Fix complex dot product bugs: add np.conj() to match MATLAB x'*x#58
Open
drrelyea wants to merge 1 commit intobold-shamirfrom
Open
Fix complex dot product bugs: add np.conj() to match MATLAB x'*x#58drrelyea wants to merge 1 commit intobold-shamirfrom
drrelyea wants to merge 1 commit intobold-shamirfrom
Conversation
The original Python port incorrectly translated MATLAB's x'*x (conjugate transpose times x) to np.dot(x, x), which doesn't conjugate. For complex data, this produces complex results instead of real values. Fixed 3 source code bugs in spgl1.py: - Line 1342: f = f + (mu/2) * np.dot(x, x) -> np.dot(np.conj(x), x) - Line 1537: Same fix in tau update section - Line 2023: Fixed both np.dot(r, r) and np.dot(x, x) in rnorm computation Also added np.real() to 8 expressions that are mathematically real but may have tiny imaginary residuals from floating-point arithmetic: - rtr, f_dual, gap, aerror_dual, gtd, sts, sty computations Fixed corresponding bugs in test_mu_parameter.py where tests computed expected values using the same incorrect formula. Added regression tests (TestMuComplexData) that would have caught these bugs: - test_mu_complex_objective_is_real: Verifies rnorm is real - test_mu_complex_matches_manual_computation: Verifies formula - test_mu_complex_matches_matlab: Compares with MATLAB for complex data Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The original Python port incorrectly translated MATLAB's x'*x (conjugate transpose times x) to np.dot(x, x), which doesn't conjugate. For complex data, this produces complex results instead of real values.
Fixed 3 source code bugs in spgl1.py:
Also added np.real() to 8 expressions that are mathematically real but may have tiny imaginary residuals from floating-point arithmetic:
Fixed corresponding bugs in test_mu_parameter.py where tests computed expected values using the same incorrect formula.
Added regression tests (TestMuComplexData) that would have caught these bugs: