Over the next decade, telescopes will map billions of galaxies across unprecedented stretches of sky, depth, and object count. This project builds the analytical toolkit needed to extract the physics hidden in that avalanche of data. The standard model of cosmology—Lambda-Cold Dark Matter—works remarkably well, but it leaves fundamental questions unanswered. What is dark energy? What drove the universe’s early expansion? The new surveys will test these ideas, but only if researchers can model subtle signals from scalar fields (the most common type of proposed new physics) and separate them from astrophysical noise. Current statistical methods, which rely on two-point correlations, throw away valuable non-Gaussian information. This project develops higher-order statistics, physically consistent priors, and fast emulators to squeeze every bit of information from the data. If successful, the pipeline will produce the most precise constraints yet on new physics from Stage III and Stage IV surveys, including Euclid. The techniques themselves—auto-differentiable likelihoods and high-dimensional parameter searches—will be reusable for other cosmological datasets. This is fundamental science: it will not change a single power grid or navigation system tomorrow. But the same kind of precision cosmology that emerged from earlier surveys now underpins GPS corrections and satellite timing. A deeper understanding of the universe’s composition may, in time, reshape technologies that rely on gravity and spacetime.
View original technical description
A new generation of “Stage IV” surveys will map regions of the cosmos which are unprecedented, in terms of their area, depth and the number of objects that they contain. The analysis of such data will help us understand new physics, beyond the standard paradigm: the Lambda-Cold Dark Matter model. While we are going to see a step change in the quality and quantity of cosmological data, there is still substantial work that needs to be done to reap the scientific rewards. We need more accurate models of the fundamental physics we want to probe – in our case, scalar fields (which make up the large majority of the proposed models of new physics) – and better models for the astrophysical and observational effects. We need clever methods to squeeze out the maximum statistical power of the data, beyond the standard two point methods which are currently used. Finally, given the high dimensionality of the models, we need accurate emulators for the summary statistics and differentiable pipelines that can accurately determine the posteriors for the parameters we are interested in. In this project, we will build on ideas that we have begun developing over the past few years, and integrate them into a pipeline. There are five work packages, which will unfold in parallel. The first work package will develop well defined and physically consistent priors for the fundamental physics. The second will develop a new approach for using power spectra of higher order maps so as to extract extra, non-Gaussian information, supplementing the usual two point statistics. The third involves developing a new set of emulators, for the relevant summary statistics. The fourth will bring these developments into constructing an auto-differentiable likelihood, optimized for searches in high dimensional parameter spaces. The final, fifth work package will involve applying the resulting pipeline (which incorporates all the aspects of the previous work packages) to the final compilation of Stage III data and to the available Stage IV Euclid data. I plan, from this project, to have two types of deliverables which will have in an impact in cosmology. The first will be a new set of techniques and tools that can be applied more widely to other data sets. The second set will be scientific constraints on new physics which will inform future model building in cosmology and particle physics.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know