This paper presents t-roo, a new sampler that can tell which type of compact object pair produced a gravitational-wave signal while it fits the source parameters. In plain terms, t-roo can jump between different models — for example, a binary black hole, a binary neutron star, or a neutron star–black hole system — and report both the odds between those models and the best parameter estimates in a single analysis.

The key idea is to use reversible jump Markov chain Monte Carlo (RJMCMC). RJMCMC is a version of the common MCMC (Markov chain Monte Carlo) method that can move between models with different numbers of parameters. That is important for these signals because models for neutron-star systems include extra physics, like tidal effects, that add parameters. By sampling over the model label as part of the inference, t-roo spends more time in the model that fits the data best and fewer steps in bad models, which can cut the total computing cost.

t-roo is built on the eryn sampler and uses utilities from Bilby, two existing tools for gravitational-wave inference. It implements ensemble sampling moves such as the stretch move and uses parallel tempering to help explore complicated likelihood shapes. The authors validated t-roo on simulated signals (injections) and found agreement with results from the nested sampler dynesty. They also applied t-roo to two real events, GW190425 and GW230529, where the nature of the components is uncertain and a neutron star could not be confirmed by parameters alone.

This approach matters because standard practice runs separate analyses for each model and then compares the model evidences. A single transdimensional run can reduce manual setup and may save substantial computing time, especially when many models are compared or when data are highly informative. The authors also highlight that t-roo could be useful for next-generation detectors, where running many separate analyses will become even more costly.