Release 0.21.0

New features

• Adds the Takagi decomposition (#363)

• Adds the Montrealer and Loop Montrealer functions (#363).

Improvements

• Tighten power-trace bound of odd loop Hafnian. (#362)

• Simplifies the internal working of Bloch-Messiah decomposition (#363).

• Simplifies the internal working of Williamson decomposition (#366).

• Improves the handling of an edge case in Takagi (#373).

• Adds extra tests for the Takagi decomposition (#377)

Contributors

This release contains contributions from (in alphabetical order):

Yanic Cardin (@yaniccd), Gregory Morse (@GregoryMorse), Nicolas Quesada (@nquesada)

Release 0.20.0

New features

• Implementation of gaussian boson sampling and gaussian boson sampling with threshold detectors. (#343)
• New function to produce Bloch-Messiah decomposition of symplectic matrices. (#352)

Improvements

• Added function to extend single mode symplectic to act on multiple modes. (#347)
• Added function to compute grouped (total) click probabilities for GBS setups using threshold detectors. The function uses the positive P-distribution simulation method of Drummond et al.. (#348)

Bug fixes

• Remove redundant call of Qmat, Amat from generate_hafnian_sample. (#343)

Documentation

• The centralized Xanadu Sphinx Theme is now used to style the Sphinx documentation. (#341)

Contributors

This release contains contributions from (in alphabetical order):

Mikhail Andrenkov (@Mandrenkov), Sebastián Duque (@sduquemesa), Jacob Hastrup (@JacobHast), Antonín Hoskovec (@thonic), Martin Houde (@MHoude2), Benjamin Lanthier (@benjaminlanthier), Dominic Leclerc (@dleclerc33), Filippo Miatto (@ziofil), Will McCutcheon, Brandon Turcotte (@brandonpolymtl), Jiaqi Zhao (@JQZ1111)

Release 0.19.0

New features

• New functions for calculating properties of distinguishable squeezed states of light having passed through an interferometer. #326

• New function ltor is added which allows threshold_detector_prob to act more consistently on displaced and zero-mean Gaussian states. #317

• New functions for threshold detection probabilities of Fock states, the Bristolian (brs) and the Unitary Bristolian (ubrs). #316

• Entanglement measures entanglement_entropy and log_negativity for bipartite Gaussian states are added to the quantum submodule. #332

• New functions, recursive_hafnian and solve added in the _hafnian module. #325

• New function to check if a matrix is symplectic is_symplectic. #334.

• Adds support for Python 3.10. #337

Improvements

• Update methods for calculating threshold detector probabilities of Gaussian states, now using ltor function within threshold_detection_prob #317

• numba_tor now can benefit from numba parallelization #317

• Recursive Torontonian added for faster computation based on paper "Polynomial speedup in Torontonian calculation by a scalable recursive algorithm" by Ágoston Kaposi, Zoltán Kolarovszki, Tamás Kozsik, Zoltán Zimborás, and Péter Rakyta. #321

• Recursive Loop Torontonian added for faster computation based on combining recursive Torontonian improvement and new loop Torontonian feature. #332

• Hafnians of odd-sized matrices are calculated roughly twice as fast. #329

• The new Hafnian functions now use the Labudde method to calculate power traces instead of using diagonalization. #333

Bug fixes

• Permanent algorithms handle 0x0 cases correctly. #320

Contributors

This release contains contributions from (in alphabetical order):

Jake Bulmer, Luke Helt, Martin Houde, Theodor Isacsson, Benjamin Lanthier, Fabian Laudenbach, Dominic Leclerc, Gregory Morse, Nicolas Quesada, Brandon Turcotte, Jiaqi Zhao

Release 0.18.0

New features

• Python module for the La Budde method of computing characteristic polynomials. #304

Improvements

• Permanent algorithms are implemented in Python using Numba just-in-time compilation. #300

• Hafnian algorithms are implemented in Python using Numba just-in-time compilation. #311

• Documentation is updated to include the characteristic polynomials and decompositions modules. #312

Bug fixes

• Makes modules reachable via the global namespace, instead of requiring importing the modules explicitly. #312

  import thewalrus as tw
  tw.samples.generate_torontonian_sample

Breaking Changes

• The Walrus is no longer dependent on C++, and all C++-related code and documentation is removed. Instead, all code has been ported to Python using just-in-time compilation to improve performance. #311

Contributors

This release contains contributions from (in alphabetical order):

Theodor Isacsson, Benjamin Lanthier, Dominic Leclerc, Nicolas Quesada, Brandon Turcotte, Trevor Vincent, Jiaqi Zhao

Release 0.17.0

Improvements

• Python installation no longer requires repoze.lru. #293

• Multidimensional Hermite polynomials are now implemented in Numba, hence reducing the C++ dependencies of The Walrus. #295

