Repository logo

Bayesian Modelling of Galaxy Clusters with the Arcminute Microkelvin Imager



Change log


Elwood, Patrick 


The Arcminute Microkelvin Imager (AMI) is a dual-array interferometer comprising the Small Array (SA) and Large Array (LA), operating between 13 GHz–18 GHz, situated at Lord’s Bridge, near Cambridge. It was designed to observe galaxy clusters using the Sunyaev Zel’dovich (SZ) effect. The SA is used to observe secondary SZ anisotropies in the Cosmic Microwave Background (CMB), whilst the longer baselines of the LA make it suitable for the detection of contaminating point-like sources in the cluster field. In 2015, AMI’s analogue correlator was upgraded to a digital correlator, which was intended to improve the dynamic range and to reduce artefacts in the maps, particularly in the presence of bright sources. In this thesis I present cluster observation and analysis with the new correlator, to investigate the improvements in AMI’s cluster observing capabilities.

I describe the AMI observation and data reduction procedures, and test the performance of the new LA correlator for source detection on LA rasters. I find the performance of the new correlator to be comparable to that of the old correlator, however precise comparison isn’t possible due to source variability between observation epochs. I also find that new correlator flux estimates are systematically lower, which I attribute to a change in the primary calibration procedure that was implemented between the old and new observation epochs.

Galaxy clusters observed with the SA are analysed in a fully Bayesian manner with the in-house software package McAdam, using a parametrised model for the cluster whilst simultaneously modelling the point source environment using priors based on LA estimates. The detection of clusters is reported based on their Bayesian evidence.

I introduce the AMI observational model used to describe AMI data and present the pipeline that I developed to analyse cluster samples. A sample of Planck SZ cluster detections was originally observed by AMI with the old correlator and results were presented in Perrott et al. (2015), with 99 of the clusters detected. This sample had been subject to a number of selection cuts based on the radio source environments. A sample comprising 54 of these clusters has been observed with the new correlator and I analyse this sample with the observational model and present the results. I find that the strength of classifications is systematically improved with the new correlator, with the majority of the previously undetected clusters now detected.

A previously unobserved sample of 56 Planck clusters is then selected based on AMI observing limits, but is not subject to any selection cuts based on the source environments around the clusters. I observe this sample using the new correlator and find that the majority of these clusters are detected. In the sub-sample of clusters that break the point source selection criteria applied previously, some detections are judged to be spurious, the result of bright central sources in the cluster field, however the majority of detections are judged to be legitimate.

I introduce the AMI physical cluster model and use it to estimate physical cluster parameters, such as mass and temperature, for the Planck clusters that were detected with the observational model. I find that detection classifications from the physical model are in good agreement with those from the observational model but that the observational model typically leads to stronger classifications. I compare the AMI and Planck mass estimates, finding that they are in good agreement for very strong detections, whereas for weaker detections the AMI mass estimates are systematically lower. I investigate this discrepancy and attribute it, at least in part, to the log-uniform mass prior biasing posteriors downwards in cases where the posterior is not tightly constrained.

Finally, I use the results of the physical cluster modelling to calculate estimates of emission weighted temperature, which can be compared to X-ray temperatures as a method for determining which clusters are recent mergers. X-ray temperatures of a sample of Planck clusters were presented in Lovisari et al. (2020). I use the 5 clusters from this X-ray sample that have already been observed with the new AMI correlator as a test of my methodology. There are an additional 23 clusters in the X-ray sample that lie within the AMI observing limits, however observation of additional clusters has not been possible due to COVID-19, and therefore observation and analysis of these clusters is suggested as an area for future work.





Saunders, Richard
Green, Dave
Perrot, Yvette


Galaxy Clusters, Bayesian, Arcminute Microkelvin Imager, Cosmic Microwave Background, CMB, Sunyaev Zel'dovich


Doctor of Philosophy (PhD)

Awarding Institution

University of Cambridge
STFC (1803590)
Magdalene College; Cambridge Philosophical Society