Strawberryfields Versions Save

Post release to address changes to the documentation.

New features

• Program Xanadu’s new Borealis hardware device via Strawberry Fields and Xanadu Cloud. (#714)

  Borealis is a cloud-accessible photonic quantum computer, offering full programmability over all of its gates and capable of quantum advantage. Its hardware is based on time-domain multiplexing (TDM); a single squeezed-light source emits batches of 216 time-ordered squeezed-light pulses that interfere with one another with the help of optical delay loops, programmable beamsplitters, and phase shifters.

  Learn more about Borealis via our website and tutorials.

• GBS data visualization functions are added. (#714)

• A set of TDM compilers are added, including a Borealis compiler which compiles and validates programs against the hardware specification and calibration certificate. (#714)

• A remove_loss utility function is added to the program_utils module, allowing for the removal of LossChannels from Strawberry Fields programs. (#714)

• Cropping vacuum modes from TDM program results is now possible by passing crop=True as a run option. (#714)

  n, N = get_mode_indices(delays)
  prog = sf.TDMProgram(N)
  
  with prog.context(*gate_args) as (p, q):
      ops.Sgate(p[0]) | q[n[0]]
      for i in range(len(delays)):
          ops.Rgate(p[2 * i + 1]) | q[n[i]]
          ops.BSgate(p[2 * i + 2], np.pi / 2) | (q[n[i + 1]], q[n[i]])
      ops.MeasureX | q[0]
  
  eng = sf.Engine("gaussian")
  results = eng.run(prog, crop=True)
  

• Resulting samples from TDM jobs return only the non-empty mode measurements when setting the crop option to True in the program run_options or as a keyword argument in the engine run method. (#714)

• Realistic loss can be added to a Borealis circuit for local simulation execution. (#714)

  compile_options = {
      "device": device,  # hardware device object needed
      "realistic_loss": True,
  }
  
  eng = sf.Engine("gaussian")
  results = eng.run(prog, compile_options=compile_options)
  

• Utility functions are added to allow for easier Borealis program and parameter creation. (#714)

• Functions are added for analyzing GBS results for comparisons with classical simulations. (#714)

Improvements

• A locked program can now be (un)rolled, and automatically restores the lock if previously in place. (#703)

• Rolling and unrolling now only happens in place, and no longer returns the (un)rolled circuit. (#702)

• Program.assert_number_of_modes and Program.assert_max_number_of_measurements are combined into a single assert_modes method. (#709)

• Job results can now be retrieved without converting integers to np.int64 objects by setting integer_overflow_protection=False (default True) when running a program via RemoteEngine.run(). (#712)

• The TDM module is refactored to contain program.py, with the TDMProgram class, and utils.py, with various utility functions. (#714)

• The Compiler base class is updated to allow for setting a rigid circuit layout to validate a program during compilation. (#714)

• The Compiler base class now contains methods that can be overwritten to provide subclass compilers with loss-additions (e.g., to add realistic loss to a circuit) and program parameter updates. (#714)

Bug fixes

• Trying to unroll an already unrolled program with a different number of shots works as expected. (#702)

• Fixed bug with vacuum modes missing. (#702)

• Validating parameters now works with nested parameter arrays. (#711)

• Store correct rolled circuit before unrolling (fixes issue when rolled circuit has changed due to e.g., compilation). (#710)

Documentation

• The centralized Xanadu Sphinx Theme is now used to style the Sphinx documentation. (#701)

• The documentation on Gaussian circuit operations is fixed so that it's properly rendered. (#714)

Contributors

This release contains contributions from (in alphabetical order):

Mikhail Andrenkov, Sebastian Duque, Luke Helt, Theodor Isacsson, Josh Izaac, Fabian Laudenbach

New features since last release

• Device.layout and Device.gate_parameters may now return None. This can happen when a remote simulator device is used. (#661)

• A new interferometer decomposition method is implemented following the proposal of the paper Simple factorization of unitary transformations. (#665)

  import numpy as np
  
  import strawberryfields as sf
  from strawberryfields import ops
  
  U = np.array([[-0.39302099+0.28732291j,  0.83734522+0.24866248j],
                [ 0.00769051+0.87345344j, -0.3847068 +0.29836325j]])
  
  prog = sf.Program(2)
  
  with prog.context as q:
      ops.Interferometer(U, mesh="sun_compact") | q
  

• A Device.certificate method is added which returns the hardware device certificate. (#679)

  >>> import strawberryfields as sf
  >>> eng = sf.RemoteEngine("X8")
  >>> print(eng.device.certificate)
  {'target': 'X8_01' ... }
  

• Setting shots=None in the engine or program run options will not execute any measurements applied on the circuit. (#682)

  import strawberryfields as sf
  from strawberryfields import ops
  
  prog = sf.Program(1)
  eng = sf.Engine("gaussian")
  
  with prog.context as q:
      ops.Sgate(0.5) | q[0]
      ops.MeasureFock() | q
  
  results = eng.run(prog, shots=None)
  
  # samples will output an empty list []
  print(results.samples)
  
  # the resulting Gaussian state is still accessible
  # via its vector of means and covariance matrix
  print(results.state.means())
  print(results.state.cov())
  

