Below are some brief summaries of research areas I am currently working in, or have recently worked on.

Machine Learning & Astronomy

I am keenly interested in the interface between simulations and machine learning methods. Whilst numerical models obviously do not represent the true universe, they do model the complex non-linear spatial and time dependent interactions of populations of objects. This can be important for accurately predicting intrinsic properties, something that traditional spectral energy distribution (SED) fitting techniques do not take into account. Training machines to learn these relationships, then applying these to observations, can provide unique predictions that complement existing techniques.

I recently worked with Prof. Viviana Acquaviva at City University New York applying this method to the prediction of Star Formation Histories (SFH) in the SDSS catalogue. We trained a Convolutional Neural Network to learn the relationship between spectra and SFH in the EAGLE and Illustris simulations. The paper is available here, and below is a talk I gave on this research at the Royal Astronomical Society meeting, “Machine Learning and Artificial Intelligence applied to astronomy” in March 2019 (slides available here)

Galaxy Protoclusters

Galaxy clusters are the largest collapsed objects in the universe, comprising of a highly evolved galaxy population embedded in a hot, rarefied InterCluster Medium (ICM). Their pre-collapse progenitors, known as galaxy protoclusters, are host to some of the most extreme objects (in terms of mass, star formation rate and nuclear activity) at these early times. Protoclusters are of significant interest for understanding the environmental dependence of galaxy evolution at early times, as well as the build-up, enrichment and heating of the ICM.

The dark matter distribution in a Protocluster at $z \sim 5$ simulated with the EAGLE code

Protoclusters do not yet host an X-ray emitting ICM, and so are primarily identified through 3D galaxy overdensities. In a recently accepted paper (Lovell et al. 2018) I studied in detail the relationship between galaxy overdensity and the presence and descendant mass of protoclusters in the L-galaxies semi-analytic model. The motivation for this work was to explore the systematic issues that have the greatest impact on protocluster identification. Surface overdensities of galaxies seen in narrow band photometric surveys are typically compared to simulations in order to evaluate their protocluster probability and estimate their descendant mass. I developed a more rigorous method for generating these statistics that takes in to account the completeness and purity of the protocluster galaxy population, the galaxy distribution shape, redshift space distortions and redshift uncertainties, as well as the coincidence of AGN with protoclusters.

Numerical Simulations

Cosmological hydrodynamic simulations have, in recent years, become capable of matching key distribution functions in the local universe, such as those of stellar mass and star formation rate. However, high resolution, large volume simulations have rarely been tested in the high redshift ($z > 5$) regime, particularly in the most overdense environments. Creating models that fit both high redshift and low redshift observables self consistently is a significant challenge, but key to understanding the properties of galaxies in the first billion years of the universe’s history, and how this affects their latter evolution. Such models are also necessary to make detailed predictions, and plan observations, for upcoming space based instruments, such as JWST, WFIRST and Euclid.

The distribution of gas in a Protocluster at $z \sim 5$, simulated with the EAGLE code

I have worked extensively on EAGLE, a state-of-the-art cosmological hydrodynamic simulation that has been tuned to a small number of distribution functions in the local universe; results at high redshifts represent predictions of the model. I have led a new simulation project during this period, First Light And Reionisation Epoch Simulations (FLARES), a suite of ‘zoom’ simulations using a modified version of the EAGLE physics code, of regions selected at high redshift.

Spectral Energy Distribution Modelling

Since hydrodynamic simulations do not resolve individual stars or HII regions a number of subgrid models and assumptions must be employed to accurately determine the galaxy SED, which can have a significant impact on the predicted emission. One example is the choice of stellar population synthesis (SPS) model, which links the initial mass, age and metallicity of a star particle in the simulation ($M_{*} \sim 10^{6} \, M_{\odot}$) to its intrinsic SED. In recent years a number of advanced SPS models have been developed, including the effects of binary interactions, post-AGB stars and nebular emission. We demonstrated in Wilkins et al. 2016 that the production efficiency of ionising radiation can vary by up to a factor of 4 due to the choice of SPS model, and it can also have a significant effect on predicted magnitudes in the rest-frame UV for high-$z$ objects.

Median SED of galaxies in the EAGLE simulation at $z = 8$, both intrinsic and dust attenuated, with JWST NIRCAM filters overlayed

The dust contribution at high redshift is also highly uncertain, but is key for predicting realistic observed spectra. Dust modelling can vary significantly in sophistication, from simple screen models linked to the mass and metallicity of star forming gas, to full radiative transfer solutions taking account of the spatial distribution of dust and the orientation of the observer. Nebular emission is another important component in the SED of high-$z$ galaxies. It is obviously necessary for predicting the presence and strength of individual emission lines, but such lines can also have a significant impact on broad band magnitudes.

I have recently been performing detailed modelling of galaxy SEDs in hydrodynamic simulations in order to carry out close comparisons with HST observations of the rest-frame UV luminosity, and make predictions for JWST.