JEGJonas El Gammal
Open-source PTA inference

discoverysamplers

A lightweight interoperability layer connecting JAX-enabled Discovery pulsar-timing-array models to several complementary Bayesian inference backends.

Why a bridge package?

Sampler libraries make different assumptions about likelihood signatures, parameter ordering, prior transforms, output formats, and accelerator support. Those interface differences make it unnecessarily difficult to compare inference strategies on the same PTA model. discoverysamplers keeps the scientific model and prior specification fixed while adapting them to each backend.

Supported backends

ErynEnsemble MCMC, parallel tempering, and reversible-jump support
nessaiFlow-based nested and importance nested sampling
JAX-NSPure-JAX nested sampling with accelerator support
GPryActive-learning Gaussian-process posterior emulation

A consistent model boundary

The package provides a unified prior format and handles the transformations between Discovery-style callables and the APIs expected by the samplers. This makes backend comparisons less vulnerable to accidental differences in parameter conventions or model wiring.

Current scope

The package is intended for researchers using Discovery who want to benchmark or exchange inference engines without rewriting the surrounding analysis. Further end-to-end JAX integration—including gradient-based samplers—is an active direction.