Differentiable and GPU ready phase-space mappings

This repository contains a refactored version of the code used in our publication Differentiable MadNIS-Lite.

Supported mappings

TorchSpace contains and supports a selection of different phase-space mappings which are useful for collider physics:

• Chili [https://arxiv.org/abs/2302.10449]
• Rambo [Comput. Phys. Commun. 40 (1986) 359-373]
• Rambo on diet [https://arxiv.org/abs/1308.2922]
• Mahambo [https://arxiv.org/abs/2305.07696]
• Diagram based [see this paper]
• 1 $\to$ 3 block [https://freidok.uni-freiburg.de/data/154629]
• including trainable variants of the above

Installation

# clone the repository
git clone https://github.com/madgraph-ml/torchspace.git
# then simply install (editable if needed with flag "-e")
cd torchspace
pip install .

Usage example

For an example usage of, for instance Mahambo, see examples/rambo/rambo_example.py

Citation

If you use this code or parts of it, please cite:

@article{Heimel:2024wph,
author = "Heimel, Theo and Mattelaer, Olivier and Plehn, Tilman and Winterhalder, Ramon",
title = "{Differentiable MadNIS-Lite}",
eprint = "2408.01486",
archivePrefix = "arXiv",
primaryClass = "hep-ph",
reportNumber = "IRMP-CP3-24-23",
doi = "10.21468/SciPostPhys.18.1.017",
journal = "SciPost Phys.",
volume = "18",
number = "1",
pages = "017",
year = "2025"}