Decompositions

[josh146]

Description of the Change

Adds the following as frontend functionality to Strawberry Fields:

• Provides a module of decomposition functions, covering the Takagi, Clements, Williamson, and Bloch-Messiah decompositions.
• Creates a new abstract base class strawberryfields.ops.Decompositions, which is used to create quantum operations which target the decompositions. The included operation are:
  • CovarianceState: uses the Williamson decomposition to decompose a Gaussian covariance matrix into a symplectic transformation acting on a set of thermal states
  • GaussianTransform: uses the Bloch-Messiah decomposition to decompose a Gaussian symplectic transformation to two linear interferometers and one-mode squeezers
  • Interferometer: decomposes a unitary representing a multi-mode linear interferometer into a combination of beamsplitters and rotation gates, using the Clements decomposition.
• Documentation has also been updated, with an additional section added to the conventions, detailing the various decompositions.

Benefits

• Provides front-facing user ability to decompose arbitrary Gaussian states and unitaries into the set of quantum gates provided by Strawberry Fields
• Supported by all backends when performing numerical simulations
• Provides a class structure that can be updated with additional decompositions down the line, for example non-Gaussian decompositions for Bose-Hubbard etc.

Possible Drawbacks

• At the moment, the only way for a user to see the decomposed gates is to apply the decomposition operation and run the engine, then inspect the output of eng.print_applied(). This is consistent with how existing gates are decomposed in Strawberry Fields.
• The decomposition class only accepts NumPy objects as arguments, not tensorflow objects. If attempting to initialize a decomposition with a tensorflow object, the user will receive a NotImplementedError.

Related GitHub Issues: None

[codecov]

Codecov Report

Merging #7 into master will increase coverage by 0.16%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master       #7      +/-   ##
==========================================
+ Coverage   96.36%   96.53%   +0.16%     
==========================================
  Files          21       22       +1     
  Lines        3302     3464     +162     
==========================================
+ Hits         3182     3344     +162     
  Misses        120      120

Flag	Coverage Δ
#b0	94.6% <100%> (+0.26%)	⬆️
#b1	66.1% <98.83%> (+3.08%)	⬆️
#fock	46.99% <98.83%> (+3.99%)	⬆️
#gaussian	42.49% <100%> (+2.82%)	⬆️
#m0	96.33% <100%> (+0.17%)	⬆️
#m1	95.32% <100%> (+0.22%)	⬆️
#tf	67.75% <98.83%> (+2.97%)	⬆️

Impacted Files	Coverage Δ
strawberryfields/backends/gaussianbackend/ops.py	93.33% <ø> (-0.19%)	⬇️
...fields/backends/gaussianbackend/gaussiancircuit.py	99.57% <100%> (ø)	⬆️
strawberryfields/decompositions.py	100% <100%> (ø)
strawberryfields/backends/shared_ops.py	97.08% <100%> (+0.61%)	⬆️
strawberryfields/utils.py	100% <100%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 92633c7...4497434. Read the comment docs.