iiinova — where your digital self keeps evolving

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 02/06/2026

Bonus result: when you split by star-forming vs. quiescent, the environmental trends within each group are flat. The environment isn't reshaping galaxies within a class — it's accelerating transitions between classes. Quenching and bulge-building are coupled. (8/n)

Four panels at z ≥ 0.5 showing how galaxy structure varies with environment, split into star-forming galaxies (top row) and quiescent galaxies (bottom row), across two redshift bins. Within each subpopulation, the trends across environmental density are essentially flat — star-forming galaxies stay disk-dominated and quiescent galaxies stay bulge-dominated, regardless of whether they live in voids or overdense regions. The cluster points for quiescent galaxies sit slightly elevated. Two mass bins are shown per panel as colored points with 5σ error bars: lower-mass in purple, higher-mass in green, with cluster environments marked as diamonds.

100

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 02/06/2026

Why the shift with cosmic time? At earlier epochs, galaxies were gas-rich, merger rates were higher, and group preprocessing was more efficient. Galaxies got transformed even before they reached a cluster. By z < 0.5, only the densest cluster cores retain that punch. ⏳ (7/n)

100

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 02/06/2026

At z < 0.5: large-scale overdensity does ~nothing once you control for mass. But rich clusters still produce bulgier galaxies → cluster-specific physics is doing the work. At z ≥ 0.5, the effect operates across all scales. Remember that schematic from the first tweet? (6/n)

Schematic diagram showing how galaxy structural transformation mechanisms vary across cosmic time, overlaid on an HSC overdensity map of the cosmic web. Two panels at top: the left panel, labeled "Significant only at z ≥ 0.5," shows larger-scale preprocessing mechanisms — mergers, gas stripping, cosmic-web stripping, and starvation — operating in infalling galaxy groups along filaments. The right panel, labeled "All epochs, dominant mode at z < 0.5," shows cluster-scale transformation via ram-pressure stripping and repeated tidal interactions inside a massive cluster halo. Below the panels, a section of the HSC overdensity map shows bright filamentary structures and clusters at 0.6 ≤ z < 0.7, with arrows pointing from the map to the relevant mechanism panels.

100

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 02/06/2026

We confirm with >5σ that galaxy structure depends on environment even at fixed stellar mass. But — and this is the key — how it depends on environment changes with redshift, mass, AND environmental scale. And, this is why past literature has diverged on results! (5/n)

Four panels showing the median deviation in bulge-to-total light ratio at fixed stellar mass, ΔL_B/L_T, as a function of environment across the four redshift slices from 0.3 ≤ z < 0.7. The x-axis in each panel spans environmental bins from underdense (σ = −2) to overdense (σ = 3), with a separate "Cluster" point at the right. Each panel shows two mass bins as colored points with 5σ error bars: lower-mass galaxies in purple and higher-mass galaxies in green. Cluster environments are marked with diamonds. At z < 0.5, the large-scale environment points hover around ΔL_B/L_T = 0 with no monotonic trend, but the cluster points sit clearly above zero. At z ≥ 0.5, the higher-mass green points rise monotonically from underdense to cluster environments, while the lower-mass purple points show a weaker but similar trend, with elevated cluster values in both mass bins.

100

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 02/06/2026

Mass first: L_B/L_T correlates strongly with stellar mass at all redshifts — and the distribution is bimodal Two evolutionary tracks: secular growth vs. rapid transformation The bridge between them is more visible at higher z (4/n)

Schematic illustration of galaxy evolutionary pathways overlaid on the bulge-to-total light ratio versus stellar mass plane for galaxies at 0.3 ≤ z < 0.4. The background shows the observed distribution of galaxies as a hex-binned density plot. Two arrows mark distinct evolutionary tracks starting from a low-mass, disk-dominated population: a blue arrow labeled "Secular Growth" moves rightward and slightly upward, representing disk-dominated galaxies that build stellar mass while remaining disk-dominated. A green dashed arrow labeled "Rapid Transition" moves up and to the right, representing galaxies undergoing mergers or violent disk instabilities that quench star formation and produce bulge-dominated systems in the upper-right of the plane.

110

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 02/06/2026

Our approach: 3 million HSC galaxies, z < 0.7, with two environment tracers: 🌌 Large-scale overdensity (10 cMpc) 🪐 Rich clusters (CAMIRA, ~2 cMpc) and full posteriors on structural parameters. Sample size + posteriors = the key for untangling weak correlations (3/n)

100

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 02/06/2026

The argument is about fixed stellar mass: once you control for mass, does environment still shape galaxy structure? Some studies say yes. Some say no. Some say "only for massive galaxies." Some say "only for star-forming ones." 👇 (2/n)

A table titled "Selected Studies of Morphology-Environment Correlation," summarizing 15+ prior studies from 1980 to 2025. Columns list reference, survey or instrument, redshift range, mass or magnitude limit, sample size, density measure, correlation with environment, and whether a correlation persists at fixed stellar mass. The final-column results vary widely across studies: some find positive correlations for early-type galaxies, some find no correlation, others find correlations only at specific masses or redshifts or for specific galaxy subpopulations. The bottom row shows the present work, which uses 3 million HSC galaxies and finds cluster-scale correlations at all redshifts and masses, plus large-scale correlations only at z ≥ 0.5 for massive galaxies.

100

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 02/06/2026

Paper Acceptance Day! 🎉 Galaxies in denser regions of our universe are structurally different. Alan Dressler showed this in 1980. 45 years later, we're still arguing about why. With ~3 million HSC galaxies, we try to untangle it. 🧵 (1/n) #astronomy 🔭 🧪

130

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 22/05/2026

To learn more, check out: 🌐 arXiv: arxiv.org/abs/2605.18959 💻 Code: github.com/lincc-frameworks/hyrax 📚 Docs & 🎓 Tutorials: : hyrax.readthedocs.io Co-led with Drew Oldag and Michael Taurso, with an amazing team across @dirac-institute.bsky.social, CMU, Pitt, UMN, Yale & more! (12/n)

030

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 22/05/2026

Showcase #4: ⚡ Catching transients early in collaboration with the AppleCiDEr Team Light curves + spectra + images + metadata, all fused into one model running inside Hyrax. Strong early classification across SNe, CVs, AGN ... 🌟 Getting ready for the LSST firehose 🚒 (11/n)

110

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 22/05/2026

Showcase #3: 🌌 Dwarf galaxy hide-and-seek in collaboration with @theotherpsf.bsky.social Semi-resolved dwarfs are fuzzy and easily missed. So we injected fake ones into HSC, trained a CNN with Hyrax, and let it loose. 0.2% false positives! (10/n)

110

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 22/05/2026

Showcase #2: 🔭 Cluster-Scale Lens hunting with Hyrax in collab with @khullar.bsky.social UMAP + DBSCAN found a galaxy cluster overdensity in @vrubinobs.bsky.social DP1 → human-in-the-loop spotted a lensed arc behind it → Euclid VIS confirmed presence Gemini follow-up incoming 👀 (9/n)

132

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 22/05/2026

Showcase #1: 🪞Mergers & LSBs hiding in @vrubinobs.bsky.social Trained unsupervised models on ~400k DP1 galaxies. No labels. No supervision. New merger candidates + diffuse LSB galaxies that Euclid Q1 and DES missed — all found by poking around in latent space 🔭✨ (8/n)

120

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 22/05/2026

Hyrax is completely science- and data-type-agnostic, and we have tested it across a wide variety of scientific cases. Below is an incomplete list: (7/n)

110

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 22/05/2026

We also provide the necessary tools to quickly search for similar objects in your latent space (shown below for the @vrubinobs.bsky.social Data Preview 1 galaxies) These quick searches are powered by industry-standard databases and the Hierarchical Navigable Small World algorithm (6/n)

110

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 22/05/2026

Need to run your latent space exploration inside a Jupyter Notebook instead? No worries! We have got you covered there as well! (5/n)

132

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 22/05/2026

Given the recent rise in popularity of representation learning, Hyrax provides you with a wide swathe of tools to explore these latent spaces. Our 3D Latent Space Explorer is shown below. Head over to link.ghosharitra.com/hyrax-3d-demo for a live demo (4/n)

110

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 22/05/2026

Hyrax wraps all of the above in a small set of verbs: hyrax. download() → train() → infer() → visualize() - Multi-GPU Support - Multi-modal Data (images/spectra/light-curves) - Automatic Hyperparameter tuning using Optuna - Run monitoring and logging using MLFlow (3/n)

110

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 22/05/2026

Astronomy ML is often framed as “build a better model.” But in practice, the model itself is a tiny piece of the work. The rest is data wrangling, configuration, tracking, scaling, and visualization. (2/n)

110

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 22/05/2026

🔭 Paper Day! 🧪 Introducing Hyrax — an astro-focused ML framework built for the era of Rubin, Roman, and Euclid Handles the entire ML lifecycle: from data acquisition → training → inference → interactive latent-space discovery. pip install hyrax (1/n) 🧵 #astronomy #astrocoding #extragalactic

1163

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 09/04/2026

Thrilled to be part of the 2026 RCSA Fellows cohort — grateful for the opportunity and looking forward to learning from this amazing community! ❤️

000

Reposted by Aritra Ghosh I অরিত্র ঘোষ

NSF–DOE Rubin Observatory @rubinobservatory.org · 12/03/2025

Lights, camera, action! The world's largest digital camera has been installed at NSF–DOE Rubin Observatory! 🤩 The LSST Camera was the final major component of the observatory. With it in place, Rubin officially enters its final phase of testing!🔭 🔗: rubinobservatory.org/news/lsst-camera-installed

521570

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 25/11/2024

🔭 Has someone already made an Astro X ML/AI starter pack?

