Deng Energy and Nanotechnology Group

for energy conversion and nanomaterial synthesis

ReacTorch: A Differentiable Reacting Flow Simulation Package in PyTorch

Description
ReacTorch is a package for simulating chemically reacting flows in PyTorch. The capability of auto-differentiation enables us to efficiently compute the derivatives of the solutions to all of the species concentrations (obtaining Jacobian matrix) as well as model parameters (performing sensitivity analysis) at almost no cost. It also natively supports GPU computation with PyTorch. In addition, the capability of differentiating the entire reacting model is the foundation of adopting many recent hybrid physics-neural network algorithms. This package is aimed at providing an easily accessible platform for implementing those emerging hardware and software infrastructures from the deep learning community in chemically reacting flow simulations.

Credits
ReacTorch was initially developed in Deng Energy and Nanotechnology Group in 2020, and we welcome the contribution from the community. While the package is open source, we would appreciate it if you could cite ReacTorch, as you find it helpful in your research. This helps the community reproduce and further improve upon your work, as well as giving credits to the many authors who have contributed their time to developing ReacTorch. The recommended citation for ReacTorch is as follows:

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

Download and Instructions
The ReacTorch package can be downloaded from GitHub, and your feedback is welcome.

Arrhenius.jl

Description
Arrhenius.jl is a differentiable combustion simulation package in Julia. Classical autoignition and one-dimensional flame simulations can be done with this package. Due to the auto-differentiation feature, sensitivity to all kinetic parameters can be evaluated. Demos include using this package to automonously learn kinetic model using neural network, conducting uncertainty quantification, and performing mechanism reduction.

Credits
Arrhenius.jl was initially developed in Deng Energy and Nanotechnology Group in 2021, and we welcome the contribution from the community. While the package is open source, we would appreciate it if you could cite Arrhenius.jl, as you find it helpful in your research. This helps the community reproduce and further improve upon your work, as well as giving credits to the many authors who have contributed their time to developing ReacTorch. The recommended citation for Arrhenius.jl is as follows:

"W. Ji, X. Su, B. Pang, Y. Li, Z. Ren, S. Deng, SGD-based optimization in modeling combustion kinetics: Case studies in tuning mechanistic and hybrid kinetic models, Fuel 324 (2022) 124560."

Download and Instructions
The Arrhenius.jl package can be downloaded from GitHub, and your feedback is welcome.