Release v0.6.0 (#285)

# Release 0.6.0 (current release) ### New features * Added a new method to discretize Wigner functions that revolves Clenshaw summations. This method is expected to be fast and reliable for systems with high number of excitations, for which the pre-existing iterative method is known to be unstable. Users can select their preferred methods by setting the value of `Settings.DISCRETIZATION_METHOD` to either `interactive` (default) or `clenshaw`. * Added the `PhaseNoise(phase_stdev)` gate (non-Gaussian). Output is a mixed state in Fock representation. It is not based on a choi operator, but on a nonlinear transformation of the density matrix. [(#275)](#275) ### Breaking changes * The value of `hbar` can no longer be specified outside of `Settings`. All the classes and methods that allowed specifying its value as an input now retrieve it directly from `Settings`. * Certain attributes of `Settings` can no longer be changed after their value is queried for the first time. ### Improvements * Tensorflow bumped to v2.14 with poetry installation working out of the box on Linux and Mac. [(#281)](#281) ### Bug fixes * Fixed a bug about the variable names in functions (apply_kraus_to_ket, apply_kraus_to_dm, apply_choi_to_ket, apply_choi_to_dm). [(#271)](#271) * Fixed a bug that was leading to an error when computing the Choi representation of a unitary transformation. ### Documentation ### Contributors [Filippo Miatto](https://github.com/ziofil), [Samuele Ferracin](https://github.com/SamFerracin), [Yuan Yao](https://github.com/sylviemonet), [Zeyue Niu](https://github.com/zeyueN) --------- Co-authored-by: Sebastián Duque Mesa <[email protected]> Co-authored-by: JacobHast <[email protected]> Co-authored-by: elib20 <[email protected]> Co-authored-by: Luke Helt <[email protected]> Co-authored-by: zeyueN <[email protected]> Co-authored-by: Robbe De Prins <[email protected]> Co-authored-by: Robbe De Prins (UGent-imec) <[email protected]> Co-authored-by: Yuan <[email protected]> Co-authored-by: Ryk <[email protected]> Co-authored-by: Gabriele Gullì <[email protected]> Co-authored-by: Yuan Yao <[email protected]> Co-authored-by: SamFerracin <[email protected]> Co-authored-by: Yuan Yao <[email protected]> Co-authored-by: heltluke <[email protected]>
XanaduAI · Oct 14, 2023 · 4d47e8e · 4d47e8e
1 parent df94d49
commit 4d47e8e
Show file tree

Hide file tree

Showing 29 changed files with 2,525 additions and 1,626 deletions.
diff --git a/.github/CHANGELOG.md b/.github/CHANGELOG.md
@@ -1,3 +1,50 @@
+# Release 0.6.0 (current release)
+
+### New features
+
+* Added a new method to discretize Wigner functions that revolves Clenshaw summations. This method is expected to be fast and
+reliable for systems with high number of excitations, for which the pre-existing iterative method is known to be unstable. Users
+can select their preferred methods by setting the value of `Settings.DISCRETIZATION_METHOD` to either `interactive` (default) or
+`clenshaw`.
+  [(#280)](https://github.com/XanaduAI/MrMustard/pull/280)
+
+* Added the `PhaseNoise(phase_stdev)` gate (non-Gaussian). Output is a mixed state in Fock representation.
+  It is not based on a choi operator, but on a nonlinear transformation of the density matrix.
+  [(#275)](https://github.com/XanaduAI/MrMustard/pull/275)
+
+### Breaking changes
+
+* The value of `hbar` can no longer be specified outside of `Settings`. All the classes and 
+  methods that allowed specifying its value as an input now retrieve it directly from `Settings`.
+  [(#278)](https://github.com/XanaduAI/MrMustard/pull/278)
+
+* Certain attributes of `Settings` can no longer be changed after their value is queried for the
+  first time.
+  [(#278)](https://github.com/XanaduAI/MrMustard/pull/278)
+
+### Improvements
+
+* Tensorflow bumped to v2.14 with poetry installation working out of the box on Linux and Mac.
+  [(#281)](https://github.com/XanaduAI/MrMustard/pull/281)
+
+### Bug fixes
+
+* Fixed a bug about the variable names in functions (apply_kraus_to_ket, apply_kraus_to_dm, apply_choi_to_ket, apply_choi_to_dm).
+  [(#271)](https://github.com/XanaduAI/MrMustard/pull/271)
+
+* Fixed a bug that was leading to an error when computing the Choi representation of a unitary transformation.
+  [(#283)](https://github.com/XanaduAI/MrMustard/pull/283)
+
+* Fixed the internal function to calculate ABC of Bargmann representation (now corresponds to the literature) and other fixes to get the correct Fock tensor.
+  [(#255)](https://github.com/XanaduAI/MrMustard/pull/255)
+
+### Documentation
+
+### Contributors
+[Filippo Miatto](https://github.com/ziofil), [Samuele Ferracin](https://github.com/SamFerracin), [Yuan Yao](https://github.com/sylviemonet), [Zeyue Niu](https://github.com/zeyueN)
+
+---
+
 # Release 0.5.0
 
 ### New features
@@ -35,7 +82,7 @@
   # Or, in command line: `tensorboard --logdir={tb_cb.logdir}` and open link in browser.
   ```
 
-* Gaussian states support a `bargmann` method for returning the bargmann representation. 
+* Gaussian states support a `bargmann` method for returning the bargmann representation.
   [(#235)](https://github.com/XanaduAI/MrMustard/pull/235)
 
 * The `ket` method of `State` now supports new keyword arguments `max_prob` and `max_photons`.
@@ -50,7 +97,7 @@
   ket = Gaussian(2).ket(max_prob=0.99, max_photons=3)
   ```
 
-* Gaussian transformations support a `bargmann` method for returning the bargmann representation. 
+* Gaussian transformations support a `bargmann` method for returning the bargmann representation.
   [(#239)](https://github.com/XanaduAI/MrMustard/pull/239)
 
 * BSGate.U now supports method='vanilla' (default) and 'schwinger' (slower, but stable to any cutoff)
@@ -82,10 +129,11 @@
   [(#239)](https://github.com/XanaduAI/MrMustard/pull/239)
 
 * More robust implementation of cutoffs for States.
-[(#239)](https://github.com/XanaduAI/MrMustard/pull/239)
+
+  [(#239)](https://github.com/XanaduAI/MrMustard/pull/239)
 
 * Dependencies and versioning are now managed using Poetry.
-[(#257)](https://github.com/XanaduAI/MrMustard/pull/257)
+  [(#257)](https://github.com/XanaduAI/MrMustard/pull/257)
 
 ### Bug fixes
 

diff --git a/mrmustard/__init__.py b/mrmustard/__init__.py
@@ -14,89 +14,11 @@
 
 """This is the top-most `__init__.py` file of MrMustard package."""
 
-import numpy as np
-import rich.table
+
 from rich import print
 
 from ._version import __version__
-
-
-# pylint: disable=too-many-instance-attributes
-class Settings:
-    """Settings class."""
-
-    def __new__(cls):  # singleton
-        if not hasattr(cls, "instance"):
-            cls.instance = super(Settings, cls).__new__(cls)
-        return cls.instance
-
-    def __init__(self):
-        self._backend = "tensorflow"
-        self.HBAR = 2.0
-        self.CHOI_R = 0.881373587019543  # np.arcsinh(1.0)
-        self.DEBUG = False
-        self.AUTOCUTOFF_PROBABILITY = 0.999  # capture at least 99.9% of the probability
-        self.AUTOCUTOFF_MAX_CUTOFF = 100
-        self.AUTOCUTOFF_MIN_CUTOFF = 1
-        self.CIRCUIT_DECIMALS = 3
-        # use cutoff=5 for each mode when determining if two transformations in fock repr are equal
-        self.EQ_TRANSFORMATION_CUTOFF = 3  # 3 is enough to include a full step of the rec relations
-        self.EQ_TRANSFORMATION_RTOL_FOCK = 1e-3
-        self.EQ_TRANSFORMATION_RTOL_GAUSS = 1e-6
-        # for the detectors
-        self.PNR_INTERNAL_CUTOFF = 50
-        self.HOMODYNE_SQUEEZING = 10.0
-        # misc
-        self.PROGRESSBAR = True
-        self._seed = np.random.randint(0, 2**31 - 1)
-        self.rng = np.random.default_rng(self._seed)
-        self.DEFAULT_BS_METHOD = "vanilla"  # can be 'vanilla' or 'schwinger'
-
-    @property
-    def SEED(self):
-        """Returns the seed value if set, otherwise returns a random seed."""
-        if self._seed is None:
-            self._seed = np.random.randint(0, 2**31 - 1)
-            self.rng = np.random.default_rng(self._seed)
-        return self._seed
-
-    @SEED.setter
-    def SEED(self, value):
-        """Sets the seed value."""
-        self._seed = value
-        self.rng = np.random.default_rng(self._seed)
-
-    @property
-    def BACKEND(self):
-        """The backend which is used.
-
-        Can be either ``'tensorflow'`` or ``'torch'``.
-        """
-        return self._backend
-
-    @BACKEND.setter
-    def BACKEND(self, backend_name: str):
-        if backend_name not in ["tensorflow", "torch"]:  # pragma: no cover
-            raise ValueError("Backend must be either 'tensorflow' or 'torch'")
-        self._backend = backend_name
-
-    # use rich.table to print the settings
-    def __repr__(self):
-        """Returns a string representation of the settings."""
-        table = rich.table.Table(title="MrMustard Settings")
-        table.add_column("Setting")
-        table.add_column("Value")
-        table.add_row("BACKEND", self.BACKEND)
-        table.add_row("SEED", str(self.SEED))
-        for key, value in self.__dict__.items():
-            if key == key.upper():
-                table.add_row(key, str(value))
-        print(table)
-        return ""
-
-
-settings = Settings()
-"""Settings object."""
+from .settings import *
 
 
 def version():

diff --git a/mrmustard/lab/abstract/state.py b/mrmustard/lab/abstract/state.py
@@ -94,7 +94,7 @@ def __init__(
         self._norm = _norm
         if cov is not None and means is not None:
             self.is_gaussian = True
-            self.is_hilbert_vector = np.allclose(gaussian.purity(self.cov, settings.HBAR), 1.0)
+            self.is_hilbert_vector = np.allclose(gaussian.purity(self.cov), 1.0)
             self.num_modes = cov.shape[-1] // 2
         elif eigenvalues is not None and symplectic is not None:
             self.is_gaussian = True
@@ -140,7 +140,7 @@ def purity(self) -> float:
         """Returns the purity of the state."""
         if self._purity is None:
             if self.is_gaussian:
-                self._purity = gaussian.purity(self.cov, settings.HBAR)
+                self._purity = gaussian.purity(self.cov)
             else:
                 self._purity = fock.purity(self._dm)
         return self._purity
@@ -209,7 +209,10 @@ def fock(self) -> ComplexTensor:
         r"""Returns the Fock representation of the state."""
         if self._dm is None and self._ket is None:
             _fock = fock.wigner_to_fock_state(
-                self.cov, self.means, shape=self.shape, return_dm=not self.is_hilbert_vector
+                self.cov,
+                self.means,
+                shape=self.shape,
+                return_dm=not self.is_hilbert_vector,
             )
             if self.is_mixed:
                 self._dm = _fock
@@ -223,7 +226,7 @@ def fock(self) -> ComplexTensor:
     def number_means(self) -> RealVector:
         r"""Returns the mean photon number for each mode."""
         if self.is_gaussian:
-            return gaussian.number_means(self.cov, self.means, settings.HBAR)
+            return gaussian.number_means(self.cov, self.means)
 
         return fock.number_means(tensor=self.fock, is_dm=self.is_mixed)
 
@@ -233,7 +236,7 @@ def number_cov(self) -> RealMatrix:
         if not self.is_gaussian:
             raise NotImplementedError("number_cov not yet implemented for non-gaussian states")
 
-        return gaussian.number_cov(self.cov, self.means, settings.HBAR)
+        return gaussian.number_cov(self.cov, self.means)
 
     @property
     def norm(self) -> float:
@@ -508,7 +511,9 @@ def __and__(self, other: State) -> State:
         cov = gaussian.join_covs([self.cov, other.cov])
         means = gaussian.join_means([self.means, other.means])
         return State(
-            cov=cov, means=means, modes=self.modes + [m + self.num_modes for m in other.modes]
+            cov=cov,
+            means=means,
+            modes=self.modes + [m + self.num_modes for m in other.modes],
         )
 
     def __getitem__(self, item) -> State:
@@ -581,11 +586,15 @@ def __eq__(self, other) -> bool:  # pylint: disable=too-many-return-statements
             return True
         try:
             return np.allclose(
-                self.ket(cutoffs=other.cutoffs), other.ket(cutoffs=other.cutoffs), atol=1e-6
+                self.ket(cutoffs=other.cutoffs),
+                other.ket(cutoffs=other.cutoffs),
+                atol=1e-6,
             )
         except TypeError:
             return np.allclose(
-                self.dm(cutoffs=other.cutoffs), other.dm(cutoffs=other.cutoffs), atol=1e-6
+                self.dm(cutoffs=other.cutoffs),
+                other.dm(cutoffs=other.cutoffs),
+                atol=1e-6,
             )
 
     def __rshift__(self, other: Transformation) -> State:
@@ -617,7 +626,8 @@ def __rmul__(self, other):
         """
         if self.is_gaussian:
             warnings.warn(
-                "scalar multiplication forces conversion to fock representation", UserWarning
+                "scalar multiplication forces conversion to fock representation",
+                UserWarning,
             )
             if self.is_pure:
                 return State(ket=self.ket() * other)

diff --git a/mrmustard/lab/abstract/transformation.py b/mrmustard/lab/abstract/transformation.py
@@ -250,6 +250,7 @@ def choi(self, cutoffs: Sequence[int]):
                 f"Invalid number of cutoffs: {len(cutoffs)} (expected {self.num_modes} or {4*self.num_modes})"
             )
         if self.is_unitary:
+            shape = shape[: 2 * self.num_modes]
             U = self.U(shape[: self.num_modes])
             Udual = self.U(shape[self.num_modes :])
             return fock.U_to_choi(U, Udual)

diff --git a/mrmustard/lab/gates.py b/mrmustard/lab/gates.py
@@ -19,9 +19,9 @@
 """
 
 from typing import List, Optional, Sequence, Tuple, Union
-
+import numpy as np
 from mrmustard import settings
-from mrmustard.lab.abstract import Transformation
+from mrmustard.lab.abstract import State, Transformation
 from mrmustard.math import Math
 from mrmustard.physics import fock, gaussian
 from mrmustard.training import Parametrized
@@ -45,6 +45,7 @@
     "Attenuator",
     "Amplifier",
     "AdditiveNoise",
+    "PhaseNoise",
 ]
 
 
@@ -92,7 +93,7 @@ def __init__(
 
     @property
     def d_vector(self):
-        return gaussian.displacement(self.x.value, self.y.value, settings.HBAR)
+        return gaussian.displacement(self.x.value, self.y.value)
 
     def U(self, cutoffs: Sequence[int]):
         r"""Returns the unitary representation of the Displacement gate using
@@ -787,11 +788,11 @@ def __init__(
 
     @property
     def X_matrix(self):
-        return gaussian.loss_XYd(self.transmissivity.value, self.nbar.value, settings.HBAR)[0]
+        return gaussian.loss_XYd(self.transmissivity.value, self.nbar.value)[0]
 
     @property
     def Y_matrix(self):
-        return gaussian.loss_XYd(self.transmissivity.value, self.nbar.value, settings.HBAR)[1]
+        return gaussian.loss_XYd(self.transmissivity.value, self.nbar.value)[1]
 
 
 class Amplifier(Parametrized, Transformation):
@@ -844,11 +845,11 @@ def __init__(
 
     @property
     def X_matrix(self):
-        return gaussian.amp_XYd(self.gain.value, self.nbar.value, settings.HBAR)[0]
+        return gaussian.amp_XYd(self.gain.value, self.nbar.value)[0]
 
     @property
     def Y_matrix(self):
-        return gaussian.amp_XYd(self.gain.value, self.nbar.value, settings.HBAR)[1]
+        return gaussian.amp_XYd(self.gain.value, self.nbar.value)[1]
 
 
 # pylint: disable=no-member
@@ -897,4 +898,69 @@ def __init__(
 
     @property
     def Y_matrix(self):
-        return gaussian.noise_Y(self.noise.value, settings.HBAR)
+        return gaussian.noise_Y(self.noise.value)
+
+
+class PhaseNoise(Parametrized, Transformation):
+    r"""The phase noise channel.
+
+    The phase noise channel is a non-Gaussian transformation that is equivalent to
+    a random phase rotation.
+
+    Args:
+        phase_stdev (float or List[float]): the standard deviation of the (wrapped) normal
+            distribution for the angle of the rotation
+        modes (optional, list(int)): the single mode this gate is applied to (default [0])
+    """
+
+    def __init__(
+        self,
+        phase_stdev: Union[Optional[float], Optional[List[float]]] = 0.0,
+        phase_stdev_trainable: bool = False,
+        phase_stdev_bounds: Tuple[Optional[float], Optional[float]] = (0.0, None),
+        modes: Optional[List[int]] = None,
+    ):
+        super().__init__(
+            phase_stdev=phase_stdev,
+            phase_stdev_trainable=phase_stdev_trainable,
+            phase_stdev_bounds=phase_stdev_bounds,
+        )
+        self._modes = modes or [0]
+        self.is_unitary = False
+        self.is_gaussian = False
+        self.short_name = "P~"
+
+    # need to override primal because of the unconventional way
+    # the channel is defined in Fock representation
+    def primal(self, state):
+        idx = state.modes.index(self.modes[0])
+        if state.is_pure:
+            ket = state.ket()
+            dm = fock.ket_to_dm(ket)
+        else:
+            dm = state.dm()
+
+        # transpose dm so that the modes of interest are at the end
+        M = state.num_modes
+        indices = list(range(2 * M))
+        indices.remove(idx)
+        indices.remove(idx + M)
+        indices += [idx, idx + M]
+        dm = math.transpose(dm, indices)
+
+        coeff = math.cast(
+            math.exp(
+                -0.5
+                * self.phase_stdev.value**2
+                * math.arange(-dm.shape[-2] + 1, dm.shape[-1]) ** 2
+            ),
+            dm.dtype,
+        )
+
+        for k in range(-dm.shape[-2] + 1, dm.shape[-1]):
+            diagonal = math.diag_part(dm, k=k)
+            diagonal *= coeff[k + dm.shape[-2] - 1]
+            dm = math.set_diag(dm, diagonal, k=k)
+
+        # transpose dm back to the original order
+        return State(dm=math.transpose(dm, np.argsort(indices)), modes=state.modes)