Joel Leja

Intro

I am an associate professor of astronomy & astrophysics at The Pennsylvania State University. I hold the Dr. Keiko Miwa Ross Mid-Career endowed chair. My research focuses on understanding the processes of galaxy formation through combining modern, deep galaxy surveys with telescopes like JWST with statistics and machine learning.

In particular, I specialize in interpreting galaxy photometry and spectra using complex models for the stars, gas and dust living within them, in building and exploring analytical and empirical models of the evolving galaxy population as a whole, and in using astrostatistics and machine learning to better understand the universe. I’ve harnessed millions of hours of supercomputer time running specialized code to build a more complete picture of how galaxies form and evolve. I’ve worked on a wide range of science ranging from understanding the stellar origins of kilonova explosions to developing new methods for wide-field infrared galaxy surveys. I am very lucky to have been afforded the privilege of asking the big questions of the Universe!

Some of my current active projects include the Prime Focus Spectrograph galaxy survey, searching for first-light galaxies in the deep JWST survey UNCOVER, finding and understanding rare, red things in the early universe in the RUBIES (Red Unknowns: Bright Infrared Extragalactic Survey) survey with JWST, and developing software to analyze galaxy spectra with JWST. I’m also an active member of the Institute for Gravitation and the Cosmos and the Institute for Computational and Data Sciences at Penn State.

I believe we should all try to leave the world a better place than we found it. Here are my mentoring notes for my research group, loosely following a similar document from my postdoctoral mentor, Charlie Conroy.

Press

• Aug 2026 Hidden stars suggest that distant galaxies are more massive than they appear, PSU/Leiden Release (Science Daily, Space.com, Universe Today, Astronomy Now, The Debrief, Phys.org)

  Cheng et al., Nature Astronomy, led by Mariska Kriek at Leiden: a bottom-heavy initial mass function hides enough low-mass stars to make the earliest massive galaxies three to four times heavier than they look.

• Sep 2025 Mysterious 'red dots' in early universe may be 'black hole star' atmospheres, PSU/Max Planck Release (Phys.org, Universe Magazine)

  de Graaff et al., Astronomy & Astrophysics: the little red dots may not be galaxies at all, but dense envelopes of cool gas around a feeding supermassive black hole — a black hole star.

• June 2024 Little red dots in early universe baffle scientists (Space.com, Universe Today, Sky & Telescope, IFLScience, Earth.com, Mashable)

  and an AAS Journal Series Author Series interview.

• Feb 2024 'Cosmic lighthouses' that cleared primordial fog identified with JWST (Science Daily)
• Nov 2023 JWST discovery of the second- and fourth-most distant galaxies (PopSci, Business Insider, Space.com, Newsweek, Daily Mail, MSN)
• July 2023 Cosmic Front

  Featured in an hour-long documentary on JWST’s first year of science, NHK / Japanese National Television.

• March 2023 Machine learning in astrophysics, PSU ICDS Feature Story
• Feb 2023 Massive early galaxies defy prior understanding of the universe, ANU/PSU/NASA/Nature (NPR, the Guardian, the Atlantic, CNN, WTAJ, Fraser Cain interview)

  plus a KUSI television interview.

• Feb 2023 NASA's James Webb Telescope uncovers new details in Pandora's Cluster, NASA/STScI/PSU Release
• Nov 2022 Bright light from early universe 'opens new chapter in astronomy', NASA/STScI/PSU Release
• Nov 2022 Tracing the origins of rare, cosmic explosions, Keck/Northwestern/PSU Release
• Sep 2021 Early, massive galaxies running on empty, STScI/ALMA/PSU Release

My Research Group

Current Group Members

Former Group Members

Graduate Students & Postdocs

• Bingjie Wang, postdoctoral researcher, 2022–2025 — little red dots and ML-powered inference → Hubble Fellow at Princeton University
• Yijia Li, Ph.D. Penn State 2025, 2020–2025 — neural net-powered emission line modeling, author of cue → postdoctoral researcher at Northwestern University
• Kanishk Pandey, graduate student, 2023–2024
• Gautam Nagaraj, graduate student, 2021–2023 → postdoctoral researcher at EPFL
• Will Bowman, Ph.D. Penn State 2022, 2021–2022 — covariant data in galaxy inference → postdoctoral researcher at Yale University
• Imad Pasha, graduate student, Yale University, 2019–2020
• Jonathan Cohn, graduate student, Texas A&M, 2017–2018

