ChE solution cuts computational cost nearly 100-fold, increases accuracy in predicting reaction outcomes, study shows
Determining whether materials will be both stable during their use and easy to break down afterward is difficult and costly. Predicting these properties poses a major roadblock on the path from materials discovery to applications.
But a newly published article in Nature Computational Science by Purdue Chemical Engineering researchers presents a breakthrough in a Purdue-led Multidisciplinary University Research Initiative (MURI) project. The article introduces yet another reaction program (YARP) – an automated computational method that marks the first step in bringing materials stability into the realm of predictability. YARP both reduces computational cost and improves reaction coverage when compared with current methods.
Co-authors Brett Savoie, the Charles Davidson Assistant Professor of Chemical Engineering and principal investigator, and graduate student Qiyuan Zhao developed this new approach to predicting reaction outcomes from scratch. YARP is the first step in a suite of computational tools the group plans on developing as part of the MURI project.
“YARP achieved a nearly 100-fold reduction in computational cost relative to the state-of-the-art method applied to the same problem,” Savoie said. “This essentially means that for the same investment of computational resources, YARP is able to characterize almost 100 times as many reactions.”
Due to the high cost of reaction characterization, researchers often use intuition to narrow the scope of reactions that are studied. However, this method may cause researchers to miss important reactions, leading to erroneous conclusions. In the context of materials stability, missing a reaction could mean failing to identify an important susceptibility. By lowering the computational cost, YARP enables broader reaction coverage and, thus, greater accuracy.
“A major source of the speedup that we have achieved is using inexpensive models to form approximate solutions before refining with more expensive and accurate models,” Savoie said. “This mixed approach proved to be very effective, as it led to no loss in accuracy and allowed us to explore more reactions.”
According to the article, “This combination of ultra-low cost and high reaction coverage creates opportunities to explore the reactivity of larger systems and more complex reaction networks for applications such as chemical degradation, where computational cost is a bottleneck.”
The study was inspired by advances in the computational chemistry community. Savoie’s team built YARP on a concept that treats chemicals like “graphs,” which allows chemical reactions to be described in a way that computers can interpret and automate.
YARP enhancements are continuing. “For example, we have recently extended YARP to account for conformational sampling, charge transfer reactions, and reactions involving ions and radicals,” Savoie said. “These are relatively technical developments, but for practitioners, each of these areas poses unique challenges that are important to tackle.”
While YARP can determine reaction susceptibility of materials, much work still must be done to predict how a specific material degrades under different conditions, such as being subjected to heat, oxygen or moisture. The five-year, $7.5 million MURI effort is attempting to fill this gap through collaboration among Purdue, Argonne National Laboratory, Carnegie Mellon University and University of Pittsburgh researchers.
“We are working with experimental groups that are collecting detailed degradation data in real materials in different use cases, as well as theoretical groups that are addressing different aspects of simulating degradation steps,” Savoie said.
Full Article: Simultaneously improving reaction coverage and computational cost in automated reaction prediction tasks (Nature Computational Science)