Pluggable Clifford synthesis for RB circuits #1288

itoko · 2023-10-17T12:14:09Z

Summary

Make the Clifford synthesis algorithm for RB circuits pluggable (implementing it as a HighLevelSynthesisPlugin).
Fixes #1279 and #1023.
Change to accept Clifford elements consisting only of instructions supported by the backend for interleaved_element option in InterleavedRB.
Speed up 2Q RB/IRB for backends with unidirectional 2q gates, e.g. IBM's 127Q Eagle processors.

Details and comments

Previously, for 3Q+ RB circuits, entire circuit is transpiled at once and hence for each of the resulting Cliffords, the initial and the final layout may differ, that means sampled Cliffords are changed during the transpilation. Also in the worst case, the resulting circuit may use physical qubits not in the supplied physical_qubits. To avoid that, this commit changes to transpile an RB sequence Clifford by Clifford. The Clifford synthesis algorithm (rb_default) is implemented as a HighLevelSynthesisPlugin (see RBDefaultCliffordSynthesis in clifford_synthesis.py), which forces the initial layout (i.e. guarantees the initial layout = the final layout) and physical qubits to use. As a byproduct, the performance of 2Q RB/IRB for backends with directed 2q gates (e.g. IBM's 127Q Eagle processors) is drastically improved. For those cases, previously we had to rely on transpile function to make generated circuits comply with the coupling map, however, after this commit, we can synthesize Cliffords with considering the 2q-gate direction and go through the fast path introduced in #982.

Depends on Qiskit/qiskit#10477 (qiskit 0.45)

itoko · 2023-10-26T09:13:24Z

No performance regression by this commit:

Comparable with the main branch for 1q and 2q
About 2x speed-up for 3q (by the improvement in cafe761)

itoko · 2023-10-26T09:14:36Z

qiskit_experiments/library/randomized_benchmarking/clifford_synthesis.py

+            QiskitError: If basis_gates is not supplied.
+        """
+        # synthesize cliffords
+        circ = synth_clifford_full(high_level_object)


I’ve examined synth_clifford_depth_lnn (as suggested by @alexanderivrii) using random RB circuits on 3 linear qubits (i.e. no need for SabreSwap routing) and found that it is a bit faster as expected but surprisingly it produces circuits with more CNOT counts and deeper depths (about 2x/1.5x) than synth_clifford_full + SabreSwap. So I’ll stick to synth_clifford_full in the PR.

synth_clifford_depth_lnn:

total cx counts = 2961, depth = 4418, max_length=10, time_transpiled_circuits: 1.233635034004692 total cx counts = 6286, depth = 9193, max_length=20, time_transpiled_circuits: 1.9450378390029073 total cx counts = 8659, depth = 12850, max_length=30, time_transpiled_circuits: 2.565633656005957 total cx counts = 11074, depth = 16313, max_length=40, time_transpiled_circuits: 3.270593570006895 total cx counts = 13727, depth = 20190, max_length=50, time_transpiled_circuits: 3.964288625997142

synth_clifford_full:

total cx counts = 1589, depth = 3092, max_length=10, time_transpiled_circuits: 1.8188774910086067 total cx counts = 2915, depth = 6009, max_length=20, time_transpiled_circuits: 2.883968324007583 total cx counts = 4120, depth = 8318, max_length=30, time_transpiled_circuits: 3.982568758990965 total cx counts = 5439, depth = 10996, max_length=40, time_transpiled_circuits: 4.483078286997625 total cx counts = 6493, depth = 13110, max_length=50, time_transpiled_circuits: 5.633502009004587

synth_clifford_depth_lnn is asymptotically better as it guarantees a total cx-depth bounded by 7n+2 (when n is large), but for n=3 qubits perhaps it's less useful.

ShellyGarion · 2023-10-31T13:05:03Z

Note that for the final inverse Clifford, you can reduce the depth and CX count by using synth_stabilizer_depth_lnn or synth_stabilizer_layers, since you don't need the entire Clifford, but only part of it. See an example in the following test:

https://github.com/Qiskit/qiskit/blob/7c89b68668b6ba74926bf767293dc8dd51ba3cfb/test/python/synthesis/test_stabilizer_synthesis.py#L78

itoko · 2023-11-09T01:04:01Z

@ShellyGarion Thanks, good to know, however, I think it's too much for RB to have one more Clifford synthesis option only for the last inversion Clifford.

…m pulse gates

alexanderivrii

I have looked at all of the code and everything looks good. Personally I have especially liked the idea of creating coupling-map and basis-gates aware synthesis plugins by running any synthesis routine, followed by SabreSwap , undoing layout change, and running BasisTranslation.

I would be also interested in hearing your feedback on the high level synthesis interface - do you think it could be simplified or extended?

alexanderivrii · 2024-01-03T11:40:38Z

qiskit_experiments/library/randomized_benchmarking/clifford_utils.py

+    qc = QuantumCircuit(clifford.num_qubits, name=str(clifford))
+    qc.append(clifford, qc.qubits)
+    return _synthesize_clifford_circuit(
+        qc,
+        basis_gates=basis_gates,
+        coupling_tuple=coupling_tuple,
+        synthesis_method=synthesis_method,
+    )


This is really minor. For a clifford over 3+ qubits, the code would first construct a quantum circuit with this clifford, then convert it back to Clifford (as done on line 191) and then against construct a quantum circuit with this new clifford (as also done on line 191). Possibly these extra translations may be avoided.

I think there's a slight misunderstanding. The extra translations you mentioned happen only when synthesizing a 1- or 2-qubit Clifford using a custom synthesis method. I think it's rare and users who has a Clifford circuit can directly call _synthesize_clifford_circuit as I do so here. Also _synthesize_clifford is used only in the constructor of InterleavedRB and not repeatedly called so much. I think there is no clear need for its performance improvement at least for now.

Somewhat related to this, if HighLevelSynthesis had a method such that it takes a Clifford and returns a QuantumCircuit (namely synthesize?), it would be very helpful. Actually, if it was, I didn't need to wrap a Clifford with a QuantumCircuit here just because HighLevelSynthesis.run requires a DAGCircuit and support only the QuantumCircuit-to-QuantumCircuit conversion through PassManager.run.

alexanderivrii · 2024-01-03T11:48:37Z

qiskit_experiments/library/randomized_benchmarking/clifford_utils.py

+    # for 3q+ or custom synthesis method, synthesizes clifford circuit
+    hls_config = HLSConfig(clifford=[(synthesis_method, {"basis_gates": basis_gates})])
+    pm = PassManager([HighLevelSynthesis(hls_config=hls_config, coupling_map=coupling_map)])
+    circuit = pm.run(circuit)


This is great! You are using all of the functionality offered by the high level synthesis interface, including passing additional options (like "basis_gates") for the plugin. Not related to the PR, but I would like to hear your feedback on this: do you think the interface could be simplified? extended?

Thank you for providing this useful HighLevelSynthesisPlugin interface. I like it and it's sufficient for this PR.
Let me share a minor point that could be simplified. I was a bit confused with the fact that HighLevelSynthesisPlugin.run takes over coupling_map option set up via HighLevelSynthesis.__init__, but it does not take over basis_gates option set up via HighLevelSynthesis.__init__, so I had to set it via hls_config here. Is this (run does not have basis_gates option) intentional?
It was not clear to me which options should be given through hls_config (HLSConfig) and which options should be given directly (like coupling_map) in the current API. Probably, more explanation on that point in the API reference doc would be helpful for developers like me.

As for an extension, it may be worth considering an introduction of HighLevelSynthesis.synthesize for the ease of custom Clifford-to-QuantumCircuit conversion as I mentioned above.

alexanderivrii · 2024-01-03T12:03:06Z

qiskit_experiments/library/randomized_benchmarking/clifford_synthesis.py

+        pm = PassManager(
+            [
+                SabreSwap(coupling_map),
+                undo_layout_change,
+                BasisTranslator(sel, basis_gates),
+                CheckGateDirection(coupling_map),
+            ]
+        )
+        pm.append([GateDirection(coupling_map)], condition=_direction_condition)
+        pm.append([Optimize1qGatesDecomposition(basis=basis_gates)])


This is clever. This allows to turn any synthesis plugin into one that adheres to the coupling map and to the basis gates.

coruscating

I'm good with this, just some minor comments.

releasenotes/notes/plugable-rb-clifford-synthesis-0e66c62fa3088fba.yaml

coruscating · 2023-12-19T20:32:20Z

requirements.txt

@@ -1,6 +1,6 @@
 numpy>=1.17
 scipy>=1.4
-qiskit>=0.44
+qiskit>=0.45


With the version bump here, the qiskit>=0.45.0 pin in requirements-dev.txt can be removed.

