With the release of the largest Gravitational Wave (GW) catalog so far — the LIGO-Virgo-KAGRA collaboration’s GWTC-4.0, containing more than 200 candidates — the race to measure the Hubble constant with dark sirens has accelerated.
But as our measurements of the universe’s expansion improve, new questions emerge: How much do the data themselves drive the results? How robust are our inferences? And to what extent does the choice of cosmological model shape what we conclude?
In our recent work, we take the dark siren approach one step further by moving beyond the usual ΛCDM (Dark Energy and Dark Matter dominated universe) assumption. Instead, we introduce a fully non-parametric method that lets the data speak for themselves. Using 137 binary BH events from GWTC-4.0, we directly reconstruct the cosmic expansion history, the Hubble parameter H(z).
This work provides the first non-parametric GW-based reconstruction of the Hubble parameter, but it also highlights something deeper: the cosmological model we assume shapes the inferred expansion rate. To make this clearer, we combine our reconstruction with independent measurements from baryon acoustic oscillations in DESI DR2. These external “anchors” help us identify where GW data alone are driving the inference, and where model choices matter most.
Together with this publication, we also release Gsirens a version of icarogw written in Jax and numpyro to allow for the possibility of sampling hundreds of population parameters.
Figure: Inferred H(z) from the 137 GW sources of the GWTC-4.0 catalog, using our non-parametric approach (left) and the standard ΛCDM model (right).
