Speaker
Description
Stellar streams are among the most sensitive probes of dark matter substructure on small scales, where the microphysical nature of dark matter could leave distinct signatures. When a stream is subjected to many encounters with low-mass substructure over its lifetime, the cumulative effect is well described by a diffusion regime in which velocity kicks accumulate as a random walk. We present a JAX-based, fully differentiable forward model that operates in this regime. Rather than resolving individual substructure encounters, an approach that becomes prohibitively expensive at low perturber masses, our simulator models the collective statistical effect of the entire substructure population in the diffusion limit via the velocity injection formalism, while retaining an accurate numerical treatment of orbital dynamics and stream formation. The only input characterizing the perturbing environment is the power spectrum of the substructure density field, which can be computed for any dark matter model, including scenarios such as fuzzy dark matter where a description in terms of discrete halos does not apply, as well as baryonic contributions. We validate the framework against analytical predictions where available. Exploiting the differentiability of the model, we compute Fisher forecasts for the subhalo mass function parameters using density and velocity power spectra of a GD-1–like stream, quantifying the expected sensitivity of current and upcoming data.