Research

The Big Picture

In the broadest sense, my research seeks to explain why the Universe has order instead of being a chaotic mess. One of the best examples of this is galaxies: they started out as random quantum fluctuations in the early Universe and then became intricately organized collections of stars, gas, dust, black holes, planets, etc. Galaxies come in all shapes, sizes and colors – I use a combination of observations, theory and data science to understand how and why. It is important to decode the astrophysics of galaxies because they represent one of our most promising probes of fundamental physics via cosmology – their numbers, motions and spatial clustering have historically provided a variety of evidence for dark matter, dark energy and inflation. This has motivated ambitious upcoming telescopes such as Roman, Rubin, Euclid and Simons which will map tens of billions of luminous galaxies and their cosmic ecosystems. Without transforming galaxy formation into a precision science, as I am trying to do, it is going to be impossible to confidently interpret this data to constrain the origin, evolution and fate of our Universe.

Here is an overview of my research program and how it can help answer one of the biggest questions of them all: how did we get here?

Understanding Galaxies as Dynamical Systems

During my NSF-funded PhD at UC Santa Cruz and my NASA Hubble Postdoctoral Fellowship at Columbia University, I re-visited and modernized the foundations of how we evolve and understand galaxies as dynamical systems. My PhD thesis was the only one exclusively focused on the Simulating Multi-scale Astrophysics to Understand Galaxies (SMAUG) Project to ever come out of that collaboration. As a core member of SMAUG, I was fortunate to analyze simulations from the Feedback In Realistic Environments (FIRE) Project as an initial testbed for my research, later generalizing to other simulations and datasets. As part of the Simons Collaboration on Learning the Universe (LtU), I independently conceived and led the development of sapphire, which bridges galactic dynamics, numerics, statistics, astrophysics and cosmology in “next-generation” ways and is built on top of the JAX Python Library.

My research in this area currently falls under three themes:

Hybrid physics-informed, data-driven galaxy population simulators

One of the grand challenges of modern astrophysics is to develop a fully predictive theory of galaxy formation so that galaxies can become more robust cosmological probes. However, this requires overcoming five obstacles: (1) the relevant physical processes span a large dynamic range, (2) they are non-linearly coupled across scales leading to complicated emergent behavior, (3) many of those processes are not understood from first principles, (4) we cannot directly observe the evolution of individual galaxies on human timescales so must resort to statistical, population-level studies, and (5) we have noisy, incomplete data. I believe that currently popular modeling techniques cannot simultaneously address these challenges and we need a fresh approach for galaxy-scale simulations. Figure 1 of Pandya+26 provides a schematic overview of my new interdisciplinary dynamical systems vision for achieving a hybrid physics-informed, data-driven understanding of galaxy evolution using sapphire.

Energy flows in galactic atmospheres

Like planets and stars, galaxies have atmospheres that regulate energy balance. During my PhD and Hubble Fellowship, I prototyped modeling energy flows between galaxies and their gaseous atmospheres because the thermodynamic state of the latter may hold new clues about both astrophysics and cosmology. Figure 1 of Pandya+23 illustrates a new physical model in which eight non-linearly coupled ordinary differential equations describe how mass, metals and energy flow between galaxies and their gaseous atmospheres. This physical model was developed in collaboration with Dr. Drummond Fielding, Dr. Greg Bryan, Dr. Rachel Somerville and Dr. Chris Carr. We were guided by my earlier analysis of such flows in the “Santa Cruz” galaxy formation model originally created by my PhD co-advisor Dr. Rachel Somerville as well as in the FIRE simulations, following earlier work by Dr. Drummond Fielding and Dr. Daniel Anglés-Alcázar (Pandya+20, Pandya+21). With Prof. Mark Voit, we developed a generalized variant of this model called ExpCGM. My students Austen Gabrielpillai and Bill Robinson are adding satellite galaxies to this picture and my collaborators Dr. Bryan Terrazas and Dr. Sophie Koudmani are exploring black holes. With my student Yossi Oren and collaborator Dr. Osase Omoruyi, we validated and extended the approach from my PhD thesis to another set of simulations: IllustrisTNG. Following my earlier work with Dr. Yakov Faerman, Dr. Rachel Somerville and Dr. Amiel Sternberg, we will soon be in a position to forward model direct observables of galactic atmospheres. This model also sets the stage for understanding the spatially-resolved morphological evolution of galaxies following earlier work by my collaborator Dr. John Forbes and PhD co-advisor Dr. Kevin Bundy.