• There's a program_equivalence function in strawberryfields/program_utils.py which checks Strawberry Fields programs for equivalence. (#686)

• An equality operator is implemented for strawberryfields.Program, checking that the exact same gates and respective parameters, are applied in order. (#686)

  import strawberryfields as sf
  from strawberryfields import ops
  
  prog_1 = sf.Program(1)
  prog_2 = sf.Program(1)
  
  with prog.context as q:
      ops.Sgate(0.42) | q[0]
      ops.MeasureFock() | q
  
  with prog.context as q:
      ops.Sgate(0.42) | q[0]
      ops.MeasureFock() | q
  
  assert prog_1 == prog_2
  

• A Program.equivalence convenience method is added which calls the program_equivalence utility function. (#686)

  prog_1 = sf.Program(1)
  prog_2 = sf.Program(1)
  
  with prog.context as q:
      ops.Sgate(1.1) | q[0]
      ops.MeasureFock() | q
  
  with prog.context as q:
      ops.Sgate(0.42) | q[0]
      ops.MeasureFock() | q
  
  assert prog_1.equivalence(prog_2, compare_params=False)
  

• A Device.validate_target static method is added which checks that the target in the layout is the same as the target field in the specification. This check is also performed at Device initialization. (#687)

• Tests are run in random order and the seed for NumPy's and Python's random number generators are set by pytest-randomly. (#692)

• Adds support for Python 3.10. #695

Breaking Changes

• DeviceSpec is renamed to Device, which now also contains more than only the device specification. (#679)

  >>> import strawberryfields as sf
  >>> eng = sf.RemoteEngine("X8")
  >>> isinstance(eng.device, sf.Device)
  True
  >>> print(eng.device.target)
  X8_01
  

Bug fixes

• It's now possible to show graphs using the plot apps layer when not run in notebooks. (#669)

• program.compile now raises an error if the device specification contains gate parameters but no circuit layout. Without a layout, the gate parameters cannot be validated against the device specification. (#661)

• The teleportation tutorial examples/teleportation.py now uses the correct value (now phi = 0 instead of phi = np.pi / 2) for the phase shift of the beamsplitters. (#674)

• Program.compile() returns a deep copy of the program attributes, except for the circuit and the register references. (#688)

Contributors

This release contains contributions from (in alphabetical order):

Sebastian Duque, Theodor Isacsson, Jon Schlipf, Hossein Seifoory

New features since last release

• A Result.metadata property is added to retrieve the metadata of a job result. (#663)

• A setter method for Result.state is added for setting a state for a local simulation if a state has not previously been set. (#663)

• Functions are now available to convert between XIR and Strawberry Fields programs. (#643)

  For example,

  prog = sf.Program(3)
  eng = sf.Engine("gaussian")
  
  with prog.context as q:
      ops.Sgate(0, 0) | q[0]
      ops.Sgate(1, 0) | q[1]
      ops.BSgate(0.45, 0.0) | (q[0], q[2])
      ops.MeasureFock() | q[0]
  
  xir_prog = sf.io.to_xir(prog)
  

  resulting in the following XIR script

  >>> print(xir_prog.serialize())
  Sgate(0, 0) | [0];
  Sgate(1, 0) | [1];
  BSgate(0.45, 0.0) | [0, 2];
  MeasureFock | [0];
  

Bug fixes

• The TDMProgram.compile_info and TDMProgram.target fields are now set when a TDMProgram is compiled using the "TDM" compiler. (#659)

• Updates Program.assert_max_number_of_measurements to expect the maximum number of measurements from the device specification as a flat dictionary entry instead of a nested one. (#662)

  "modes": {
      "pnr_max": 20,
      "homodyne_max": 1000,
      "heterodyne_max": 1000,
  }
  

  instead of

  "modes": {
      "max": {
          "pnr": 20,
          "homodyne": 1000,
          "heterodyne": 1000,
      }
  }
  

Documentation

• README has been ported to Markdown. (#664)

Contributors

This release contains contributions from (in alphabetical order):

Theodor Isacsson

New features since last release

• The generic multimode Gaussian gate Ggate is now available in the sf.ops module with the backend choice of tf. The N mode Ggate can be parametrized by a real symplectic matrix S (size 2N * 2N) and a diplacement vector d (size N). You can also obtain the gradients of the Ggate gate via TensorFlow's tape.gradient (#599) (#606)

  from thewalrus.random import random_symplectic
  
  num_mode = 2
  cutoff = 10
  S = tf.Variable(random_symplectic(num_mode))
  d = tf.Variable(np.random.random(2 * num_mode))
  
  eng = sf.Engine("tf", backend_options={"cutoff_dim": cutoff})
  prog = sf.Program(2)
  
  with prog.context as q:
      sf.ops.Ggate(S, d) | (q[0], q[1])
  
  state_out = eng.run(prog).state.ket()
  

  Note that in order to update the parameter S by using its gradient, you cannot use gradient descent directly (as the unitary would not be symplectic after the update). Please use the function sf.backends.tfbackend.update_symplectic which is designed specifically for this purpose.