• Updates missing figures in the "Basics of Hafnians and Loop Hafnians" documentation. #288

Contributors

This release contains contributions from (in alphabetical order):

Mikhail Andrenkov, Sebastián Duque

Release 0.16.2

Bug fixes

• hermite_multidimensional_numba can now handle a cutoff of type np.ndarray with shape=[]. #283

Contributors

This release contains contributions from (in alphabetical order):

Filippo Miatto

Release 0.16.1

Improvements

• Faster implementation of hermite_multidimensional_numba and hermite_multidimensional_numba_grad. #280

Bug fixes

• Updates the samples.generate_torontonian_sample function to ensure probabilities are normalized. #250

• Pins Numba to version <0.54 to avoid binary imcompatibilities with the 1.21 release of NumPy. #250

Contributors

This release contains contributions from (in alphabetical order):

Josh Izaac, Filippo Miatto, Nicolas Quesada.

Release 0.16.0

New features

• Adds the function hafnian_sparse to compute sparse loop hafnians (pure Python implementation). #245

• The symplectic.squeezing function is now generalized to multiple modes of single mode squeezing. #249

• Adds a function symplectic.passive_transformation which allows for Gaussian states to be transformed by arbitrary non-unitary, non-square linear optical transformations. #249

• The torontonian_sample_state function can now sample displaced Gaussian states. #248

• Adds the function hafnian_banded to calculate the hafnian of a banded matrix. #246

• Adds the functions hermite_multidimensional_numba and grad_hermite_multidimensional_numba to calculate renormalized multidimensional Hermite polynomials and its gradients using Numba. #251

• Adds the functions mzgate and grad_mzgate to calculate the Fock representation of the Mach-Zehnder gate and its gradients. #257

• Adds the ability to calculate n-body photon number distributions using the function n_body_marginals. #253

• Adds the ability to calculate cumulants and arbitrary expectation values of products of powers of photon numbers with the functions photon_number_cumulant and photon_number_moment respectively. #264

• Adds support for calculating the permanent using the BBFG algorithm and changes this to the default method for calculating permanents. #267

• Adds the ability to calculate click cumulants in threshold detection with the function click_cumulant. #264

Improvements

• Speeds up the calculation of photon number variances/covariances. #244

• Updates documentation for the the tor function. #265

• Numba methods for multidimensional hermite can now detect dtype automatically. #271

Bug fixes

• Corrects bug in the function photon_number_covar that gave incorrect results when the covariance between two modes with finite displacements was calculated. #264

• Fixes a bug in setup.py that would cause the build to fail when using miniforge for M1 macs. #273

• Updates the samples.generate_hafnian_sample function to renormalize probabilities. #250

Breaking changes

• Torontonians and approximations to the hafnian for non-negative matrices are no longer calculated in C++ using the Eigen software library. Instead, they are now calculated in pure Python using Numba. These changes have the nice result of making The Walrus compilable from source using only a C++ compiler. #262 #259.

Contributors

This release contains contributions from (in alphabetical order):

Ali Asadi, Jake Bulmer, Timjan Kalajdzievski, Filippo Miatto, Nicolas Quesada, Yuan Yao

Release 0.15.1

Bug fixes

• Builds The Walrus binaries against an older version of NumPy, to avoid a breaking ABI change in NumPy 1.20. #240

Contributors

This release contains contributions from (in alphabetical order):

Josh Izaac

Version 0.15.0

New features

• Adds the function random_banded_interferometer to generate unitary matrices with a given bandwidth. #208

• Adds the function tvd_cutoff_bounds to calculate bounds in the total variation distance between a Fock-truncated and an ideal GBS distribution. #210

• Adds function for calculating threshold detection probabilities for Gaussian states with displacement. #220

• Adds new functions total_photon_number_distribution and characteristic_function to study properties of the total photon number distribution of a k identical lossy squeezers. #230

• Adds new functions xxpp_to_xpxp and xpxp_to_xxpp in the symplectic module to swap the ordering of the quadrature operators in vectors and matrices. #237

Improvements

• The hafnians and loop hafnians of diagonal matrices are now calculated in polynomial time. #212

• Refactors setup.py to avoid issues with CFLAGS. #229

• The fidelity function in quantum/gaussian_checks.py is rewritten to add clarity. #226

• Simplifies logic of normal_ordered_expectation by removing mutually cancelling np.conj. #228

Bug fixes

• Removes unnecessary np.real_if_close statements in quantum/fock_tensors.py causing the probabilities to not be normalized. #215

• Fixes the prefactor in pure_state_amplitude. #231

Contributors

This release contains contributions from (in alphabetical order):

Jack Brown, Jake Bulmer, Rachel Chadwick, Stefano Paesani, Nicolas Quesada