledgments: The mathematical approach used in this derivation was heavily inspired by the beautiful coordinate-free matrix calculus taught by Professors Alan Edelman and Steven G. Johnson in their MIT course, Matrix Calculus for Machine Learning and Beyond. Their presentation of matrix calculus through differentials and linear operators was instrumental in shaping the mathematical viewpoint of this solver. You can find a link to the course from MIT OCW here
If you'd like to dive deeper into the underlying thermodynamics and numerical methods, I wrote a paper on Gibbs minimization a while ago. Please feel free to refer to it:
MIT