Context-aware dynamical decoupling

[Cryoris]

Summary

Implement context-aware dynamical decoupling (DD) from arXiv:2403.06852.

Details and comments

When selecting decoupling sequences for an idle qubit, context-aware DD takes into account what actions are occuring on neighboring qubits. It ensures that neighboring sequences are mutually orthogonal (to avoid collisions, like in staggered DD) but also whether the qubit is a CX or ECR target/control spectator. This is nicely summarized in a figure of the paper, where the different wire colors denote different decoupling sequences:

This particularly leads to improvement if a lot of qubits are idle at the same time or if they are next to ECR/CX gates. One important category of circuits in this class are the layers in twirled circuits. This pass also works best if the ECR/CX gate times are the same for all connections, which is almost always the case on IBM devices.

Some simple experiments show that this pass performs better than the current standard in Qiskit, PadDynamicaDecoupling with an X-X sequence. Here's some basic benchmarks, see the paper for more:

This code is based off @kdk and @ajavadia 🙂

[qiskit-bot]

One or more of the following people are relevant to this code:

• @Qiskit/terra-core

[coveralls]

Pull Request Test Coverage Report for Build 11685932835

Warning: This coverage report may be inaccurate.

This pull request's base commit is no longer the HEAD commit of its target branch. This means it includes changes from outside the original pull request, including, potentially, unrelated coverage changes.

• For more information on this, see Tracking coverage changes with pull request builds.
• To avoid this issue with future PRs, see these Recommended CI Configurations.
• For a quick fix, rebase this PR at GitHub. Your next report should be accurate.

Details

• 332 of 355 (93.52%) changed or added relevant lines in 4 files are covered.
• 30 unchanged lines in 5 files lost coverage.
• Overall coverage increased (+0.03%) to 88.802%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
qiskit/transpiler/passes/scheduling/padding/context_aware_dynamical_decoupling.py	329	352	93.47%

Files with Coverage Reduction	New Missed Lines	%
crates/accelerate/src/unitary_synthesis.rs	1	92.2%
qiskit/primitives/backend_sampler_v2.py	2	98.53%
crates/qasm2/src/lex.rs	6	91.98%
qiskit/primitives/backend_estimator.py	9	96.12%
crates/qasm2/src/parse.rs	12	97.15%

Totals
Change from base Build 11669445061:	0.03%
Covered Lines:	77120
Relevant Lines:	86845

---

💛 - Coveralls

[miamico]

Anything holding this up from being merged? Happy to lend a hand if needed to speed up the process @Cryoris @eliarbel

[eliarbel]

Thanks Julien, overall seems well written and documented. I didn't read the paper so the review is not about the actual algorithmic part of the DD method itself. I suggest getting input from Kevin or Ali if you want another pair of eyes on this aspect.
I've left line comments, most of which are minor but still require addressing.
In terms of performence - do you have a sense about the overhead of the block collection and manipulation (e.g. _collect_adjacenet_delay_blocks )?

[eliarbel]

There are several validation checks across the code. I suggest enabling and using them in test_context_aware_dd.py

[ElePT]

Thanks @Cryoris, as you know I have ran this pass and noticed a few painpoints I wanted to point out. In addition to the documentation suggestions, I have noticed that min_joinable_duration is a key parameter to avoid 1q gate explosion in some circuits, I wonder if you had considered setting it no a non-0 default value?

[ElePT]

I would find it very helpful if the documentation mentioned something that could be linked easier to "minimal delay time" and what units are used, something like (I still find it a bit clunky):

Suggested change

	min_joinable_duration: Do not join delays with others below this duration. This
	can be useful, e.g. if a big delay block would be interrupted and split into
	smaller blocks due to a very short, adjacent delay.
	min_joinable_duration: Minimal duration to join delays (in dt units). Do not join delays with
	others below this duration . This can be useful, e.g. if a big delay block would be interrupted
	and split into smaller blocks due to a very short, adjacent delay.

[ElePT]

In what units is skip_dd_threshold? I have noticed that min_joinable_duration is an int but this one is a float, so maybe it's not dt?

[ElePT]

Same comment about units

[ElePT]

Another comment, I believe there is a bug in the instruction duration assignation, because the timeline drawer fails after running this pass complaining about missing gate durations:

Traceback (most recent call last):
  File "/Users/ept/qiskit_workspace/qiskit/debug_dd.py", line 84, in <module>
    timeline_drawer(bound_scheduled_circuit, filename="out.png")
  File "/Users/ept/qiskit_workspace/qiskit/qiskit/utils/deprecation.py", line 183, in wrapper
    return func(*args, **kwargs)
  File "/Users/ept/qiskit_workspace/qiskit/qiskit/utils/deprecation.py", line 183, in wrapper
    return func(*args, **kwargs)
  File "/Users/ept/qiskit_workspace/qiskit/qiskit/visualization/timeline/interface.py", line 379, in draw
    canvas.load_program(program=program)
  File "/Users/ept/qiskit_workspace/qiskit/qiskit/visualization/timeline/core.py", line 193, in load_program
    for datum in gen(gate_source, formatter=self.formatter):
  File "/Users/ept/qiskit_workspace/qiskit/qiskit/visualization/timeline/generators.py", line 158, in gen_sched_gate
    if gate.duration > 0:
TypeError: '>' not supported between instances of 'NoneType' and 'int'

[eliarbel]

Thanks @Cryoris for applying all the changes. This looks good and I have no further comments. There are still some comments from @ElePT.

[eliarbel]

Fine from my side. Please address the remaining comments from @ElePT