Measurement of the Hubble constant using the Dark Energy Survey Year 6 Gold galaxy catalog and the fourth Gravitational-Wave Transient Catalog 2026 Measurement of the Hubble constant using the Dark Energy Survey Year 6 Gold galaxy catalog and the fourth Gravitational-Wave Transient Catalog – 3 Isaac McMahon, Danny Laghi, Marcelle Soares-Santos, Kendall Ackley, Gergely Dálya, Yavuz Gençel, David Sánchez-Cid, Felipe Andrade-Oliveira, Sean MacBride, Christian Chapman-Bird, Rachel Gray, Alexander Papadopoulos Abstract Gravitational wave (GW) standard sirens enable independent measurements of the Hubble constant H 0 H_{0} . In the absence of electromagnetic counterparts, the "dark siren" method statistically correlates GW events with potential host galaxies. We present a measurement of H 0 H_{0} using 142 compact binary coalescences from the fourth Gravitational-Wave Transient Catalog (GWTC-4.0) combined with the Dark Energy Survey Year 6 Gold photometric galaxy catalog. Using the gwcosmo pipeline, we jointly infer cosmological and GW population parameters. We analyze the impact of galaxy catalog properties on the inference, identifying significant features in the galaxy redshift distribution which can introduce biases. By restricting the galaxy catalog to 0.05 < z < 0.35 0.050.05 in the K -band, too shallow to provide significant information at typical GW distances. Other analyses with the DESI ( 25 ) , DESI Legacy Survey ( 85 ) , and DELVE ( 30 ) consider only a small selection of available GW events and assume fixed mass distribution and merger rate parameters. In this work, we adapt the DES Year 6 Gold photometric galaxy catalog ( 26 ) for the dark siren method and obtain an H_{0} measurement with galaxies out to z=0.5 (this limitation is discussed in Section 3.5 ). We explain our choices to make the galaxy catalog data cosmology-ready with object selection and classification (Sections 2.2 & 3.3 ). We measure the luminosity function of DES galaxies out to z=0.5 with photometric redshift to quantify the completeness of our galaxy sample for dark sirens (Section 3.4 ). We describe our choices for computing the LOS ( LOS ) redshift prior (Section 3.2 ). We present our computed LOS redshift priors and describe the features in the redshift distribution that we observe (Section 3.7 ). Finally, we show our measurement for H_{0} and discuss potential sources of bias which arise from making different choices in the analysis (Section 4 ). If an assumed reference cosmology is required, we adopt best-fit values from Planck 2015 ( 16 ) unless otherwise stated, as these are the defaults used in our chosen software (see Section 3.1 ). 2 Data 2.1 Gravitational-wave events The LVK network of observatories ( 2 ) consists of two LIGO ( 1 , LIGO ;) detectors in the USA and the Virgo Observatory ( 15 ) in Italy, as well as the KAGRA Observatory ( 22 ) in Japan. The GWTC-4.0 ( GWTC-4.0 ) ( 6 ; 4 ; 2 ) , consisting of approximately seven months of data from 24/05/2023 to 16/01/2024, is the latest set of GW candidate data released by LVK . When combined with the previous GWTC-2.1 and GWTC-3.0 ( 10 ; 11 ) data releases, they consist of all GW candidate data released by LVK to date. We use a candidate list which is consistent with the O4a LVK cosmology analysis ( 3 ) , totaling 142 CBC GW candidates. 137 of these candidates are believed to have an origin in a binary black hole collision, and the remaining 5 candidates could have a neutron star component. The event GW170817, one of these 5, is the only event with a confirmed electromagnetic counterpart ( 37 ; 98 ) , and is thus treated separately in this analysis as a bright siren. The GW information which is used as an input for our analysis comes in the form of parameter estimation posterior samples and HEALPix skymaps ( 54 ) . The posterior samples encode information obtained only from the GW signal itself, including location, mass, spin, and other parameters which affect the gravitational waveform. The skymap is created from these samples and encodes posterior probability information both on the projected celestial sphere as well as luminosity distance in the radial direction ( 97 ) . For all events, we use the publicly available data released on Zenodo (relevant links found in Section Data Availability ). In addition to the real GW candidate data, we require simulated GW injections to model GW selection effects due to detector sensitivity ( 45 ) . We use the injections with simulated detector sensitivity up to O4a which were also released with GWTC-4.0 . 