v1.1.0 The Round Table

New features

• Python 3.14 is now supported.
• JobGroup
  • Two new async methods to launch jobs, launch_async_jobs and relaunch_async_failed_jobs, that launch as many jobs as possible.
  • New cancel_all method.
  • Jobs coming from a JobGroup should now be grouped together on the Quandela Cloud interface using the name of the JobGroup.
• PPNR detectors have a new wire_efficiency parameter, allowing to represent losses directly within the detector.
• The cloud target url can now be saved in RemoteConfig.

Improvements

• PersistentData folder can now be specified using a PCVL_PERSISTENT_PATH environment variable, or falls back to a temp folder if the default path is not writable.
• Sped-up computation with Heralds and non-perfect detectors.
• HTTP errors received during the use of a RemoteProcessor or a RemoteJob are more explicit.
• SLAP backend can now directly simulate feed-forward in usual cases.

Bug fixes

• Duplicated Parameter names are now correctly detected when inserting a new component into a Processor.
• Processor.get_parameters now successfully gives the parameters defined in any inserted FFCircuitProvider.
• RemoteProcessor.estimate_required_shots and RemoteProcessor.estimate_expected_samples now work with non-unitary experiments.
• Conversion from BSDistribution to BSCount or BSSamples was forgetting the 0-photon states.
• JobGroup was unable to take a Job containing an iteration list containing an input state or a noise model.
• PNR detectors can now limit the number of photons.
• Tomography will now work with RemoteProcessor.

Breaking changes

Even though the changes in the API brought by the 1.1.0 version were targeted to be minimal, there are still a few breaking changes:

• Following Python cycle of life, python 3.9 is no longer supported.
• Linux version requirement has been increased. We now support:
  • Debian 10+
  • Ubuntu 18.10+
  • Fedora 29+
  • CentOS/RHEL 8+
• RemoteConfig.set_cloud_maximal_job_count is deprecated as JobGroup now retrieves the number of launchable jobs directly from the cloud.
• The order of the parameters in Circuit.compute_unitary was inconsistent with the order for this method on components, so it has been changed.
• The .copy() methods now perform deepcopies (was previously something between shallow and deep copies). This method has been removed for Processors.

A detailed documentation on how to update your Perceval code to version 1.1 is available here: https://perceval.quandela.net/docs/v1.1/legacy.html

You can contact us on Quandela community forum

v1.0.1 - The Grail Encounter

Improvements

• In legacy, we clarified that BasicState with distinguishability noise should use the new NoisyFockState format (i.e. "|{0}>" instead of "|{_:0}>")

Bug fixes

• Fix Expression and Parameter copy.
• Circuit and Experiment containing several times the same Parameter can now be copied properly using .copy().
• Simulator allowed AnnotatedFockState inputs when it shouldn't
• BasicState can be initialised from a numpy.array again
• Fixed Two-particle bosonic-fermionic quantum walk example notebook
• Fixed typos and missing links in the user documentation

v1.0.0 - The Grail Encounter

After a long wait, Perceval has finally reached version 1!

New features and reasons to upgrade

• The user documentation has been completely revamped.

  • Its improved structure is more natural, especially for newcomers,
  • The tutorials section was split in lessons divided into three levels (beginner, advanced and expert),
  • A particular care was given to Perceval and exqalibur code reference which now covers all existing features!

• It is now possible to send Experiments containing feed-forward to the Cloud. Please note that only a limited list of platforms can handle feed-forward (at the time of its release, only "sim:slos" has this capability).

• Noisy simulations were optimised a lot, through an improved preprocessing of noisy input states, the use of light-weight FockState representation during computations and an improved algorithm. It is now consuming way less memory and shows speed-ups up to 50x on larger noisy simulations.

Improvements

• CircuitOptimizer now supports an initial guess

• A Herald can be added on only one side of an Experiment, making it asymmetrical (user input state and output state won't have the same size anymore)

• Better probability threshold management in chained tensor products of distributions

• Use herald values during an imperfect Detector simulation

• Heralds and PostSelect are now transmitted between simulation layers for LC and TD components

• Optimized usage of an FSMask in SLAP

• Optimized the SLAP.prob_distribution call

• Optimized the Source model algorithm (by moving it to exqalibur)

• NoiseModel now provides pythonic access to its attributes

• A request payload may now be generated without connecting to a Cloud provider

• Enhanced Scaleway provider code

• Shor's algorithm example notebook has been modernised

• Rendering

  • pdisplay handles BSSamples
  • Sped up the rendering of large SVDistribution

• Remote computing

  • Perceval now checks a new constraint from remote platforms, telling whether they can handle multiple input photons per mode
  • Read RemoteProcessor.status property to know the current status of the corresponding remote platform
  • Add field cloud_maximal_job_count to RemoteConfig, limiting the number of concurrent Job run at once when calling JobGroup.run_parallel
  • JobGroup.list_existing() method was renamed to list_locally_saved(), for clarity
  • The behaviour of get_results is now consistent for failed jobs (LocalJob or RemoteJob).

Bug fixes

• Experiment composition

  • When composing two experiments or processors, the right-hand-side detectors now have priority
  • Adding a single component to an Experiment on non-consecutive modes correctly applies an inverted permutation after it

• Circuit and Processor rendering

  • Modes are now rendered below the input ports
  • Fixed a bug where a subcircuit and its container box would be shifted
  • Fixed a bug in specific feed-forwarded circuits rendering

• Fix handling of min_detected_photons in the NoisySamplingSimulator

• Fix Expression.fixed property

• Fix a crash that could occur in Clifford & Clifford 2017 sampling back-end

• Internal logger can now be setup to write in a file at a path containing non-ASCII characters

• A Job does not forget its name anymore when reloaded from a JobGroup

• Fix the result of PS.definition()

Breaking changes

With the release of Perceval version 1 comes a list of updated features and syntax. Here's a list of the major potential code breaks:

• To achieve large performance gains in noisy simulations, the FockState class was split into three. While the pcvl.BasicState retains most of its versatility, user code might be broken by this code.
• In some computations, physical_perf and logical_perf scores have been merged into a global_perf
• Gate based to LO circuits converters were moved from Perceval to a new package perceval-interop. Read its documentation.

A detailed documentation on how to update your Perceval code to version 1 is available here: https://perceval.quandela.net/docs/v1.0/legacy.html

You can contact us on Quandela community forum

v0.13.2 - Castle of Wonders

Bug fixes

• Fixed copy of Experiment when holding a defined state
• Fixed a memory leak in Experiment

v0.13.1 - Castle of Wonders

New Features

• The StateGenerator is now able to generate photonic states based on new rules

Improvements

• The catalog items now expose a build_experiment method that allows creating an Experiment without creating a Processor, allowing easier integration into RemoteProcessor.
• Non BasicState input in a Processor require setting the min_detected_photons_filter. The error message when omitting it was not human readable. The message is now clearer.

Bug fixes

• Fixed copy of Experiment and Processor that now give a deep copy of the original object.

v0.13.0 - Castle of Wonders

New features and reasons to upgrade

• Noisy states aren't completely computed anymore for sampling simulations, allowing for 15+ noisy photon samplings.

• Perceval Parameter class now supports math expressions with the new Expression class
  See: https://perceval.quandela.net/docs/v0.13/reference/utils.html#expression

• Perceval handles proxies configuration
  See: https://perceval.quandela.net/docs/notebooks/Remote_computing.html

• Complete overhaul of the JobGroup feature, to improve its reliability and reduce the number of http requests to a minimum

• Introduce a new noise parameter in NoiseModel: the phase error
  See: https://perceval.quandela.net/docs/reference/noise_model.html

Improvements

• Native optimised library (exqalibur) is now available for Linux ARM64, making Perceval installable on such an environment (i.e. you can now install Perceval in a Linux virtual machine running on a recent Mac computer)
• Detectors may now be simulated in non-unitary Processors
• Perceval component catalog now contains all major 1-qubit gates
  See: https://perceval.quandela.net/docs/reference/catalog.html
• JobGroup now has __len__ and __getitem__ operators
• Simulator masks as much of the Fock space as possible to improve memory consumption and computation times (in case of superposed and/or noisy input)
• PostSelect.merge method ignores duplicate conditions
• Components containing variable parameters can now be inverted
• Beam splitter reflectivity to theta conversion manages variable parameters
• Removed several meaningless user warnings

Bug fixes

• Fixed Perceval logging system using a Python logger
• Circuit and Processor renderers wrap and/or reduce text size when it's too long
• Circuits using the same named variable multiple times will now be correctly serialised/deserialised

Other changes

• Perceval no longer deduces the min_detected_photon_filter for noisy and remote simulations. It is now mandatory to pass a value.
• min_detected_photons_filter value in a processor containing heralds is now the minimum expected photon count ignoring any ancillary mode and herald
• Source is no longer a valid parameter when instantiating a Processor

For more information, see: https://perceval.quandela.net/docs/v0.13/legacy.html

You can contact us on the Perceval forum

v0.12.2 - The Bleeding Lance

Improvements

• Detector simulation now uses the optional progress callback
• Renamed JobGroup method list_unfinished_jobs to list_unsuccessful_jobs for clarity
• Removed a warning which always occurred when injecting noise in a converter

Bug fixes

• Fixed post-selection conditions composition in converters
• Fixed heralds display in highly nested circuits
• Fixed JobGroup not saving the refreshed jobs status in a parallel run

v0.12.1 - The Bleeding Lance

New features

• A KL divergence method to compare two distributions.
• Two methods that reduce the number of states in a BSDistribution to help comparing distributions.
• A qloq ansatz in the catalog to help building ansatz with qudit encoding.
  See https://perceval.quandela.net/docs/v0.12/notebooks/QLOQ_QUBO_tutorial.html

Improvements

• The catalog now includes usual 1-qubit gates.
• The CircuitOptimizer can now ignore columns of the unitary given a list of modes that won't receive input photons.
• Perceval's Qiskit bridge is now using Qiskit 1.3.1
• Feed-forward simulation is now faster when using heralds.
• Noisy input state generation is now faster and is now only dependant on the number of photons.

Bug fixes

• LC and TD simulation now succeed if there are empty modes at the end of the Processor.
• Fixed performance computation when using LC and TD.
• Fixed potentially wrong results when using several layers of feed-forward.
• The Sampler iterator no longer need heralded modes to be defined in the input state.
• Fixed a crash when trying to display a circuit having empty sub-circuits.

You can contact us on the Perceval forum

v0.12.0 - The Bleeding Lance

New features

• Feed-forward local simulation
  You can now build a Processor containing detectors and feed-forward controllers configuring unitary circuits and obtain strong simulation results!
  See: https://perceval.quandela.net/docs/notebooks/Quantum_teleportation_feed_forward.html
  See also: https://perceval.quandela.net/docs/reference/feed_forward_configurator.html

• Detector imperfections can now be simulated.
  Four different types of detectors are currently supported:

  • Photon number resolving (PNR) perfect detectors,
  • Threshold detectors,
  • Interleaved pseudo-PNR detectors,
  • Pseudo-PNR detectors using layers of beam splitters and threshold detectors.

• A new job management system helps you group jobs, check statuses and retrieve results.
  See: https://perceval.quandela.net/docs/reference/job_group.html

• A new strong simulation back-end is available: SLAP.
  See: https://perceval.quandela.net/docs/backends.html#slap

• New usage example - Quantum kernel methods
  See: https://perceval.quandela.net/docs/notebooks/quantum_kernel_methods.html

• Resource estimator for gate-based conversion, supporting dual rail and qudit encodings.

• A TVD function between probability distributions.

Improvements

• Perceval now supports Python 3.13.
• PostSelect syntax now supports nested logical expressions of 'and', 'or', 'xor' and 'not' operators.
• Gate-based circuits to linear optics converters now save heralded photons (by using as many post-selected CNOT as possible).
• Strong simulation now automatically uses heralds to mask the Fock space when possible, saving computation time and memory.
• Failed and cancelled jobs may now be programmatically restarted.
• Barrier components can now be invisible (used only for alignments).
• Improved circuit rendering.
• DisplayConfig enabling to select a rendering skin can now be persistent.
• Modernized some example notebooks.

Bug fixes

• Tensor product between BasicState and SVDistribution now always succeeds

Other information & known issues

• Python 3.8 is no longer supported.
• Perceval is compatible with both numpy versions 1 and 2. However, installing a fresh Perceval, then a dependency requiring numpy<2 may result in a pip conflict. If it happens, try to install them both in a single pip command.
• You may have to update pip and setuptools to the latest version before installing Perceval in a Python 3.13 virtual environment.

See syntax changes here: https://perceval.quandela.net/docs/v0.12/legacy.html
You can contact us on the Perceval forum

v0.11.2 - Sir Gawain

Improvements

• Add a Symmetric MZI in the component catalog
• Rectangular GenericInterferometer can now autogenerate any 1-mode component whenever the first or last mode is MZI-free
• Fix typos in logging documentation
• Remove deprecated calls from user documentation

Bug fixes

• Remove a small sleep time in local sampling simulation (that would hinder low photon count sampling)
• Fix how unit tests requiring extra dependencies are run