Kaustav Mitra

Ongoing Research:

My current research focus at Argonne National Lab involves differentiable modelling of galaxy formation and evolution, with the goal to jointly forward-model complex multi-wavelength multi-survey data from different telescopes, to better constrain galaxy formation physics and cosmological parameters. A crucial service work generated by this research is the creation of extremely realistic mock catalogues that are invaluable for the calibration and validation pipelines in different surveys, such as DESI, Rubin, Roman, etc. This invited talk at the Mock-NYC 2026 conference, offers a snapshot of where I was in my postdoctoral research as of mid-January, 2026.

Learn more about my ongoing research work HERE.

Past Research:

This recorded video of an invited talk [Feb 04, 2026] at Clemson University's Department of Physics and Astronomy online seminar gives a very brief overview of my PhD thesis work at Yale.

The following are a few of my key highlights from my past research work: either demonstrating some particular methodology or showcasing a specific result.

Please visit this SciX page to see my list of publications (only papers, i.e., excluding conference proceedings).

Selected publications

• BASILISK IV. No S8 Tension with Satellite Kinematics.
  Kaustav Mitra, Frank C. van den Bosch, Josephine Baggen, Johannes U. Lange.
  (in review).
• The Impact of Baryonic Effects on the Dynamical Masses Inferred Using Satellite Kinematics. Josephine Baggen, Frank C. van den Bosch, Kaustav Mitra.
  OJAp, February 2026.
• Constraining the Low-mass End of the Stellar-to-halo Mass Relation with Surveys of Satellite Galaxies. J. Sebastian Monzon, Frank C. van den Bosch, Kaustav Mitra.
  ApJ, December 2024.
• BASILISK II. Improved constraints on the galaxy-halo connection from satellite kinematics in SDSS. Kaustav Mitra, Frank C. van den Bosch, Johannes U. Lange.
  MNRAS, September 2024.
• rms-flux relation and disc-jet connection in blazars in the context of the internal shocks model. Aritra Kundu, Ritaban Chatterjee, Kaustav Mitra, Sripan Mondal.
  MNRAS, March 2022.

Motivation:

The traditional techniques of modelling the galaxy-halo connection, such as the HOD / CLF modelling or abundance matching, were mostly developed in the first decade of this millenium, primarily with SDSS Main Galaxy Sample in mind: a low redshift sample where galaxies do not evolve significantly and a simple r-band cut results in a fairly complete sample. But the nature of galaxy survey data is changing drastically with DESI, Rubin, ROMAN, DESI-2 and other ongoing and upcoming surveys. The traditional approach to galaxy-halo connection will soon prove to be insufficient to capture the unprecedented dynamic range and the complexity in the joint distribution of multi-wavelength multi-survey data across vast redshift ranges. Galaxies evolve drastically between redshifts 10 and 0.1, and moreover, key features in galaxy SEDs move in and out of survey filters; as a result of these two together, a simple magnitude or color cut will produce very different galaxy samples at different redshifts.

Modelling this complex data is not only crucial to improve our understanding of galaxy formation and evolution, but is also an essential step in using these galaxy surveys for (a) unbiased cosmology inference and (b) probing beyond-standard-model physics.

One can go beyond simplicity of HOD models or abundance matching by using a physics-driven galaxy formation model. Simulations and semi-analytic models (SAM) of galaxy formation produce realistic and complex populations of galaxies. However, they are not efficient enough to be used in an inference pipeline for modelling raw data with survey systematics, and to perform a full-fledged MCMC to marginalize over all the model uncertainties in the subgrid physics.

My ongoing research:

To address this dire necessity, my ongoing work focuses on empirical modelling of galaxy formation and evolution, with the goal to jointly forward-model complex multi-wavelength multi-survey data from different telescopes, to better constrain galaxy formation physics and cosmological parameters.

For example, the following plot shows the apparent magnitude distributions (top row), color-magnitude diagrams (middle row), and conditional color PDF for bins of i-band apparent magnitude cut (bottom row), of COSMOS-2020 data across different redshift bins (different columns). Overlayed with the data are the fits from diffsky, an empirical galaxy formation model that uses modern AI/ML library called JAX to enable end-to-end differentiable forward modelling of population-level galaxy SEDs for any type of galaxy survey.