Exascale Catalytic Chemistry (ECC)

Led by a team from Sandia National Laboratories, and in collaboration with Argonne National Laboratory, Pacific Northwest National Laboratory, and Brown University, our group in the Chemical Engineering department at Northeastern University is pleased to begin work on an $8M project to develop a suite of computational tools that will allow scientists and engineers to leverage the next generation exa-scale computers to build predictive models of complex chemical systems including heterogenous catalysis coupled with gas-phase reactions.

Our efforts at Northeastern will focus on developing our AutoTST software that automates transition state theory calculations of reaction kinetics, and our RMG-Cat software that is a fully automated Reaction Mechanism Generator for Heterogeneous Catalysis.

In September 2018, we began work on an exciting new project in collaboration with Mike Burke’s group at Columbia University, that we call the “AutoScience” project. The goal is to couple automated calculations and automated experiments together using automatically generated models that are automatically analyzed. Then we can all retire!

The collaborative 3-year project is supported by Computational and Data-Enabled Science and Engineering program (CDS&E) at the NSF, and the project abstract at the NSF is like this:

To meet pressing societal needs for more cost-effective and sustainable energy, future combustion engines need to be more fuel-efficient, produce less emissions, and operate on a variety of fuels, including alternative fuels. Engineers often use computer models of fuel combustion chemistry to design engines with improved performance and determine the suitability of a certain fuel in an engine. In producing combustion models for engineers to use, scientists usually start by creating a trial model, then generate computational and experimental data to test the model, and improve and validate the model against the data. The latter two tasks are often repeated until the resulting model is sufficiently accurate for reliable use. Present techniques for developing reliable, validated models for transportation-relevant fuels typically involve combining the efforts of multiple research groups, taking multiple years or even decades to obtain enough data. The present approach for developing fuel combustion chemistry models is insufficient to address pressing energy needs in a timely and effective manner, particularly as many potential modern fuels have not been well characterized. This project will create and test the performance of a new autonomous system that creates trial models, generates data, and makes model improvements to rapidly converge on a reliable, validated, fuel chemistry model. Successful implementation of the novel autonomous system will provide an advanced model development tool for combustion kinetics and an accelerated means of understanding the oxidation behavior of the many alternative fuels, which governs their viability. Finally, this project will engage undergraduate and graduate students in research and create novel teaching modules for data science applied to combustion kinetics. The modules will enhance proficiency of younger generations of students in the scripting and data science tools necessary to ensuring a competitive STEM program in the U.S.

The technical objective of this project is to create an autonomous system for studying fuel oxidation chemistry and evaluate its performance relative to current time-intensive approaches. This autonomous system will use a multi-physics uncertainty quantification framework, MultiScale Informatics, to integrate an automated kinetic model construction platform, Reaction Mechanism Generator, an adaptable automated High-Throughput Jet Stirred Reactor experiment, and an algorithm for performing automated quantum chemistry, statistical thermodynamics, and transition state theory calculations (AutoTST). By linking the uncertainties both in experimental observables in the Jet Stirred Reactor and in Quantities of Interest, such as onset of ignition in an engine, to physically meaningful parameters in the kinetic model, such as barrier heights of a reaction, calculations and experiments can be optimally designed to improve the model’s accuracy for predicting Quantities of Interest. This project seeks to (1) create the autonomous platform, (2) use it to generate a model for n-heptane, for which previous data and models are relatively mature, to assess its performance, and (3) apply it to diisobutylene, a promising biofuel recently identified in the DOE’s Co-Optima program. This project will create a new data-driven approach for combustion research at an accelerated pace, contribute to scientific understanding for n-heptane and diisobutylene, and, more broadly, contribute to understanding of autonomous science.