CIBER times galaxy cross-correlations reveal a bright, low-redshift NIR background

This paper presents the first tomographic analysis of near-infrared (NIR) extragalactic background light (EBL) anisotropies, cross-correlating CIBER 1.1 and 1.8 µm imager data with photometric galaxy catalogs from DESI Legacy Survey DR8 and Hyper-Suprime-Cam Ultra-Deep Survey. The authors measure significantly higher cross-power than expectations from an integrated galaxy light (IGL) model on scales l < 2000, concentrated at low redshift (z ≲ 0.6).

For multipoles 304 < l < 2000, the authors detect cross-power at 9.8σ/9.2σ significance from CIBER × DESI-LS and 5.5σ/7.1σ with CIBER × HSC. For l < 2000, the measurements exceed predictions at 7.4σ/6.2σ significance for CIBER × DESI-LS 1.1/1.8 µm and 4.3σ/5.3σ for CIBER × HSC. Extending to lmax = 4000, the disagreement rises to 10.3σ/13.2σ and 7.2σ/9.4σ significance for the same combinations.

The strongest cross-correlation signals arise at low redshift (0.1 < z < 0.5), exceeding baseline IGL predictions by factors as large as 5−10 in the z ∈ [0.2, 0.3) and z ∈ [0.3, 0.4) bins. The cross-correlation coefficient rlIg peaks at z ∼ 0.2 − 0.4 and then declines toward z = 1. In redshift bin z ∈ [0.2, 0.3), rlIg peaks in both bands at 0.40 − 0.45, compared to predicted rlIg ∼ 0.25 − 0.30.

Cluster member galaxies (CMGs) and associated structure account for 15−20% of the large-angle cross-power for multipoles 1000 < l < 10000, indicating that group- and galaxy-scale halos contribute the bulk of the signal. The CMG sample spans a redshift range that peaks at z = 0.6 and resides largely below z = 1.

Through a parametric halo model decomposition, the authors detect two-halo and one-halo clustering in cross-power at high significance, with amplitudes that decline smoothly across z = 0−1. The one-halo component is detected at > 3σ significance for nearly all DESI-LS combinations, peaking at 6.4σ for 0.2 5σ for bins 0.2 < z < 0.4 and 0.4 < z < 0.6.

The inferred one-halo cross-power is of similar amplitude between DESI-LS and the deeper HSC catalog, implying a scenario in which low-redshift EBL fluctuations are amplified by contributions from lower-mass halos with satellites and/or diffuse intra-halo light (IHL).

Converting two-halo fits into estimates of bI × dI/dz, the authors find that standard IGL predictions underestimate measurements, even when assuming an intensity bias as high as 3, similar to that of large SZ clusters. Across four bias configurations (characteristic halo masses log Mh ∈ 12.0, 13.0, 14.0 and tSZ-like bias), the predicted bI × dI/dz varies by a factor of four, yet even the maximal tSZ-like bias predictions are unable to reproduce measurements at z < 0.6 by a large factor.

Using a coherence-based estimator, the authors reconstruct the NIR auto-power ClII associated with correlated large-scale structure (LSS) at redshifts z < 1. They find that it exceeds IGL predictions on scales 304 < l < 2,000 by a factor of 5 − 10, accounting for a substantial portion of the CIBER auto-power spectrum excess from earlier work when combined with estimates of ISL and DGL. The bounds from HSC are higher than from DESI-LS by roughly a factor of two in both CIBER bands.

On small angular scales (l > 10,000, θ ≲ 1′), the auto- and cross-spectra agree closely with IGL predictions, reflecting a model that is well-calibrated in the Poisson-dominated regime for both tracers. The small-scale cross-spectra have a sub-Poissonian slope relative to expectation, measuring DlIg ∝ l1.6 for CIBER × DESI-LS and ∝ l1.5 for CIBER × HSC.

The authors discuss two primary explanations for the excess: (1) a more luminous satellite population than baseline models assume, and (2) diffuse intra-halo light from tidal stripping. They note that the satellite and IHL scenarios are not mutually exclusive and are difficult to separate observationally. The results imply that a higher dI/dz is required to resolve existing discrepancies, placing estimates in direct tension with constraints from galaxy counts and γ-ray measurements but in closer agreement with absolute photometric measurements from HST, DIRBE, and CIBER low-resolution spectrometer.

These results identify low-redshift LSS as a significant and previously unappreciated contributor to near-IR EBL fluctuation measurements, setting the stage for cross-correlation science with CIBER-2, SPHEREx and a variety of LSS tracers. The authors note that fluctuation measurements targeting scales near l ∼ 1000, e.g., EoR searches using the Lyman-α line or Lyman break feature, require either substantially larger survey area or improved foreground mitigation to achieve their stated statistical precision.

