Reducing the complexity of chemical networks via interpretable autoencoders

Abstract

In many astrophysical applications, the cost of solving a chemical network represented by a system of ordinary differential equations (ODEs) grows significantly with the size of the network, and can often represent a significant computational bottleneck, particularly in coupled chemo-dynamical models. Although standard numerical techniques and complex solutions tailored to thermochemistry can somewhat reduce the cost, more recently, machine learning algorithms have begun to attack this challenge via data-driven dimensional reduction techniques. In this work, we present a new class of methods that take advantage of machine learning techniques to reduce complex data sets (autoencoders), the optimization of multi-parameter systems (standard backpropagation), and the robustness of well-established ODE solvers to to explicitly incorporate time-dependence. This new method allows us to find a compressed and simplified version of a large chemical network in a semi-automated fashion that can be solved with a standard ODE solver, while also enabling interpretability of the compressed, latent network. As a proof of concept, we tested the method on an astrophysically-relevant chemical network with 29 species and 224 reactions, obtaining a reduced but representative network with only 5 species and 12 reactions, and a x65 speed-up.

accepted A&A, code available at https://bitbucket.org/tgrassi/latent_ode_paper/

References in corpus (15)

• Array Programming with NumPy
• A Tutorial on Principal Component Analysis
• Hidden Physics Models: Machine Learning of Nonlinear Partial Differential Equations
• Complex organic molecules in protoplanetary disks
• DiffEqFlux.jl - A Julia Library for Neural Differential Equations
• Global Hydromagnetic Simulations of Protoplanetary Disks with Stellar Irradiation and Simplified Thermochemistry
• Chemical Modelling of Young Stellar Objects, I. Method and Benchmarks
• The chemistry of protoplanetary fragments formed via gravitational instabilities
• The chemistry of episodic accretion in embedded objects. 2D radiation thermo-chemical models of the post-burst phase
• An investigation of GPU-based stiff chemical kinetics integration methods
• The 3D structure of CO depletion in high-mass prestellar regions
• The challenges of modelling microphysics: ambipolar diffusion, chemistry, and cosmic rays in MHD shocks
• Chemical network reduction in protoplanetary disks
• Exploiting Network Topology for Accelerated Bayesian Inference of Grain Surface Reaction Networks
• MaNN: Multiple Artificial Neural Networks for modelling the Interstellar Medium

Cited by in corpus (11)

• The GRETOBAPE gas-phase reaction network: the importance of being exothermic
• 3D simulations of AGB stellar winds -- II. Ray-tracer implementation and impact of radiation on the outflow morphology
• Neural network-based emulation of interstellar medium models
• Quantifying the informativity of emission lines to infer physical conditions in giant molecular clouds. I. Application to model predictions
• Mapping Synthetic Observations to Prestellar Core Models: An Interpretable Machine Learning Approach
• Using a neural network approach to accelerate disequilibrium chemistry calculations in exoplanet atmospheres
• Radiative Transfer as a Bayesian Linear Regression problem
• NeuralPDR: Neural Differential Equations as surrogate models for Photodissociation Regions
• Neural-Network Chemical Emulator for First-Star Formation: Robust Iterative Predictions over a Wide Density Range
• Beetroots: spatially-regularized Bayesian inference of physical parameter maps -- Application to Orion
• Systematic selection of surrogate models for nonequilibrium chemistry