Skip to content

Welcome to the QuanEstimation

QuanEstimation is an open-source Python–Julia toolkit for quantum parameter estimation. It provides calculations of quantum metrological tools and quantum resources, optimizations over probe states, controls, and measurements, and comprehensive optimizations in quantum metrology. QuanEstimation can generate both optimal quantum parameter estimation schemes and adaptive measurement schemes. The package structure of QuanEstimation is shown in the following figure:

Image title

The package structure of QuanEstimation. The blue and light blue boxes represent folders and Python files. The orange and gray boxes represent classes and functions. Functions circled by dotted lines represent Julia methods wrapped by the Python interface and evaluated in Julia.

The package contains several widely used metrological tools, such as quantum (classical) Cramér-Rao bounds, the Holevo Cramér-Rao bound and Bayesian versions of quantum (classical) Cramér-Rao bounds, the quantum Ziv-Zakai bound, and Bayesian estimation. Users can use these bounds as objective functions to optimize the probe state, control, and measurement, as well as to perform simultaneous optimizations over them. The optimization methods include gradient-based algorithms such as the gradient ascent pulse engineering (GRAPE), GRAPE based on automatic differentiation (auto-GRAPE), and automatic differentiation (AD), as well as gradient-free algorithms such as particle swarm optimization (PSO), differential evolution (DE), and the reverse iterative (RI) algorithm.

The Python interface wraps the Julia core: most numerical work is executed in Julia for computational efficiency. The same core is also available through the native Julia interface, QuanEstimation.jl.

Julia-first / thin-shell direction

The long-term architecture is Julia-first: numerical kernels live in QuanEstimationBase, and Python remains a thin layer for input validation, data conversion, and convenience APIs. The core Fisher-information kernels (QFIM/CFIM/SLD/RLD/LLD/FIM) are delegated to Julia in 0.5; specialized kernels and the optimization wrappers will follow. New numerical features should be implemented in Julia first. This keeps the Python and Julia results on the same numerical implementation and avoids duplicating numerical policy in Python.