Add measure propagation rules by joshuasn · Pull Request #369 · Qiskit/samplomatic

joshuasn · 2026-05-28T20:23:09Z

Summary

This PR refactors how measure is treated during the samplex build process. Rather than using its own dedicated mechanism, this approach treats the measure as a propagation node.

Details and comments

Some Opus + Sonnet 4.6.

Add infrastructure for propagating virtual Paulis through measurements: - PreMeasurePropagate node type in pre_samplex graph - Z2->Pauli conversion in z2_register - add_measure_propagate() method on PreSamplex - Lowering logic (Fork A: Z2 output, Fork B: continued Pauli) - Twirl emission moved to parse(None) in LeftBoxBuilder with deferred leftward connection via connect_emit_leftward() Current state: the deferred leftward connection works correctly (same graph topology as original for the left collector). However, emitting the twirl's rightward dangler at parse(None) causes hard gates to propagate it rightward, changing what the right collector receives. This causes 52 integration test failures. The fundamental tension is that measurements need the rightward dangler to exist during parse(), but hard gates should NOT propagate it rightward (they should only create leftward propagation nodes for the adjacent right box). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

ihincks

Thanks for the PR, @joshuasn !

I'd love to see some more integration tests, it seems like we could add some neat ones.

Also, I'm not sure what's going on with the node serializer change, and I'm wondering if that should be pulled out into a separate PR?

ihincks · 2026-06-04T16:14:16Z

                "new_name": obj.new_name,
                "new_type": obj.new_type.value,
-                "num_subsystems": obj.num_subsystems,
+                "num_subsystems": str(obj.num_subsystems),


Is this a bugfix? If so, it should receive a changelog entry. Also, how was it working before??

Changed the lowering rule to use SliceRegisterNode for now, bugfix to follow.

ihincks · 2026-06-04T16:19:31Z

        assert samplex == samplex_new

-    @pytest.mark.parametrize("ssv", SUPPORTED_SSVS)
+    @pytest.mark.parametrize("ssv", [s for s in SUPPORTED_SSVS if s >= 3])


This change and the one below it are suspicious. They make it look like this PR breaks serialization for ssv=1 and ssv=2.

ihincks · 2026-06-04T18:20:34Z

-        style = super().get_style().append_data("Direction", self.direction.name)
-        style.marker = "bowtie"
-        style.color = "purple"
+    def to_key(self) -> Hashable:


Can you remind me who uses it? I can't help but notice the other ones includes things like direction, and I'm wondering if that's relevant here

merge_parallel_pre_propagate_nodes and replaced it with a merge_parallel_nodes method that acts on all the nodes of a given type. Direction isn't relevant for measure because it only propagates rightwards. The other attributes are lists of names and offsets which we simply combine then the lowering handles dispatching to separate collect nodes.

We could also rewrite the merging so each pre-measure has one register name.

ihincks · 2026-06-04T18:21:44Z

+
+    @classmethod
+    def from_cluster(cls, nodes: list["PreMeasure"]) -> "PreMeasure":
+        combined_subsystems = QubitIndicesPartition.union(*(n.subsystems for n in nodes))


.union is order preserving?

yeah for QubitPartition

ihincks · 2026-06-16T20:22:39Z

@@ -0,0 +1,2 @@
+Measurement twirling is now implemented by treating measure operations as virtual gate propagators. Previously, measurement twirling was implemented by collecting bit flips associated with virtual gates exclusively in the right hand side of a left-dressed box. The new approach allows for more flexible measurement twirling, e.g., sequential measurements of the same qubit to different clbits or measurements interwoven with other operations in the same box.


It looks like the changes also remove " measurements interwoven with other operations in the same box". Is that because this is not new behaviour? I see for example this existing integration test:

def test_gates_and_measure_all(self, save_plot): circuit = QuantumCircuit(3) with circuit.box([Twirl(dressing="left")]): circuit.x(0) circuit.h(1) circuit.cx(0, 1) circuit.measure_all() sample_simulate_and_compare_counts(circuit, save_plot)

ihincks · 2026-06-17T13:47:42Z

 from samplomatic.ssv import SSV

-SUPPORTED_SSVS = set(range(1, SSV))
+SUPPORTED_SSVS = set(range(1, SSV + 1))


good catch, ty

ihincks · 2026-06-17T13:51:35Z

+        with circuit.box([Twirl(dressing="right")]):
+            circuit.noop(range(2))
+
+        _, samplex = build(circuit)
+        samplex_new = samplex_from_json(samplex_to_json(samplex, ssv=ssv))
+        samplex_new.finalize()
+
+        assert samplex == samplex_new
+
+    @pytest.mark.parametrize("ssv", SUPPORTED_SSVS)
+    def test_measure_twirl_circuit(self, ssv):
+        """Test a circuit with measurement twirling."""
+        circuit = QuantumCircuit(2)


Okay, one last question on the serialization: can we now just undo all changes to these tests that this PR makes (aside from the fix SSV -> SSV+1)? I'm not sure I understand the benefit of splitting this into two cases, one with measurements, and one without, unless perhaps it exercises more nodes in the serial format?

I undid. It's not necessary anymore, it to ensure ChangeBasis circuits had tests in all SSVs with the restricted measure twirl SSVs.

joshuasn and others added 14 commits May 20, 2026 09:05

Change twirl location in left-dressed boxes.

d4b964a

Simplify check.

22e4cde

Add status summary for measure-twirl branch

292343a

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Merge branch 'main' into change-left-twirl-location

340ab80

Changelog.

002bd17

Merge branch 'main' into measure-twirl

c39b7f1

Merge branch 'change-left-twirl-location' into measure-twirl

29e3561

Merge branch 'main' into measure-twirl

4f2b5cf

Finishing up.

144b01a

Remove remnants of old implementation.

619c121

Updating docs.

ebeb4fa

Merge.

b44ff8f

Changelog.

ee95e5e

joshuasn marked this pull request as ready for review June 2, 2026 14:59

ihincks reviewed Jun 4, 2026

View reviewed changes

joshuasn added 4 commits June 5, 2026 16:14

Merge branch 'main' into measure-twirl

9c925f8

CR.

d86164d

Better changelog.

24e982f

Simplify.

fcd1527

ihincks reviewed Jun 16, 2026

View reviewed changes

Early exit for terminal measurement twirling in lowering.

de22a95

joshuasn requested a review from ihincks June 17, 2026 12:57

ihincks reviewed Jun 17, 2026

View reviewed changes

Simplify.

881cf11

ihincks approved these changes Jun 17, 2026

View reviewed changes

joshuasn merged commit bbd6efd into main Jun 17, 2026
8 checks passed

joshuasn deleted the measure-twirl branch June 17, 2026 14:28

		@@ -0,0 +1,2 @@
		Measurement twirling is now implemented by treating measure operations as virtual gate propagators. Previously, measurement twirling was implemented by collecting bit flips associated with virtual gates exclusively in the right hand side of a left-dressed box. The new approach allows for more flexible measurement twirling, e.g., sequential measurements of the same qubit to different clbits or measurements interwoven with other operations in the same box.

Conversation

joshuasn commented May 28, 2026

Summary

Details and comments

Uh oh!

ihincks left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants