|
Promotionskolloquium |
Davit Alkhanishvili
| SCHEDULED |
AIfA, Bonn
The Lambda-CDM model is the standard in cosmology, with large-scale
structure clustering measurements playing a key role in parameter
constraints. Next-generation galaxy surveys highlight the growing
importance of higher-order statistics like the bispectrum, which offers
improved constraints but introduces computational challenges. This
thesis focuses on modeling the bispectrum using perturbation theory to
enhance its role in extracting cosmological information. First, we test
next-to-leading order perturbation theory expansions using N-body
simulations, finding that effective field theory (EFT) provides the most
accurate small-scale predictions. We also assess the impact of
systematic and statistical errors. Next, we demonstrate how deep neural
networks can be employed to model survey geometry effects on the power
spectrum and bispectrum, achieving high accuracy with efficient
computation. Lastly, we evaluate a third-order galaxy bias expansion
against synthetic Eucilid-like survey catalogues, demonstrating that
combining the power spectrum and bispectrum allows accurate cosmological
parameter extraction up to mildly non-linear scales, improving
constraints by a factor of 2-5 over the power spectrum alone.