电催化剂
纳米颗粒
计算机科学
还原(数学)
人工神经网络
材料科学
选择性
刮擦
机器学习
分子动力学
纳米技术
计算模型
吸附
实现(概率)
可扩展性
生物系统
水准点(测量)
人工智能
计算复杂性理论
工作(物理)
蒸汽重整
格式化
电化学
铂纳米粒子
曲面(拓扑)
化学
催化作用
转化(遗传学)
Atom(片上系统)
作者
WooSeok Lee,Dong Hyeon Mok,Hyun Dong Jung,Hyun‐Seok Cho,Seoin Back
标识
DOI:10.1021/acscatal.6c06348
摘要
Abstract The conventional computational approach to electrocatalyst evaluation has long been limited by its reliance on simplified slab models and adsorption energies on bare surfaces, which fail to capture the structural complexity of realistic nanoparticles (NPs) and the presence of adsorbates that decorates electrode surfaces under working conditions. To bridge this gap, we present a multi-scale computational framework that combines machine learning potential (MLP)-driven molecular dynamics (MD) simulations with the explicit realization of potential-dependent surface coverage and high-throughput evaluation of active sites. Applying this framework to the CO2 reduction reaction (CO2RR) on Cu–Pd alloys, we generated realistic 8 nm NP morphologies using a SevenNet MLP trained from scratch and determined the coverage states and binding energies of key intermediates (CO*, H*, and OH*) across all surface sites using a zero-shot universal MLP. Our coverage-resolved evaluation reveals a systematic transformation of product selectivity across the composition–potential space, from formate production on Cu-rich NPs to CO evolution and the competing hydrogen evolution reaction (HER) on their Pd-rich counterparts. Notably, a discernible C1+ fraction emerges on the Cu3Pd alloy at intermediate overpotentials, reflecting the promotional effect of dilute Pd incorporation. This work establishes a methodology that bridges the gap between atomistic predictions and the structural and electrochemical complexity of working nanocatalysts, paving the way for more accurate in-silico discovery of electrocatalysts.
科研通智能强力驱动
Strongly Powered by AbleSci AI