Automated first-principles highways for Mars exploration

The ambitious goal of a human mission to Mars brings forth new, challenging scientific problems posed by the planet's harsh environment. Central is the scarcity of natural resources on the planet, leaving its atmosphere and its regolith as the only sources of substances, such as oxygen and metals. The lack of fossil fuels particularly demands a technological paradigm shift, having electricity as the core driving force. A full electrochemical approach is therefore the only viable option, mainly based on molten salts electrolysis, but current solutions are not yet capable of selectively recovering metals. Moreover, targeted metal extraction highly depends on the composition of the local regolith, calling for fast and reliable recognition techniques, for instance, through vibrational spectra. Acceleration in these fields would be facilitated by having large databases of materials properties of interest, which help engineering optimal conditions, and comparing spectra. To date, both the design of metal electrowinning reactors and the recognition of minerals from vibrational spectra are largely limited by the scarce availability of data. First-principles simulations can be highly valuable to tackle this scarcity scenario and to provide a systematic framework to create such databases at large scale. However, several specific computational and methodological challenges for the generation of vibrational fingerprints and thermochemical data must be overcome. The central challenge lies in the lack of automated, accurate, and efficient methods designed to work together. In this thesis, we address this by developing a fully automated collection of novel approaches and workflows suitable for high-throughput applications, each tailored to a specific area, and, more importantly, designed to interoperate seamlessly. We start by focusing on the foundational issue of the predictive accuracy of density-functional theory (DFT) for transition metal compounds. We employ Hubbard-corrected DFT in its extended formulation, DFT+U+V, and develop an automated workflow that can compute from first-principles structurally and electronically self-consistent Hubbard parameters. Building on this foundation, we introduce a functional-agnostic, ab-initio workflow for infrared and Raman spectroscopy that combines finite fields and finite displacements to obtain Born effective charges, dielectric and Raman tensors, and phonons. To enable engineering selective electroreduction, we derive a grand-potential geometric formalism that generalizes corrosion stability diagrams in molten salts, explicitly parameterized by oxide-ion activity and applied potential. To improve the accuracy of finite-temperature free energies for the prediction of solid-dissolved species electrochemical equilibria, we then develop two distinct approaches for their efficient calculation. For solids, we couple the stochastic self-consistent harmonic approximation with on-the-fly actively trained machine-learning interatomic potentials. To address the thermodynamics of dissolved species, we introduce an automated framework that couples molecular dynamics methods with equivariant neural network force fields and ensemble uncertainty quantification schemes. Together, these contributions deliver a coherent highway of automation for a scarcity-driven paradigm.