In my research, I focus on the following fundamental open questions in cosmology: What is the expansion rate and geometry of the Universe? Can the Universe be accurately described by the standard CDM model, or do we need new physics? What is the nature of dark matter and dark energy, and how does dark energy evolve over time? The answers to these questions will have great implications for our understanding of the inner-workings, composition and evolution of the Universe.
To address the above questions, I am using cosmic distance indicators and galaxy clusters to constrain cosmological parameters, construct the expansion history of the Universe and test potential new models that aim to describe it. More details about my work can be found below.
systematics or new physics behind a short sound horizon?
The Hubble constant () and the sound horizon () are two fundamental cosmological parameters, that respectively describe the cosmic expansion rate and set the distance scale in the early Universe. Recently, a tension has emerged between measurements of and obtained by local measurements (from type Ia SNe, Cepheids and gravitationally lensed quasars) and those inferred from observations of the cosmic microwave background (CMB) radiation. This tension might suggest hidden systematics in either of the measurements, or it can point to new physics beyond the standard CDM model.
In a project carried out with Radek Wojtak and Adriano Agnello, we provide new measurements of and in a cosmology-independent way (Arendse et al., 2019). This was done using data of type Ia supernovae, baryon acoustic oscillations, Cepheid variable stars and a sample of six gravitationally lensed quasars. Cepheids and lensed quasars provide absolute distance scales, while the role of type Ia supernovae is to link cosmological signals coming from sources spread over a range of distances. The values that result from this analysis based on the local Universe are in tension with CMB-based observations.
Several modifications of CDM have been proposed in the literature to resolve the tension by altering the model-dependent contours of the CMB and shifting them closer to the local values of and . We investigated the effects of four commonly used extensions and found none of them to be very promising (Arendse et al., 2020). Models that change the physics after recombination, i.e. late-time modifications, fail to provide a solution because they are incapable of changing the value of the sound horizon, such that the tension in persists. Early time modifications, which change the physics before recombination, slightly decrease the tension, but do not manage to dissipate it completely.
Gravitationally lensed quasars
to constrain cosmological parameters
When the configuration is just right, light rays from a supernova behind a massive foreground galaxy or cluster can be lensed to produce multiple images of the same source. On their trajectory to the lens galaxy, the light rays are affected by a geometric and gravitational time delay, causing the images to appear delayed by hours up to years. The time delays are primarily sensitive to the Hubble constant, but also weakly to several other cosmological parameters. Although only three such systems have been observed to date, upcoming surveys such as the Legacy Survey of Space and Time (LSST) and WFIRST are poised to discover orders of magnitude more. This will allow us to peer further into the Universe as supernovae which were otherwise too faint will become magnified, and this will help us to put stringent constraints on cosmological parameters.
However, the quest for using gravitationally lensed supernovae (glSNe) for cosmology is not without challenges. Firstly, distinguishing between lensed supernovae and unlensed ones in the large quantities of data is a complicated task, especially considering the lack of glSNe observations that could serve as training data for a machine learning model. Secondly, uncertainties in the lens mass profile and lensing contributions from microlensing by substructures in the lens galaxy and line-of sight-structures can lead to additional systematic errors in the cosmological parameters.
Together with my collaborators Radek Wojtak and Doogesh Kodi Ramanah, I am working on a series of exciting new undertakings centred around the use of glSNe in future surveys such as LSST. We are currently developing a deep learning framework to infer cosmological constraints from simulated glSNe images, as well as one to distinguish between images of lensed and unlensed supernovae in astronomical surveys.
Dynamical cluster mass inference
using convolutional neural networks
Galaxy clusters, the most massive gravitationally bound systems in the Universe, provide another powerful way to constrain cosmological parameters and to study the formation and evolution of cosmic large-scale structures. Traditional methods of cluster mass estimation often rely on scaling relations and are limited by several assumptions, primarily involving dynamical equilibrium and spherical symmetry. With the upcoming unprecedented volumes of data from surveys such as DESI, LSST and eROSITA, and the increasing scale of state-of-the-art cosmological simulations that provide large amounts of training data, machine learning approaches offer a promising alternative to standard mass estimation techniques.
In work carried out with Doogesh Kodi Ramanah and Radek Wojtak, we used a 3D convolutional neural network to infer dynamical mass estimates of galaxy clusters, employing information about their observed line-of-sight velocities and positions in the sky (Ramanah, Wojtak & Arendse, 2020). We used a simulation-based inference approach to obtain robust uncertainties for the mass estimates. We then applied the method to a realistic mock catalogue generated specifically to emulate the Sloan Digital Sky Survey (SDSS) Legacy spectroscopic observations and showed that the resulting mass estimates were fully consistent with recent mass measurements from Abdullah et al (2020).
This work is the first to exploit the full 3D phase-space distribution of galaxy clusters, to use a simulation-based inference approach to infer uncertainties, and to be applied to real cluster data from SDSS. Future work in this direction can take the exciting avenue of using the cluster mass function to infer constraints on the matter density and clustering amplitude that are complementary to traditional approaches.