Preserve sigma-point pair symmetry under rounding

[FlorianPfaff]

Summary

Fix a floating-point symmetry bug in the Merwe and Julier sigma-point generators.

Both generators originally formed each pair independently as x + U[:, i] and x - U[:, i]. For small spreads, the represented offsets can differ by one ULP. With the standard CircularUKF Merwe parameters (alpha=1e-3), the large off-center weights amplify that tiny asymmetry enough to break the existing identity-prediction regression.

Fix

Construct each pair through the representable displacement and re-realize that displacement once after forming the negative point:

positive_point = x + U[:, i]
realized_offset = positive_point - x
negative_point = x - realized_offset
realized_offset = x - negative_point
positive_point = x + realized_offset

The second realization matters: a one-pass reflection can itself re-round on x - realized_offset, leaving the final stored pair one ULP asymmetric. Rebuilding the positive point from the negative point's final representable displacement removes that residual asymmetry without using the overflow-prone expression 2*x - point.

The implementation is shared by MerweScaledSigmaPoints and JulierSigmaPoints through _symmetric_sigma_points.

Regression coverage

tests/test_sigma_points_pair_symmetry.py checks exact pair-offset equality for small-spread Merwe and Julier cases. The existing CircularUKF.test_predict_nonlinear_identity_function independently checks the downstream numerical effect.

Numerical reproduction for the affected Merwe coordinate:

• original construction: pair offsets differ (0.0008366600265340973 vs 0.0008366600265341528 in the negative-mean coordinate)
• one-pass reflection: can still differ by one ULP in the same coordinate
• two-stage realization: final stored offsets are exactly equal for the regression case

The branch changes only the sigma-point implementation and the targeted regression test.

[github-actions]

✅MegaLinter analysis: Success

Descriptor	Linter	Files	Fixed	Errors	Max errors	Warnings	Elapsed time
✅ COPYPASTE	jscpd	yes		no		no	8.58s
✅ JSON	prettier	7	0	0		0	0.55s
✅ JSON	v8r	7		0		0	4.31s
✅ MARKDOWN	markdownlint	68	0	0		0	1.95s
✅ MARKDOWN	markdown-table-formatter	68	0	0		0	0.42s
✅ PYTHON	black	1990	15	0		0	59.18s
✅ PYTHON	isort	1990	28	0		0	2.87s
✅ REPOSITORY	betterleaks	yes		no		no	1.18s
✅ REPOSITORY	checkov	yes		no		no	35.34s
✅ REPOSITORY	git_diff	yes		no		no	0.12s
✅ REPOSITORY	secretlint	yes		no		no	100.59s
✅ REPOSITORY	syft	yes		no		no	3.45s
✅ REPOSITORY	trivy-sbom	yes		no		no	2.58s
✅ YAML	prettier	11	0	0		0	0.71s
✅ YAML	v8r	11		0		0	16.34s
✅ YAML	yamllint	11		0		0	1.0s

Notices

⚠️ Your configuration references items that have been removed from MegaLinter and are ignored: MAKEFILE_CHECKMAKE. See Removed linters to find their replacements.

See detailed reports in MegaLinter artifacts

Your project could benefit from a custom flavor, which would allow you to run only the linters you need, and thus improve runtime performances. (Skip this info by defining FLAVOR_SUGGESTIONS: false)

• Documentation: Custom Flavors
• Command: npx mega-linter-runner@10.0.0 --custom-flavor-setup --custom-flavor-linters PYTHON_BLACK,PYTHON_ISORT,COPYPASTE_JSCPD,JSON_V8R,JSON_PRETTIER,MARKDOWN_MARKDOWNLINT,MARKDOWN_MARKDOWN_TABLE_FORMATTER,REPOSITORY_CHECKOV,REPOSITORY_GIT_DIFF,REPOSITORY_BETTERLEAKS,REPOSITORY_SECRETLINT,REPOSITORY_SYFT,REPOSITORY_TRIVY_SBOM,YAML_PRETTIER,YAML_YAMLLINT,YAML_V8R

Show us your support by starring ⭐ the repository