Skip to content

Differentiable and hardware accelerated chemical kinetics simulations in python using JAX

License

Notifications You must be signed in to change notification settings

comocheng/jax-reactor

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

40 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

JAX-reactor

Simulating large chemical mechanisms is vital in combustion, atmospheric chemistry and heterogeneous catalysis. JAX-reactor is a package written in python to simulate large kinetic models leveraging just-in-time (JIT) compilation, automatic differentiaion and vectorization capabilities of awesome JAX package. JAX uses XLA to JIT compile python code to CPU, GPU and TPU. JAX can automatically differentiate python and numpy functions allowing us to efficiently calculate Jacobians of large chemical systems.

JAX-reactor uses Cantera's recently developed YAML input format to read large detailed kinetic models. Currently, JAX-reactor provides a basic JAX implementation of backward differentiation formula (BDF) solver to integrate stiff chemical systems. JAX-reactor is heavily inspired by a similar package written in PyTorch called reactorch. JAX-reactor is a research project and is in early stages of development.

Installation

First set up a conda environment

conda create --name jax python=3.7 scons cython boost numpy=>1.5 ruamel_yaml scipy=>1.4 matplotlib jupyter

Install the dev build of Cantera

conda activate jax

conda install -c cantera/label/dev cantera

CPU only version of JAX can be easily installed using pip

pip install --upgrade pip
pip install --upgrade jax jaxlib  # CPU-only version

For GPU version of JAX please follow the official installation instructions at JAX GPU installation

Known JAX GPU build issue

If you want to use JAX on CentOS-7 you need to build JAX from source following instructions at jax-ml/jax#2083

Once the enviroment is setup install JAX-reactor:

git clone https://github.com/comocheng/jax-reactor.git
export PYTHONPATH=<full-path-to-cloned-folder>:$PYTHONPATH

References

1. Weiqi Ji, Sili Deng. ReacTorch: A Differentiable Reacting Flow Simulation Package in PyTorch, https://github.com/DENG-MIT/reactorch, 2020.

About

Differentiable and hardware accelerated chemical kinetics simulations in python using JAX

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages