Efficient, reliable, and well-tested derivatives for scientific forecasting.
DerivKit provides a unified framework for numerical derivatives, Fisher/DALI expansions, and model forecasting — designed for cosmology, but general-purpose across scientific domains.
DerivKit grew out of practical needs in cosmological inference, combining flexible derivative estimators with rigorous error control and clean, modern APIs. While it was developed with cosmology in mind, DerivKit is fully problem-agnostic: it can be used for numerical derivatives, inference, and forecasting across a wide range of scientific and technical applications.
If you use DerivKit in your research, please cite:
@misc{sarcevic2026derivkitstablenumericalderivatives,
title = {DerivKit: stable numerical derivatives bridging Fisher forecasts and MCMC},
author = {Šarčević, Nikolina and van der Wild, Matthijs and Trendafilova, Cynthia},
year = {2026},
eprint = {2602.08078},
archivePrefix = {arXiv},
primaryClass = {astro-ph.IM},
url = {https://arxiv.org/abs/2602.08078}
}|
Core Python package for numerical derivatives, Fisher forecasting, DALI expansions, and scientific inference workflows. |
Tutorials, worked examples, and practical demonstrations for learning and testing DerivKit in real workflows. |
DerivKit is released under the MIT License and actively maintained by
@nikosarcevic, @lonbar and the DerivKit team.
