DerivKit

Reliable numerical derivatives and derivative-based forecasting for scientific computing.

DerivKit is a modular toolkit designed for workflows where numerical derivatives must be computed explicitly and robustly. It provides a common foundation for derivative-based methods, together with higher-level tools for calculus, forecasting, and likelihood analysis.

DerivKit is particularly suited to applications in physics, astronomy, and cosmology, where models are often noisy, computationally expensive, or non-smooth, and where automatic differentiation may be unavailable or inappropriate.

What does DerivKit provide?

DerivKit is organized into four interoperable layers:

• DerivativeKit for robust numerical differentiation

• CalculusKit for gradients, Jacobians, Hessians, and mixed partials

• ForecastKit for Fisher matrices, Fisher bias, Laplace, and DALI expansions

• LikelihoodKit for lightweight likelihood utilities that integrate with derivative-based workflows

Each layer can be used independently, while sharing a common derivative backend to ensure consistent numerical behavior throughout the library.

If you want to jump straight to usage, start with Examples.