Quantum mechanics and machine learning to accurately predict chemical reactions at high temperatures



Scheme of bridging the cold quantum world and high-temperature metal extraction with machine learning. Photo credit: Rodrigo Ortiz de la Morena and Jose A. Garrido Torres / Columbia Engineering

Method combines quantum mechanics with machine learning to accurately predict oxide reactions at high temperatures when experimental data are not available; could be used to design clean, CO2-neutral processes for steelmaking and metal recycling.

Extraction of metals from oxides at high temperatures is essential not only for the production of metals such as steel, but also for recycling. Since current extraction processes are very carbon intensive and emit large amounts of greenhouse gases, researchers are looking for new approaches to develop “greener” processes. This work was particularly challenging in the laboratory because it requires expensive reactors. Building and running computer simulations would be an alternative, but currently there is no computer method that can accurately predict oxide reactions at high temperatures in the absence of experimental data.

A team at Columbia Engineering reports that it has developed a new computational technique that, through the combination of quantum mechanics and machine learning, can accurately predict the temperature of reduction of metal oxides to their base metals. Their approach is computationally just as efficient as conventional calculations at zero temperature and their tests are more accurate than computationally intensive simulations of temperature effects using quantum chemical methods. The study, led by Alexander Urban, Junior Professor of Process Engineering, was carried out on December 1, 2021 by. released Nature communication.

“Decarbonising the chemical industry is critical to moving towards a more sustainable future, but developing alternatives to established industrial processes is very costly and time-consuming,” said Urban. “A computer-aided process design from the bottom up that does not require any initial experimental input would be an attractive alternative, but has not yet been implemented. To the best of our knowledge, this new study is the first time that a hybrid approach that combines computation with AI has been attempted for this application. And it is the first evidence that quantum mechanical calculations can be used for the design of high-temperature processes. “

The researchers knew that quantum mechanical calculations at very low temperatures can accurately predict the energy that chemical reactions require or release. They supplemented this zero temperature theory with a machine learning model that learned the temperature dependency from publicly available high temperature measurements. They designed their approach, which focused on extracting metal at high temperatures, to also predict the change in “free energy” with temperature, whether it was high or low.

“Free energy is a key variable in thermodynamics and other temperature-dependent variables can in principle be derived from it,” says José A. Garrido Torres, first author of the work, postdoctoral fellow in Urban’s laboratory and now a research fellow at Princeton. “We therefore expect that our approach will also be useful, for example, to predict melting temperatures and solubilities for the design of clean electrolytic metal extraction processes that run on renewable electrical energy.”

“The future just got a little closer,” said Nick Birbilis, deputy dean of the Australian National University University of Technology and Computer Science and an expert in material design with a focus on corrosion resistance, who was not involved in the study. “Much of the human effort and sunken capital of the past century has gone into creating materials that we use every day – and rely on for our performance, flight and entertainment. Material development is slow and costly, which makes machine learning a crucial development for future material design. In order for machine learning and AI to develop their potential, models must be mechanistically relevant and interpretable. This is exactly what the work of Urban and Garrido Torres shows. In addition, for the first time, the work pursues an overall system approach in which atomistic simulations on the one hand are linked with engineering applications on the other – using advanced algorithms. “

The team is now working to extend the approach to other temperature-dependent material properties such as solubility, conductivity, and melting, which are needed to develop electrolytic metal extraction processes that are carbon-free and run on clean electrical energy.

Reference: “Augmenting Zero-Kelvin Quantum Mechanics with Machine Learning for Predicting Chemical Reactions at High Temperatures” by Jose Antonio Garrido Torres, Vahe Gharakhanyan, Nongnuch Artrith, Tobias Hoffmann Eegholm and Alexander Urban, December 1, 2021, Nature communication.
DOI: 10.1038 / s41467-021-27154-2


Comments are closed.