projects
Current and recent research projects
Modelling Nebular Emission Lines
Stellar population synthesis models are essential for studying galaxy evolution: they let physical parameters such as stellar mass and star formation rate be inferred from spectrophotometric data, and link theoretical models to their predicted emission. JWST has revealed large numbers of very young, high-redshift galaxies whose light is dominated by strong emission lines, making accurate nebular modelling essential, as shown in Vijayan et al. (2026), to which I was a major contributor.
Motivated by this, I used the photoionisation code Cloudy to add nebular emission to new Maraston stellar population models, which account for stellar rotation, and compared the results with JWST data. Models built on different physical assumptions diverged substantially, with some single-star models predicting more high-energy line emission, notably [O III], than binary-star models. These discrepancies mean that galaxy properties inferred from observations, and predictions from cosmological simulations, can shift significantly depending on the stellar model assumed, a source of uncertainty that had not previously been systematically quantified and that highlights the need for careful model selection.
Relevant papers
Testing the Newest Dust Prescriptions
To interpret observed galaxy populations, I combine stellar and AGN population synthesis models with cosmological simulations and semi-analytic models (SAMs) of galaxy formation. These tools bridge small-scale physical processes, such as star formation and feedback, with large-scale structure formation in the Universe.
I am preparing a publication testing whether relationships linking dust behaviour to a galaxy's physical properties, learned from the IllustrisTNG50 and TNG100 simulations (Sommovigo et al. 2025), can be applied to semi-analytic models (Somerville et al. 2008). SAMs simulate the formation and evolution of galaxies using simplified physical recipes rather than the full physics solved by a cosmological simulation, and are consequently far cheaper to run. The key test is whether, when combined with these models, the dust relationships produce galaxy population statistics that match those observed in the Universe. Our physically grounded dust relationships reproduce observed galaxy population statistics well, showing that fast, cheap SAMs can now generate realistic mock observations directly comparable to real survey data, without needing to run a full, expensive cosmological simulation each time.
Relevant papers
Applying Simulation-Based Inference to Large Simulation Suites
Traditional approaches to parameter inference rely on explicit likelihood functions, which are often intractable in complex astrophysical models. Simulation-Based Inference (SBI) offers a powerful alternative by directly learning the mapping between simulated data and model parameters using machine learning. This framework allows one to extract posteriors even when the likelihood is unknown, by training neural networks on forward-modeled simulations.
I am currently applying SBI across several major cosmological simulation suites, including IllustrisTNG, SIMBA and EAGLE, to determine the dust prescriptions required to reproduce key observational constraints, such as the abundance of bright galaxies at different epochs in the Universe's history. SBI makes it possible to efficiently explore the high-dimensional parameter space of the dust models. By constraining the dust prescriptions for each flagship simulation, my work enables a publication in preparation by the Synthesizer team that will produce an unprecedented library of synthetic observations spanning a wide range of galaxy formation models. This will provide a valuable resource for interpreting the wealth and diversity of galaxies revealed by the latest wide-field surveys, allowing us to systematically connect observed galaxy populations to the underlying physics of galaxy formation and evolution.