Research

Always under construction!

CV (Last updated mid 2025).

Galaxies remember inflation:

Inflation is our model for the Universe in its earliest moments. There is a zoo of possible inflationary models, which we can only sort through based on their downstream impact on cosmological measurements. For the forseeable future, the best way to characterize an especially interesting subset of these models (multi-field models) is with galaxy surveys. This is because galaxy surveys are especially well-suited for measuring local primordial non-Gaussianity (LPNG), which is an observable signature in the statistics of galaxies that is closely associated with multi-field inflation.

Some of my recent work has focused on using the influence of early-Universe physics on galaxy formation through the phenomenon of galaxy bias, which connects the statistics of primordal flucturations to the statistics of galaxies:

  • Galaxy and halo assembly can tell us a lot about LPNG: The number of galaxies that form in the Universe is modulated by the strength of LPNG. LPNG also influences what happens inside dark matter halos and galaxies, influencing their density profiles, assembly history, and observed photometric colors. We showed that when using multiple populations of galaxies that are defined using these properties, it is possible to obtain much tighter constraints on the amplitude of LPNG (see here).
  • Connecting LPNG to tracer time evolution: Living in a Universe with (positive) LPNG is like living in a Universe where time has been fast forwarded. This insight, which we made precise, allows for one to estimate the response of a particular galaxy population to LPNG. Since lack of knowledge of this response is the main uncertainty impacting constraints on the amplitude parameter for LPNG, it is now possible to accurately inform the size of this response in a data-driven way (see here for more).
  • Working directy at the field level: LPNG affects summary statistics of the galaxy distribution, such as the power spectrum and bispectrum. But the presence of LPNG is most directly understood as a Taylor expansion in the in the gravitational potential. As a result, its influence is not limited to summary statistics. We performed a first study of the LPNG response of simulated dark matter halos beyond summary statistics by using all the information in the field (see here for more).

High-redshift - the future of galaxy surveys

  • Star-forming galaxies and their clustering: The future of cosmological surveys is at high redshifts ($z\gtrsim 2$), with several planned surveys targeting star-forming galaxies for cosmological analysis. It is possible to robustly model the clustering of these galaxies for cosmological inference using the Effective Field Theory of LSS (EFT). We found that current versions of such models will likely have a similar or increased regime of validity for high-redshift star-forming galaxies, which will enable improved cosmological constraints (see here for more).

The Lyman-\(\alpha\) Forest in Effective Field Theory

  • The Lyman-\(\alpha\) forest at the field level: Producing Lyman-\(\alpha\) forest flux transmission fields is extremely challening. An accurate description of the small- and large-scale gas distribution demands excessively high-resolution hydrodynamical simulations. Recently, we developed a field-level model for the Lyman-\(\alpha\) forest flux transmission field based in the Effective Field Theory of large-scale structure, which permits quick simulation of large-scale mock Lyman-\(\alpha\) forest using information from small-scale, high-resolution hydrodynamical simulations (see here).

Understanding Large-scale Structure

  • The information content of tracers: The night of large-scale structure is dark, but it offers lampposts in the form of luminous tracers, like galaxies and hot gas. It is not very easy to describe how these tracers are located in space, especially when they are almost close enough to overlap. I spent some time working on this directly (see here), but am thinking about it all the time!
  • Massive neutrinos: More than ten million neutrinos from the early universe are passing through your body right now. Neutrinos are sub-atomic particles predicted by the Standard Model, but what was not predicted is that they have mass. It turns out they do have mass, and cosmological measurements can place the tightest upper bound on that mass - that is - if we understand how neutrinos affect large-scale structure. I worked toward this goal by improving the treatment of massive neutrinos in cosmological simulations (see here).

Statistical and Numerical Tools for Cosmological Inference

  • Differentiable forward models, optimization, and sampling: Suppose you have to climb a hill as fast as possible. Going straight up seems a pretty good option, but what direction is “up”? Gradient-based optimization (which I worked on here) and sampling methods use this directional information to move around spaces of very large dimension. When this information is avaliable, it makes the process of finding the best parameters for a given physical model and their uncertainties much easier. To do this, however, the model has to be differentiable - and I’ve worked on developing this capability for one model or another the entire time I was at Berkeley.