Undergraduate Students

• Senti Bo, Nanjing University, 2025
• Si Rui, Nanjing University, 2025
• Nathan Cristello, Penn State, 2023–2024
• Junyu Zhang, B.S. Astronomy & Astrophysics Penn State 2023, 2021–2023 — finding rejuvenating galaxies, published in ApJ → Astronomy & Astrophysics graduate student at the University of Arizona
• Liam Schwartz, Penn State, 2021
• Leah Zuckerman, Brown University, 2020–2021 — published in ApJ
• Yuxin Dong, Brown University, 2019–2021 — published in ApJ
• Evan Haze Nunez, Smithsonian Astrophysical Observatory REU, 2018 — poster at the AAS
• Michael Bueno, Banneker Institute, 2017 — poster at the AAS
• Christopher Bradshaw, undergraduate thesis, Yale University, 2014–2015

Research

Distant galaxies reach us as light. Measuring that light alone is difficult and requires advanced optics. Turning those pinpricks of light into a mass, an age, and a formation history is an even harder part — and doing it better and better continually overturns what we thought we knew. That work is being rapidly accelerated and deepened by our group’s use of GPUs and modern machine learning, which have accelerated inference pipelines by ~2 million in the past half-decade.

It’s a particularly exciting time to be working on this, when JWST keeps finding things that should not exist. Our group builds the models that decide whether they are real.

How and When Galaxies Form Their Stars

Galaxies are star factories — they turn gas into stars. They do not work at a steady rate, however, and pretending otherwise can bias almost everything else you infer about them. Young stars outshine old ones by factors of tens of thousands, so a recent burst can hide the bulk of a galaxy’s stellar mass — the “outshining” problem.

With Bingjie Wang we built population models for star formation timescales as a first step at solving it, after showing how much the answers move once you take burstiness, the initial mass function and nebular physics seriously as unknowns rather than as fixed assumptions. Emilie Burnham has since developed a forward model that infers how bursty a whole galaxy population is, rather than fitting galaxies one at a time and hoping the errors cancel.

This continues a longer thread. Our census of the 0.2 < z < 3 universe (covering most of cosmic time!) rebuilt the stellar mass function and star-forming sequence from flexible models, and found a universe that formed its stars earlier and more calmly than the standard picture implied. We’re working to extend this coherence throughout all of cosmic time.

The Machinery

None of the above is possible without the tools, and building them is a large part of what the group does. Prospector, developed with Charlie Conroy and Ben Johnson, fits stars, gas, dust and black holes to galaxy photometry and spectra within a single Bayesian framework.

The bottleneck was never the physics but instead our implementation of it on silicon: a careful fit could take days on a single core. So we have been replacing the expensive parts with learned approximations. SBI++ brought simulation-based inference to astronomical data with its awkward realities — missing bands, wildly varying uncertainties. Yijia Li built cue, a neural emulator for nebular emission that frees line modelling from a handful of pre-computed grids. Elijah Mathews and Emilie Burnham work on high-dimensional emulators for both photometry and spectra of both resolved and unresolved stellar populations, and Lishan Shi is applying ultra-fast neural posterior estimators to survey-scale data.

The practical result has not been incremental but instead transformational: inference that once ran overnight per galaxy now runs in a fraction of a second. This unlocks new types of enormously detailed investigations (hundreds or thousands of parameters per galaxy) and new scales of survey science.

Little Red Dots and the Early Universe

The most confusing objects JWST has found are the little red dots: tiny, intensely red points of light that mostly live in the first billion years. Originally interpreted as ordinary galaxies, they imply stellar masses large enough to strain the cosmological model — but that turned out to be the wrong angle.

Working with Bingjie Wang (now a Hubble Fellow at Princeton) and our fantastic collaborators in the UNCOVER/RUBIES teams, we have argued that they are something stranger. Their ionizing spectra resemble massive stars rather than the hard radiation expected from an accreting black hole, and our most recent work finds water absorption, the signature of genuinely cool gas. Together with colleagues at the Max Planck Institute for Astronomy, that points toward a black hole star: a supermassive black hole feeding so hard that it wraps itself in a dense, cool envelope and shines like a stellar atmosphere.

The same modelling underpins our JWST survey work — the UNCOVER catalog of galaxies behind Abell 2744, and RUBIES, which turned up massive galaxies with extended formation histories at redshift 7–8. Nikko Cleri, Jakob Helton and Agrim Gupta are pushing on their mysterious ionizing sources, the most distant galaxies known, and the earliest quiescent systems.

What's Next

The coming surveys are large enough that per-galaxy fitting stops being an option. The Prime Focus Spectrograph will collect millions of galaxy spectra, and the interesting questions there are about populations: how bursty star formation is as a function of mass and epoch, how attenuation and the initial mass function vary, what the distributions look like rather than the individual draws.