150

Reposted by Aritra Ghosh I অরিত্র ঘোষ

Ava Polzin @avapolzin.bsky.social · 14/11/2024

Figured I would get in on this whole starter pack thing and list (in whatever order bsky chooses) scientists working on studying galaxies (broadly defined, including galaxy formation/evolution + near-field cosmology). If I missed you and you'd like to be added, let me know! 🧪🔭 go.bsky.app/5ZZtqDy

218617

Reposted by Aritra Ghosh I অরিত্র ঘোষ

James Davenport @jradavenport.com · 25/11/2024

Are you on the Postdoc market and excited about early science with @vrubinobs.bsky.social, especially time domain, solar system, near field cosmology, or SETI? Well check out the @dirac-institute.bsky.social fellowship position that just got posted! apply.interfolio.com/158267

apply.interfolio.com

Apply - Interfolio {{$ctrl.$state.data.pageTitle}} - Apply - Interfolio

11815

Reposted by Aritra Ghosh I অরিত্র ঘোষ

Daniel | ScienceSocks et al. @sciencesocks.bsky.social · 25/11/2024

OMG look how glorious 🤯 #NancyGraceRoman #SpaceTelescope #astronomy

0273

Reposted by Aritra Ghosh I অরিত্র ঘোষ

Ava Polzin @avapolzin.bsky.social · 07/11/2024

Perhaps more timely now than ever: We are hosting a scientific symposium (among other events -- see pic) at UChicago, focused on promoting and retaining women in astrophysics. Symposium registration (link in next tweet) is open through Jan. 30 and we offer hybrid attendance. 🔭👩‍🔬🧪

A calendar showing the Picture an Astronomer events:
January 30 - Kick-off event
February 3 - Dara Norman Public Lecture
February 7 - Katie Mack Public Lecture
February 19 - Jocelyn Bell Burnell Public Lecture
February 25 - Anna Frebel Public Lecture
March 4 - 6 - Symposium

2268

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 15/08/2024

A huge shoutout to my co-authors:- Meg Urry, Meredith Powell, Rhythm Shimakawa, Frank vdBosch, Daisuke Nagai, Kaustav Mitra, and Andy Connolly! (11/n)

010

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 15/08/2024

My own blog post on this, including links to our public code, models, and public catalogs:- ghosharitra.com/hsc_rad_den.html (10/n)

100

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 15/08/2024

A wonderful write-up on this from James Urton at UW News summarizing how our work helps to settle some long-standing questions while raising new ones:- www.washington.edu/news/2024/08/14/… (8/n)

100

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 15/08/2024

If I have got you intrigued, take a look at the paper and the press releases! ApJ Link: doi.org/10.3847/1538-4357/ad596f arXiv Link: arxiv.org/abs/2408.07128 (7/n)

100

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 15/08/2024

... should depend on environmental density (at a fixed stellar mass and morphology). Although some previous theoretical works can be selectively invoked to explain some of the correlations, our work demonstrates the need for more comprehensive theoretical investigations! (6/n)

100

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 15/08/2024

Due to its dynamic nature, galaxy structure is a powerful probe of both dark matter halos and the baryonic processes inside them. There is no straightforward explanation as to why the sizes of galaxies across different morphological types and star-formation states ... (5/n)

110

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 15/08/2024

An alternative way to answer the same question is to investigate the size-mass relationship for galaxies in different environments. We did that and found that even at higher redshifts, the size-environment correlation exists beyond a certain stellar mass threshold! (4/n)

100

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 15/08/2024

How did we reach this conclusion? We correlated the deviation in effective radius from the average size against density excess measured using a top hat of 10 co-moving Mpc. We find the correlation to be strongest at lower redshifts and weaken in the last two z-slices. (3/n)

100

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 15/08/2024

This comprehensive study is an important step in resolving decades of contradictory results on this topic (see table below). It also sheds new light on how the structure of galaxies is connected with their dark matter halos; as well as their merger history. (2/n)

100

Aritra Ghosh I অরিত্র ঘোষ @aritraghsh09.bsky.social · 15/08/2024

🎉 🎉 Paper + Press Release Day 🎉 🎉 Using 3 million Hyper Suprime-Cam galaxies, we demonstrate with >5σ confidence that galaxies in denser environments are up to 25% larger than their counterparts with similar mass and morphology in less dense regions of the Universe. (1/n) 🔭, #astro

153

Reposted by Aritra Ghosh I অরিত্র ঘোষ

Diane Curtis @dianecurtis.bsky.social · 11/05/2024

Eclipse: You shall wait for years, make careful plans, drive hundreds of miles for five miraculous minutes. Aurora: U up?

101613364

Reposted by Aritra Ghosh I অরিত্র ঘোষ

James Davenport @jradavenport.com · 07/11/2023

Looking for a #astronomy postdoc this cycle? Excited about Rubin/LSST, big data/AI, building tools, finding the weird/wild/wonderful things in surveys? Check out the DiRAC Fellowship! jobregister.aas.org/ad/9be8c8f3

024