Chemistry is discussed and rationalized in terms of concepts rooted in quantum mechanics that have their physical counterpart in large, intricate and often opaque data produced by solving the Schrödinger equation. This thesis project aims at developing a machine-learning framework for the characterization of chemical bonding based on volumetric data of molecular electron densities obtained by state-of-the-art quantum-chemistry calculations, with a focus on latent relationships between hidden data features and well-established concepts (e.g., donation, backdonation, hydrogen bonding). Applications will include comparison with available bond-analysis methods [1, 2] and the characterization of molecular systems exhibiting controversial features.

Generative artificial-intelligence (AI) models have recently emerged as a new paradigm for molecular discovery [1, 2]. Within these models, a discrete molecular space is typically mapped into a continuous latent space which can then be explored for molecule extrapolation or interpolation driven by some optimization criteria. This thesis project will focus on the development and application of generative AI models for the discovery of new molecules featuring specific desired properties, with applications including drug discovery (e.g., targeting high efficiency for specific therapeutic purposes), advanced materials for energy harvesting, storage and conversion, and the design of sustainable deep-eutectic solvents for raw-material recycle.

Modern spectroscopic techniques may produce a large amount of data which can be difficult to analyse and interpret with conventional techniques. In fluorescence lifetime imaging microscopy (FLIM) [1, 2] of biological cells, for instance, for each pixel of a targeted area a time-resolved signal is produced, resulting in a rich data structure intimately connected with the cell activity. This thesis project will focus on the definition and training of machine-learning models for the analysis and interpretation of spectroscopic experimental results obtained by FLIM measurements.

Solving the time-dependent molecular Schrödinger equation typically involves the expansion of the molecular wavefunction in a set of time-independent electronic functions weighed by time-dependent nuclear functions. A promising emerging alternative approach is based on the idea of an exact factorization of the molecular wavefunction in a product of a single time-dependent nuclear function and a single time-dependent electronic function [1]. This thesis project will focus on the exploration of exact-factorization techniques for treating reactive quantum scattering [2] problems with applications to the modeling of nonadiabatic elementary chemical reactions.

The physics and chemistry of organic molecules adsorbed at metal surfaces plays an important role in many applications of scientific and technological interest including catalysis, light-emitting diodes, single-molecule junctions, molecular sensors and switches, and photovoltaics [1]. This thesis project will focus on the computational modeling of the structure and dynamics of such hybrid organic/inorganic systems via Density Functional Theory (DFT), with emphasis on the interpretation of experimental results including Scanning Tunneling Microscopy (STM) [2], capable of rendering images of the adsorbed molecules with atomic-level resolution.