Differentiable, GPU-accelerated, interpretable galaxy evolution

I finished my PhD in 2021 before the hype of large language models. Ever since, I have been fascinated by automatic differentiation and GPU/TPU parallelization which, together with algorithmic advances, underpin the recent AI/ML revolution. Thanks to JAX and Julia, I have been exploring how to leverage this technology to accelerate and improve the interpretability and causal identifiability of our models. Figure 4 of Pandya+26 shows a Jacobian matrix with interpretable, non-random structures that encode the astrophysical sensitivity and locally linearized dynamics of a model Milky Way-like galaxy. These gradients provide a fundamentally new way of understanding the behavior, successes and limitations of our models. They also unlock previously inaccessible techniques to perform very fast Bayesian inference for galaxy formation, as is standard in precision cosmology. With my collaborators Dr. Lucas Makinen, Dr. Matthew Ho, Dr. Kartheik Iyer, Dr. Chris Lovell and others in LtU, we are exploring implicit likelihood inference and synthetic observations.

Galaxy Morphology, Kinematics and Scaling Relations

We don’t know how, but we think that galaxies evolve according to some set of physical laws because they can be grouped into a handful of morphological classes and exhibit remarkably tight correlations both globally, locally and across time. Since the evolution of individual galaxies is not observable on human timescales, these population-level trends provide essential information about the dynamical phase space behavior of galaxies. I have a wide variety of research interests in this area spanning observations, theory and data science.

3D shapes of protogalaxies

As a core member of the Cosmic Evolution Early Release Science (CEERS) Survey, with Dr. Haowen Zhang and others, I used the James Webb Space Telescope to show that most galaxies in the early Universe are preferentially prolate, not disks as commonly assumed. This has dramatic implications for astrophysics and possibly offers a new tracer or nuisance for cosmology. Figure 5 of Pandya+24 illustrates different 3D shapes and their joint distribution of projected sizes and axis ratios. With Marina Dunn and Dr. Marc Huertas-Company we are extending this to Euclid. With my student Arnav Shah, Prof. Steve Finkelstein, and Prof. Raymond Simons we explored gas kinematics as a tracer for 3D shape.

Intrinsic alignments of prolate protogalaxies

I originally got interested in prolate galaxies because early during my PhD with Profs. Joel Primack, Avishai Dekel, Sandy Faber, and David Koo, we proposed that these galaxies may align with each other on large scales, tracing out dark matter filaments of the cosmic web. Figure 1 of Pandya+19 illustrates such alignments, which could be a new cosmological probe of the collapse of dark matter filaments in the early Universe. With my collaborators Dr. Farhanul Hasan and Dr. Haowen Zhang and others, we showed that the Roman Space Telescope can help us study these alignments.

Weak gravitational lensing

My interests in intrinsic alignments and 3D shapes of early Universe galaxies led to a collaboration with Prof. Avi Loeb on the implications for weak gravitational lensing, which is a major effort in precision cosmology. Using JWST-CEERS as one example “blank field” we found evidence for correlated ellipticities and orientations suggesting the presence of dark overdensities in the foreground, though we could not rule out intrinsic alignments. Figure 12 of Pandya+25 shows polarization-like patterns in the large-scale orientations of galaxies in some sub-regions of the survey.

Star formation and morphological evolution

Like stars, galaxies follow tight “scaling relations” and one of the most famous is the star-forming main sequence: the rate at which they form new stars vs. their mass in existing stars. This is one of numerous observational examples suggesting that galaxies self-regulate their star formation and morphological evolution through feedback loops that dynamical systems approaches can help us understand. Figure 5 from Pandya+17a illustrates various trajectories that galaxies can take to oscillate on or depart from this main sequence, using models from Dr. Rachel Somerville and Dr. Ena Choi.

Interstellar gas in local giant ellipticals

As part of the MASSIVE Survey with Prof. Jenny Greene I performed emission line spectroscopy to find substantial reservoirs of warm ionized interstellar gas in local giant ellipticals. Figure 2 of Pandya+17b shows one such example which is surprising because the galaxy is otherwise “red and dead” and not forming many, if any, new stars.

Ancient stars in ultra-diffuse galaxies

I performed the first spectral energy distribution (SED) fitting to understand the stellar populations and formation histories of ultra-diffuse galaxies, combining ground-based optical imaging and space-based Spitzer IR imaging. This work was done with Dr. Aaron Romanowsky, Dr. Jean Brodie and the SAGES team at UC Santa Cruz. Figure 3 from Pandya+18 shows some example posterior predictive checks using a very early version of the now popular prospector Bayesian code.

Intermediate-mass black holes in ultra-compact dwarf galaxies

Besides studying the biggest and fluffiest galaxies above, I have also searched for intermediate-mass black holes in the tiniest galaxies using X-ray and radio observations with Dr. John Mulchaey and Prof. Jenny Greene. Figure 1 from Pandya+16 shows faint X-rays from several ultra-compact dwarf galaxies.

Miscellaneous Projects

One of the beautiful things about galaxy formation is that it involves pretty much every subfield of astrophysics and cosmology, and has deep connections to the fundamentals of dynamics, numerics, statistics and interpretable AI/ML. I chose to work in this field because it meant that I could pursue interdisciplinary research after being trained as a generalist during my doctoral and post-doctoral fellowships. I like to stay grounded about how much we still don’t know about the various constituents of galaxies and the Universe by collaborating across research domains. Here are some example projects: