🔭 astrophysics

IRMaGiC: Extending Luminous Red Galaxy Selection into the Infrared with Joint Rubin Observatory's Large Survey of Space Time and Roman's High Latitude Imaging Survey

The paper introduces IRMaGiC, an algorithm that extends the RedMaGiC method to the $1 \leq z \leq 2$ redshift range by combining Rubin LSST optical data with Roman Space Telescope infrared observations, thereby significantly improving the selection and photometric redshift accuracy of Luminous Red Galaxies for future cosmological surveys.

Original authors: Zhiyuan Guo (for the LSST Dark Energy Science Collaboration), Chris. W. Walter (for the LSST Dark Energy Science Collaboration), Eli S. Rykoff (for the LSST Dark Energy Science Collaboration)

Published 2026-04-22

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Zhiyuan Guo (for the LSST Dark Energy Science Collaboration), Chris. W. Walter (for the LSST Dark Energy Science Collaboration), Eli S. Rykoff (for the LSST Dark Energy Science Collaboration)

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Finding the Universe's "Old Stars"

Imagine the universe is a giant, sprawling city that has been growing for billions of years. In this city, there are two main types of neighborhoods:

The "Party Districts": These are young, blue galaxies where new stars are being born constantly. They are loud, bright, and chaotic.
The "Quiet Retirement Communities": These are Luminous Red Galaxies (LRGs). They are old, massive, and have stopped making new stars. They are red because their stars are aging (like how hair turns gray).

Astronomers love these "Retirement Communities" because they are the perfect signposts to map out the structure of the universe. They are bright, easy to spot, and their colors tell us exactly how far away they are (their "redshift").

The Problem: The "Redshift Blind Spot"

For a long time, astronomers had a tool called RedMaGiC (Red Galaxy Mapper) that was great at finding these old galaxies. But it had a blind spot.

Think of the universe's expansion like a giant rubber band stretching out. As the rubber band stretches, light from distant objects gets stretched too, shifting from blue to red.

For nearby galaxies, their light is still in the visible spectrum (what our eyes see).
For very distant galaxies (farther away than 1 billion light-years), their light gets stretched so much that it shifts into the infrared (heat radiation), which human eyes can't see.

Previous tools were like a camera that only had visible light lenses. They could see the "Retirement Communities" up to a certain distance, but once the galaxies got too far away, the camera went blind. The "4000 Angstrom break" (a specific fingerprint in the galaxy's light) moved out of the camera's view, making it impossible to tell exactly how far away the galaxy was.

The Solution: IRMaGiC (The Infrared Upgrade)

This paper introduces IRMaGiC (Infrared MaGiC). It's an upgrade to the old tool that combines two powerful telescopes:

LSST (Rubin Observatory): A giant ground-based telescope with a huge camera that sees the universe in visible light (like a standard camera).
Roman Space Telescope: A space telescope that sees the universe in infrared light (like a night-vision camera).

The Analogy:
Imagine you are trying to identify a specific type of rare bird in a forest.

LSST is like looking at the bird through a window during the day. You can see its shape and general color.
Roman is like putting on night-vision goggles. You can see the heat signature and details you missed in the daylight.

By combining the "daylight view" (LSST) with the "night-vision view" (Roman), IRMaGiC can see the "Retirement Communities" much further away, all the way to redshifts of $z=2$ (which is looking back in time to when the universe was much younger).

How It Works: The "Red Sequence" Recipe

The algorithm works like a master chef following a recipe to identify the perfect ingredients.

The "Seed" Ingredients: First, the team needed a small, perfect sample of these old galaxies to learn what they look like. They used a special "spectroscopy" mode on the Roman telescope (like a high-tech scanner) to get the exact distance of a few thousand galaxies. These are the "Seeds."
The "Recipe" (Calibration): Using these seeds, they created a mathematical "recipe" (called a Red-Sequence Template). This recipe predicts exactly what color an old galaxy should be at any specific distance.
- Analogy: It's like knowing that a ripe apple is always red. If you see a red object, you know it's likely an apple. If you see a galaxy that matches the "red apple" color recipe, you know it's an old galaxy.
The "Search" (Selection): The algorithm then scans millions of galaxies. If a galaxy matches the "red apple" recipe and is bright enough, it gets selected.
The "Refinement" (Afterburner): Sometimes the recipe isn't perfect. The team uses a "post-processing" step (an afterburner) to tweak the distance estimates, making them even more accurate.

