Update pass_operations_over with fixed conjugated_by #7123
Conversation
babacry
force-pushed
the
conj
branch
2 times, most recently
from
a21cd5a to
4292cea
Compare
babacry
changed the title
| _assert_pass_over([op0], ps_before, ps_after) | ||
|
|
||
| op0 = cirq.PauliInteractionGate(Y, t_or_f1, X, t_or_f2)(q0, q1) | ||
| ps_before = cirq.PauliString({q0: Z, q2: Y}, sign) | ||
| ps_after = cirq.PauliString({q0: Z, q2: Y, q1: X}, sign) | ||
| ps_after = cirq.PauliString({q0: Z, q2: Y, q1: X}, -sign if t_or_f2 else sign) |
There was a problem hiding this comment.
Previous test case doesn't preserve unitaries of expected PauliInteractionGate here).
It's either a deeper issue from the e.g., decompose func of PauliInteractionGate or the implementation error of the old implementation of pass_operations_over.
babacry
changed the title
…y preserve the order of qubits.
|
|
||
| # Initialize the conjugation of Pc. | ||
| conjugated: 'cirq.DensePauliString' = ( | ||
| dense_pauli_string.DensePauliString(pauli_mask=[identity.I for _ in op.qubits]) | ||
| * self.coefficient | ||
| * ps.coefficient |
There was a problem hiding this comment.
This is a bug of the previous pr I submitted. We should roll against ps in the iteration, coeff should be from ps not self, found this by enforcing unitary checks in tests for all conjugated_by tests, previously if expected is given, unitary isn't checked. Fix it here.
* Avoid broken wheels for qcs-sdk-python The installation of `qcs-sdk-python==0.21.13` fails on Linux Python 3.10 with `zipfile.BadZipFile: Bad CRC-32 for file 'qcs_sdk/qcs_sdk.cpython-310-x86_64-linux-gnu.so'` Block new versions of qcs-sdk-python until this is fixed. Ref: rigetti/qcs-sdk-rust#531 * Also adjust Dockerfile to avoid the bad wheel
… necessary for the test
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #7123 +/- ##
==========================================
- Coverage 98.15% 98.12% -0.04%
==========================================
Files 1093 1093
Lines 95420 95427 +7
==========================================
- Hits 93664 93637 -27
- Misses 1756 1790 +34 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
|
@Strilanc , I noticed #2351 when I tried to fix conjugation related crashes / failures in PauliString. Fortunately, we have tableau repr of Clifford now, so I rewrite conjugated_by with Clifford gates' tableau. I figure you have more context of the conjugation, the implementation histories and their usages, wonder if you could help review this and point out the places need to be adjusted? Thanks! |
conjugated_by(fixed in Fix conjugated_by of PauliString. #7065) frompass_operations_overbased on tableau repr of Cliffords.Context1:
pass_operations_overis supposed to be replaced by conjugated_by as being described in #2351. Unless it is widely used, we should deprecate this function as the order of the CliffordOps input is very confusing (like being suggested in #2351).Context2 (existing implementations failed in a couple of cases, including some test cases): current implementation of
pass_operations_overmight crash like #6946. And it produced wrong results for e.g.,(Z(q0)*Y(q1)).pass_operations_over([PauliInteractionGate(Z, True, X, True)(q0, q1)])=Y(q1), but should be-Y(q1).Followup is to start deprecating
pass_operations_over. (I choose to not start deprecating the function as I want to make sure all the pass_operations_over tests would pass under the new implementation, a separate pr to isolate the deprecation process would be cleaner given that there are dependencies in the code base)Fix previous tests of
pass_operation_over, 24 test cases in the previous unit test doesn't preserve unitaries using previous implementations ofpass_operation_over. The CliffordOps in those failed test cases are constructed fromPauliInteractionGatewith different parameters. Updated the test so unitaries are preserved.Also fixed a bug in conjugated_by.
Nit fixes to pauli_string.py. e.g., remove
import cirq.