Estimating the Hubble constant with a federated unscented Kalman filter

The Hubble tension, the persistent discrepancy between early-universe predictions and late-time measurements of the expansion rate (H0), remains one of the confounding puzzles in modern cosmology. Addressing this challenge demands reliable statistical frameworks capable of synthesizing diverse observational probes while accounting for complex model non-linearities. Here, we develop a federated unscented Kalman filter (FUKF) framework to combine Type Ia supernova (SNe Ia), cosmic chronometer (CC), and baryon acoustic oscillation (BAO) data within flat and non-flat ΛCDM models. Unlike Markov chain Monte Carlo (MCMC), the FUKF is a recursive estimator rather than a full posterior sampler. We report H0 = 68.18 ± 0.90 km s−1 Mpc−1 for the flat model and H0 = 66.80 ± 1.43 km s−1 Mpc−1 for the non-flat case. These values are lower than the SH0ES distance-ladder estimate and closer to the Planck 2018 ΛCDM inference. The comparison with MCMC indicates that the FUKF recovers the main posterior region at substantially lower computational cost and remains robust in the presence of parameter degeneracies.