  def overlap_loss(state, objective):
      return -tf.abs(tf.reduce_sum(tf.math.conj(state) * objective)) ** 2
  
  def norm_loss(state):
      return -tf.abs(tf.linalg.norm(state)) ** 2
  
  def loss(state, objective):
      return overlap_loss(state, objective) + norm_loss(state)
  
  num_mode = 1
  cutoff = 10
  
  S = tf.Variable(random_symplectic(num_mode))
  d = tf.Variable(np.random.random(2 * num_mode))
  kappa = tf.Variable(0.3)
  objective = tf.Variable(np.eye(cutoff)[1], dtype=tf.complex64)
  
  adam = tf.keras.optimizers.Adam(learning_rate=0.01)
  eng = sf.Engine("tf", backend_options={"cutoff_dim": cutoff})
  prog = sf.Program(1)
  
  with prog.context as q:
      sf.ops.Ggate(S, d) | q
      sf.ops.Kgate(kappa) | q
  
  loss_vals = []
  for _ in range(200):
      with tf.GradientTape() as tape:
          state_out = eng.run(prog).state.ket()
          loss_val = loss(state_out, objective)
  
      eng.reset()
      grad_S, gradients_d, gradients_kappa = tape.gradient(loss_val, [S, d, kappa])
      adam.apply_gradients(zip([gradients_d, gradients_kappa], [d, kappa]))
      update_symplectic(S, grad_S, lr=0.1)  # update S here
      loss_vals.append(loss_val)
  

Breaking Changes

• Complex parameters of the Catstate operation are expected in polar form as two separate real parameters. (#441)

• The sf CLI has been removed in favour of the Xanadu Cloud Client. (#642)

  1. Configuring account credentials using:

     • Strawberry Fields v0.19.0

       $ sf configure --token "foo"
       

     • Strawberry Fields v0.20.0

       $ xcc config set REFRESH_TOKEN "foo"
       Successfully updated REFRESH_TOKEN setting to 'foo'.
       

  2. Verifying your connection to the Xanadu Cloud using:

     • Strawberry Fields v0.19.0

       $ sf --ping
       You have successfully authenticated to the platform!
       

     • Strawberry Fields v0.20.0

       $ xcc ping
       Successfully connected to the Xanadu Cloud.
       

  3. Submitting a Blackbird circuit to the Xanadu Cloud using:

     • Strawberry Fields v0.19.0

       $ # Version 0.19.0
       $ sf run "foo.xbb"
       Executing program on remote hardware...
       2021-11-02 03:04:05,06 - INFO - The device spec X8_01 has been successfully retrieved.
       2021-11-02 03:04:05,07 - INFO - Compiling program for device X8_01 using compiler Xunitary.
       2021-11-02 03:04:05,08 - INFO - Job b185a63c-f302-4adb-acf8-b6e4e413c11d was successfully submitted.
       2021-11-02 03:04:05,09 - INFO - The remote job b185a63c-f302-4adb-acf8-b6e4e413c11d has been completed.
       [[0 0 0 0]
       [0 0 0 0]
       [0 0 0 0]
       [0 0 0 0]]
       

     • Strawberry Fields v0.20.0

       $ xcc job submit --name "bar" --target "X8_01" --circuit "$(cat foo.xbb)"
       {
           "id": "0b0f5a46-46d8-4157-8005-45a4764361ba",  # Use this ID below.
           "name": "bar",
           "status": "open",
           "target": "X8_01",
           "language": "blackbird:1.0",
           "created_at": "2021-11-02 03:04:05,10",
           "finished_at": null,
           "running_time": null,
           "metadata": {}
       }
       $ xcc job get 0b0f5a46-46d8-4157-8005-45a4764361ba --result
       {
           "output": [
               "[[0 0 0 0]\n[0 0 0 0]\n[0 0 0 0]\n[0 0 0 0]]"
           ]
       }
       

• The sf.api.Connection class has been replaced with the xcc.Connection class. (#645)

  Previously, in Strawberry Fields v0.19.0, an sf.RemoteEngine can be instantiated with a custom Xanadu Cloud connection as follows:

  import strawberryfields as sf
  import strawberryfields.api
  
  connection = strawberryfields.api.Connection(
    token="Xanadu Cloud API key goes here",
    host="platform.strawberryfields.ai",
    port=443,
    use_ssl=True,
  )
  engine = sf.RemoteEngine("X8", connection=connection)
  

  In Strawberry Fields v0.20.0, the same result can be achieved using

  import strawberryfields as sf
  import xcc
  
  connection = xcc.Connection(
    refresh_token="Xanadu Cloud API key goes here",  # See "token" argument above.
    host="platform.strawberryfields.ai",
    port=443,
    tls=True,                                        # See "use_ssl" argument above.
  )
  engine = sf.RemoteEngine("X8", connection=connection)
  

• The sf.configuration module has been replaced with the xcc.Settings class. (#649)

  This means that Xanadu Cloud credentials are now stored in exactly one location, the path to which depends on your operating system:

  1. Windows: C:\Users\%USERNAME%\AppData\Local\Xanadu\xanadu-cloud\.env

  2. MacOS: /home/$USER/Library/Application\ Support/xanadu-cloud/.env

  3. Linux: /home/$USER/.config/xanadu-cloud/.env

  The format of the configuration file has also changed to .env and the names of some fields have been updated. For example,

  # Strawberry Fields v0.19.0 (config.toml)
  [api]
  authentication_token = "Xanadu Cloud API key goes here"
  hostname = "platform.strawberryfields.ai"
  port = 443
  use_ssl = true
  

  is equivalent to

  # Strawberry Fields v0.20.0 (.env)
  XANADU_CLOUD_REFRESH_TOKEN='Xanadu Cloud API key goes here'
  XANADU_CLOUD_HOST='platform.strawberryfields.ai'
  XANADU_CLOUD_PORT=443
  XANADU_CLOUD_TLS=True
  

  Similarly, the names of the configuration environment variables have changed from

  # Strawberry Fields v0.19.0
  export SF_API_AUTHENTICATION_TOKEN="Xanadu Cloud API key goes here"
  export SF_API_HOSTNAME="platform.strawberryfields.ai"
  export SF_API_PORT=443
  export SF_API_USE_SSL=true
  

  to

  # Strawberry Fields v0.20.0
  export XANADU_CLOUD_REFRESH_TOKEN="Xanadu Cloud API key goes here"
  export XANADU_CLOUD_HOST="platform.strawberryfields.ai"
  export XANADU_CLOUD_PORT=443
  export XANADU_CLOUD_TLS=true
  

  Finally, strawberryfields.store_account() has been replaced such that

  # Strawberry Fields v0.19.0
  import strawberryfields as sf
  sf.store_account("Xanadu Cloud API key goes here")
  

  becomes

  # Strawberry Fields v0.20.0
  import xcc
  xcc.Settings(REFRESH_TOKEN="Xanadu Cloud API key goes here").save()
  

• The sf.api.Job class has been replaced with the xcc.Job class. (#650)

  A Job object is returned when running jobs asynchronously. In previous versions of Strawberry Fields (v0.19.0 and lower), the Job object can be used as follows:

  >>> job = engine.run_async(program, shots=1)
  >>> job.status
  'queued'
  >>> job.result
  InvalidJobOperationError
  >>> job.refresh()
  >>> job.status
  'complete'
  >>> job.result
  [[0 1 0 2 1 0 0 0]]
  

  In Strawberry Fields v0.20.0, the Job object works slightly differently:

  >>> job = engine.run_async(program, shots=1)
  >>> job.status
  'queued'
  >>> job.wait()
  >>> job.status
  'complete'
  >>> job.result
  {'output': [array([[0 1 0 2 1 0 0 0]])]}
  

  The job.wait() method is a blocking method that will wait for the job to finish. Alternatively, job.clear() can be called to clear the cache, allowing job.status to re-fetch the job status.

• The sf.api.Result class has been updated to support the Xanadu Cloud Client integration. (#651)

  While Result.samples should return the same type and shape as before, the Result.all_samples property has been renamed to Result.samples_dict and returns the samples as a dictionary with corresponding measured modes as keys.

  >>> res = eng.run(prog, shots=3)
  >>> res.samples
  array([[1, 0], [0, 1], [1, 1]])
  >>> res.samples_dict
  {0: [np.array([1, 0, 1])], 1: [np.array([0, 1, 1])]}
  

  The samples dictionary is only accessible for simulators.

• The sf.api.DeviceSpec class has been updated to support the Xanadu Cloud Client integration. (#644)

  It now works as a container for a device specification dictionary. There are no more API connection usages, and DeviceSpec.target is retrieved from the device specification rather than passed at initialization.

• The api subpackage has been removed and the contained DeviceSpec and Result classes have been moved to the root strawberryfields folder. (#652)

  They can now be imported as follows:

  import strawberryfields  as sf
  # sf.DeviceSpec
  # sf.Result
  

Documentation

• Strawberry Fields interactive has been removed from the documentation. (#635)

Contributors

This release contains contributions from (in alphabetical order):

Mikhail Andrenkov, Sebastián Duque Mesa, Theodor Isacsson, Josh Izaac, Filippo Miatto, Nicolás Quesada, Antal Száva, Yuan Yao.

New features since last release

• Compact decompositions as described in https://arxiv.org/abs/2104.07561, (rectangular_compact and triangular_compact) are now available in the sf.decompositions module, and as options in the Interferometer operation. (#584)

  This decomposition allows for lower depth photonic circuits in physical devices by applying two independent phase shifts in parallel inside each Mach-Zehnder interferometer. rectangular_compact reduces the layers of phase shifters from 2N+1 to N+2 for an N mode interferometer when compared to e.g. rectangular_MZ.

  Example:

  import numpy as np
  from strawberryfields import Program
  from strawberryfields.ops import Interferometer
  from scipy.stats import unitary_group
  
  M = 10
  
  # generate a 10x10 Haar random unitary
  U = unitary_group.rvs(M)
  
  prog = Program(M)
  
  with prog.context as q:
      Interferometer(U, mesh='rectangular_compact') | q
  
  # check that applied unitary is correct
  compiled_circuit = prog.compile(compiler="gaussian_unitary")
  commands = compiled_circuit.circuit
  S = commands[0].op.p[0] # symplectic transformation
  Uout = S[:M,:M] + 1j * S[M:,:M] # unitary transformation
  
  print(np.allclose(U, Uout))
  

• A new compiler, GaussianMerge, has been added. It is aimed at reducing calculation overhead for non-Gaussian circuits by minimizing the amount of Gaussian operations in a circuit, while retaining the same functionality. (#591)

  GaussianMerge merges Gaussian operations, where allowed, into GaussianTransform and Dgate operations. It utilizes the existing GaussianUnitary compiler to merge operations and Directed Acyclic Graphs to determine which operations can be merged.

  modes = 4
  cutoff_dim = 6
  
  # prepare an intial state with 4 photons in as many modes
  initial_state = np.zeros([cutoff_dim] * modes, dtype=complex)
  initial_state[1, 1, 1, 1] = 1
  
  prog = sf.Program(4)
  
  with prog.context as q:
      ops.Ket(initial_state) | q  # Initial state preparation
      # Gaussian Layer
      ops.S2gate(0.01, 0.01) | (q[0], q[1])
      ops.BSgate(1.9, 1.7) | (q[1], q[2])
      ops.BSgate(0.9, 0.2) | (q[0], q[1])
      # Non-Gaussian Layer
      ops.Kgate(0.5) | q[3]
      ops.CKgate(0.7) | (q[2], q[3])
      # Gaussian Layer
      ops.BSgate(1.0, 0.4) | (q[0], q[1])
      ops.BSgate(2.0, 1.5) | (q[1], q[2])
      ops.Dgate(0.01) | q[0]
      ops.Dgate(0.01) | q[0]
      ops.Sgate(0.01, 0.01) | q[1]
      # Non-Gaussian Layer
      ops.Vgate(0.5) | q[2]
  
  prog_merged = prog.compile(compiler="gaussian_merge")
  

• A new operation, PassiveChannel has been added. It allows for arbitrary linear/passive transformations (i.e., any operation which is linear in creation operators). Currently only supported by the gaussian backend. (#600)

  from strawberryfields.ops import PassiveChannel, Sgate
  import strawberryfields as sf
  from scipy.stats import unitary_group
  import numpy as np
  
  M = 4
  
  circuit = sf.Program(M)
  U1 = unitary_group.rvs(M)
  U2 = unitary_group.rvs(M)
  losses = np.random.random(M)
  
  T = U2 @ np.diag(losses) @ U1
  
  eng = sf.Engine(backend='gaussian')
  circuit = sf.Program(M)
  with circuit.context as q:
      for i in range(M):
          ops.Sgate(1) | q[i]
      ops.PassiveChannel(T) | q
  
  cov = eng.run(circuit).state.cov()
  

• A new compiler, passive, allows for a circuit which only consists of passive elements to be compiled into a single PassiveChannel. (#600)

  from strawberryfields.ops import BSgate, LossChannel, Rgate
  import strawberryfields as sf
  
  circuit = sf.Program(2)
  with circuit.context as q:
      Rgate(np.pi) | q[0]
      BSgate(0.25 * np.pi, 0) | (q[0], q[1])
      LossChannel(0.9) | q[1]
  
  compiled_circuit = circuit.compile(compiler="passive")
  

  >>> print(compiled_circuit)
     PassiveChannel([[-0.7071+8.6596e-17j -0.7071+0.0000e+00j]
     [-0.6708+8.2152e-17j  0.6708+0.0000e+00j]]) | (q[0], q[1])
  

Improvements

• backends/tfbackend/ops.py is cleaned up to reduce line count, clarify function similarity across backend ops, and replace tensorflow.tensordot with broadcasting. (#567)

• Support is added for using a TDMProgram to construct time-domain circuits with Fock measurements and multiple loops. (#601)

• measure_threshold in the gaussian backend now supports displaced Gaussian states. (#615)

• Speed improvements are addded to gaussian_unitary compiler. (#603)

• Adds native support in the Fock backend for the MZgate. (#610)

• measure_threshold is now supported in the bosonic backend. (#618)

Bug fixes

• Fixes an unexpected behaviour that can result in increasing memory usage due to sympy.lambdify caching too much data using linecache. (#579)

• Keeps symbolic expressions when converting a Strawberry Fields circuit to a Blackbird program by storing them as blackbird.RegRefTransforms in the resulting Blackbird program. (#596)

• Fixes a bug in the validation step of strawberryfields.tdm.TdmProgram.compile which almost always used the wrong set of allowed gate parameter ranges to validate the parameters in a program. (#605)

• The correct samples are now returned when running a TDMProgram with several shots, where timebins % concurrent_modes != 0. (#611)

• Fixes the formula used for sampling generaldyne outcomes in the gaussian backend. (#614)

• Measurement arguments are now stored as non-keyword arguments, instead of keyword arguments, in the resulting Blackbird program when using the io.to_blackbird() converter function. (#622)

• Factorials of numbers larger than 170 are now calculated using long integer arithmetic, using the flag exact=True in scipy.special.factorial, when calling sf.apps.similarity.orbit_cardinality. (#628)

Documentation

• References to the simulon simulator target have been rewritten to simulon_gaussian to reflect changes made on the Xanadu Quantum Cloud. The language has been modified to imply that multiple simulators could be available on XQC. (#576)

Contributors

This release contains contributions from (in alphabetical order):

J. Eli Bourassa, Jake Bulmer, Sebastian Duque, Theodor Isacsson, Aaron Robertson, Jeremy Swinarton, Antal Száva, Federico Rueda, Yuan Yao.

Documentation

• References to the simulon simulator target have been rewritten to simulon_gaussian to reflect changes made on the Xanadu Quantum Cloud. The language has been modified to imply that multiple simulators could be available on XQC. (#576)

Contributors

This release contains contributions from (in alphabetical order):

Jeremy Swinarton.

New features since last release

• Adds the Bosonic backend, which can simulate states represented as linear combinations of Gaussian functions in phase space. (#533) (#538) (#539) (#541) (#546) (#549)

  It can be regarded as a generalization of the Gaussian backend, since transformations on states correspond to modifications of the means and covariances of each Gaussian in the linear combination, along with changes to the coefficients of the linear combination. Example states that can be expressed using the new backend include all Gaussian, Gottesman-Kitaev-Preskill, cat and Fock states.

  prog = sf.Program(1)
  eng = sf.Engine('bosonic')
  
  with prog.context as q:
      sf.ops.GKP(epsilon=0.1) | q
      sf.ops.MeasureX | q
  
  results = eng.run(prog, shots=200)
  samples = results.samples[:, 0]
  
  plt.hist(samples, bins=100)
  plt.show()
  

• Adds the sf.ops.GKP operation, which allows the Gottesman-Kitaev-Preskill state to be initialized on both the Bosonic and Fock backends. (#553) (#546)

  GKP states are qubits, with the qubit state defined by:

  

  where the computational basis states are

  

• Adds the measurement-based squeezing gate MSgate; a new front-end operation for the Bosonic backend. (#538) (#539) (#541)

  MSgate is an implementation of inline squeezing that can be performed by interacting the target state with an ancillary squeezed vacuum state at a beamsplitter, measuring the ancillary mode with homodyne, and then applying a feed-forward displacement. The channel is implemented either on average (as a Gaussian CPTP map) or in the single-shot implementation. If the single-shot implementation is used, the measurement outcome of the ancillary mode is stored in the results object.

  prog = sf.Program(1)
  eng = sf.Engine('bosonic')
  
  with prog.context as q:
      sf.ops.Catstate(alpha=2) | q
      r = 0.3
      # Average map
      sf.ops.MSgate(r, phi=0, r_anc=1.2, eta_anc=1, avg=True) | q
      # Single-shot map
      sf.ops.MSgate(r, phi=0, r_anc=1.2, eta_anc=1, avg=False) | q
  
  results = eng.run(prog)
  ancilla_samples = results.ancilla_samples
  
  xvec = np.arange(-5, 5, 0.01)
  pvec = np.arange(-5, 5, 0.01)
  wigner = results.state.wigner(0, xvec, pvec)
  
  plt.contourf(xvec, pvec, wigner)
  plt.show()
  

• The tf backend now accepts the Tensor DType as argument. (#562)

  Allows high cutoff dimension to give numerically correct calculations:

  prog = sf.Program(2)
  eng  = sf.Engine("tf", backend_options={"cutoff_dim": 50, "dtype": tf.complex128})
  with prog.context as q:
      Sgate(0.8) | q[0]
      Sgate(0.8) | q[1]
      BSgate(0.5,0.5) | (q[0], q[1])
      BSgate(0.5,0.5) | (q[0], q[1])
  state = eng.run(prog).state
  N0, N0var = state.mean_photon(0)
  N1, N1var = state.mean_photon(1)
  print(N0)
  print(N1)
  print("analytical:", np.sinh(0.8)**2)
  

Improvements

• Program compilation has been modified to support the XQC simulation service, Simulon. (#545)

• The sympmat, rotation_matrix, and haar_measure functions have been removed from backends/shared_ops.py. These functions are now imported from The Walrus. In addition, various outdated functionality from the shared_ops.py file has been removed, including the caching of beamsplitter and squeezing pre-factors. (#560) (#558)

• Sample processing in the TDMProgram is now more efficient, by replacing calls to pop with fancy indexing. (#548)

• No VisibleDeprecationWarning is raised when using the state wigner method. (#564)

• The backend utility module shared_ops.py has been removed, with all of its functionality now provided by The Walrus. (#573)

Breaking changes

• Removes support for Python 3.6. (#573)

Bug fixes

• Connection objects now send requests to the platform API at version 0.2.0 instead of the incorrect version number 1.0.0. (#540)

• TDM programs now expect a flat (not nested) dictionary of modes in device specifications obtained from the XQC platform API. (#566)

• Fixes a bug in the CatState operation, whereby the operation would return incorrect results for a high cutoff value. (#557) (#556)

Documentation

• The "Hardware" quickstart page has been renamed to "Xanadu Quantum Cloud" to encompass both hardware and cloud simulators. A new "Cloud simulator" entry has been added, describing how to submit programs to be executed via the XQC simulator. (#547)

• Cleanup docs to make contribution easier. (#561)

• Add development requirements and format script to make contribution easier. (#563)

Contributors

This release contains contributions from (in alphabetical order):

J. Eli Bourassa, Guillaume Dauphinais, Ish Dhand, Theodor Isacsson, Josh Izaac, Leonhard Neuhaus, Nicolás Quesada, Aaron Robertson, Krishna Kumar Sabapathy, Jeremy Swinarton, Antal Száva, Ilan Tzitrin.

New features since last release

• TDMProgram objects can now be compiled and submitted via the API. (#476)

• Wigner functions can be plotted directly via Strawberry Fields using Plot.ly. (#495)

  prog = sf.Program(1)
  eng = sf.Engine('fock', backend_options={"cutoff_dim": 10})
  
  with prog.context as q:
      gamma = 2
      Vgate(gamma) | q[0]
  
  state = eng.run(prog).state
  
  xvec = np.arange(-4, 4, 0.01)
  pvec = np.arange(-4, 4, 0.01)
  mode = 0
  
  sf.plot_wigner(state, mode, xvec, pvec, renderer="browser")
  

• Fock state marginal probabilities can be plotted directly via Strawberry Fields using Plot.ly. (#510)

  prog = sf.Program(1)
  eng = sf.Engine('fock', backend_options={"cutoff_dim":5})
  
  with prog.context as q:
      Sgate(0.5) | q[0]
  
  state = eng.run(prog).state
  state.all_fock_probs()
  
  modes = [0]
  
  sf.plot_fock(state, modes, cutoff=5, renderer="browser")
  

• Position and momentum quadrature probabilities can be plotted directly via Strawberry Fields using Plot.ly. (#510)

  prog = sf.Program(1)
  eng = sf.Engine('fock', backend_options={"cutoff_dim":5})
  
  with prog.context as q:
      Sgate(0.5) | q[0]
  
  state = eng.run(prog).state
  
  modes = [0]
  xvec = np.arange(-4, 4, 0.1)
  pvec = np.arange(-4, 4, 0.1)
  
  sf.plot_quad(state, modes, xvec, pvec, renderer="browser")
  

• Strawberry Fields code can be generated from a program (and an engine) by calling sf.io.generate_code(program, eng=engine). (#496)

Improvements

• Connection objects now send versioned requests to the platform API. (#512)

• TDMProgram allows application of gates with more than one symbolic parameter. #492

• The copies option, when constructing a TDMProgram, has been removed. Instead, the number of copies of a TDM algorithm can now be set by passing the shots keyword argument to the eng.run() method. (#489)

  >>> with prog.context([1, 2], [3, 4]) as (p, q):
  ...     ops.Sgate(0.7, 0) | q[1]
  ...     ops.BSgate(p[0]) | (q[0], q[1])
  ...     ops.MeasureHomodyne(p[1]) | q[0]
  >>> eng = sf.Engine("gaussian")
  >>> results = eng.run(prog, shots=3)
  

  Furthermore, the TDMProgram.unrolled_circuit attribute now only contains the single-shot unrolled circuit. Unrolling with multiple shots can still be specified via the unroll method: TDMProgram.unroll(shots=60).

• The Result.samples returned by TDM programs has been updated to return samples of shape (shots, spatial modes, timebins) instead of (shots, spatial modes * timebins). (#489)

• A sample post-processing function is added that allows users to move vacuum mode measurements from the first shots to the last shots, and potentially crop out the final shots containing these measurements. (#489)

• pytest-randomly is added to the SF tests. (#480)

• TDMProgram objects can now be serialized into Blackbird scripts, and vice versa. (#476)

Breaking Changes

• Jobs are submitted to the Xanadu Quantum Cloud through a new OAuth based authentication flow using offline refresh tokens and access tokens. (#520)

Bug fixes

• Fixes a bug where Dgate, Coherent, and DisplacedSqueezed do not support TensorFlow tensors if the tensor has an added dimension due to the existence of batching. (#507)

• Fixes an issue with reshape_samples where the samples were sometimes reshaped in the wrong way. (#489)

• The list of modes is now correctly added to the Blackbird program when using the io.to_blackbird function. (#476)

• Fixes a bug where printing the Result object containing samples from a time-domain job would result in an error. Printing the result object now correctly displays information about the results. (#493)

• Removes the antlr4 requirement due to version conflicts. (#494)

• TDMProgram.run_options is now correctly used when running a TDM program. (#500)

• Fixes a bug where a single parameter list passed to the TDMProgram context results in an error. (#503)

Documentation

• TDMProgram docstring is updated to make it clear that only Gaussian programs are allowed. (#519)

• Clarifies special cases for the MZgate in the docstring. (#479)

Contributors

This release contains contributions from (in alphabetical order):

Tom Bromley, Jack Brown, Theodor Isacsson, Josh Izaac, Fabian Laudenbach, Tim Leisti, Nicolas Quesada, Antal Száva.