Improvement 1: Redshift-Aware Foreground Subtraction for EoR and Line-Intensity Mapping

The AI system can now incorporate the measured low-redshift (z < 0.6) excess cross-power into its foreground models. When analyzing future data from SPHEREx, CIBER-2, or Lyman-α intensity mapping, the AI will automatically down-weight or subtract the z < 0.6 contribution using the provided redshift-dependent kernels (bI × dI/dz). This enables more accurate isolation of high-redshift signals (e.g., EoR) by predicting and removing the dominant low-z contamination, reducing false detections by up to 10σ on scales l < 2000.

Improvement 2: Halo-Model Calibration for Galaxy Evolution Simulations

The AI can now refine its sub-grid prescriptions for satellite galaxies and intra-halo light (IHL) in cosmological simulations. Using the measured one-halo and two-halo cross-power amplitudes (with >3σ and >5σ detections), the AI will adjust the mass–luminosity relation, satellite luminosity function, and tidal stripping efficiency to match the observed excess at z < 0.6. This yields more realistic mock catalogs for predicting NIR background fluctuations, improving the fidelity of synthetic sky maps used for survey planning and instrument calibration.

Improvement 3: Bias-Intensity Joint Inference for Galaxy Surveys

The AI can now jointly infer the galaxy bias and the redshift-dependent intensity kernel (dI/dz) from cross-correlation data, rather than assuming a fixed IGL model. By incorporating the four bias configurations (log Mh = 12, 13, 14, and tSZ-like), the AI will produce posterior distributions that account for the factor-of-four variation in bI × dI/dz. This allows the AI to flag discrepancies between photometric galaxy counts and NIR absolute photometry, and to propose new observational strategies (e.g., deeper surveys or narrower redshift bins) to resolve the tension.

Improvement 4: Coherence-Based Auto-Power Reconstruction for Background Decomposition

The AI can now use the coherence-based estimator (from the paper) to decompose total NIR auto-power into contributions from correlated LSS, intra-halo light, and diffuse Galactic light. This enables the AI to automatically separate astrophysical components in future CIBER-2 or SPHEREx data, providing a real-time pipeline that outputs redshift-resolved power spectra and identifies which component dominates at each angular scale (e.g., Poisson vs. clustering). This improves the accuracy of cosmological parameter extraction from background fluctuations.

Improvement 5: Sub-Poissonian Slope Modeling for Small-Scale Cross-Spectra

The AI can now model the observed sub-Poissonian slope (DlIg ∝ l 1.5–1.6) on scales l > 10,000. This allows the AI to correct for the non-standard scaling when predicting small-scale signals from galaxy catalogs, improving the calibration of point-source subtraction and the detection of faint, extended structures in future surveys. The AI will also use this slope to distinguish between satellite-dominated and IHL-dominated regimes, enabling more precise constraints on halo mass functions at low z.

We perform the first tomographic analysis of near-IR extragalactic background light (EBL) anisotropies, cross-correlating CIBER 1.1 and 1.8 mu m imager data with photometric galaxy catalogs from DESI Legacy Survey DR8 and Hyper-Suprime-Cam Ultra-Deep Survey. We measure significantly higher cross-power than expectations from an integrated galaxy light (IGL) model on scales < 2000, concentrated at low redshift (z 0.6). Cluster member galaxies and associated structure account for 15-20% of the large-angle cross-power, indicating that group- and galaxy-scale halos contribute the bulk of the signal. Through a parametric halo model decomposition, we detect two-halo and one-halo clustering in cross-power at high significance, with amplitudes that decline smoothly across z=0 - 1. The inferred one-halo cross-power is of similar amplitude between DESI-LS and the deeper HSC catalog, implying a scenario in which low-redshift EBL fluctuations are amplified by contributions from lower-mass halos with satellites and/or diffuse intra-halo light (IHL). Converting our two-halo fits into estimates of b I times dI/dz, we find that standard IGL predictions underestimate our measurements, even when assuming an intensity bias as high as 3, similar to that of large SZ clusters, suggesting that a higher dI/dz is required to reconcile observed discrepancies. Lastly, we find that correlated large-scale structure (LSS) at z<1 accounts for a substantial fraction of the CIBER auto-power reported in earlier work. These results identify low-redshift LSS as a significant and previously unappreciated contributor to near-IR EBL fluctuation measurements, setting the stage for cross-correlation science with CIBER-2, SPHEREx and a variety of LSS tracers.