Adds the quantum fourier transform and controlled phase operations #1064
Conversation
Codecov Report
@@ Coverage Diff @@
## master #1064 +/- ##
==========================================
- Coverage 97.75% 97.74% -0.01%
==========================================
Files 153 153
Lines 11517 11579 +62
==========================================
+ Hits 11258 11318 +60
- Misses 259 261 +2
Continue to review full report at Codecov.
|
pennylane/ops/qubit.py
Outdated
| from pennylane.operation import (AnyWires, DiagonalOperation, Observable, | ||
| Operation) | ||
| from pennylane.templates import template | ||
| from pennylane.templates.state_preparations import (BasisStatePreparation, | ||
| MottonenStatePreparation) | ||
| from pennylane.utils import expand, pauli_eigs |
trbromley
changed the title
|
|
trbromley
added
the
review-ready 👌
| def decomposition(wires): | ||
| num_wires = len(wires) |
There was a problem hiding this comment.
When running the QFT, is there a way to make it print out the full decomposition when using draw? Currently it draws this:
0: ──╭QFT──╭┤ State
1: ──├QFT──├┤ State
2: ──╰QFT──╰┤ State
but for the sake of testing it'd be convenient to see the innards.
There was a problem hiding this comment.
This could be the moment that our new qml.draw() shines!
Since it is not dependent on the QNode state (unlike qnode.draw()), we can add some logic to qml.draw() to always decompose the QFT operation (or maybe any operation that has a draw_decomp flag or something).
There was a problem hiding this comment.
It'd be great to have it at the level of the draw function itself -
qml.draw(circuit, draw_decomp=True)There was a problem hiding this comment.
Or even
qml.draw(qnode, decompose=[“QFT”, “Rot”])(params)
Co-authored-by: Olivia Di Matteo <2068515+glassnotes@users.noreply.github.com>
|
|
||
| * Number of wires: 2 | ||
| * Number of parameters: 1 | ||
| * Gradient recipe: :math:`\frac{d}{d\phi}f(CR_\phi(\phi)) = \frac{1}{2}\left[f(CR_\phi(\phi+\pi/2)) - f(CR_\phi(\phi-\pi/2))\right]` |
There was a problem hiding this comment.
I did check this and can confirm that it's unlike the other CRX, CRY, CRZ gates because the 2-parameter shift works:
import pennylane as qml
from pennylane import numpy as np
wires = 2
dev = qml.device("default.qubit", wires=range(wires))
w1 = np.random.random((4, 2, 3))
w2 = np.random.random((4, 2, 3))
@qml.qnode(dev)
def f(x):
qml.templates.StronglyEntanglingLayers(w1, wires=range(wires))
# qml.CRX(x, wires=[0, 1])
qml.ControlledPhaseShift(x, wires=[0, 1])
qml.templates.StronglyEntanglingLayers(w2, wires=range(wires))
return qml.expval(qml.PauliZ(0))
x = 0.3
g1 = qml.grad(f)(x)
g2 = (f(x + np.pi / 2) - f(x - np.pi / 2)) / 2
np.allclose(g1, g2)
|
This should be good to merge, I did a local coverage run and coverage looks good - the only bit I'm unsure about is incomplete coverage on |
Context:
Adds two new gates:
QFTControlledPhaseShift(which is required in the decomposition of QFT)Description of the Change:
These two gates have been added as operations in
pennylane/ops/qubit.py. Both have a defined decomposition and matrix representation. Also added these gates to the supported operations indefault.*devices.Possible Drawbacks:
We use the
ControlledPhaseShiftwhich looks similar to the CV gateControlledPhase. However, the first sentence of theControlledPhaseShiftdocstring reads "A qubit controlled phase shift." to hopefully remove some confusion.