New features since last release

• Adds the ability to construct time domain multiplexing algorithms via the new sf.TDMProgram class, for highly scalable simulation of Gaussian states. (#440)

  For example, creating and simulating a time domain program with 2 concurrent modes:

  >>> import strawberryfields as sf
  >>> from strawberryfields import ops
  >>> prog = sf.TDMProgram(N=2)
  >>> with prog.context([1, 2], [3, 4], copies=3) as (p, q):
  ...     ops.Sgate(0.7, 0) | q[1]
  ...     ops.BSgate(p[0]) | (q[0], q[1])
  ...     ops.MeasureHomodyne(p[1]) | q[0]
  >>> eng = sf.Engine("gaussian")
  >>> results = eng.run(prog)
  >>> print(results.all_samples)
  {0: [array([1.26208025]), array([1.53910032]), array([-1.29648336]),
  array([0.75743215]), array([-0.17850101]), array([-1.44751996])]}
  

  For more details, see the code documentation.

• Adds the function VibronicTransition to the apps.qchem.vibronic module. This function generates a custom Strawberry Fields operation for applying the Doktorov operator on a given state. (#451)

  >>> from strawberryfields.apps.qchem.vibronic import VibronicTransition
  >>> modes = 2
  >>> p = sf.Program(modes)
  >>> with p.context as q:
  ...     VibronicTransition(U1, r, U2, alpha) | q
  

• Adds the TimeEvolution function to the apps.qchem.dynamics module. This function generates a custom Strawberry Fields operation for applying a time evolution operator on a given state. (#455)

  >>> modes = 2
  >>> p = sf.Program(modes)
  >>> with p.context as q:
  ...     sf.ops.Fock(1) | q[0]
  ...     sf.ops.Interferometer(Ul.T) | q
  ...     TimeEvolution(w, t) | q
  ...     sf.ops.Interferometer(Ul) | q
  

  where w is the normal mode frequencies, and t the time in femtoseconds.

• Molecular data and pre-generated samples for water and pyrrole have been added to the data module of the Applications layer of Strawberry Fields. For more details, please see the data module documentation (#463)

• Adds the function read_gamess to the qchem module to extract the atomic coordinates, atomic masses, vibrational frequencies, and normal modes of a molecule from the output file of a vibrational frequency calculation performed with the GAMESS quantum chemistry package. (#460)

  >>> r, m, w, l = read_gamess('../BH_data.out')
  >>> r # atomic coordinates
  array([[0.0000000, 0.0000000, 0.0000000],
         [1.2536039, 0.0000000, 0.0000000]])
  >>> m # atomic masses
  array([11.00931,  1.00782])
  >>> w # vibrational frequencies
  array([19.74, 19.73, 0.00, 0.00, 0.00, 2320.32])
  >>> l # normal modes
  array([[-0.0000000e+00, -7.5322000e-04, -8.7276210e-02,  0.0000000e+00,
       8.2280900e-03,  9.5339055e-01],
     [-0.0000000e+00, -8.7276210e-02,  7.5322000e-04,  0.0000000e+00,
       9.5339055e-01, -8.2280900e-03],
     [ 2.8846925e-01, -2.0000000e-08,  2.0000000e-08,  2.8846925e-01,
      -2.0000000e-08,  2.0000000e-08],
     [ 2.0000000e-08,  2.8846925e-01, -2.0000000e-08,  2.0000000e-08,
       2.8846925e-01, -2.0000000e-08],
     [-2.0000000e-08,  2.0000000e-08,  2.8846925e-01, -2.0000000e-08,
       2.0000000e-08,  2.8846925e-01],
     [-8.7279460e-02,  0.0000000e+00,  0.0000000e+00,  9.5342606e-01,
      -0.0000000e+00, -0.0000000e+00]])
  

Improvements

• When jobs submitted to the Xanadu Quantum Cloud are canceled, they will now display a cancel_pending JobStatus until the cancellation is confirmed. (#456)

Bug fixes

• Fixed a bug where the function reduced_dm in backends/tfbackend/states.py gives the wrong output when passing it several modes. (#471)

• Fixed a bug in the function reduced_density_matrix in backends/tfbackend/ops.py which caused the wrong subsystems to be traced out. (#467) (#470)

• Fixed a bug where decompositions to Mach-Zehnder interferometers would return incorrect results on NumPy 1.19. (#473)

• The Walrus version 0.14 introduced modified function names. Affected functions have been updated in Strawberry Fields to avoid deprecation warnings. (#472)

Documentation

• Adds further testing and coverage descriptions to the developer documentation. This includes details regarding the Strawberry Fields test structure and test decorators. (#461)

• Updates the minimum required version of TensorFlow in the development guide. (#468)

Contributors

This release contains contributions from (in alphabetical order):

Juan Miguel Arrazola, Tom Bromley, Theodor Isacsson, Josh Izaac, Soran Jahangiri, Nathan Killoran, Fabian Laudenbach, Nicolás Quesada, Antal Száva, ‪Ilan Tzitrin.