Addressed at 3361a6a

releasenotes/notes/plugable-rb-clifford-synthesis-0e66c62fa3088fba.yaml

ShellyGarion · 2024-01-07T13:42:44Z

test/library/randomized_benchmarking/test_clifford_utils.py

+        """Check if clifford synthesis does not change Clifford"""
+        basis_gates = tuple(["rz", "h", "cz"])
+        coupling_tuple = (
+            None if num_qubits == 1 else tuple((i, i + 1) for i in range(num_qubits - 1))


Perhaps it's worth to test also a non-linear connectivity (for 4 qubits).
What happens if the qubits are not near each other? do you get an error?

Added such test cases at 3efc58a and confirmed that it works well with non-linear (star and cycle) connectivities.
I don't understand what you mean by "the qubits are not near each other". Does it mean "disconnected" (e.g. ((0, 1), (2, 3))) or "not full" (e.g. (0, 2), (2, 3), (3, 4))? In both cases, _synthesize_clifford fails with TranspilerError: The circuit has an invalid layout as two qubits need to interact in disconnected components of the coupling map... or Fewer qubits in the circuit (4) than the coupling map (5).... I think these are expected results for users.

Co-authored-by: Helena Zhang <[email protected]>

### Summary Add a new experiment class to measure layer fidelity, which is a holistic benchmark to characterize the full quality of the devices at scale (https://arxiv.org/abs/2311.05933) Example notebook: [run_lf_qiskit_experiments_large.ipynb (Gist)](https://gist.github.com/itoko/28c7cc117614c67e2a1899a3757d4ad1) ### Experimental features: - Exceptionally `circuits()` method returns circuits on physical qubits (not virtual qubits as usual) - Add `reason` as an extra analysis result entry to tell users why the `quality` of the analysis was "bad" ### Follow-up items: - Add API for customizing DD (e.g. register DD sequence generator by name and specify the name in experiment_options) ``` def dd_func1(delay_length, backend) -> list[Instruction]; LayerFidelity.dd_functions = { "dd1": dd_func1, "dd2": dd_func2, } ``` ### Features decided not to include: - `full_sampling` option (`StandardRB` has). `LayerFidelity` behaves as if setting always `full_sampling==True` to avoid correlation between sequences, which RB theory does not consider. - `replicate_in_parallel` option that allows to use a common direct RB sequence for all qubit pairs. It turned out that `replicate_in_parallel==True` may underestimate some types of errors, suggesting it should always be set to `False`. ### Issues to be addressed separately - Poor interface for querying figures: No pointer to a relevant figure (or data used for fitting) is stored in `AnalysisResult` (i.e. users who find a bad fitting result in `AnalysisResult` cannot easily look at a figure relevant to the result) ==> For now, you can query a figure by its name; e.g. `exp_data.figure("DirectRB_Q83_Q82")` ### Dependencies: - [x] #1288 for Clifford synthesis with unidirectional 2q gates (e.g. IBM Eagle processors)

itoko mentioned this pull request Oct 17, 2023

Cleaner specification for RB experiments #1030

Closed

itoko added the on hold On hold until something else is done. label Oct 24, 2023

itoko force-pushed the plugable-rb-clifford-synthesis branch 3 times, most recently from 7662b10 to b04681d Compare October 24, 2023 01:03

itoko commented Oct 26, 2023

View reviewed changes

wshanks mentioned this pull request Oct 30, 2023

Allow StandardRB to handle a V2 backend without a coupling map #1293

Merged

itoko force-pushed the plugable-rb-clifford-synthesis branch from d707f5c to fd9c575 Compare November 9, 2023 03:33

itoko removed the on hold On hold until something else is done. label Nov 9, 2023

itoko force-pushed the plugable-rb-clifford-synthesis branch 2 times, most recently from 1d822f4 to cdb9483 Compare November 9, 2023 06:00

wshanks mentioned this pull request Nov 16, 2023

Replace qiskit-terra references with qiskit references #1318

Merged

itoko force-pushed the plugable-rb-clifford-synthesis branch from cdb9483 to 0923c0e Compare November 28, 2023 13:37

itoko added 9 commits November 28, 2023 22:53

Make RB Clifford synsthesis plugable

e6e42fe

Change allowed interleaved elements in InterleavedRB

ce9388b

Clean up codes and docs

a696f5f

Fix a bug where calibrations are not added to RB circuit for 3q custo…

241750a

…m pulse gates

Improve performance

bbb16e9

Fix argumentis mismatch

a6845a2

Fix dependent qiskit version

5c227c7

Fix for supporting BackendV2 simulators

4949853

Add release note

23277cb

itoko force-pushed the plugable-rb-clifford-synthesis branch from 0923c0e to 23277cb Compare November 28, 2023 13:54

itoko mentioned this pull request Nov 28, 2023

Add layer fidelity experiment #1322

Merged

1 task

itoko and others added 3 commits November 29, 2023 22:36

Use Target more safely

d51b456

Merge branch 'main' into plugable-rb-clifford-synthesis

76b736d

Merge branch 'main' into plugable-rb-clifford-synthesis

33dddf8

itoko added a commit to itoko/qiskit-experiments that referenced this pull request Nov 30, 2023

Fixes for backends with ECRGate employing qiskit-community#1288

2846b0f

itoko requested a review from wshanks November 30, 2023 01:21

itoko requested review from coruscating and removed request for wshanks December 18, 2023 05:43

itoko added a commit to itoko/qiskit-experiments that referenced this pull request Dec 22, 2023

Fixes for backends with ECRGate employing qiskit-community#1288

5ac51fe

alexanderivrii approved these changes Jan 3, 2024

View reviewed changes

coruscating approved these changes Jan 4, 2024

View reviewed changes

ShellyGarion reviewed Jan 7, 2024

View reviewed changes

itoko and others added 5 commits January 16, 2024 22:57

Apply suggestions from code review

c116a60

Co-authored-by: Helena Zhang <[email protected]>

Remove duplicated qiskit version constraint in requirements-dev.txt

3361a6a

Add more unittests

3efc58a

Remove unused code

93eff2e

Lint

5dbb32f

itoko added this pull request to the merge queue Jan 18, 2024

Merged via the queue into qiskit-community:main with commit 48c9e67 Jan 18, 2024
11 checks passed

itoko added a commit to itoko/qiskit-experiments that referenced this pull request Jan 19, 2024

Fixes for backends with ECRGate employing qiskit-community#1288

6530278

itoko added a commit to itoko/qiskit-experiments that referenced this pull request Feb 14, 2024

Fixes for backends with ECRGate employing qiskit-community#1288

e4a8510

itoko added a commit to itoko/qiskit-experiments that referenced this pull request Feb 29, 2024

Fixes for backends with ECRGate employing qiskit-community#1288

a6c0326

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Pluggable Clifford synthesis for RB circuits #1288

Pluggable Clifford synthesis for RB circuits #1288

itoko commented Oct 17, 2023 •

edited

Loading

itoko commented Oct 26, 2023

itoko Oct 26, 2023 •

edited

Loading

itoko Oct 26, 2023

ShellyGarion Jan 3, 2024

ShellyGarion commented Oct 31, 2023

itoko commented Nov 9, 2023

alexanderivrii left a comment

alexanderivrii Jan 3, 2024

itoko Jan 17, 2024 •

edited

Loading

itoko Jan 17, 2024

alexanderivrii Jan 3, 2024

itoko Jan 17, 2024

alexanderivrii Jan 3, 2024

coruscating left a comment

coruscating Dec 19, 2023

itoko Jan 17, 2024

ShellyGarion Jan 7, 2024

itoko Jan 17, 2024

Pluggable Clifford synthesis for RB circuits #1288

Pluggable Clifford synthesis for RB circuits #1288

Conversation

itoko commented Oct 17, 2023 • edited Loading

Summary

Details and comments

itoko commented Oct 26, 2023

itoko Oct 26, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ShellyGarion commented Oct 31, 2023

itoko commented Nov 9, 2023

alexanderivrii left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

itoko Jan 17, 2024 • edited Loading

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

coruscating 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

itoko commented Oct 17, 2023 •

edited

Loading

itoko Oct 26, 2023 •

edited

Loading

itoko Jan 17, 2024 •

edited

Loading