🔭 astrophysics

Initial Characterization of Stellar Photometry of Roman images from the OpenUniverse Simulations

This paper presents an initial benchmark characterization of stellar photometry in Roman Space Telescope simulations using OpenUniverse, demonstrating that a library of effective PSFs and sub-SCA division can achieve 0.6–1.2% flux precision while revealing residual non-linearity and color dependence that require further algorithmic refinement to meet mission requirements.

Original authors: Lauren Aldoroty, Daniel Scolnic, Arun Kannawadi, Rob Knop, Benjamin Rose, Rebekah Hounsell, Michael Troxel

Published 2026-04-17

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Lauren Aldoroty, Daniel Scolnic, Arun Kannawadi, Rob Knop, Benjamin Rose, Rebekah Hounsell, Michael Troxel

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

Imagine the Nancy Grace Roman Space Telescope as a giant, ultra-powerful camera being built to take the most detailed "family photo" of the universe ever attempted. Its main job is to study Dark Energy, the mysterious force pushing the universe apart, by watching distant exploding stars (Supernovae) over billions of years.

To do this, the telescope needs to measure the brightness of these stars with extreme precision. If the camera is even slightly off, the scientists might think the universe is expanding faster or slower than it actually is.

This paper is like a test drive for the camera's software before the real mission begins. The authors built a library of "effective" lenses (called ePSFs) to help the computer understand exactly how the telescope sees light.

Here is a breakdown of their findings using simple analogies:

1. The "Fuzzy" Lens Problem

When you look at a star through a telescope, it doesn't look like a perfect dot. It looks like a fuzzy smudge. This smudge is called the Point Spread Function (PSF).

The Problem: The Roman telescope has a very wide field of view. Because of this, the "fuzziness" of the star changes depending on where you look in the picture. A star in the top-left corner looks different than a star in the bottom-right. It's like looking through a window that is warped differently in every corner.
The Challenge: If you use one single "average" lens to measure every star in the photo, you will get the brightness wrong.

2. The Solution: The "Mosaic" Approach

To fix this, the authors didn't try to make one perfect lens for the whole image. Instead, they cut the image into a grid, like a pizza sliced into smaller and smaller slices.

The Analogy: Imagine trying to describe the texture of a giant, bumpy rug. If you describe the whole rug with one word, you'll be wrong. But if you divide the rug into 64 small squares (an 8x8 grid) and describe the texture of each square individually, you get a much better picture.
The Result: By creating a specific "lens model" for each of these 64 small squares, they could measure the brightness of stars with incredible accuracy. They found that using this detailed grid improved their measurements by up to 20% compared to using just one big, average model.

3. The "Color" Glitch

The team also discovered that the "fuzziness" of the lens depends on the color of the star.

The Analogy: Think of a prism. Blue light bends differently than red light. The telescope's lens does something similar.
The Finding: In the bluer filters (like the R062 band), the lens model struggled more. It was like trying to take a photo of a blue object with a lens that was slightly tuned for red objects. The authors suggest that in the future, the software might need to have different lens models for different colors to get perfect results.

4. The "Pixel Phase" Bias

The telescope's camera is made of pixels (tiny squares of light sensors). Sometimes, a star lands perfectly in the middle of a pixel, and sometimes it lands right on the line between two pixels.

The Problem: If the software doesn't know exactly where the star is, it might guess the brightness wrong. This is called "pixel-phase bias." It's like trying to weigh a coin on a scale that only has markings every inch; if the coin is between the lines, you have to guess.
The Fix: The authors tested how well their system could find the exact center of the star. They found that for most colors, they could pinpoint the location very well. However, for the bluer, fuzzier images, the system was a bit "wobbly," leading to slightly less precise measurements.

5. The Bottom Line

Success: They proved that by using their new "grid" method, they can measure star brightness with less than 1% error. This is the gold standard needed for the Roman mission to succeed.
Warning: While they are very close, there is still a tiny bit of "nonlinearity" (a slight wobble in the math) that needs to be fixed before the telescope launches.
Future: They recommend that when the real telescope is in space, it should take multiple pictures of the same spot with slight rotations (dithering). This will help the computer figure out exactly where the stars are and correct for the "wobbly" lens effects.

In summary: This paper is a successful "dress rehearsal." It showed that if the Roman telescope uses a grid-based system to account for the fact that its lens looks different in every corner of the sky, it will be able to take the incredibly precise measurements needed to solve the mystery of Dark Energy.

1. Problem Statement

The Nancy Grace Roman Space Telescope (Roman) is designed to study dark energy using Type Ia Supernovae (SNe Ia) via the High-Latitude Time Domain Survey (HLTDS). A critical requirement for these cosmological measurements is photometric precision better than 1% when sky noise is subdominant.

However, achieving this precision is complicated by the nature of Roman's Wide Field Instrument (WFI):

Undersampling: The Point-Spread Function (PSF) is undersampled by the detector pixels.
Spatial Variation: The PSF varies significantly across the focal plane due to optical aberrations, with Full Width at Half Maximum (FWHM) variations of ~1% to 10% across a single Sensor Chip Assembly (SCA).
Chromatic Dependence: The PSF shape is expected to vary with stellar color.
Pixel-Phase Bias: In undersampled images, the measured position and flux of a star depend on its sub-pixel location relative to the pixel grid.

Standard single-PSF models applied to entire images introduce systematic errors that exceed the required precision. The paper addresses the need to characterize and mitigate these effects using the OpenUniverse2024 simulations, which provide realistic end-to-end Roman survey data.

2. Methodology

The authors developed a library of effective PSFs (ePSFs) and tested their ability to recover stellar fluxes and positions.

Data Source: 180 simulated images (10 pointings $\times$ 7 photometric bands: R062, Z087, Y106, J129, H158, F184, K213) from the OpenUniverse2024 HLTDS simulation.
ePSF Generation Strategy:
- Instead of using a single PSF for an entire image, the authors divided each SCA (4088 $\times$ 4088 pixels) into a grid of $N_{grid} \times N_{grid}$ elements.
- Tested grid sizes: $1\times1$ , $2\times2$ , $4\times4$ , and $8\times8$ .
- Star Selection: Used stars with true magnitudes $19 < m_{truth} < 21.5$ (unsaturated and sky-noise limited).
- Stacking: Since a single image does not contain enough stars per grid element to build a robust PSF, they stacked stars from $N_{stack}$ different images (totaling $\sim$ 100 stars per grid element) to construct the ePSF using photutils.
- Oversampling: Final ePSFs were oversampled by a factor of 3.
Photometric Analysis:
- Applied the ePSFs to stars not used in their construction to avoid bias.
- Used photutils for PSF fitting, allowing coordinates to vary as fit parameters (simulating real detection software behavior).
- Calculated the modified median absolute deviation ( $\hat{\sigma}$ ) of the fractional flux difference: $(F_{truth} - F_{fit}) / F_{truth}$ .
Performance Metrics:
- Flux Precision: Measured via $\hat{\sigma}$ .
- Nonlinearity: Measured the slope ( $s_{NL}$ ) of flux residuals vs. magnitude.
- Color Dependence: Measured the slope ( $s_{color}$ ) of residuals vs. stellar color ( $J129 - H158$ ).
- Positional Accuracy: Measured the offset between recovered and true coordinates.
- Detection Efficiency: Used Source Extractor to determine the magnitude limit for 50% detection ( $m_{50}$ ).

3. Key Contributions

ePSF Library for OpenUniverse: Created and validated a library of ePSFs compatible with Roman HLTDS simulations, specifically designed to handle spatial variations by subdividing the detector.
Grid Optimization: Demonstrated that subdividing the detector into an $8\times8$ grid is optimal for the OpenUniverse simulations, improving flux precision by 9%–30% compared to a single global PSF ( $1\times1$ grid).
Quantification of Systematics:
- Identified that R062 and Z087 bands suffer from the largest systematic errors due to severe undersampling and PSF asymmetry.
- Quantified the impact of PSF asymmetry and ellipticity on coordinate recovery and flux precision.
- Measured photometric nonlinearity and color dependence, highlighting the need for color-dependent PSF models in specific bands.
Detection Limits: Established that Source Extractor achieves $\sim$ 100% detection efficiency down to magnitude 26 in all bands.

4. Key Results

Flux Precision:
- The $8\times8$ grid achieved flux precision between 0.6% and 1.2% across most bands.
- R062 ( $\hat{\sigma} = 1.22\%$ ) and Z087 ( $\hat{\sigma} = 1.40\%$ ) failed to meet the strict $<1\%$ requirement, primarily due to undersampling and asymmetry.
- Redder bands (e.g., K213) performed better ( $\sim0.73\%$ ).
Spatial Variation:
- Moving from a $1\times1$ to an $8\times8$ grid reduced flux scatter by up to 20%.
- Positional recovery precision was generally $<0.03$ pixels (except for R062/Z087), correlating loosely with flux precision.
Nonlinearity:
- Measured nonlinearity slope $|s_{NL}| < 1.93 \times 10^{-3} \text{ dex}^{-1}$ .
- This exceeds the detector count-rate nonlinearity requirement of $<0.3\%$ over 11 magnitudes, indicating that photometric methodology (not just detector physics) is a source of systematic error that needs further reduction.
Color Dependence:
- A significant color dependence was found in the R062 band ( $s_{color} \approx -4.73 \times 10^{-2}$ , $\sim8\sigma$ significance).
- This suggests that a color-dependent PSF model is necessary for the bluer bands to achieve required precision.
Asymmetry & Ellipticity:
- Strong correlation found between PSF asymmetry index ( $\alpha$ ) and flux scatter.
- R062, Z087, and Y106 showed the highest asymmetry and ellipticity, leading to larger coordinate offsets.
Detection Efficiency:
- 50% detection limit ( $m_{50}$ ) ranges from 25.26 (K213) to 26.70 (Y106).
- Nearly 100% of stars are detected down to magnitude 26.

5. Significance and Future Work

Benchmarking: This work serves as a critical "benchmark" for the Roman community, establishing baselines for photometric performance before the telescope launches.
Methodology Validation: It validates the necessity of spatially varying PSF models (grid-based ePSFs) rather than global models for Roman data analysis.
Calibration Strategy: The results suggest that future ePSF generation strategies (both for simulations and on-orbit calibration) must:
- Utilize fine grids (e.g., $8\times8$ ) to capture spatial variations.
- Account for color dependence, particularly in blue bands.
- Prioritize observations in regions of high PSF asymmetry or rapid FWHM change.
Limitations & Next Steps:
- The current study used simulated images without dithering (rotational dithering is planned for the actual mission). Future work must incorporate dithered observations to break the degeneracy between star position and PSF shape (pixel-phase bias).
- The nonlinearity due to photometry remains higher than requirements; further refinement of the fitting algorithms is needed.
- Real detector effects (e.g., time-evolving PSF) were not included; these will be addressed during the mission's calibration phase using "touchstone" fields.

In conclusion, the paper demonstrates that while Roman's spatially varying and undersampled PSF poses significant challenges, a carefully constructed, grid-based ePSF library can recover stellar fluxes to near the required 1% precision, provided that color dependence and specific band limitations (like R062) are explicitly modeled.