Optimizing the Roman Space Telescope High-Latitude Wide Area Survey for mitigating chromatic PSF effects on shear measurement

Abstract

Chromatic point-spread-function (PSF) effects arise from differences between the spectral energy distributions (SEDs) of stars, used to model the PSF, and galaxies, used to measure shape distortions due to weak gravitational lensing, or shear. For the Roman Space Telescope, these effects can bias shear measurement and cosmological inference, making them an important systematic effect for shear calibration. These biases depend sensitively on survey design choices, particularly filter coverage and the availability of color information. In this work, we investigate how different Roman survey strategies affect the ability to mitigate chromatic PSF effects and whether residual biases in shear propagate into cosmological inference. Using realistic image simulations, we infer per-galaxy near-infrared SED slopes via radial basis function regression for four-, three-, two-, and single-band survey configurations. We quantify residual shear calibration biases under representative and non-representative training assumptions and propagate these biases into Markov Chain Monte Carlo analyses of cosmic shear and $3\times2$-point statistics. We find that three- and four-band strategies can reduce residual shear biases to $|m|\lesssim10^{-3}$, lowering the induced shifts in the lensing amplitude from $ΔS_8 \sim 0.6σ$ (cosmic shear) and $ΔS_8 \sim 0.7σ$ ($3\times2$-pt) in the uncorrected case to $ΔS_8 \lesssim 0.07σ$. Single-band surveys remain limited, with residual shear biases reaching or exceeding $|m|\sim 2\times 10^{-3}$ in some tomographic bins. Average, sample-wide corrections reduce but do not eliminate chromatic systematics, leaving residual biases of $ΔS_8 \sim 0.1σ$. Overall, our results demonstrate that we can robustly correct for these effects in the recommended three-band medium tier, but may encounter residual biases in a single-band wide tier.

Figures (7)

• Figure 1: The redshift distribution for the weak lensing galaxy sample from the first Roman Cosmology data challenge (blue) and our OU24 sample (orange) for comparison. There are a total of 8 tomographic bins, with the highest reaching $z = 4$ for DC1. We note that our OU24 sample reaches a maximum redshift of $z = 3.06$. For visualization purposes, the OU24 distribution is smoothed using a Gaussian kernel density estimator.
• Figure 2: Mean $J-H$ color as a function of redshift for galaxies with $H<25$, comparing the OpenUniverse2024 (green) catalog with added photometric noise to the COSMOS2020 Farmer catalog (orange). Solid lines denote the mean $J-H$ color when binned over redshift, while the shaded regions indicate the $1\sigma$ scatter. We can clearly see that at most redshifts, the OU24 NIR color-redshift relation is much tighter than that observed in real data. We confirm that the spread of the COSMOS2020 $J-H$ color is not driven by outliers, but rather it is due to features of the real data that are not present in simulations.
• Figure 3: Shear multiplicative bias $m$ as a function of redshift for the 8 Roman tomographic bins, measured from $H$-band shape measurements. Black solid lines show the uncorrected bias, while the black dashed lines correspond to applying a single true average correction to all galaxies. Colored lines indicate the residual bias after applying survey-specific corrections inferred using the RBF method with a representative training sample to predict the $H$-band SED slope from available photometric information. The left panel is for an analysis of Roman photometry alone, while the right panel uses LSST and Roman photometry. For the Wide tier, which has only a single Roman NIR band, an SED-based correction cannot be inferred on a per-object basis from Roman data alone. Instead, we can apply an average correction calibrated from the Medium tier, or use complementary LSST photometry to infer the SED-based correction for every galaxy. For this reason, the per-galaxy Wide-tier result is shown only in the right panel. Shaded gray bands indicate the relaxed requirement, described in Sec. \ref{['BiasRequirements']}, for comparison. We can see that most strategies fall within or close to the relaxed requirement when including LSST photometry, with the exception of the wide tier. When only Roman colors are used, the correction can fall outside of requirements for some redshift bins, but without any systematic trend towards positive or negative biases, with the exception of Wide-JH.
• Figure 4: Similarly to Fig. \ref{['fig:mbias_repsample']}, but with a bright training sample ($i < 24$) for the RBF correction. We see a clear systematic residual bias for the wide tier at higher redshifts, where the training sample has less coverage. Other configurations with 2+ NIR bands can stay within or close to the relaxed requirement boundary in the case where both Roman and LSST photometry is used for the correction. In the case of Roman-only information, all 3+ band strategies perform similarly, with only a single tomographic bin in the Medium-JHF scenario exceeding the requirements. For the Wide-JH scenario, biases are found outside or at the requirement in three tomographic bins.
• Figure 5: Left: Contour plot of the redshift vs. $H$-band magnitude for COSMOS20 (filled blue) and JWST NIRSpec (dashed red) galaxies. The adjacent 1D histograms show the respective distributions. Right: Contour plot of the redshift vs. $J-H$ color for the same galaxies. The upper panel shows the $J-H$ color and its $1\sigma$ dispersion as a function of redshift. The JWST galaxies have a higher average redshift, yet follow similar color-redshift evolution and occupy a similar region of color space. All 2-D contours show 16th, 50th, and 84th percentiles of the data distribution.