[WIP] Adds decomposition to `DoubleExcitation` operator (and update gradient recipes) #1303

glassnotes · 2021-05-12T21:22:42Z

Context: Currently DoubleExcitation is decomposed into DoubleExcitationPlus and DoubleExcitationMinus, neither of which are decomposed into elementary gates, such that DoubleExcitation cannot be implemented on all devices.

Description of the Change: Adds the decomposition into elementary gates based on this paper to DoubleExcitation. Furthermore, updates the gradient recipe of both SingleExcitation and DoubleExcitation to use the four-term parameter-shift rule of the aforementioned paper.

Benefits: DoubleExcitation can now run on any device; Single and Double excitation gradients are improved.

Possible Drawbacks: None

Related GitHub Issues: #1275 #1278

codecov · 2021-05-13T14:47:39Z

Codecov Report

Merging #1303 (0554aea) into master (9687c13) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #1303   +/-   ##
=======================================
  Coverage   98.15%   98.15%           
=======================================
  Files         148      148           
  Lines       11357    11359    +2     
=======================================
+ Hits        11147    11149    +2     
  Misses        210      210

Impacted Files	Coverage Δ
pennylane/ops/qubit.py	`98.36% <100.00%> (+<0.01%)`	⬆️

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 9687c13...0554aea. Read the comment docs.

tests/ops/test_qubit_ops.py

…nyLaneAI/pennylane into decompose_double_excitations

glassnotes · 2021-05-13T15:58:29Z

pennylane/ops/qubit.py

@@ -1707,14 +1707,13 @@ class SingleExcitation(Operation):

    * Number of wires: 2
    * Number of parameters: 1
-    * Gradient recipe: Obtained from its decomposition in terms of the
-      :class:`~.SingleExcitationPlus` and :class:`~.SingleExcitationMinus` operations
+    * Gradient recipe: The SingleExcitation operator satisfies a four-term parameter-shift rule


@dwierichs just to confirm, this is what you had in mind for enabling both operations to use the four-term shift rules?

yes, exactly :)

tests/ops/test_qubit_ops.py

antalszava

Looks good to me! 💯 🥇 this is very helpful, thank you for going so far to get this one! 🎉

antalszava · 2021-05-13T19:05:09Z

.github/CHANGELOG.md

+* The `qml.SingleExcitation` and `qml.DoubleExcitation` operations now
+  have decompositions over elementary gates, and their gradient recipes
+  have been updated to use the four-term parameter-shift rules.
+  [(#1303)](https://github.com/PennyLaneAI/pennylane/pull/1303)


🥳 🥳 🥳

pennylane/ops/qubit.py

antalszava · 2021-05-13T19:18:55Z

tests/ops/test_qubit_ops.py

+        mats = []
+        for i in reversed(decomp):
+            if i.wires.tolist() == [1, 0] and isinstance(i, qml.CRY):
+                new_mat = np.array(


(minor): We could potentially have a separate function e.g., cry_target_first in the gate_data.py file that could be called here and in the other test cases. This is out of scope, as it means a bit of organization also for test_single_excitation_plus_decomp and test_single_excitation_minus_decomp.

This would just be a step towards making this file slightly more organized, as the matrix representation of controlled-operations where the wires are in a non-ascending order start coming up in the tests for decompositions.

That would be super useful to have just in general. We could actually have something very generic, like:

def non_adjacent_controlled_mat(U, wires, num_wires): ...

And then return the matrix where C-U is applied to wires in a num_wires-qubit system. (Basically, what I have implemented in the DoubleExcitation test; and similarly for a random single-qubit gate sitting on a qubit amidst other qubits, if that doesn't already exist somewhere).

I will make an issue for this so we don't forget to address it in the future 👍

tests/ops/test_qubit_ops.py

antalszava · 2021-05-13T19:23:24Z

tests/ops/test_qubit_ops.py

+        P0 = np.array([[1, 0], [0, 0]])
+        P1 = np.array([[0, 0], [0, 1]])


How about moving these into gate_data and maybe naming them state_zero_projector and state_one_projector?

The pro is that it'll make the test shorter, although it will also be less explicit. So if it would look less readable (as the names are not descriptive enough), then leaving this as is the way to go.

I didn't know about gate_data 😅 that's a good idea. These projectors are quite commonly used, so I think it's worth having them in there. I will call them perhaps StateZeroProjector and StateOneProjector to match the existing capitalization conventions if that works.

Co-authored-by: antalszava <antalszava@gmail.com>

ixfoduap

Looks great! @glassnotes: you make this look so easy! 🧙

ixfoduap · 2021-05-14T14:09:18Z

pennylane/ops/qubit.py

        decomp_ops = [
-            DoubleExcitationPlus(theta / 2, wires=wires),
-            DoubleExcitationMinus(theta / 2, wires=wires),
+            qml.CNOT(wires=[wires[2], wires[3]]),


Yikes! Makes this building block look less like a fundamental one 😅
Still, not too bad in the end!

ixfoduap · 2021-05-14T14:11:17Z

tests/ops/test_qubit_ops.py

-        decomposed_matrix = mats[0] @ mats[1]
+        """Tests that the SingleExcitation operation calculates the correct decomposition.
+
+        Need to consider the matrix of CRY separately, as the control is wire 1


Why is this change to the single excitation gate being done in this PR? Just sneaking it in?

It's because we are updating the gradient recipes for both gates to the four-term shift rules; so the SingleExcitation decomposition is adjusted accordingly.

tests/ops/test_qubit_ops.py

ixfoduap · 2021-05-14T14:17:58Z

tests/ops/test_qubit_ops.py

+
+        from functools import reduce
+
+        # To compute the matrix for CX on an arbitrary number of qubits, use the fact that


Co-authored-by: ixfoduap <40441298+ixfoduap@users.noreply.github.com>

Olivia Di Matteo added 3 commits May 12, 2021 17:03

Adds decomposition to DoubleExcitationOperator.

c9be627

Add new (correct) decomposition and test.

bdc5784

Add four-term param shift rule for grad recipes.

2d84e80

Olivia Di Matteo and others added 3 commits May 13, 2021 10:48

Rewrite SingleExcitation decomposition and related test.

d5518c1

Update CHANGELOG.

491e4fe

Merge branch 'master' into decompose_double_excitations

5e4ce2d

glassnotes marked this pull request as ready for review May 13, 2021 14:53

Merge branch 'master' into decompose_double_excitations

a18eaea

glassnotes changed the title ~~[WIP] Adds decomposition to DoubleExcitation operator.~~ [WIP] Adds decomposition to DoubleExcitation operator (and update gradient recipes) May 13, 2021

glassnotes commented May 13, 2021

View reviewed changes

tests/ops/test_qubit_ops.py Outdated Show resolved Hide resolved

Update tests/ops/test_qubit_ops.py

38b705f

glassnotes commented May 13, 2021

View reviewed changes

tests/ops/test_qubit_ops.py Outdated Show resolved Hide resolved

glassnotes commented May 13, 2021

View reviewed changes

tests/ops/test_qubit_ops.py Outdated Show resolved Hide resolved

glassnotes and others added 3 commits May 13, 2021 10:57

Apply suggestions from code review

26b3c3a

Fix docstring.

39e071b

Merge branch 'decompose_double_excitations' of https://github.com/Pen…

b79e655

…nyLaneAI/pennylane into decompose_double_excitations

glassnotes requested review from ixfoduap and antalszava May 13, 2021 15:57

glassnotes commented May 13, 2021

View reviewed changes

tests/ops/test_qubit_ops.py Outdated Show resolved Hide resolved

Update tests/ops/test_qubit_ops.py

9c11ad4

antalszava approved these changes May 13, 2021

View reviewed changes

glassnotes and others added 3 commits May 13, 2021 17:41

Apply suggestions from code review

5201a81

Co-authored-by: antalszava <antalszava@gmail.com>

Add projectors to gate_data and include in test.

f4783ef

Merge branch 'master' into decompose_double_excitations

c777d7e

ixfoduap approved these changes May 14, 2021

View reviewed changes

Update tests/ops/test_qubit_ops.py

0554aea

Co-authored-by: ixfoduap <40441298+ixfoduap@users.noreply.github.com>

glassnotes mentioned this pull request May 14, 2021

Add utility functions for matrix construction of controlled gates on non-adjacent wires #1313

Closed

glassnotes merged commit 86e427d into master May 14, 2021

glassnotes deleted the decompose_double_excitations branch May 14, 2021 14:38

glassnotes mentioned this pull request May 14, 2021

Define decomposition for single and double excitation operations #1275

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[WIP] Adds decomposition to `DoubleExcitation` operator (and update gradient recipes) #1303

[WIP] Adds decomposition to `DoubleExcitation` operator (and update gradient recipes) #1303

glassnotes commented May 12, 2021 •

edited

Loading

codecov bot commented May 13, 2021 •

edited

Loading

glassnotes May 13, 2021

dwierichs May 13, 2021

antalszava left a comment

antalszava May 13, 2021

antalszava May 13, 2021

glassnotes May 13, 2021

glassnotes May 14, 2021

antalszava May 13, 2021

glassnotes May 13, 2021

ixfoduap left a comment

ixfoduap May 14, 2021

ixfoduap May 14, 2021

glassnotes May 14, 2021

ixfoduap May 14, 2021

		P0 = np.array([[1, 0], [0, 0]])
		P1 = np.array([[0, 0], [0, 1]])


		from functools import reduce

		# To compute the matrix for CX on an arbitrary number of qubits, use the fact that

[WIP] Adds decomposition to DoubleExcitation operator (and update gradient recipes) #1303

[WIP] Adds decomposition to DoubleExcitation operator (and update gradient recipes) #1303

Conversation

glassnotes commented May 12, 2021 • edited Loading

codecov bot commented May 13, 2021 • edited Loading

Codecov Report

Choose a reason for hiding this comment

Choose a reason for hiding this comment

antalszava left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ixfoduap left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

[WIP] Adds decomposition to `DoubleExcitation` operator (and update gradient recipes) #1303

[WIP] Adds decomposition to `DoubleExcitation` operator (and update gradient recipes) #1303

glassnotes commented May 12, 2021 •

edited

Loading

codecov bot commented May 13, 2021 •

edited

Loading