2.2 Galaxy catalog The Dark Energy Survey ( 103 ; 12 ) was conducted using the Dark Energy Camera (DECam) at the 4-m Blanco telescope at Cerro Tololo Inter-American Observatory in Chile. The Year 6 Gold data release ( 26 ) is curated for cosmology analysis, covering 4923 deg 2 in the southern sky and achieving a photometric uniformity of <2 mmag in the griz photometric filters over a total of 669 million objects. The 10\sigma multi-epoch galaxy magnitude depth in r -band is 23.9. Included in the catalog are photometric redshift (photo-z) measurements from Directional Neighborhood Fitting (DNF) ( 40 ) , with state-of-the-art average uncertainties \sigma_{z}/(1+z)<3\% for z<1 . Using the r -band luminosity function (see Section 3.4 ) and the r -band apparent magnitude limit, we estimate that DES is approximately 100% complete to z\sim 0.5 for an absolute magnitude limit M_{r}<-17.5 , and complete to z\sim 0.4 for a limit M_{r}<-17 ( 51 ) . The data release includes three different models for photometric flux measurements: PSF (Point Spread Function), BDF (Bulge+Disk Fixed), and GAp (Gaussian Aperture). In this analysis we choose BDF as the representative photometry because the model performs better for galaxies. Specifically, we use the BDF_MAG_*_CORRECTED columns which have been normalized to the global MAG_APER_8 photometric calibration system and dereddened for Galactic extinction. We apply the following recommended cuts based on quality flags for each source: • Non-NaN redshift and photometry in the griz filters • \texttt{FLAGS\_FOOTPRINT}==1 • \texttt{FLAGS\_FOREGROUND}==0 • \texttt{FLAGS\_GOLD}==0 FLAGS_FOOTPRINT is a flag that equals 1 for all sources which have data from at least 2 exposures in each of the griz bands, defined by a NSIDE=16384 HEALPix map of exposure areas. A cut of \texttt{FLAGS\_FOOTPRINT}==1 excludes 3.3% of the full catalog. FLAGS_FOREGROUND is a flag for sources that are impacted by the presence of nearby bright astrophysical foreground objects. The minimum area masked by the foreground mask is the area of one HEALPix NSIDE=4096 pixel, 0.86 arcmin 2 . A cut of \texttt{FLAGS\_FOREGROUND}==0 excludes 9.7% of the full catalog. FLAGS_GOLD is a flag for photometric processing failures, objects with unphysical colors, noisy detections, and other issues. A cut of \texttt{FLAGS\_GOLD}==0 excludes 8.1% of the full catalog. In total, these quality cuts combine to exclude 19.1% of the sources in the full catalog. Out of all 141 GW event candidates used in our analysis as dark sirens, 14 candidates (10%) have more than 50% probability to be in the DES footprint (see Figure 1 , upper panel). These candidates are well-distributed among the redshifts to which DES is most sensitive. A total of \sim\!14\% of the probability of all 141 candidates falls within the DES footprint. Although most of the candidates fall outside of the DES survey footprint or redshift depth, we still include them in our analysis as they provide crucial cosmology information mainly through the shape of the GW mass distribution. Figure 1: Top : GW event candidate 50% and 90% probability contours included in our analysis which have more than 50% probability to be within the DES Y6 Gold survey footprint (shown in blue). Bottom : The photometric redshift distribution of galaxies \text{d}N/\text{d}z in DES Y6 Gold (red) compared to a uniform in comoving volume distribution (dotted black), scaled to have the same total number of galaxies. More discussion can be found in Section 3.7 . On the same x-axis are shown redshift measurements (90% confidence interval) for the events, converted from luminosity distance and ordered by increasing localization volume. 3 Methodology 3.1 Analysis framework We jointly infer cosmology and CBC population parameters using a Bayesian framework as implemented in the gwcosmo python package ( 55 ; 56 ) . Given a set of GW event data \{d\} from a number N_{\text{det}} of detections, \theta^{\,\text{det}} are the parameters of the detected CBC measured only from the GW data in the detector frame of reference. The joint posterior probability on CBC population parameters \Lambda and cosmological parameters \Lambda_{\text{c}} given a set of GW events (individually labeled with i ) can thus be written as: \begin{split}&p\left(\Lambda,\Lambda_{\text{c}}|\{d\},N_{\text{det}}\right)\propto\pi(\Lambda)\,\pi(\Lambda_{\text{c}})\,\xi(\Lambda,\Lambda_{\text{c}})^{-N_{\text{det}}}\\ &\;\;\times\prod_{i=1}^{N_{\text{det}}}\int\!\!\text{d}\theta^{\text{\,det}}_{i}\,\frac{p\big({\theta}^{\text{\,det}}_{i}|d_{i}\big)}{\pi(\theta^{\text{\,det}}_{i})}\times\Big|\frac{\text{d}\theta^{\text{\,det}}_{i}(\theta^{\text{\,src}}_{i},\Lambda_{\text{c}})}{\text{d}\theta^{\text{\,src}}_{i}}\Big|^{-1}p(\theta^{\text{\,src}}_{i}|\Lambda)\end{split} (1) where \pi denotes a prior and the GW source-frame parameters \theta^{\;\text{src}} are given by a function of the detector-frame parameters and the cosmological parameters, \theta^{\text{\,src}}=\theta\,(\theta^{\text{\,det}},\Lambda_{\text{c}}) . The term \xi(\Lambda,\Lambda_{\text{c}}) is the expected fraction of GW events which are detected, required because the GW detectors will not be able to measure the entire true population of GW events due to some selection effects. It can be written as: \begin{split}\xi({\Lambda,\Lambda_{\text{c}}})=&\int\!\text{d}\theta^{\text{\,det}}\,P(\text{det}|\theta^{\text{\,det}})\,\times\Big|\frac{\text{d}\theta^{\text{\,det}}(\theta^{\text{\,src}},\Lambda_{\text{c}})}{\text{d}\theta^{\text{\,src}}}\Big|^{-1}p(\theta^{\text{\,src}}|\Lambda)\end{split} (2) where P(\text{det}|\theta^{\,\text{det}}) is the probability of detection given the measured GW parameters \theta^{\,\text{det}} , dependent on the sensitivity of the detectors. The expected true underlying distribution of events p(\theta^{\,\text{src}}|\Lambda) depends on the chosen model for the population of CBC mergers and its parameters \Lambda . 3.2 Line-of-sight redshift prior In Equation 1 , the probability of a set of GW parameters \theta_{i}^{\,\text{src}} given \Lambda depends on the distribution of potential host galaxies: p(\theta_{i}^{\,\text{src}}|\Lambda)\propto\frac{\text{d}N^{\,\text{eff}}_{\text{gal,cat}}}{\text{d}z\text{d}\Omega}+\frac{\text{d}N^{\,\text{eff}}_{\text{gal,out}}}{\text{d}z\text{d}\Omega} (3) The first term \text{d}N^{\,\text{eff}}_{\text{gal,cat}}/\text{d}z\text{d}\Omega represents the spatial distribution of galaxies in a chosen galaxy catalog across redshift z and angular sky position \Omega . Each galaxy is represented with a gaussian likelihood defined by its photometric redshift z and the redshift error \sigma_{z} at its sky location \Omega . While photometric redshift uncertainties are not truly gaussian in general ( 86 ; 85 ; 104 ; 30 ) , we use this approximation in this work. The gaussian likelihood of each galaxy is assigned a host probability weight w_{j} as defined by: w_{j}\,(\epsilon,M_{j})=10^{-0.4\,\epsilon\,(M_{j}-M^{*})} (4) where M^{*} is the reference absolute magnitude of the Schechter luminosity function, described in Section 3.4 . The parameter \epsilon is chosen to change how the galaxies are weighted with respect to each other. When \epsilon=0 , all galaxies are weighted equally as potential host galaxies for the GW event. When \epsilon=1 , this corresponds to a logarithmic weight in absolute magnitude, or equivalently a linear weight in luminosity. The extent to which the probability of a galaxy being a