Johnny Esteves smiling, in a dark blazer and light blue shirt, in front of a chalkboard

Johnny H. Esteves

Astrophysicist
Postdoctoral Researcher
Harvard University

I weigh galaxy clusters to find out what dark energy is doing, and I help keep the Vera C. Rubin Observatory’s images sharp.

See my research Get in touch

Rubin Observatory on Cerro Pachón at sunset. Photo: J. Esteves.

About me

Hi, and welcome! I’m an astrophysicist and a postdoctoral researcher at Harvard University, in Christopher Stubbs’s group. I studied physics in Brazil and France, did my M.Sc. at the University of São Paulo, and finished my Ph.D. at the University of Michigan with Marcelle Soares-Santos in 2024.

I study how matter clumps together in the universe and what that tells us about dark energy. My main tool is galaxy clusters. I work on both ends of the problem: the cosmology, with the Dark Energy Survey and Rubin Observatory’s Legacy Survey of Space and Time, and the instrument, helping Rubin deliver sharp images from its mountaintop in Chile.

Mentoring is one of my favorite parts of the job. I have worked with Ph.D. students and undergraduates on projects from cluster cosmology to building a calibration instrument; you can read about it in Mentoring & outreach.

Hover over or tab to any dotted term for a one-line definition; all of them are collected in the glossary.

Research

Weighing galaxy clusters to test dark energy

In plain words. A galaxy cluster is a city of galaxies held together by dark matter. How many exist, and how heavy they are, depends on how fast the universe has been growing, and dark energy changes that growth. Weighing clusters well is a direct test.

Optical selection bias paper to appear on arXiv, November 2026

We find clusters in survey images by counting the galaxies that belong to them: the richness. But galaxies that merely lie in front of or behind a cluster along the same line of sight inflate the count, so the sample favors clusters with extra material lined up with them. That bias reaches the weak-lensing measurements used to weigh the clusters. In DES Year 3 it was serious enough that the team set aside the lensing signal close to the cluster centers, over 70% of its signal-to-noise, in order to combine clusters with the other probes. I derived the bias from first principles and showed that the data can calibrate it directly, without simulations.

The DES key project

I lead one of the two Dark Energy Survey cluster-cosmology key projects, which measures how much the cosmological answer shifts once this selection effect is modeled correctly.

My Ph.D.: stellar mass versus richness

For my thesis I asked whether the bias comes from which galaxies we count. I compared richness (λ) with an alternative built from stellar mass (μ_⋆, the total stellar mass of the member galaxies). Both showed similar behavior around the cluster aperture, which points to line-of-sight projection, not galaxy color, as the cause.

Galaxy clusters: who belongs?

In plain words. Before you can study a cluster you have to decide which galaxies are really in it. Two tools I built help with that.

Copacabana gives every galaxy a probability of belonging to a cluster, and does not rely on color, so blue galaxies are not missed (MNRAS 2025).
MStar is a neural network that estimates galaxy stellar masses to about 0.06 dex (roughly 15%) out to redshift 2, without fitting full galaxy spectra.

Fast code for big analyses

In plain words. Fitting a cluster model to survey data means running a program millions of times. On ordinary processors, one full run took more than a month.

With DES collaborators I worked on moving the heavy part to graphics processors (GPUs). The result, y3_cluster_cpp, takes about 0.7 seconds per step on NERSC Perlmutter and finishes a full run in about three hours.

Rubin from the inside: sharper images

In plain words. Weak lensing reads tiny stretches in galaxy shapes, so a blurry image costs real science. Rubin’s blur comes from the atmosphere, from warm air in the dome, and from the optics. Since 2022, partly on Cerro Pachón in Chile, I have worked on the parts I can help fix.

Twilight temperature forecast. A deep-learning model (N-BEATSx) predicts the evening temperature so the dome can be prepared in time. It is used in Rubin operations (SPIE 2026).
Guider pipeline. Real-time software for the camera’s corner guiders, which follow stars several times a second (designed for up to 9 times a second). It feeds Rubin’s monitoring dashboards and GuiderDIMM, a seeing monitor led by Z. Zhang and A. Roodman.
Camera characterization. My first-author paper describes how the LSST Camera’s sensors perform with projected artificial stars (PASP 2023).
Calibration instruments. PANDORA, a portable calibrator for telescope throughput, built with an undergraduate mentee (SPIE 2025).

Johnny Esteves smiling in a white hard hat with a headlamp, with the Rubin telescope's steel structure behind him — Inside the Rubin telescope on Cerro Pachón. Photo: J. Esteves.

Closing the Rubin dome during sunset. Video: J. Esteves. Watch on YouTube. The video loads from YouTube only when you press play.

Software

Code I wrote or helped write, mostly open source.

Copacabana

Probabilistic galaxy-cluster membership, independent of galaxy color.

MStar

Neural-network stellar masses for galaxies, with documentation on Read the Docs.

y3_cluster_cpp

GPU-accelerated cluster counts and lensing for DES, developed with Marc Paterno and DES collaborators.

CLensPy

Python tools for cluster gravitational lensing: halo profiles and projected densities.

Rubin’s Oracle

Twilight temperature forecasts for the Rubin dome, with a live dashboard.

mcp-ads-arxiv

A server that lets AI assistants read arXiv papers from their LaTeX source and search NASA ADS.

Xpipe

A pipeline for galaxy clusters in Chandra X-ray observations (2019).

sanzo-wada-colors

Sanzo Wada’s 171 colors and 581 combinations as matplotlib and seaborn palettes.

Publications

Selected papers, with a line on why each matters. The full list is on ADS.

The Halo Model of Optically Selected Galaxy Clusters

Esteves, J. H. · to appear on arXiv in mid-November · 2026

Explains optical selection bias from first principles and shows how it can be calibrated from the data itself.
Forward analytical model for the optical selection bias on galaxy cluster lensing profiles

Costanzi, Wu, Esteves, Grandis, To, Aguena · Phys. Rev. D 113, 103508 · 2026

A forward model of the bias that the DES small-scale analysis builds on.
Forecast for dome thermal control: a hybrid deep learning framework for the Vera C. Rubin Observatory

Esteves, Stubbs, Urbach, Venegas, et al. · Proc. SPIE 14155 · 2026

Predicts the evening temperature so Rubin's dome is ready for sharper images.
Dark Energy Survey Year 3 results: Cosmological constraints from cluster abundances, weak lensing, and galaxy clustering

DES Collaboration (Abbott, T. M. C., et al.) · Phys. Rev. D 112, 083535 · 2025

The first DES combination of clusters with the 3x2pt measurements, which required discarding the small-scale lensing signal.
Copacabana: a probabilistic membership assignment method for galaxy clusters

Esteves, Pereira, Soares-Santos, et al. · MNRAS 536, 931 · 2025

Decides which galaxies belong to a cluster without relying on color alone.
PANDORA: new avenues in sub-percent photometric precision for ground-based astronomy

Sjoberg, Esteves, Stubbs · Proc. SPIE 13624 · 2025

A portable calibrator for telescope throughput, built with an undergraduate mentee.
Photometry, Centroid and Point-spread Function Measurements in the LSST Camera Focal Plane Using Artificial Stars

Esteves, Utsumi, Snyder, Roodman, et al. · PASP 135, 115003 · 2023

How the LSST Camera's sensors perform when test stars are projected onto them.
Optical selection bias and projection effects in stacked galaxy cluster weak lensing

Wu, Costanzi, To, Esteves, et al. · MNRAS 515, 4471 · 2022

Showed that projection along the line of sight can bias stacked cluster lensing.

All papers, by year

Papers with a large collaboration are listed with the first author or the collaboration name. The ADS library has the complete author lists.

Mentoring & outreach

Mentoring

Two Ph.D. students at the University of Sussex (ongoing): the DES Year 3 cluster-cosmology pipeline and optical selection bias.
A Harvard undergraduate (2025): building the PANDORA calibrator, which won the 2025 Goldberg Prize and led to a refereed SPIE paper and an AAS talk. They are now a Ph.D. student at Caltech.
Two University of Michigan undergraduates (2022 to 2023): a stellar-mass validation on COSMOS, and a stellar-mass–observable relation for SPT-DES clusters that won a Blue Ribbon poster award at the UROP symposium.
Teaching assistant in physics: introductory electromagnetism labs at Michigan (2020 to 2021) and introductory astrophysics at Brandeis (2018 to 2019).

Outreach

Co-founded Astrotubers, a Brazilian astronomy YouTube channel with weekly videos and interviews (2017 to 2018).
Led night-sky observing sessions and astronomy lectures for students from elementary school to college at the University of São Paulo (IAG/USP, 2017 to 2018).

Johnny Esteves smiling next to a chalkboard with a chalk drawing of a spiral galaxy and the word Astro — Astronomy, one chalkboard at a time.

Talks

Sep 2026 Survey Science Group, Princeton University
Jun 2026 SPIE Astronomical Telescopes + Instrumentation: forecasting dome temperature for Rubin
Apr 2026 Astronomy seminar, University of Sussex
May 2025 Domain Astronomy in the Rubin Era, Rio de Janeiro
May 2023 Physics seminar, University of Wisconsin–Madison
Jan 2023 DES Collaboration Meeting, plenary talk, Portsmouth
Nov 2022 Fermilab Survey Group
Sep 2022 Physics seminar, University of Brasília

Johnny Esteves presenting a slide about the Vera C. Rubin Observatory to a seminar room — Presenting Rubin and LSST at the University of Michigan.

The full list is in my CV; a public version will be posted here soon.

Glossary

Plain-language definitions of the terms on this page.

Dark energy: The name for whatever is making the expansion of the universe speed up. We do not know what it is. Measuring how fast structure grew is one way to test it.
Galaxy cluster: The biggest structures held together by gravity: from a few dozen to thousands of galaxies, plus a lot of dark matter, in one clump.
Weak gravitational lensing: Mass bends light. A cluster’s gravity slightly stretches the shapes of galaxies behind it. Averaging over thousands of galaxies weighs the cluster, dark matter included.
Richness (λ): A headcount of the galaxies that belong to a cluster. More members usually means a heavier cluster.
Selection bias, projection effects: Clusters are picked by counting galaxies, so unrelated galaxies on the same line of sight can push a system over the cut. The sample then favors clusters with extra material in front or behind.
Mass calibration: Turning something we can measure, such as richness, into a cluster’s mass.
redMaPPer: The algorithm DES uses to find clusters. It looks for the tight “red sequence” of old, red galaxies that clusters contain.
DES: The Dark Energy Survey: a six-year imaging survey of about an eighth of the sky, from the 4-meter Blanco telescope in Chile.
Rubin, LSST, LSSTCam: The Vera C. Rubin Observatory will image the southern sky every few nights for ten years (the Legacy Survey of Space and Time) with LSSTCam, a 3.2-gigapixel camera.
Image quality, seeing: How sharp stars look. Turbulence in the air, warm air in the dome and imperfect optics blur them. Weak lensing needs sharp images.
Guider: Small sensors at the corners of Rubin’s camera that follow stars to keep the telescope pointing steady.
GPU: A graphics processor. It does many calculations at once, so each step of a fit runs much faster.

Contact

Happy to talk about cluster cosmology, Rubin image quality, GPU inference, and AI tools for science. Students and early-career colleagues are welcome to write.

admin (at) johnnyesteves.com

CV: a public version is coming soon.