Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[labs.dla 3] structure_constants_dense #6376

Open
wants to merge 45 commits into
base: dense_util
Choose a base branch
from

Conversation

Qottmann
Copy link
Contributor

Building on top of #6371

An implementation of structure_constants using dense matrices, which sometimes is advantageous in certain scenarios.

The projection routine can be optimized still

[sc-75525]

Copy link
Contributor

Hello. You may have forgotten to update the changelog!
Please edit doc/releases/changelog-dev.md with:

  • A one-to-two sentence description of the change. You may include a small working example for new features.
  • A link back to this PR.
  • Your name (or GitHub username) in the contributors section.

Copy link

codecov bot commented Oct 10, 2024

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.29%. Comparing base (0b886ad) to head (ee1cbfe).

Additional details and impacted files
@@             Coverage Diff             @@
##           dense_util    #6376   +/-   ##
===========================================
  Coverage       99.29%   99.29%           
===========================================
  Files             454      454           
  Lines           43262    43267    +5     
===========================================
+ Hits            42959    42964    +5     
  Misses            303      303           

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

@dwierichs dwierichs marked this pull request as draft October 10, 2024 15:06
@Qottmann Qottmann changed the title [labs] structure_constants_dense [labs.dla 3] structure_constants_dense Nov 11, 2024
@Qottmann Qottmann changed the base branch from lie_closure_dense to dense_util November 11, 2024 17:52
@Qottmann Qottmann marked this pull request as ready for review November 11, 2024 17:54
Copy link
Contributor

@dwierichs dwierichs left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice to see this polished up so neatly after using it internally already :D 💯
I again only had small comments, and one request for another test case.

pennylane/labs/dla/dense_util.py Outdated Show resolved Hide resolved
pennylane/labs/dla/lie_closure_dense.py Outdated Show resolved Hide resolved
pennylane/labs/dla/structure_constants_dense.py Outdated Show resolved Hide resolved
Comment on lines 65 to 87
@pytest.mark.parametrize("dla", [Ising3, XXZ3])
@pytest.mark.parametrize("use_orthonormal", [False, True])
def test_structure_constants_elements(self, dla, use_orthonormal):
r"""Test relation :math:`[i G_\alpha, i G_\beta] = \sum_{\gamma = 0}^{\mathfrak{d}-1} f^\gamma_{\alpha, \beta} iG_\gamma`."""

d = len(dla)
dla_dense = np.array([qml.matrix(op, wire_order=range(3)) for op in dla])
if use_orthonormal:
gram_inv = sp.linalg.sqrtm(
np.linalg.pinv(np.tensordot(dla_dense, dla_dense, axes=[[1, 2], [2, 1]]).real)
)
dla_dense = np.tensordot(gram_inv, dla_dense, axes=1)
dla = [(scale * op).pauli_rep for scale, op in zip(np.diag(gram_inv), dla)]
ad_rep = structure_constants_dense(dla_dense, is_orthonormal=use_orthonormal)
for i in range(d):
for j in range(d):

comm_res = 1j * dla[i].commutator(dla[j])
res = sum((c + 0j) * dla[gamma] for gamma, c in enumerate(ad_rep[:, i, j]))
res.simplify()
assert set(comm_res) == set(res) # Assert equal keys
if len(comm_res) > 0:
assert np.allclose(*zip(*[(comm_res[key], res[key]) for key in res.keys()]))
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think the tests are nice. We should add one on a non-orthonormal set of operators, though, that checks exactly this defining property of the structure constants :)

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

(Can do this just by applying some random coefficients to the existing DLAs.)

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I guess to be truly non-orthonormal we should also break the orthogonality, so doing some sums of paulis

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah yeah, I meant a non-diagonal coefficients matrix :)
Like

coeffs = np.random.random((len(dla), len(dla)))
new_dla = [qml.sum(*(c * op for c, op in zip(_coeffs, dla))) for _coeffs in coeffs]

or so

Copy link
Contributor Author

@Qottmann Qottmann Nov 14, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I added a tested orthonormalize function but somehow the test now fails for the cases where I orthonormalize the basis 🤔 I am confused

edit: talking about the test_structure_constants_elements test, the tests for orthonormalize work as expected

Copy link
Contributor Author

@Qottmann Qottmann Nov 14, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think the problem is the inconsistency between the trace inner product for matrices and pauli sentences that is assumed. I.e. the pauli sentence one is normalized whereas in the matrix case we dont. So I think to have consistent adjoint representations we should fix one convention and stick to that. Not sure which one though?

x = X(0) @ X(1)
x = x.pauli_rep
(x @ x).trace()
# 1.0
xm = qml.matrix(x, wire_order=range(2))
np.trace(xm @ xm)
# 4.0

I feel like the normalized one is neater because it should not depend on the (potentially redundant) matrix dimensions. E.g. in the example if you set the wire_order to range(3) the inner product is 8

In particular, I would suggest the matrix inner product to be

xm = qml.matrix(x, wire_order=range(4))
np.trace(xm @ xm)/len(xm)

to match those pauli sentences.

I pushed some suggestions but the sructure_constants test is still failing when the operators are being orthonormalized 🤔

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fully agree that we should be using $\text{tr}[A^\dagger B] / d$

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants