Computational models in heterogeneous catalysis have traditionally treated catalysts as static entities with well-defined active sites. While this approach has led to important insights, it overlooks the influence of operative conditions, such as temperature, pressure, and reactant flow, on the structure and behavior of catalysts. Growing experimental evidence indicates that catalysts are inherently dynamic systems, whose activity depends on their ability to undergo structural transformations under reaction conditions. This perspective calls for a revision of traditional computational approaches. With increasing computational power and the integration of machine learning algorithms, computational catalysis is undergoing a paradigm shift. Simulations are now capable of exploring catalyst behavior under realistic operating conditions, providing atomistic insights into their dynamic nature. Within this emerging framework, catalysis must be regarded as a collective, dynamic, and non-local phenomenon, in which the entire system evolves in response to external variables and actively participates in the reactive process. This thesis contributes to this vision through the use of molecular dynamics simulations, enhanced by machine learning interatomic potentials, as the primary computational tool. Three case studies are presented, in which the structural dynamics of the catalysts play a central role in determining reactivity. The first case investigates ammonia synthesis catalyzed by barium hydride (BaH2) in a chemical looping process. Simulations reveal that, under operative conditions, the material undergoes transformations involving the entire bulk and not just the surface, indicating that the whole solid actively participates in the catalytic process. The second case addresses the electrochemical reduction of CO2 at the Ag(111)/H2O interface. The applied electrochemical potential induces a strong reorganization of the interfacial water layer, altering its orientation and hydrogen-bond network. This collective behavior of the solvent contributes to stabilizing intermediates, facilitating proton transfer, and modulating reaction kinetics. Finally, ammonia synthesis is explored in lithium-sodium alloys, proposed as liquid catalysts due to their intrinsic dynamism. The highly fluxional interface between the two metals creates a microheterogeneous environment where continuous structural rearrangements and synergy between the two components drive catalytic activity. These studies highlight the crucial role of operative conditions in shaping the structure and dynamics of catalysts. Such phenomena, invisible to traditional static models, can only be understood by adopting a microscopic and dynamic perspective, enabled by recent advances in atomistic simulations guided by machine learning.
I modelli computazionali nella catalisi eterogenea hanno tradizionalmente trattato i catalizzatori come entità statiche con siti attivi ben definiti. Sebbene questo approccio abbia portato a importanti intuizioni, trascura l’influenza delle condizioni operative, come temperatura, pressione e flusso di reagenti, sulla struttura e sul comportamento dei catalizzatori. Evidenze sperimentali sempre più numerose mostrano che i catalizzatori sono sistemi intrinsecamente dinamici, la cui attività dipende dalla capacità di subire trasformazioni strutturali sotto condizioni di reazione. Questa visione impone una revisione degli approcci computazionali tradizionali. Grazie all’aumento della potenza computazionale e all’integrazione di algoritmi di machine learning, la catalisi computazionale sta vivendo un cambio di paradigma. Le simulazioni permettono ora di esplorare il comportamento dei catalizzatori in condizioni operative realistiche, offrendo una comprensione atomistica della loro natura dinamica. In questo contesto, la catalisi deve essere considerata come un fenomeno collettivo, dinamico e non locale, in cui l’intero sistema evolve in risposta alle variabili esterne e partecipa attivamente al processo reattivo. In questa tesi si contribuisce a questa visione mediante l’uso della dinamica molecolare, potenziata da potenziali interatomici basati su machine learning, come principale strumento di simulazione. Vengono presentati tre casi di studio in cui la dinamica strutturale dei catalizzatori gioca un ruolo centrale nel determinare la reattività. Nel primo caso si studia la sintesi dell’ammoniaca catalizzata da idruro di bario (BaH2) in un processo di chemical looping. Le simulazioni rivelano che, sotto condizioni operative, il materiale subisce trasformazioni che coinvolgono l’intero bulk e non solo la superficie, indicando una partecipazione attiva di tutto il solido al processo catalitico. Il secondo caso riguarda la riduzione elettrochimica della CO2 all’interfaccia Ag(111)/H2O. Il potenziale elettrochimico applicato causa una profonda riorganizzazione dello strato d’acqua interfacciale, alterando orientamento e rete di legami a idrogeno. Questo comportamento collettivo del solvente contribuisce a stabilizzare gli intermedi, favorire il trasferimento protonico e modulare la cinetica di reazione. Infine, viene esplorata la sintesi dell’ammoniaca in leghe litio-sodio, proposte come catalizzatori liquidi grazie alla loro dinamica intrinseca. L’interfaccia altamente flussionale tra i due metalli crea un ambiente microeterogeneo dove continue riorganizzazioni strutturali e sinergia tra i due metalli guidano l’attività catalitica. Attraverso questi studi, si evidenzia il ruolo cruciale delle condizioni operative nella modulazione della struttura e della dinamica dei catalizzatori. Tali fenomeni, invisibili ai modelli statici, possono essere compresi solo adottando una prospettiva microscopica e dinamica, resa possibile dagli sviluppi recenti delle simulazioni atomistiche guidate dal machine learning.
Machine learning methods for capturing the dynamic nature of heterogeneous catalysts under operando conditions
TOSELLO GARDINI, AXEL
2026
Abstract
Computational models in heterogeneous catalysis have traditionally treated catalysts as static entities with well-defined active sites. While this approach has led to important insights, it overlooks the influence of operative conditions, such as temperature, pressure, and reactant flow, on the structure and behavior of catalysts. Growing experimental evidence indicates that catalysts are inherently dynamic systems, whose activity depends on their ability to undergo structural transformations under reaction conditions. This perspective calls for a revision of traditional computational approaches. With increasing computational power and the integration of machine learning algorithms, computational catalysis is undergoing a paradigm shift. Simulations are now capable of exploring catalyst behavior under realistic operating conditions, providing atomistic insights into their dynamic nature. Within this emerging framework, catalysis must be regarded as a collective, dynamic, and non-local phenomenon, in which the entire system evolves in response to external variables and actively participates in the reactive process. This thesis contributes to this vision through the use of molecular dynamics simulations, enhanced by machine learning interatomic potentials, as the primary computational tool. Three case studies are presented, in which the structural dynamics of the catalysts play a central role in determining reactivity. The first case investigates ammonia synthesis catalyzed by barium hydride (BaH2) in a chemical looping process. Simulations reveal that, under operative conditions, the material undergoes transformations involving the entire bulk and not just the surface, indicating that the whole solid actively participates in the catalytic process. The second case addresses the electrochemical reduction of CO2 at the Ag(111)/H2O interface. The applied electrochemical potential induces a strong reorganization of the interfacial water layer, altering its orientation and hydrogen-bond network. This collective behavior of the solvent contributes to stabilizing intermediates, facilitating proton transfer, and modulating reaction kinetics. Finally, ammonia synthesis is explored in lithium-sodium alloys, proposed as liquid catalysts due to their intrinsic dynamism. The highly fluxional interface between the two metals creates a microheterogeneous environment where continuous structural rearrangements and synergy between the two components drive catalytic activity. These studies highlight the crucial role of operative conditions in shaping the structure and dynamics of catalysts. Such phenomena, invisible to traditional static models, can only be understood by adopting a microscopic and dynamic perspective, enabled by recent advances in atomistic simulations guided by machine learning.| File | Dimensione | Formato | |
|---|---|---|---|
|
phd_unimib_896514.pdf
accesso aperto
Licenza:
Tutti i diritti riservati
Dimensione
11.36 MB
Formato
Adobe PDF
|
11.36 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/368815
URN:NBN:IT:UNIMIB-368815