MPS Synthesis #7752
MPS Synthesis #7752
Conversation
- Fixed `_parser.py` formatting issues.
- Added testers for `cp`, `c3x`, `c4x`, `csx`, and `c3sqrtx` gates. - Fixed u0 gate. - Removed `r`, `ryy`, `cu2`, and `iswap` as they are not part of qelib 1.
- Fixes gate definition for `id` to address git-blame note. - Fixes gate definition for `p` to use ZPowGate. - Adds rccx and rc3x using `cirq.MatrixGate`. - Adds missing testers for cry and crz.
- Fixed failed CIs.
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #7752 +/- ##
========================================
Coverage 99.57% 99.57%
========================================
Files 1102 1105 +3
Lines 98638 98778 +140
========================================
+ Hits 98218 98358 +140
Misses 420 420 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
|
I added single qubit state prep manually, it should be removed when cirq gets state preparation added. I'll be sure to remind in the next cirq cynq. |
|
I need to check a few things with colleagues; I will have an update early next week. |
pavoljuhas
left a comment
pavoljuhas
left a comment
There was a problem hiding this comment.
Please see inline comments. Otherwise LGTM, thank you for contributing this.
PS: I'd like to loop in @tanujkhattar for more comments after we converge on the initial review.
| derived from the input matrix. If a column is (approximately) zero, it | ||
| is replaced with a random vector. | ||
| """ | ||
| num_rows, num_columns = matrix.shape |
There was a problem hiding this comment.
The only caller of this function passes a square matrix.
Can we require the matrix to be square here?
There was a problem hiding this comment.
Okay if I add this for safety? I'll add it for now, and later before merge I can remove it if you don't want it.
if matrix.shape[0] != matrix.shape[1]:
raise ValueError(
"Input matrix is expected to be square. "
f"Got shape {matrix.shape} instead. "
"If this is not an error, please file an issue at "
"https://github.com/quantumlib/Cirq/issues"
)| unitary = np.zeros((num_rows, num_columns), dtype=np.complex128) | ||
| orthonormal_basis: list[NDArray[np.complex128]] = [] |
There was a problem hiding this comment.
Please rewrite using scipy.linalg.org and scipy.linalg.null_space
There was a problem hiding this comment.
I'm trying something like that but noticed a slight drop in fidelity.
# Current
Fidelity for N=5: 0.9763405705141897
Fidelity for N=8: 0.9376726644863744
Fidelity for N=10: 0.8896167855158122
Fidelity for N=11: 0.8809997802239913
# Requested
Fidelity for N=5: 0.9625124988989704
Fidelity for N=8: 0.8941036354519807
Fidelity for N=10: 0.8705322861238998
Fidelity for N=11: 0.8701612775480855
I followed the Gram Schmidt decomposition based on scipy's doc to get started for this. If I use scipy.linalg.qr directly (couldn't get SVD decomposition approaches like orth and null_space to work), it gets slightly worse than this, so I feel it'd be better to keep as is to get those slight extra fidelity. I can try other implementations, perhaps I'm missing something obvious.
If others have a recommended approach, would appreciate guidance on that as well.
| state = np.random.rand(2**N) + 1j * np.random.rand(2**N) | ||
| state /= np.linalg.norm(state) |
There was a problem hiding this comment.
Please reuse cirq.testing.random_superposition instead
There was a problem hiding this comment.
I checked, and random_superposition can have volume-law states which are hard to approximate with vanilla MPS (I'll add the sweeping I mentioned before soon, not in this PR though). Is it alright that for this test I keep it as is since the one I have there produces area-law states?
…s internal module only. - Added assertion to `_gram_schmidt` to raise an error should `matrix` not be square to catch edge cases not known at the time of changing this code. - Changed class name to `MPSSequential`. - Added class docstring. - Removed `convention` from class attributes. - Added indentation to docstring sections where needed. - Removed `logger` uses. - Added `mps_circuit_from_statevector` to `__init__.py` for ease of use. - Used `cirq.testing.random_superposition` in testers where possible, except `test_compile_area_law_states` which requires strictly area-law entangled states as part of the test. - Rewrote `test_compile_trivial_state_with_mps_pass` to avoid using qasm and avoid inter-module dependency. - Used `np.testing.assert_allclose` for about exact matching assertions.
|
Pushed commit that addressed as many comments as I could. 4 comments remain which need your kind insight. Main comment remaining is converging on optimal/preferred implementation for |
3 participants
Closes #7650 .