Fix: Allow subtraction of IdentityGate in PauliSum.__sub__

[RevanthGundala]

Summary

Fixes #7120

This PR fixes a bug where subtracting a GateOperation with an IdentityGate (e.g., cirq.I(q) - cirq.Z(q)) would raise a TypeError instead of returning a valid PauliSum.

Details

Updated PauliSum.sub in cirq-core/cirq/ops/linear_combinations.py to promote GateOperation(IdentityGate) to a PauliString() when performing subtraction.
This ensures arithmetic like cirq.I(q) - cirq.Z(q) now works as expected and matches the behavior for addition.
The fix addresses issue # (link to the original issue if possible).

Reproduction

Before this fix, the following would raise an error:

import cirq
q = cirq.LineQubit(0)
result = cirq.I(q) - cirq.Z(q)  # TypeErro

[daxfohl]

Lgtm, follows the existing pattern from the add method. Once you add unit tests for the test cases from the linked issue, should be good to merge.

[RevanthGundala]

Pushed the unit test from the linked issue

[daxfohl]

I think I+I and I-I will still fail here, for reasons described in the linked issue thread. If so, probably have to add a conditional check here to skip those cases.

[daxfohl]

Huh. Surprisingly I+I passes now even though it didn't in the linked issue and this PR didn't change __add__. Any idea what changed?

[RevanthGundala]

My test wraps all gates in PauliString, so operations like PauliString(I(q)) + PauliString(I(q)) are handled correctly and the test passes. If you use raw I(q) + I(q) (without PauliString), it will fail, but that's not the intent of this test.

[daxfohl]

Oh I see. But wouldn't those all also pass prior to your change? IIUC the change fixes unwrapped paulis by implicitly wrapping them during subtraction. So IIUC this test where they're explicitly being wrapped beforehand wouldn't actually test the fix. Am I missing something?

[RevanthGundala]

No you were right. Those tests didn't test anything new. Just added new tests that should cover the changes. Let me know if that looks good.

In terms of the I+I, it's still not passing on my end.

[daxfohl]

Yeah I+I being out of scope was discussed in the linked issue. It might not be that hard to create an IdentityGate.__add__/__sub__ that replaces self with an empty PauliString first (or maybe DensePauliString since it's a gate?). Now that I think of it, that approach might even remove the need for these instance checks in PauliSum.__add__/__sub__? Or it may not work at all and/or break some other tests. Haven't tried.

If you want to give that a shot and see if it works, go ahead. Otherwise, this looks good, just open up a separate issue to track the I+I case and mention the issue number in your unit test's comment about bypassing the I+I case.

[daxfohl]

FYI, check/format-incremental( --apply), check/pylint-changed-files, check/pytest(-changed-files), check/mypy, check/all from the command line are useful for avoiding having to wait for CI just to find silly mistakes. https://github.com/quantumlib/Cirq/tree/main/check

[RevanthGundala]

Sounds good, thanks! Updated the new issue to #7280

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.66%. Comparing base (c054b90) to head (603e696).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #7276      +/-   ##
==========================================
- Coverage   98.66%   98.66%   -0.01%     
==========================================
  Files        1106     1106              
  Lines       96086    96097      +11     
==========================================
+ Hits        94808    94818      +10     
- Misses       1278     1279       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[daxfohl]

Lgtm cc @pavoljuhas