Machine learning-driven molecular dynamics unveils a bulk phase transformation driving ammonia synthesis on barium hydride

Nature Communications, Journal Year: 2025, Volume and Issue: 16(1)

Published: March 12, 2025

The modern view of industrial heterogeneous catalysis is evolving from the traditional static paradigm where catalyst merely provides active sites, to that a functional material in which dynamics plays crucial role. Using machine learning-driven molecular simulations, we confirm this picture for ammonia synthesis catalysed by BaH2. Recent experiments show system acts as highly efficient catalyst, but only when exposed first N2 and then H2 chemical looping process. Our simulations reveal N2, BaH2 undergoes profound change, transforming into superionic mixed compound, BaH2−2x(NH)x, characterized high mobility both hydrides imides. This transformation not limited surface involves entire catalyst. When compound second step process, readily formed released, process greatly facilitated ionic mobility. Once all nitrogen are hydrogenated, reverts its initial state, ready next cycle. microscopic analysis underlines dynamic nature does serve platform reactions, rather it entity evolves under reaction conditions. shifting paradigm. Here, authors reactions during synthesis,

Language: Английский

---

Modeling Dynamic Catalysis at ab Initio Accuracy: The Need for Free-Energy Calculation

ACS Catalysis, Journal Year: 2024, Volume and Issue: 14(21), P. 16086 - 16097

Published: Oct. 17, 2024

InfoMetricsFiguresRef. ACS CatalysisASAPArticle This publication is free to access through this site. Learn More CiteCitationCitation and abstractCitation referencesMore citation options ShareShare onFacebookX (Twitter)WeChatLinkedInRedditEmailJump toExpandCollapse ViewpointOctober 17, 2024Modeling Dynamic Catalysis at ab Initio Accuracy: The Need for Free-Energy CalculationClick copy article linkArticle link copied!Qi-Yuan FanQi-Yuan FanEngineering Research Center of Ministry Education Fine Chemicals, School Chemistry Chemical Engineering, Shanxi Key Laboratory Coal-based Value-added Chemicals Green Synthesis, University, Taiyuan 030006, ChinaMore by Qi-Yuan FanFu-Qiang GongFu-Qiang GongState Physical Solid Surface, Collaborative Innovation Energy Materials (iChEM), College Xiamen 361005, Fu-Qiang GongYun-Pei LiuYun-Pei LiuState ChinaLaboratory AI Electrochemistry (AI4EC), IKKEM, Yun-Pei LiuHao-Xuan ZhuHao-Xuan ZhuState Hao-Xuan ZhuJun Cheng*Jun ChengState ChinaInstitute Artificial Intelligence, China*Email: [email protected]More Jun Chenghttps://orcid.org/0000-0001-6971-0797Open PDFACS CatalysisCite this: Catal. 2024, 14, XXX, 16086–16097Click citationCitation copied!https://pubs.acs.org/doi/10.1021/acscatal.4c05372https://doi.org/10.1021/acscatal.4c05372Published October 2024 Publication History Received 4 September 2024Accepted 8 2024Revised 7 2024Published online 17 2024article-commentary© American Society. available under these Terms Use. Request reuse permissionsThis licensed personal use Publications© SocietySubjectswhat are subjectsArticle subjects automatically applied from the Subject Taxonomy describe scientific concepts themes article.CatalystsChemical reactionsFree energyMetal clustersStructural dynamics1. IntroductionClick section linkSection copied!Heterogeneous catalysis plays an increasingly important role in modern chemical industry. active site, as proposed Taylor, (1) one most fundamental heterogeneous key understanding catalytic mechanisms. Over years, considerable effort has been dedicated exploring identifying atomic compositions geometric configurations sites. In traditional surface science studies conducted low temperatures, sites often regarded certain fixed structures with well-defined arrangements, such terrace, step, edge (Figure 1A), assumed remain unchanged reaction conditions. However, recent decades, it widely accepted that undergo dynamic changes proceeds, challenging notion static 1B).Figure 1Figure 1. Schematic illustration catalyst catalysis. (A) site nanocatalysts special arrangement. (B) evolution structures. (C) energy profiles stable metastable structures.High Resolution ImageDownload MS PowerPoint SlideUnderstanding crucial precise design catalysts. situ or operando experimental measurements essential capturing nature Among used techniques electron microscopy spectroscopy high spatial temporal resolution. former, exemplified transmission (TEM), can directly capture structural catalysts conditions scale, enabling real-time tracking rearrangements on including reconstruction, phase transition, size variations. latter provides detailed information about structures, charge transfer, bond vibration, coordination number, etc. For example, extended X-ray absorption fine structure (EXAFS) be obtain number interatomic distances. Combining allows a comprehensive analysis behaviors, providing novel insights into obtaining thermodynamic kinetic reactions remains challenging.In addition techniques, theoretical calculations initio accuracy statistical sampling configurational space allow us visualize study underlying mechanisms Global optimization (GO) molecular dynamics (MD) two common approaches modeling dynamics. These methods have significantly enriched our behavior GO commonly explore potential (PES) determine catalyst. instance, various approaches, genetic algorithms, basin hopping, stochastic walking (SSW), extensively developed studying cluster (2,3) Once shape identified, global minimum several other possible low-energy isomers selected further corresponding With efficient algorithm, approach reliably generate located PES; however, implies higher probability occurrence does not necessarily represent largest contribution activity finite temperatures. contrast, some high-energy may contribute more process 1C). mentioned above confirmed significance isomers, while mechanistic still depends initially using 0 K. based assumption: time scale elementary (typically order picoseconds) shorter than (which occur macroscopic scales). disparity decoupling when scales comparable, strong coupling would unavoidable. mode, becomes part coordinate. Thus, perspective required accurately understand intrinsic activity.Different GO, MD employed examine calculate properties ensemble. Classical typically employs predefined empirical potentials. due their oversimplified representations interactions, no force fields formation breaking types bonds. classical suitable simulations reactions. Ab (AIMD) assesses forces accurate electronic calculations. Consequently, exactly take account contributions reactions, thereby reflection realistic conditions.A series Cheng et al. (4,5) reinvigorated dynamics, which they described ensemble many rather solely focusing one. By combining AIMD calculation methods, rigorously calculated energies entropy caused Their findings revealed highly flexible temperatures substantial impact profiles. Notably, observed anomalous nonlinear curve increasing temperature, attributed adsorption-induced solid-to-liquid transition Importantly, entropic effect help optimize working reactions.Although successfully conditions, limited its computational cost only relatively small systems, comprising hundreds atoms tens picoseconds. inefficient processes complex Recently, machine learning methodologies, especially potentials (MLPs), emerged promising tools MLPs decompose total system sum contributions, each being function local environment. constructing one-to-one mapping environments forces, enable reproduction PES. Using accelerate simulations, i.e. (MLMD), enhance efficiency orders magnitude maintaining accuracy. achieved because trained high-quality data derived quantum mechanical calculations, density functional theory (DFT). Nowadays, shown powerful tool chemistries, (6−9) networks. (10−12)In Viewpoint, we will provide brief overview field well challenges, aiming achieve following section, first briefly introduce research terms separately. Subsequently, compare illustrate importance rigorous end, discuss effects transitions reactions.2. Structural Dynamics CatalystsClick copied!Elucidating how specific arrangements facilitate central goals chemists across generations. combination advanced makes characterize unravel level. summarize applications characterization Experimental Characterizations Catalyst DynamicsThe long rooted approach, where experiments single-crystal surfaces ultrahigh-vacuum (UHV) Ertl co-workers made outstanding research. They emphasized complexity catalysis, typical examples: ammonia synthesis CO oxidation. (13) adsorption facet Pt(110), scanning tunneling (STM) UHV 300 K missing row reconstruction clean K, showing static. (14) addition, Somorjai frequency generation (SFG), diffraction (LEED), auger (AES) monitor different kinds single crystal low-pressure environments. built high-pressure cells investigate C–H activation ethylene Pt(111) surface, (15) C–O Pt surfaces, (16) (17) indicate during those vacuum, emergence new sites.While investigations provided valuable mechanisms, highlighted limitations fully complexities real systems. To bridge gap between ideal model systems industrially relevant researchers turned toward measurements, catalysis.Over past decade, numerous metal nanoparticles (e.g., Au, Pt, Cu, Ni), shed light impacts (18−21) general, behaviors include changes, transitions, composition studies, play since direct visualization over time. Hansen (22) performed atom-resolved change supported Cu environmental microscope (ETEM). work Later, Hakeda (23) also ETEM CeO2-supported Au(100) adsorbed molecules room temperature. demonstrated Au undergoes response surrounding gas Additionally, were Wang co-workers. (24) found larger sizes rigid whereas ultrasmall clusters liquid-like disordered 2A). size-dependent critical enhancing reactivity selectivity, emphasizing designing performance.Figure 2Figure 2. CeO2(111) vacuum CO+O2 atmosphere. models images nm NPs CeO2 (a–c), <2 (d–f), ∼2 layer (g–i). melting Au561 particle temperature characterized high-angle annular dark-field STEM. Pt@MCM-22 Panel reprinted permission ref (24). Copyright 2018, National Academy Sciences. reproduced (25). Available CC-BY 4.0 license. 2019 Author Name(s). (28). 2018 Name(s).High SlideAnother factor influencing offer thermal energy, enhances ability overcome barriers reconstruction. Previous works (2–5 diameter) carbon films occurs heating stage aberration-corrected (ac-STEM); (25) elevated form solid core-liquid shell 2B), confirming transformation sensitive temperature.In photoelectron (XPS), (XAS), EXAFS, track distance, state. Tao (26) dramatic reversible core–shell Rh0.5Pd0.5 state XPS Torr pressure range. It worth noting difficult reach technique. Instead, integration multiple necessary complementary aspects, establishing correlation performance. Liu (27) recently conversion Ni@Au bimetallic CO2 hydrogenation combined variety synchrotron XAS, infrared spectroscopy, simulations. selectivity was transiently reconfigured alloy promoted adsorption.In cases, rearrangement migration accompanied sizes. Corma (28−30) confined inside zeolite among isolated species, clusters, oxidation water–gas shift 2C). Such Al2O3 TiO2-supported catalysts, (31,32) zeolite-confined PtSn clusters. (33,34)Other changes. Matsuda his (35) TEM Ni hydrocarbon reformation. observations 250 350 °C, close-packed plane expanded, dissolve nanocatalysts, leading FCC HCP Ni. Besides, optimizing (36) investigated carburization, oxidation, coking Fe particles Fischer–Tropsch (FT) high-resolution ETEM. Fe5C2, dynamically formed Fe5C2 responsible FT reaction. Luo (37) captured conditions; exposure leads roughening consequently low-coordinated atoms, O2 induces quasi-crystalline CuOx phase. highlight significant influence gaseous environments, composition, Computational Modeling DynamicsIn parallel microscopic level, giving Recent years witnessed efforts subnanometer garnering particular attention distinct physical properties. smaller intriguing counterparts unique PESs rich minima comparable energies. Under high-temperature energetically similar equilibrium, exhibiting fluxionality, interconvert configurations.In attempting methods. (3,38) (39) utilized SSW method search lowest 3A). yielded thousands candidate size, demonstrating Alexandrova Sautet interesting (40−42) isomerization identified both Pt7/Al2O3 exhibit fluxional 3B). Moreover, Pt13 methane compared lowest-energy suggesting even dominate activity. (43) A phenomenon, show Mo/Al2O3 (44) mentioning sampled usually limited, e.g., configurations. Furthermore, generally consider ensemble, comprehensively.Figure 3Figure 3. PtN search. Isomerization network connecting 30 via paths. Initial final snapshots Au20 eight adsorption. (D) workflow, composed training potential, exploration configuration spaces, labeling (39). 2016 (41). (47). 2015 aspect technique purpose. address issue perspective. Salahub (45) (QM/MM) simulate taking pressure, solvent environment account. very having great advancements application elucidated Li (24,46,47) studied ceria-supported gold varying AIMD. atom upon molecular, could reintegrate back once complete 3C). results understanding. Similar single-atom Rh exposed (48) Beyond proven invaluable nanoporous Speybroeck (49,50) within channels zeolite/MOF included stability, diffusion behavior, nucleation reactant. striking example metal-zeolite become mobilized interaction species. (51)Additionally, Monte Carlo (KMC) effective incorporates events adsorption, desorption, diffusion. advantage KMC lies inhomogeneities correlations reactant distributions (52) realm grown popularity performance scale. Xu phenomenological kinetics (XPK) method, (53,54) KMC, syngas Rh(111) surface; (55) intermediate induce optimizes own pathways, overall selectivity. carried out migrations atoms; Mavrikakis (56) Cu(111) As another Gao (57) simulated reshaping atomistic nanoparticle surface. Despite fact demonstrates technical features wide faces expense develop chemistries lack general framework effects.It coupled coordinates differs re-equilibration before after much dramatically affect mechanism. context, along coordinate obtained, owing thus influences considered. poi

Language: Английский

---

Machine Learning Potentials for Heterogeneous Catalysis

ACS Catalysis, Journal Year: 2025, Volume and Issue: 15(3), P. 1616 - 1634

Published: Jan. 15, 2025

The production of many bulk chemicals relies on heterogeneous catalysis. rational design or improvement the required catalysts critically depends insights into underlying mechanisms atomic scale. In recent years, substantial progress has been made in applying advanced experimental techniques to complex catalytic reactions operando, but order achieve a comprehensive understanding, additional information from computer simulations is indispensable cases. particular, ab initio molecular dynamics (AIMD) become an important tool explicitly address atomistic level structure, dynamics, and reactivity interfacial systems, high computational costs limit applications systems consisting at most few hundred atoms for simulation times up tens picoseconds. Rapid advances development modern machine learning potentials (MLP) now offer promising approach bridge this gap, enabling with accuracy small fraction costs. Perspective, we provide overview current state art MLPs relevant catalysis along discussion prospects use science years come.

Language: Английский

---

From “Single Sites” to Stable Nanoparticles Derived from Spray-Flame Synthesized Solid Solutions of Cobalt in MgO for Ammonia Decomposition

ACS Catalysis, Journal Year: 2025, Volume and Issue: unknown, P. 5781 - 5795

Published: March 24, 2025

Language: Английский

---

Atomistic Details of Nanocluster Formation from Machine-Learned-Potential-Based Simulations

Nano Letters, Journal Year: 2025, Volume and Issue: unknown

Published: March 31, 2025

Understanding the mechanism for formation of metal nanoclusters is an open challenge in nanoscience. Computational modeling can provide molecular details nanocluster that are otherwise inaccessible. However, simulating nucleation solution presents significant challenges, including inaccurate energy predictions and limitations on system size time scale. This work addresses these challenges by combining deep neural networks (DNNs) with well-tempered metadynamics (WT-MetaD) to model a prototypical nanocluster, Ag6(SCNH2)6 methanol. A neural-network-potential-based unbiased dynamics simulation captured cluster's dynamic behavior, while WT-MetaD simulations revealed almost barrierless transition from dispersed precursors nucleated state. The method's robustness was further demonstrated scaling up 30 randomly distributed precursors, which resulted spontaneous nucleation. study first successful DNN density-functional-theory-level accuracy, paving way advancements field.

Language: Английский

---

Everything everywhere all at once: a probability-based enhanced sampling approach to rare events

Nature Computational Science, Journal Year: 2025, Volume and Issue: unknown

Published: May 5, 2025

Language: Английский

---

ML-Accelerated Automatic Process Exploration Reveals Facile O-Induced Pd Step-Edge Restructuring on Catalytic Time Scales

ACS Catalysis, Journal Year: 2024, Volume and Issue: unknown, P. 514 - 522

Published: Dec. 20, 2024

Language: Английский

---

Data efficient machine learning potentials for modeling catalytic reactivity via active learning and enhanced sampling

npj Computational Materials, Journal Year: 2024, Volume and Issue: 10(1)

Published: Dec. 19, 2024

Language: Английский

---