Machine learning framework identifies targets for catalyst improvement

This graph shows the seven-step reaction pathway of the hydrogenation of CO to methanol over copper-based catalysts, including the reactants at each step, the schematic atomic arrangements of the intermediates, and the energy activation barriers needed to pass from one step to another. The Brookhaven Lab team presented a machine learning framework that successfully identified the steps/combinations of steps to modify to improve methanol production. Their work could help guide the design of new catalysts to achieve this goal and the framework can be applied to optimize other reactions. Credit: Brookhaven National Laboratory

Chemists at the US Department of Energy’s Brookhaven National Laboratory have developed a new machine learning (ML) framework that can focus on which steps in a multi-step chemical conversion need to be changed to improve productivity. This approach could help guide the design of catalysts, chemical “negotiators” that speed up reactions.

The team developed the method to analyze the conversion of carbon monoxide (CO) to methanol using a copper-based catalyst. The reaction consists of seven fairly simple basic steps.

“Our goal was to identify which elementary step of the reaction network or which subset of steps controls the catalytic activity,” said Wenjie Liao, the first author of a paper describing the method just published. in the journal. Catalysis Science and technology. Liao is a graduate student at Stony Brook University who worked with scientists in the Catalysis Reactivity and Structure (CRS) group in Brookhaven Lab’s Chemistry Division.

Ping Liu, the CRS chemist who led the work, said, “We used this reaction as an example of our ML framework method, but you can put any reaction in this framework in general.”

Targeting Activation Energies

Imagine a multi-step chemical reaction like a roller coaster with hills of different heights. The height of each hill represents the energy required to move from one step to the next. Catalysts lower these “activation barriers” by making it easier for the reactants to come together or by allowing them to do so at lower temperatures or pressures. To speed up the overall reaction, a catalyst must target the step or steps that have the greatest impact.

Traditionally, scientists seeking to improve such a reaction calculated how to change each activation barrier one at a time could affect the overall production rate. This type of analysis could identify which step was “limiting” and which steps determine the selectivity of the reaction, i.e. whether the reactants go to the desired product or take another route to an unwanted by-product. .

Brookhaven Lab chemist Ping Liu and Stony Brook University graduate student Wenjie Liao developed a machine learning framework to identify chemical reaction steps that could be targeted to improve reaction productivity. Credit: Brookhaven National Laboratory

But, according to Liu, “these estimates end up being very rough with a lot of errors for some groups of catalysts. It really hurt catalyst design and screening, which is what we’re trying to do,” she said.

The new machine learning framework is designed to improve these estimates so scientists can better predict how catalysts will affect reaction mechanisms and chemical production.

“Now, instead of moving one barrier at a time, we move all the barriers simultaneously. And we use machine learning to interpret this data set,” Liao said.

This approach, the team said, gives much more reliable results, including how the steps in a reaction work together.

“Under the reaction conditions, these steps are not isolated or separated from each other; they are all connected,” Liu said. “If you only do one step at a time, you miss a lot of information, the interactions between the elementary steps. That’s what was captured in this development,” she said.

Build the model

The scientists started by creating a dataset to train their machine learning model. The dataset was based on “density functional theory” (DFT) calculations of the activation energy needed to transform one arrangement of atoms into another through the seven reaction steps. Next, the scientists ran computer simulations to explore what would happen if they changed all seven activation barriers simultaneously, some going up, some coming down, some individually and some in pairs.

“The range of data that we included was based on previous experience with these reactions and this catalyst system, within the interesting range of variation that is likely to give you better performance,” Liu said.

By simulating the variations of 28 “descriptors” – including activation energies for all seven stages plus pairs of stages changing two at a time – the team produced a comprehensive dataset of 500 data points. This dataset predicted how all of these individual adjustments and pairs of adjustments would affect methanol production. The model then scored the 28 descriptors according to their importance in methanol production.

“Our model ‘learned’ from the data and identified six key descriptors that it believed would have the most impact on production,” Liao said.

Once the important descriptors were identified, the scientists recycled the ML model using only these six “active” descriptors. This improved ML model was able to predict catalytic activity based only on DFT calculations for these six parameters.

“Rather than having to calculate all 28 descriptors, you can now calculate with just the six descriptors and get the methanol conversion rates you are interested in,” Liu said.

The team says they can also use the model to screen for catalysts. If they can design a catalyst that improves the value of the six active descriptors, the model predicts a maximum methanol production rate.

Understand the mechanisms

When the team compared their model predictions with the experimental performance of their catalyst – and the performance of alloys of various metals with copper – the predictions matched the experimental results. Comparisons of the ML approach with the previous method used to predict alloy performance showed that the ML method was far superior.

The data also revealed many details about how changes in energy barriers might affect the reaction mechanism. Of particular interest and importance is how the different reaction steps work together. For example, the data showed that in some cases lowering the energy barrier in the rate limiting step alone would not by itself improve methanol production. But changing the energy barrier one step earlier in the reaction network, while keeping the activation energy of the rate-limiting step within an ideal range, would increase methanol production.

“Our method gives us detailed information that we could use to design a catalyst that coordinates the interaction between these two steps well,” Liu said.

But Liu is very excited about the potential for applying these data-driven ML frameworks to more complex reactions.

“We used the methanol reaction to demonstrate our method. But the way it generates the database and how we train the ML model and how we interpolate the function role of each descriptor to determine the overall weight in terms of importance, this can be applied easily to other reactions “, she said.

Discovery of a new catalyst for the highly active and selective hydrogenation of carbon dioxide to methanol

More information:

Wenjie Liao et al, Improved identification of descriptors and understanding of mechanisms of catalytic activity using a data-driven framework: revealing the importance of interactions between elementary steps, Catalysis Science and technology (2022). DOI: 10.1039/D2CY00284A

Provided by
Brookhaven National Laboratory

Machine Learning Framework ID Targets for Catalyst Enhancement (2022, May 10)
retrieved 10 May 2022

This document is subject to copyright. Apart from any fair use for the purpose of private study or research, no
any part may be reproduced without written permission. The content is provided for information only.

Sherry J. Basler