Differentiable Stochastic HOD Modeling with Galaxy Intrinsic Alignments

Above: using gradients to update the galaxy field based on 1-pt statistics (galaxy number counts). In this case, the HOD parameterization was optimized towards producing a galaxy field with 500,000 galaxies.

This respository provides an implementation of differentiable halo-occuptation distribution (HOD) modeling that includes galaxy intrinsic alignment (IA) implementation. The code is written in JAX.

The HOD implementation is standard (see this paper), and the IA modeling formalism follows Van Alfen et al. 2023. The differentiable HOD implementation is largely inspired by Horowitz et al. 2022, with subtle changes, and the differentiable IA modeling implementation is novel.

Citation

@misc{pandya2026differentiablestochastichalooccupation,
      title={Differentiable Stochastic Halo Occupation Distribution with Galaxy Intrinsic Alignments}, 
      author={Sneh Pandya and Jonathan Blazek},
      year={2026},
      eprint={2602.04977},
      archivePrefix={arXiv},
      primaryClass={astro-ph.CO},
      url={https://arxiv.org/abs/2602.04977}, 
}