Software Projects

  • galtab is a key Python package that enabled much of the work in my PhD thesis by pretabulating galaxy placeholders to improve prediction efficiency of my Counts-in-Cylinders estimator.

  • JaxTabCorr is a Python package that integrates classes from TabCorr and halotools into a differentiable prediction framework made possible by JAX autodiff libraries.

  • mocksurvey is a Python package used for constructing mock galaxy catalogs and perform mock surveys seeded from the UniverseMachine empirical model.

  • I am a contributor to halotools, which is a Python package that provides a wide array of models of the galaxy-halo connection.

Data Products

Mock Galaxy Catalogs

You can download my mock catalogs for PFS here

Science Interests

Publications: See my papers on ADS

Postdoc at Argonne

I work closely with Andrew Hearin and others to develop models of the galaxy-halo connection, implemented with Python's JAX library to enable GPU acceleration and automatic differentiation. I am focusing on improving the scalability of our model to extremely large datasets by designing a framework that performs distributed parallel computation, while seamlessly preserving the advantages of JAX. My goal is to execute this framework to perform self-consistent mock observations on cosmological simulations, thereby minimizing biases in the joint inference of cosmology and galaxy formation physics.

PhD Thesis

Illuminating and Tabulating the Galaxy-Halo Connection

Part I: Illuminated the UniverseMachine to construct PFS mock catalogs

Using UniverseMachine as a model and UltraVISTA photometry as training data, I created a mock galaxy catalog specifically tailored to making predictions for the upcoming PFS survey. Using this mock, I published a paper which demonstrated that future extensions of the PFS survey should prioritize increasing the survey area to best improve scientific goals. This mock and the methods used to create it are publicly available.

Part II: Tabulated statistical estimators to be fast, precise, and differentiable

The galaxy-halo connection is typically analyzed via Markov-chain Monte Carlo (MCMC) sampling of parameter-space in order to place constraints on models. However, this process is slowed down by the stochastic nature of halo occupation distribution (HOD) models. I have improved the efficiency of this process with two open-source projects:

  • JaxTabCorr, in which I have rewritten parts of the TabCorr and halotools packages to replace certain NumPy operations with equivalent JAX operations. It can be used to calculate differentiable predictions of two-point correlation functions, which will improve the scalability of model inference as we need to push to larger and larger parameter spaces.
  • galtab, in which I have implemented a tabulation-accelerated statistic called Counts-in-Cylinders (CiC) that captures higher-order clustering information beyond that of the two-point correlation function. This code is also differentiable and automatically GPU-accelerated. I have published a paper presenting this code, as well as the new HOD constraints that it has made possible, utilizing the Early Data Release from the Dark Energy Spectroscopic Instrument (DESI).