Testing and emulating modified gravity on cosmological scales

  • Andrius Tamošiūnas

Student thesis: Doctoral Thesis


This thesis explores methods and techniques for testing and emulating models of modified gravity. In particular, the thesis can be split into two parts. The first part corresponding to chapters 2, 3 and 4 introduces a method of testing modified gravity models on galaxy cluster scales. In more detail, chapter 2 introduces the main concepts from galaxy cluster physics, which are important in the context of testing modified gravity. I present a summary of the different probes of mass in galaxy clusters. More specifically, the properties of the X-ray-emitting intracluster medium are discussed in the context of measuring hydrostatic masses. In addition, the key equations for weak lensing of background galaxies by galaxy clusters are summarized in terms of their importance for measuring cluster lensing masses.
Chapter 3 introduces a technique for detecting the modifications of gravity using a combination of X-ray and weak lensing data obtained by stacking multiple galaxy clusters. This technique first discussed in Terukina et al. (2014) and Wilcox et al. (2015) allows us to put some of the most competitive constraints on scalar-tensor theories with chameleon screening as well as the closely related f(R) gravity models. Chapter 3 also contains a discussion of the key theoretical concepts of scalar-tensor models and their relationship to f(R) gravity. Finally, the chapter is concluded by introducing novel results which update the tests done in Wilcox et al. (2015) with an improved dataset consisting of 77 galaxy clusters from the XCS and CFHTLenS surveys. The updated dataset containing less noisy tangential shear data allows us to put tight constraints on the chameleon field background value and the related fR0 parameter: φ∞ < 8 × 10−5Mpl and |fR0| < 6.5 × 10−5.
Chapter 4 expands the mentioned techniques for testing a different type of a model. In particular, the model of emergent gravity (introduced in Verlinde (2017)) is tested by using a variation of the techniques introduced in chapter 3. The key prediction of Verlinde’s emergent gravity is a scaling relation similar to the baryonic Tully-Fisher relation, which allows us to determine the dark matter distribution in a cluster directly from the baryonic mass distribution. The mentioned scaling relation was tested by determining the baryonic mass from the X-ray surface brightness data and calculating the predicted weak lensing tangential shear profile, which was then compared against the ΛCDM prediction based on the Navarro-Frenk-White profile. The test was performed for the Coma Cluster using data from Terukina et al. (2014) and for the 58 galaxy cluster stack from Wilcox et al. (2015). The obtained results indicate that according to the Coma Cluster data, the emergent gravity predictions agree with the ΛCDM predictions only in the range of r ≈ 250-700 kpc. Outside the mentioned radial range the standard model results are preferred according to the Bayesian information criterion analysis (despite needing two extra free parameters). The same general conclusion can be drawn from the 58 cluster stack data, which indicates a good agreement between the models only for r ≈ 1-2 Mpc. Outside of that radial range the standard model is strongly preferred.
The second part of the thesis, referring specifically to chapters 5 and 6, contains a study of machine learning techniques for emulating cosmological simulations. More specifically, generative adversarial networks are studied as an effective tool for emulating N-body simulation data quickly and efficiently. Chapter 5 contains a brief discussion of the different machine learning algorithms in the context of emulators. Specifically, artificial neural networks are introduced along with gradient boosting algorithms. Chapter 6 introduces a generative adversarial network algorithm for emulating cosmic web data along with weak lensing convergence map data coming from N-body simulations. The presented approach is based on the cosmoGAN algorithm first described in Mustafa et al. (2019), which allows us to generate thousands of realistic weak lensing convergence maps in a matter of seconds. The mentioned approach is then modified to allow emulating cosmic web and weak lensing data for ΛCDM and f(R) gravity with different cosmological parameters and redshifts. In addition, a similar approach was used to simultaneously emulate dark matter and baryonic simulation data coming from the Illustris simulation. The obtained results indicate a 1-20% difference between the power spectra of the emulated and original (N-body simulation) datasets depending on the training data used. Finally, the chapter contains an indepth study of the technique of latent space interpolation and how it can be applied to control the cosmological/modified gravity parameters during the emulation procedure. The obtained results illustrate that such machine learning algorithms will play an important role in producing accurate mock data in the era of future large scale observational survey
Date of AwardNov 2020
Original languageEnglish
Awarding Institution
  • University of Portsmouth
SupervisorRobert Nichol (Supervisor), David Bacon (Supervisor) & Kazuya Koyama (Supervisor)

Cite this