Machine Learning Potential (MLP) Training for Surface Reconstruction Analysis. (a) Workflows for MLP training and large-scale configuration space search. (BC) Molecular Dynamics (MD) Simulation on the Nanosecond Scale (b) Mesoscopic Size of SNOs2 x (c)SNS2 x. The inset shows structures that may have the lowest energy surface screened by large-scale sampling. Bright gray, red and yellow spheres represent Sn, O, and S, respectively. Credit: Hao Li et al.
Some of the most encouraging results of reaction-enhancing catalysts come from one material, tin (SN). The overall usefulness of SN as a catalyst is well known, but the underlying structure-performance relationship is not well understood, limiting its ability to maximize its potential.
To address this knowledge gap, researchers at Tohoku University's Advanced Materials Research Institute (WPI-AIMR) used machine learning to characterize SN catalytic activity. This work has been published in the journal Advanced functional materials.
A highly accurate simulation can be a game changer that helps researchers quickly and quickly design complex catalysts with high performance.
“The reason these catalysts are so important is that they can convert harmful carbon dioxide.2– Provide carbon-based fuels using renewable electricity, providing sustainable solutions to energy shortages and climate change,” explains Hao Li, WPI-IAMR.
“The purpose of this research is to bring our society to carbon neutrality.”
To closely examine SN catalysts, they employ machine learning possibilities to perform large-scale molecular dynamics simulations and successfully capture the reconstructed configuration of SNOs2/sns2. This approach used data from over 1,000 experimental literature sources to identify a variety of SN-based catalysts.
“Instead of spending days, months, or even years of these experiments rather than doing them in the lab, you can run these sophisticated, data-driven simulations. These simulations can help you find the attention to which lab-based experiments are attracting attention,” says Li.
The catalysts identified by the model were performed in simulations that monitored activity at different pH levels on a reversible hydrogen electrode (RHE) scale.
Researchers investigated jointly2 Reduction reactions, check how each catalyst was carried out under various conditions. These results provide new insight into the behavior of these catalysts, as calculations from previous literature struggled to accurately explain the effects of pH dependence on electrocatalytic performance.
-
pH-dependent microkinetic modeling on the RHE scale. (a) pH-dependent microkinetic modeling of co2RR for SN-based catalysts. A lower electric field corresponds to more alkaline conditions, while a higher magnetic field exhibits more acidic conditions. (b) Rate determination step analysis of HCOOH formation under acidic (dashed line) and alkaline (solid) conditions. The purple line indicates activity restricted by the first electron transfer, while the green line indicates restriction by the second electron transfer. Credit: Hao Li et al.
-
Material characterization and performance testing. (AB) XRD spectrum of SNOs2 and SNS2 Before and after co2RR. (CD) SNO XPS spectra2 and SNS2 Before and after the reaction. (EF) SEM images2: (e) before the reaction and (f) after the reaction. (GH) SEM images for SNS2: (g) before the reaction and (h) after the reaction. (IJ) SNO HRTEM images2: Before (i) and (j). (kl) SNS HRTEM images2: (k) before the reaction and (l) after the reaction. (MN)SNO current density and applicable potential2 and SNS2 at different pH levels. (op) Joint Farada efficiency2SNO's RR2 and SNS2 With different potentials under ph = 13. Credit: Hao Li et al.
Furthermore, the simulation results show excellent agreement with actual experimental observations verifying the accuracy of this machine learning technique.
This study will help to form a more comprehensive understanding of SN-based catalysts and will allow them to maximize their potential. More efficient catalysts bring affordable green fuel production closer to everyday reality.
In the future, the research group plans to optimize the training process of machine learning possibilities to develop a more accurate and universal training framework, thereby better filling the gap between experimental findings and theoretical predictions.
All related experimental and computational data have been uploaded to the Digital Catalyst Platform (DIGCAT), the largest catalyst database and digital platform developed by Hao Li Lab.
detail:
Yuhang Wang et al., Bridging Theory and Experiments: Potentially Driven Insights of Machine Learning on the pH-dependent CO₂ Reduction of SN-based Catalysts, Advanced functional materials (2025). doi:10.1002/adfm.202506314
Provided by Tohoku University
Quote: The study uses machine learning to map the pH-dependent performance (2025, July 4) of tin catalysts obtained from https://phys.org/2025-07-machine-ph-tin-catalysts.html on July 6, 2025.
This document is subject to copyright. Apart from fair transactions for private research or research purposes, there is no part that is reproduced without written permission. Content is provided with information only.
