recent projects

Monte-Carlo Markov-chain methods are widely (and wildly) used in cosmological inference and can always be mapped back onto a canonical, statistical physics system. Interesting questions we pursue concern canonical parition functions reflecting cosmological, non-Gaussian likelihoods and analytical methods for inference for these cases.

Information entropies (like Kullback-Leibler or Rényi entropies) are measures of statistical randomness of distributions, if applied to the posterior of a distribution they serve as a quantification of remaining statistical uncertainty, i.e. how well measurements have been able to improve the knowledge on a given physical model. We work on the connection between more conventional tools in statistics and inference such as likelihoods and statistical tests, with novel concepts like information entropies, with an application to cosmological data sets.

Machine learning methods can help infering fundamental laws of Nature from complex data or to design inference processes that are otherwise difficult to manage. We are trying to apply inference on inflationary potentials with machine learning methods and hopefully establish links to information geometry.

Cosmic inflation is an early phase of accelerated expansion that solves the flatness-problem in FLRW-cosmologies and is a mechanism for introducing fluctuations in the distribution of matter. Of particular interest to us are inflationary non-Gaussianities and their measurement in future large-scale structure surveys. The best way of measuring higher-order non-Gaussianities is still unclear; while there is a clear way of computing polyspectra from covariant perturbation theory, their estimation from data quickly becomes a combinatorial problem, for which we use advanced sampling methods.

The fluctuation pattern of the temperature and polarisation of the cosmic microwave background gets distorted by weak lensing deflection and changes therefore its correlation properties. Our group is interested in higher-order effects in lensing and cross-correlations between the lensing effect and other probes of the cosmic large-scale structure. From a methodical point of view closely related is the question of lensing of the 21cm background, which however involves more aspects of non-Gaussianity and reionisation history, including its non-uniformity.

Modern cosmology is a statistical science and we are interested in questions related to the information content of large-scale structure surveys, in particular in the nonlinear regime, selection of models and the effect of systematical errors on the parameter estimation and model selection process. In particular, we investigate what properties about gravity are in principle knowable from cosmological surveys, how non-Gaussian structures can be described in an efficient way and how information about fundamental physics can be extracted from non-Gaussian structures.

Weak lensing operates under the assumption of intrinsically uncorrelated galaxy shapes, which might not be true because galaxies experience correlated tidal gravitational fields and share a similar angular momentum generation. We have worked on tidal interaction models for galaxies to derive ellipticity correlations for investigating their contaminating effect in weak lensing parameter inference. Starting from models of tidal shearing and tidal torquing for the orientation of elliptical and spiral galaxies, we are constructing more elaborate models for predicting and investigating intrinsic alignments, and hope to apply them to Euclid data.