The Results: Why This Matters

The paper tested this new method using simulated data (a "virtual universe" created by computers). Here is what they found:

Sharper Vision: IRMaGiC is much better at guessing the distance of these galaxies than previous methods. It reduced the "blur" (scatter) in their measurements.
Going Deeper: It successfully extended the search for these galaxies from the "nearby" universe ( $z < 1$ ) into the "deep" universe ( $1 \le z \le 2$ ).
The Sweet Spot: They found that for this to work best, the two telescopes (LSST and Roman) need to look at the same patch of sky. If they overlap by at least 200 square degrees (about the size of 1,000 full moons), the system works like a charm.

The Bottom Line

IRMaGiC is like giving astronomers a pair of binoculars that work in both daylight and night-vision.

By combining the optical power of the Rubin Observatory with the infrared power of the Roman Space Telescope, we can finally map out the "Retirement Communities" of the universe much further back in time. This will help scientists understand how the universe is expanding, how dark energy is pushing galaxies apart, and how the cosmic web is structured. It turns a blurry, distant map into a high-definition picture of our cosmic history.

1. Problem Statement

Luminous Red Galaxies (LRGs) are critical tracers for cosmological studies, including Baryon Acoustic Oscillations (BAO) and Galaxy Clustering (GC). Existing algorithms like RedMaGiC successfully select LRGs and estimate photometric redshifts ( $z_{phot}$ ) in the optical regime ( $z \lesssim 1$ ). However, at higher redshifts ( $1 \le z \le 2$ ), the 4000 Å spectral break shifts into the near-infrared (NIR).

Limitation of Optical Surveys: Surveys like the Vera C. Rubin Observatory's LSST (optical $ugrizy$) lack sufficient NIR coverage to accurately calibrate the red-sequence template at $z > 1$ .
Limitation of Previous NIR Extensions: Previous attempts to extend LRG selection to higher redshifts (e.g., using Subaru HSC) are limited by the lack of deep NIR bands, restricting selection to $z \sim 1.25$ .
Goal: Develop a method to extend LRG selection and accurate $z_{phot}$ estimation to the range $1 \le z \le 2$ by combining LSST optical data with the NIR capabilities of the Nancy Grace Roman Space Telescope.

2. Methodology: IRMaGiC

The authors introduce IRMaGiC, an extension of the RedMaGiC algorithm, utilizing simulated data from the LSST Dark Energy Science Collaboration (DC2) and the Roman High Latitude Imaging Survey (HLIS).

A. Data Integration

Surveys: Joint analysis of LSST (optical $u,g,r,i,z,y$ ) and Roman HLIS (NIR $F106, F129, F158, F184$ , denoted as $Y, J, H, F$ ).
Cross-Matching: Objects are matched within 0.2 arcseconds, requiring significant detections ( $>5\sigma$ ) in LSST $i, z$ and Roman $Y, J, H, F$ bands.
Spectroscopic Calibration: The Roman High Latitude Spectroscopic Survey (HLSS) is assumed to provide spectroscopic redshifts ( $z_{spec}$ ) for a subset of galaxies to calibrate the red-sequence template.

B. Seed Galaxy Selection (Novel Strategy)

Traditional color-cut or Gaussian mixture methods fail at $z > 1$ due to the diminishing color bimodality between red and blue galaxies.

Approach: The authors employ a Support Vector Machine (SVM) with a radial basis function (RBF) kernel to define a non-linear decision boundary in the $(g-Y)$ vs. $(Y-J)$ color-color space.
Training: The SVM is trained on simulated "truth" data (using specific star formation rates and red-sequence flags) to separate quiescent (red) from star-forming (blue) galaxies.
Efficiency Filtering: A magnitude cut ( $m_H \le 22$ ) is applied, and a Roman grism efficiency curve (derived from Guo et al. 2025) is used to account for the probability of obtaining a reliable $z_{spec}$ from the Roman grism. This curve is applied differently for $1 \le z \le 1.5$ and $1.5 \le z \le 2$ .

C. Red-Sequence Template Calibration

The algorithm models the red-sequence as a ridge in the color-magnitude diagram:
$\langle c|z, m \rangle = a(z) + s(z)[m - m_{ref}(z)]$

Parameters: The intercept $a(z)$ , slope $s(z)$ , and intrinsic scatter $\sigma_{int}(z)$ are modeled as smooth functions of redshift using cubic spline interpolation.
Calibration Steps:
1. Estimate pivot magnitudes ( $m_{ref}$ ).
2. Perform initial estimates of mean color and width.
3. Refine the seed sample by selecting galaxies within the intrinsic scatter.
4. Jointly calibrate $a(z)$ , $s(z)$ , and $\sigma_{int}(z)$ by minimizing the negative log-likelihood.
Scaling: To simulate a full survey, the authors scale the 20 deg $^2$ simulation to a 2000 deg $^2$ overlap area, using a bootstrap method to validate statistical robustness.

D. Candidate Selection & Photo-z Refinement

Selection Criteria: Galaxies are selected based on:
1. Luminosity: $L > 0.5L^*$ (highdens) or $L > L^*$ (highlum).
2. Fit Quality: A constant $\chi^2_{red} \le 8$ cut (following Pandey et al. 2022 to avoid systematic biases associated with variable cuts).
Afterburner: A final calibration step uses a subset of galaxies with $z_{spec}$ to correct for median redshift offsets ( $\Delta z$ ) and refine error estimates ( $\sigma_z$ ) using spline interpolation.

3. Key Results

A. Photometric Redshift Performance

Accuracy: IRMaGiC significantly outperforms standard LSST DC2 photo-z estimates (derived from BPz).
- IRMaGiC (highdens): Bias $\approx 0.003$ , Scatter (NMAD) $\approx 0.026$ for $1 \le z \le 1.9$ .
- DC2 Photo-z: Bias $\approx -0.016$ , Scatter $\approx 0.069$ .
Outliers: The fraction of $5\sigma$ outliers is $\approx 1\%$ for $z < 1.9$ in the highdens sample, rising only beyond $z \approx 1.9$ due to a lack of seed galaxies for calibration.
Scatter Sources: Increased scatter at specific redshifts ( $z \approx 1, 1.3, 1.8$ ) correlates with the 4000 Å break crossing filter boundaries (e.g., $i \to z$ , $z \to Y$ ).

B. Comparison with Existing Samples

vs. Low-z RedMaGiC: IRMaGiC shows slightly higher scatter than low-z RedMaGiC samples. This is attributed to differences in truth magnitude definitions between the DC2 and Roman simulations and the use of generic SourceExtractor photometry in the simulation, rather than fundamental algorithmic flaws.
Redshift Extension: The method successfully extends reliable LRG selection from $z \sim 1$ to $z \approx 1.9$ , a range previously inaccessible to optical-only surveys.

C. Impact of Overlap Area

The study analyzed how the size of the overlapping LSST-Roman region affects performance:

Minimum Requirement: An overlap of at least 200 deg $^2$ is required to achieve reliable redshift measurements up to $z \approx 1.9$ .
Diminishing Returns: Increasing the overlap beyond 500 deg $^2$ yields minimal improvements in redshift coverage or scatter.

4. Significance and Conclusion

Cosmological Impact: By enabling accurate LRG selection and redshift estimation up to $z \approx 2$ , IRMaGiC facilitates high-redshift BAO and clustering measurements, which are essential for constraining dark energy evolution.
Synergy: The paper demonstrates that the combination of LSST's depth and Roman's NIR coverage is essential for probing the high-redshift universe.
Robustness: The use of an SVM for seed selection and a constant $\chi^2$ cut addresses known systematic issues in previous LRG selection methods.
Future Outlook: The authors note that while current simulations use generic photometry pipelines, future Roman data products with optimized pipelines are expected to further reduce scatter and improve calibration accuracy.

In summary, IRMaGiC provides a robust framework for extending the red-sequence technique into the high-redshift regime, leveraging the complementary strengths of the Rubin Observatory and the Roman Space Telescope to unlock new cosmological potential.