Staff Profile

Key Interests

• Atomistic Molecular Dynamics (MD) simulations
• Machine Learning (ML) potentials for molecular systems
• ML-driven analysis of transport processes
• Modelling linear and non-linear spectroscopies from MD
• Applications to energy-materials and medicinal-chemistry

Education & Research Experience:

• Research Associate, Machine Learning for Molecular Modelling, University of Cambridge, UK (2020-2023)
• Postdoctoral Scholar, Theoretical and Computational Chemistry, Caltech, USA (2016-2020)
• PhD, Condensed Matter Physics, University of Edinburgh, UK (2012-2016)
• BSc, Physics, Jacobs University Bremen, Germany (2009-2012)

Research Background

I joined Newcastle University as a Lecturer in April 2023 and I am excited to start my own group! My research background is in computational and data-driven science with interdisciplinary training across condensed matter physics, theoretical chemistry, and molecular modelling.

For my PhD, under the supervision of Prof. Graeme Ackland at the University of Edinburgh, I worked on understanding the behaviour of solid hydrogen and deuterium at extreme conditions. These states of matter form at very high pressures in the core of large gas planets like Jupiter. Despite the simplicity of the hydrogen atom itself, the condensed phase is surprisingly complex: as the pressure pushes the electrons out of the H₂ molecular bond, the crystal structures exhibit a plethora of exotic behaviours. Given enough pressure hydrogen is predicted to become metallic or even superconductor! These solid phases are obtained in controlled lab conditions using diamond anvil cells, but they are difficult to characterize, often time the only available fingerprints are Infrared and Raman spectra and bridging the gap between experiment and room-temperature dynamics simulations makes for an exciting challenge!

As part of my postdoctoral work with Prof. Thomas Miller, I studied the Terahertz-Raman spectra of molecular liquids - a novel type of nonlinear spectroscopy that is highly sensitive to molecular dynamics in the condensed phase. The peaks we find in linear spectra like Infrared and Raman are associated with single leaps on the ladder of vibrational quantum states and usually inform on the energy of a single transition (frequency) and the probability of that interaction (intensity). In contrast, nonlinear spectroscopy probes significantly more complex scattering events that can involve a succession of photon-vibron interactions and informs on the anharmonicity and nonlinear coupling of the normal modes. Understanding these convoluted many-body responses through theory and simulations is challenging, especially in the low-frequency Terahertz regime, but doing so can help us directly pinpoint the molecular-level processes at play in condensed-phase chemical dynamics!

In a separate project at Caltech, I developed Machine Learning approaches to investigate solvation and ion transport in thiophene-based polymers binders used in battery applications. Battery electrodes are normally polycrystalline and need to be embedded in a polymer binder which provides adhesion and mechanical stability. These thiophene-based polymers are appealing because they can simultaneously conduct both electrons and ions which could help improve the efficiency of batteries. The challenge lies in understanding the molecular mechanisms underpinning the ion transport, which could guide the design of better performing materials. Using Machine Learning approaches can help automatically classify and characterize solvation environments based on local atomic geometries, and ultimately help disentangle these complex molecular processes.

In my second postdoc, with Prof. Gábor Csányi at University of Cambridge, I developed Machine Learning interatomic potentials for liquid solvents which are also a key component of Li-ion batteries. These solvents need to simultaneously accomplish two tasks: separating the ion pairs and facilitating the ion diffusion, which imply competing requirements for the solvent properties. In practice this is achieved by using multi-component liquids involving a mixture of high-dipole moment molecules which help screen the ion interaction, and low-dipole moment molecule which decrease the overall viscosity and improve diffusion. Machine Learning interatomic potentials for these complex molecular systems present a significant challenge owing to the large separation of intra- and inter-molecular interactions. Developing new methods to address these issues is exciting and the results will likely impact other areas of modelling, including biophysics and medicinal Chemistry.

Find me on: Google Scholar, ORCID, LinkedIn, Twitter