host of a CBC merger may scale with its observables is unknown and is an active field of research ( 82 ; 17 ; 93 ; 32 ; 90 ; 100 ; 59 ; 68 ) . As host probabilities for stellar-origin CBC is expected to scale with the stellar mass of the galaxy, and galaxy stellar mass roughly scales with luminosity ( 76 ) , host probabilities are expected to also scale with luminosity. Most dark siren analyses to date consider only the cases \epsilon=0 or \epsilon=1 , with \epsilon=1 being the preferred fiducial analysis. In this work we choose \epsilon=1 . The second term \text{d}N^{\,\text{eff}}_{\text{gal,out}}/\text{d}z\text{d}\Omega , known as the "empty catalog" case, represents galaxies which are assumed to exist but are not present in the catalog due to selection effects. This term is modeled by assuming that all galaxies which are not detected by the catalog follow an isotropic distribution which is uniform in comoving volume. While this does not account for galaxy clustering, we expect that this model adequately describes the real average galaxy distribution over large scales. In gwcosmo , these terms are passed to the likelihood as a "Line-of-Sight (LOS) redshift prior" which encodes the sum of the in-catalog and out-of-catalog terms as a probability distribution in redshift for every pixel in a HEALPix map of chosen resolution ( 55 ) . For each pixel, only the cataloged galaxies in that pixel contribute to the probability distribution. This LOS redshift prior is unique for each galaxy catalog, and so is the focus of this work. We compute the LOS redshift prior for DES Year 6 Gold using gwcosmo and use it to obtain the posterior probability distributions on the population parameters \Lambda and cosmological parameters \Lambda_{\text{c}} . The spacing of the redshift array used in the calculation of the redshift prior and the dark siren likelihood must be chosen to allow for sufficient resolution to capture information from every galaxy considering its redshift error. The spacing between consecutive elements of the array must be smaller than the smallest redshift errors in the catalog at that redshift. We adopt a redshift array that satisfies these conditions for the properties of the DES catalog, with a linear spacing of 4.4\times 10^{-4} for 0z_{\text{max}} . We compute the redshift prior for several different values of z>z_{\text{max}} to determine the impact of the galaxy distribution on the resulting posterior distribution. We further discuss the features in the galaxy redshift distribution in Section 3.7 . Additionally, objects begin to saturate in the DECam sensors at apparent magnitudes of m_{r}<16 , meaning that objects which appear brighter than this are not cataloged in DES Y6 Gold. This saturation effect disproportionately affects galaxies at redshifts z<0.05 . For this reason we also apply a redshift cut on the low end of the LOS redshift prior at z_{\text{min}}=0.05 , excluding all in-catalog contribution to the redshift prior below this redshift. For all redshifts outside of the interval defined by the redshift cuts, z_{\text{min}}z_{\text{max}} , the prior reverts to the default uniform in comoving volume "Empty" catalog assumption. The y-axis is the ratio between the catalog prior and the empty catalog. The chosen values of z_{\text{max}} and z_{\text{min}} are shown as blue vertical lines, as in Figure 1 . The average LOS redshift prior for GLADE+ K -band is shown in green for comparison. The effect of applying a cut at z_{\text{min}} is shown for z_{\text{max}}=0.35 with a dashed line at low redshift. If there is no cut, the prior dramatically decreases at low redshift due to the saturation selection effect. Bottom : 1\sigma average redshift prior comparison for z_{\text{max}}=0.35,0.50 . For z_{\text{max}}\leq 0.35 , each prior is entirely within 1\sigma of the empty catalog. All z_{\text{max}}\geq 0.40 exceed 1\sigma at some point. The overdensity of galaxies in the DES catalog at 0.350.5 . Furthermore, we further impose a galaxy cut to 0.05