Skip to main content

Python implementation of Krotov's method for quantum optimal control

Project description

Krotov Python Package

Source code on Github Krotov on the Python Package Index Join the chat at Travis Continuous Integration AppVeyor Continuous Integration Codecov BSD License Documentation Status Launch Binder arXiv

Python implementation of Krotov’s method for quantum optimal control.

This implementation follows the original implementation in the QDYN Fortran library. The method is described in detail in D. M. Reich, M. Ndong, and C. P. Koch, J. Chem. Phys. 136, 104103 (2012) (arXiv:1008.5126).

The krotov package is built on top of QuTiP.

Development happens on Github. You can read the full documentation at ReadTheDocs.

If you use the krotov package in your research, please cite it.


Optimal control is a cornerstones of quantum technology: relying not just on a passive understanding of quantum mechanics, but on the active utilization of the quantum properties of matter. Quantum optimal control asks how to manipulate the dynamics of a quantum system to behave in some desired way. This is essential for the realization of quantum computers, and related technologies such as quantum sensing. See e.g. Glaser et al. Eur. Phys. J. D 69, 279 (2015) for an overview of methods, applications, and current research directions. Quantum technology and thus quantum control are the focus of several large-scale national and super-national research endeavors, such as the U.S. $1 billion National Quantum Initiative Act and the €1 billion European Quantum Flagship program.

Krotov’s method is one of the two leading gradient-based optimization algorithms used in numerical quantum optimal control. It simulates the dynamics of a quantum system under a set of initial controls, and evaluates the result with respect to an optimization functional to be minimized. It then iteratively modifies the controls to guarantee a monotonically decreasing value in the optimization functional. To date, there has not been an open source implementation of the method. This package provides that missing implementation.

The choice of Python as an implementation language is due to Python’s easy to learn syntax, expressiveness, and immense popularity in the scientific community. Moreover, the QuTiP library exists to provide the foundations of numerically describing quantum systems, and already includes basic versions of some of the other popular algorithms in quantum control, the gradient-based GRAPE and the gradient-free CRAB. The availability of the Jupyter notebook framework provides an ideal platform for showing the use of the method.

The Krotov package targets both students wishing to enter the field of quantum control, and researchers in the field. By providing a comprehensive set of examples, we enable users of our package to explore the formulation of typical control problems, and to understand how Krotov’s method can solve them. These examples are inspired by recent publications, and thus show the use of the method at the cutting edge of research. Optimal control is also increasingly important in the design of experiments, and we hope that the availability of an easy to use implementation of Krotov’s will facilitate this further.

The use of Python implies that for large-scale control problems, performance may become a significant issue. In this case, it may be necessary to implement Krotov’s method in a more efficient (compiled) language. While the method as such is relatively straightforward, there are some subtleties involved. Our implementation puts an emphasis on clarity, and the documentation provides detailed explanations of all necessary concepts. Thus, the Krotov package can serve as a reference implementation, leveraging Python’s reputation as “executable pseudocode”, and as a foundation against which to test other implementations.


The Krotov package is available for Python versions >= 3.5. Its main dependency is QuTiP (apart from the core packages of the Python scientific ecosystem). Thus, you should consider QuTiP’s installation instructions.

In any case, using some sort of virtual environment is strongly encouraged. Most packages in the Python scientific ecosystem are now available as wheels, making installation via pip easy. However, QuTiP currently does not provide wheels. Thus, on systems that do not have the necessary compilers installed (Windows, macOS), the conda package manager provides a good solution.

Assuming conda is installed (e.g. through Miniconda), the following commands set up a virtual (conda) environment into which the Krotov package can then be installed:

$ conda create -n qucontrolenv python=3.6
$ conda activate qucontrolenv
$ conda config --append channels conda-forge
$ conda install qutip


To install the latest released version of krotov into your current (conda) environment, run this command in your terminal:

$ pip install krotov

This is the preferred method to install the krotov package, as it will always install the most recent stable release.

You may also do

$ pip install krotov[dev,extras]

to install additional development dependencies, including packages required to run the example notebooks.

If you don’t have pip installed, the Python installation guide, respectively the Python Packaging User Guide can guide you through the process.

To install the latest development version of krotov from Github:

$ pip install git+


To use Krotov’s method for quantum optimal control in a Python script or Jupyter notebook, start with:

import krotov


  • define the necessary quantum operators and states using QuTiP.
  • create a list of objectives, as instances of krotov.Objective
  • call krotov.optimize_pulses to perform an optimization of an arbitrary number of control fields over all the objectives.

See Using Krotov with QuTiP and Examples for details.


0.4.0 (2019-10-08)

  • Added: Support for Python 3.7
  • Changed: The 'shape' key in pulse_options was renamed to 'update_shape', to further avoid confusion between pulse shapes and update shapes.
  • Changed: The .adjoint property of Objective is now a method
  • Added: Ability to not use QuTiP Qobj objects, but arbitrary low-level objects instead.
  • Improved: Printing an Objective now uses internal counters and a symbolic notation to identify objects shared between different objectives. (#43)
  • Improved: gate_objectives now takes into account if target states are (reshuffled) basis states and does not create unnecessary new copies.
  • Bugfix: Two Objective instances that contain numpy arrays as controls can now be compared with == (#44)
  • Bugfix: Custom attributes (such as weight) are now preserved when copying an Objective (#44)
  • Bugfix: Calling copy.deepcopy on an Objective now preserves control functions (#44)
  • Improved: The Objective.mesolve and Objective.propagate methods can now receive arguments H and c_ops to override the respective attributes of the objectives. This make is easier to analyze perform a robustness analysis, where the result of an optimization should be propagated under a perturbed Hamiltonian.
  • Improved: The print_table and print_debug_information info-hooks now flush their output buffers after each iteration. As a result, when writing to a file, that file can be watched with tail -f.
  • Changed: Redefine tau_vals as their complex conjugate, fixing a bug in chis_ss and chis_sm (#46)
  • Bugfix: Correctly calculate ∂H/∂ϵ if ϵ occurs in H multiple times (#47, thanks to @uiofgh)
  • Bugfix: Correctly calculate ∂H/∂ϵ=0 if the specific ϵ currently being updated does not occur in H (#48)
  • Added: Method objectives_with_controls for Result object.

0.3.0 (2019-03-01)

  • Added: Preprint citation information (krotov.__arxiv__, krotov.__citation__, krotov.__bibtex__)
  • Added: Ability to continue from a previous optimization (#26)
  • Added: Parameter out to print_table info-hook
  • Added: Parameter finalize to Result.load
  • Added: Ability to dump optimization result every so many iterations (dump_result check-convergence routine)
  • Added: re-entrant option for DensityMatrixODEPropagator
  • Bugfix: Discretize controls to float values (#41)
  • Bugfix: Fix overlap for non-Hermitian operators (#39)
  • Bugfix: Interface for passing tau_vals to chi_constructor (#36)
  • Added: function above_value for convergence check (#35)

0.2.0 (2019-02-14)

  • Added: Implementation of all the standard functionals
  • Added: The info_hook receives additional information, including ∫gₐ(t)dt (#32)
  • Added: Initialization of objectives for gate optimization in Liouville space
  • Added: A new propagator DensityMatrixODEPropagator for faster density matrix propagation
  • Added: Support for “stateful” propagators by subclassing from krotov.propagators.Propagator
  • Changed: more flexibility for parallelization (#29)
  • Added: Support for the second-order pulse update
  • Changed: The options for the controls (λₐ, update-shape) are now passed through a simplified dict interface, instead of a custom PulseOptions class.

0.1.0 (2018-12-24)

  • Initial release with complete implementation of first-order Krotov’s method
  • Support for state-to-state and gate optimization, for both closed and open systems

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for krotov, version 0.4.0
Filename, size File type Python version Upload date Hashes
Filename, size krotov-0.4.0-py2.py3-none-any.whl (69.1 kB) File type Wheel Python version py2.py3 Upload date Hashes View hashes
Filename, size krotov-0.4.0.tar.gz (3.7 MB) File type Source Python version None Upload date Hashes View hashes

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN SignalFx SignalFx Supporter DigiCert DigiCert EV certificate StatusPage StatusPage Status page