Strongly lensed supernovae are extremely rare and powerful probes that can give insights into high-redshift supernova physics, substructures in massive galaxies, and the expansion rate of the Universe. The lensed supernova field is at a turning point, as we will go from a handful of current discoveries to orders of magnitudes more with the advance of the Legacy Survey of Space and Time (LSST) at the Vera Rubin Observatory.

In Arendse et al (2023), we investigated the cosmological prospects of lensed type Ia supernovae in LSST by quantifying how many we expect to find each year, the impact of stellar microlensing, the feasibility of conducting follow-up observations, and how to best separate lensed and unlensed type Ia supernovae. The simulation code used in the paper, Lensed Supernova Simulator Tool (LensedSST) is publicly available at github.

I am a member of the Zwicky Transient Facility (ZTF), a survey telescope which scans the sky each night looking for newly appearing objects. On August 2022, we discovered a strongly lensed type Ia supernova: SN Zwicky. In Goobar, Johansson, Schulze, Arendse et al. (2023), we documented the discovery of SN Zwicky, a highly magnified supernova with very short time delays (less than a day) which is part of a new, undiscovered population of low mass lens systems. In Pierel, Arendse et al. (2023), we analysed observations of SN Zwicky with the Hubble Space Telescope and explored a set of different lens models. I made a promo video of the discovery story of SN Zwicky using water colours.

As upcoming survey telescopes will collect unprecedented volumes of data, machine learning techniques are emerging as promising tools to extract information from these large data sets. I’m interested in using machine learning to find rare objects, such as lensed supernovae, and to infer cosmological parameters from them.

In Kodi Ramanah, Arendse & Wojtak (2022), we used a spatio-temporal neural network to distinguish between lensed and unlensed type Ia supernovae. Our classifier is a convolutional long short-term memory network (ConvLSTM) which is able to pick up spatial and temporal correlations in time-series of lensed supernova images. By adopting variational inference, we quantified approximate Bayesian uncertainties on the network predictions.

A machine learning regression problem I worked on is constraining cosmological parameters from galaxy clusters, the most massive gravitationally bound systems in the Universe. Kodi Ramanah, Wojtak & Arendse (2020) presents a 3D convolutional neural network that infers dynamical mass estimates of galaxy clusters, from their observed line-of-sight velocities and positions in the sky. We derived reliable uncertainties on the cluster mass estimates using a simulation-based inference framework. The method was applied to 800 galaxy clusters from the SDSS Legacy Survey, from which we constructed a cluster mass function that is consistent with the standard cosmological model.

The exact value of the present-day expansion rate of the Universe, the Hubble constant, is a matter of ongoing debate. Local measurements using Cepheids and type Ia supernovae predict a higher Hubble constant than observations from the Cosmic Microwave Background (CMB) radiation.

This ‘Hubble tension’ might suggest hidden systematics in either of the measurements, or it can point to new physics beyond our standard cosmological model. Many new cosmological models have been proposed to resolve the tension by increasing the value of the Hubble constant from the CMB. Intimately linked to the Hubble constant is the sound horizon, a distance scale in the early Universe which is also imprinted in the clustering of galaxies at later times, the Baryon Acoustic Oscillations (BAO).

In Arendse, Agnello & Wojtak (2019), we used BAO observations and type Ia supernovae calibrated by gravitationally lensed quasars to constrain the Hubble constant, the sound horizon and the curvature of the Universe. Since the BAO signal constrains the product of the Hubble constant and the sound horizon, models that solve the tension by increasing the Hubble constant should simultaneously decrease the sound horizon. Arendse et al. (2020) investigated four proposed solutions to the Hubble tension and found that modifications to our standard model in the late Universe are disfavoured by BAO data.