That is the direction the group is heading — inference fast enough to run on everything, and statistical models designed for populations from the outset rather than retrofitted onto stacks of individual fits. The future is bright, and near!

How I Got Here

I started my Ph.D in 2010 at Yale with Pieter van Dokkum, just as the first wide, deep, representative surveys of galaxies in the first few billion years — 3D-HST and CANDELS — made it possible to write down how galaxies grow in a simple analytical model.

It did not work. Two independent measures of galaxy growth — the current star formation rate and the mass already locked into stars — disagreed by a factor of two. The problem turned out to be the models used to convert observations into physical properties: not enough physics, assumptions that were not self-consistent, and statistics too crude for the weak constraints available. Fixing that became Prospector-α, built with Charlie Conroy and Ben Johnson, and applying it to ~60k galaxies from 3D-HST brought the two measurements into agreement — implying an older, more quiescent universe than anyone had assumed.

Everything above is a continuation of that same argument: in galaxy evolution, the measurement is the science.

Outreach

In addition to my professional mentoring, I believe that knowing more about science and the scientific process improves the lives of everyone (and — talking to folks about it is lots of fun!).

Selected Publications

I am an author of 230 publications, 10 of them first-author, with an h-index of 79 and roughly 22,000 citations as of May 2026. The complete list is on ADS; a selection follows, newest first.

1. Water absorption confirms cool atmospheres in two little red dots
   Wang, Bingjie; Leja, Joel; et al., 2026, submitted to Nature
2. It's More Complicated Than You Think: A Forward Model to Infer the Recent Star Formation History, Bursty or Not, of Galaxy Populations
   Burnham, Emilie; Wang, Bingjie; Leja, Joel; et al., 2026, ApJ, 1001, 205
3. The Missing Hard Photons of Little Red Dots: Their Incident Ionizing Spectra Resemble Massive Stars
   Wang, Bingjie; Leja, Joel; et al., 2025, ApJ, accepted
4. Population Models for Star Formation Timescales in Early Galaxies: The First Step Towards Solving Outshining in Star Formation History Inference
   Wang, Bingjie; Leja, Joel; et al., 2025, ApJ, 987, 184
5. Cue: A Fast and Flexible Photoionization Emulator for Modeling Nebular Emission Powered by Almost Any Ionizing Source
   Li, Yijia; Leja, Joel; et al., 2025, ApJ, 986, 9
6. RUBIES: Evolved Stellar Populations with Extended Formation Histories at z ~ 7–8 in Candidate Massive Galaxies Identified with JWST/NIRSpec
   Wang, Bingjie; Leja, Joel; et al., 2024, ApJL, 969, L13
7. Quantifying the Effects of Known Unknowns on Inferred High-redshift Galaxy Properties: Burstiness, the IMF, and Nebular Physics
   Wang, Bingjie; Leja, Joel; et al., 2024, ApJ, 963, 74
8. The UNCOVER Survey: A First-look HST+JWST Catalog of Galaxy Redshifts and Stellar Population Properties Spanning 0.2 ≤ z ≤ 15
   Wang, Bingjie; Leja, Joel; et al., 2024, ApJS, 270, 12
9. SBI++: Flexible, Ultra-fast Likelihood-free Inference Customized for Astronomical Applications
   Wang, Bingjie; Leja, Joel; et al., 2023, ApJL, 952, L10
10. A New Census of the 0.2 < z < 3.0 Universe, Part II: The Star-Forming Sequence
    Leja, Joel; et al., 2022, ApJ, 936, 165
11. A New Census of the 0.2 < z < 3.0 Universe, Part I: The Stellar Mass Function
    Leja, Joel; et al., 2020, ApJ, 893, 111
12. An Older, More Quiescent Universe from Panchromatic SED Fitting of the 3D-HST Survey
    Leja, Joel; et al., 2019, ApJ, 877, 140
13. How to Measure Galaxy Star Formation Histories II: Nonparametric Models
    Leja, Joel; et al., 2019, ApJ, 876, 3
14. Deriving Physical Properties from Broadband Photometry with Prospector: Description of the Model and a Demonstration of its Accuracy
    Leja, Joel; et al., 2017, ApJ, 837, 170
15. Tracing Galaxies Through Cosmic Time with Number Density Selection
    Leja, Joel; et al., 2013, ApJ, 766, 33

Contact

My office is Davey Laboratory 515. You can reach me electronically at joel.leja@psu.edu.