Add _apply_channel_ optimizations for reset and confusion #5917
Conversation
There was a problem hiding this comment.
Haven't looked at the details yet, but this is what I was hoping we would do! Thank you, Dax!
I have one concern. When we do things this way, i.e. when we have different code paths implementing the same operation to be employed under different circumstances, we should test that the code paths agree on a few test inputs. We do this for _unitary_ and _apply_unitary_ using this function. By analogy, we should have assert_has_consistent_apply_channel.
In fact, if you look at our present consistency checks for channels, there are surprisingly few. In fact, we have an issue - #4579 - about adding more (albeit the issue is about consistency between different representations such as Choi/Kraus/etc and doesn't mention _apply_channel_ explicitly).
How do you feel about adding a assert_has_consistent_apply_channel that would verify that _kraus_ and _apply_channel_ yield the same results? (I'd do it in a separate PR to be merged before this one so that we could make use of the new assert helper in here, but that's up to you of course.)
|
+1 I had the same concern. Thanks for the suggestion on the approach. I will try to do it as part of this PR. If it gets hairy I will open a separate PR for it. |
|
Done. I have it calculate the superoperator from the Kraus, and separately calculating the superoperator by calling apply_channel on the identity tensor, and then comparing the results of each. I also added an apply_channel for phase_damp. |
| return 1 | ||
|
|
||
| with pytest.raises(AssertionError): | ||
| cirq.testing.assert_has_consistent_apply_channel(NoKraus()) |
There was a problem hiding this comment.
Optional: I think it'd be nice to make this case work by having protocol.kraus compute the Kraus representation when _apply_channel_ is provided but _kraus_ isn't. Probably a good idea for a follow-up rather than this PR.
viathor
added
the
automerge
CirqBot
added
the
front_of_queue_automerge
CirqBot
removed
automerge
…#5917) Confusion matrix and reset channels both can be viewed as starting with a zero DM tensor and then copying (adding) in scaled slices from the original DM tensor. Thus we make a helper function that does this and add `_apply_channel_` optimizations to those gates. Fixes quantumlib#5901. Starts quantumlib#5900 though I haven't looked at all gates. Also starts quantumlib#4579 but there's likely more to do there as well. I didn't add the new test function to the primary test suite because creating superoperators is likely computationally expensive (granted most if not all gates that use this would be three or fewer qubits, which is still cheap), and the test not relevant for most gates.
Confusion matrix and reset channels both can be viewed as starting with a zero DM tensor and then copying (adding) in scaled slices from the original DM tensor. Thus we make a helper function that does this and add
_apply_channel_optimizations to those gates.Fixes #5901.
Starts #5900 though I haven't looked at all gates.
Also starts #4579 but there's likely more to do there as well. I didn't add the new test function to the primary test suite because creating superoperators is likely computationally expensive (granted most if not all gates that use this would be three or fewer qubits, which is still cheap), and the test not relevant for most gates.