Loreti, Alessandro
(2026)
Broadening the horizons of quantum-mechanical methods through synergistic approaches, [Dissertation thesis], Alma Mater Studiorum Università di Bologna.
Dottorato di ricerca in
Chimica, 38 Ciclo. DOI 10.48676/unibo/amsdottorato/13327.
Documenti full-text disponibili:
Abstract
Accurately simulating photo-induced dynamics of molecular systems and their spectroscopic fingerprints under environmentally relevant conditions remains a major challenge in theoretical chemistry, requiring a balance between efficient sampling of nuclear motion and accurate electronic structure description for reliable computation of observables. Single-reference methods provide affordable on-the-fly dynamics but cannot describe near-degenerate regions involving the ground state, nor capture spectroscopic transitions beyond singly-excited character; multi-reference approaches recover this physics at a cost that limits dynamical sampling. A similar tension affects environment treatment: implicit models gain speed by sacrificing atomistic detail, while explicit schemes must manage a dramatic increase in degrees of freedom.
This thesis offers three methodological contributions addressing these gaps through algorithms that reconcile different quantum-chemical methods or environment descriptions. The first part introduces a diagnostic tool computing approximate wavefunction overlaps between single- and multi-reference levels of theory at reduced cost, allowing multi-reference-quality spectroscopic signals to be extracted on top of efficient single-reference on-the-fly dynamics, demonstrated for the internal conversion mechanism of acetylacetone. The second part presents an algorithm coupling a fully quantum-mechanical treatment of the system with an active molecular-mechanics environment, in which the bidirectional coupling between the two regions is computed dynamically throughout the simulation. Applied to excitation energy transfer in the NADH coenzyme, it reveals a fast population-transfer channel inaccessible to static environment descriptions. The third part leverages machine learning to extend detection of the S1/S0 coupling to single-reference frameworks, predicting multi-reference-quality overlaps directly from molecular geometries via neural network architectures. Using ethylene as a benchmark, the feasibility of this approach is established, laying the groundwork for its future integration into nonadiabatic dynamics frameworks. Collectively, the general applicability of these protocols demonstrates sufficient flexibility to be tailored to broader problems, paving the way for accurate and efficient dynamic simulations of complex molecular systems.
Abstract
Accurately simulating photo-induced dynamics of molecular systems and their spectroscopic fingerprints under environmentally relevant conditions remains a major challenge in theoretical chemistry, requiring a balance between efficient sampling of nuclear motion and accurate electronic structure description for reliable computation of observables. Single-reference methods provide affordable on-the-fly dynamics but cannot describe near-degenerate regions involving the ground state, nor capture spectroscopic transitions beyond singly-excited character; multi-reference approaches recover this physics at a cost that limits dynamical sampling. A similar tension affects environment treatment: implicit models gain speed by sacrificing atomistic detail, while explicit schemes must manage a dramatic increase in degrees of freedom.
This thesis offers three methodological contributions addressing these gaps through algorithms that reconcile different quantum-chemical methods or environment descriptions. The first part introduces a diagnostic tool computing approximate wavefunction overlaps between single- and multi-reference levels of theory at reduced cost, allowing multi-reference-quality spectroscopic signals to be extracted on top of efficient single-reference on-the-fly dynamics, demonstrated for the internal conversion mechanism of acetylacetone. The second part presents an algorithm coupling a fully quantum-mechanical treatment of the system with an active molecular-mechanics environment, in which the bidirectional coupling between the two regions is computed dynamically throughout the simulation. Applied to excitation energy transfer in the NADH coenzyme, it reveals a fast population-transfer channel inaccessible to static environment descriptions. The third part leverages machine learning to extend detection of the S1/S0 coupling to single-reference frameworks, predicting multi-reference-quality overlaps directly from molecular geometries via neural network architectures. Using ethylene as a benchmark, the feasibility of this approach is established, laying the groundwork for its future integration into nonadiabatic dynamics frameworks. Collectively, the general applicability of these protocols demonstrates sufficient flexibility to be tailored to broader problems, paving the way for accurate and efficient dynamic simulations of complex molecular systems.
Tipologia del documento
Tesi di dottorato
Autore
Loreti, Alessandro
Supervisore
Co-supervisore
Dottorato di ricerca
Ciclo
38
Coordinatore
Settore disciplinare
Settore concorsuale
Parole chiave
QM/MM nonadiabatic dynamics WFO overlap machine learning
DOI
10.48676/unibo/amsdottorato/13327
Data di discussione
15 Luglio 2026
URI
Altri metadati
Tipologia del documento
Tesi di dottorato
Autore
Loreti, Alessandro
Supervisore
Co-supervisore
Dottorato di ricerca
Ciclo
38
Coordinatore
Settore disciplinare
Settore concorsuale
Parole chiave
QM/MM nonadiabatic dynamics WFO overlap machine learning
DOI
10.48676/unibo/amsdottorato/13327
Data di discussione
15 Luglio 2026
URI
Statistica sui download
Gestione del documento: