Skip to content

Make the non-grav placeholder negligible and the discarded refinement visible - #540

Merged
matthewholman merged 2 commits into
mainfrom
fix/nongrav-reporting
Sep 3, 2026
Merged

Make the non-grav placeholder negligible and the discarded refinement visible#540
matthewholman merged 2 commits into
mainfrom
fix/nongrav-reporting

Conversation

@matthewholman

Copy link
Copy Markdown
Collaborator

Closes #539.

Placeholder 1e-15 → 1e-20 au/day², and the discarded-refinement message moves from DEBUG to WARNING and now names the flag the joint fit actually returned — 6 is a weakly-constrained solution, not a failure to converge, and 2 is a fit rejected on chi-square.

🤖 Generated with Claude Code

matthewholman and others added 2 commits September 3, 2026 17:22
… visible

Two small defects in the same path, both of which hide a failed non-grav fit.

ASSIST skips the non-gravitational block entirely when A1=A2=A3=0, so a zero
seed would leave the parameter's column zero. The fit substitutes a placeholder
to avoid that, and the partial does not depend on its magnitude -- but the value
chosen, 1e-15 au/day^2, is the order of a real Yarkovsky amplitude. On (6489)
Golevka it is 6% of the true A2 and raises the starting chi-square from 12,888
to 138,416 before the first iteration, which is enough to trip the chi-square
gate and discard the fit. Use 1e-20, which restores the gravity-only chi-square
exactly.

When the joint fit does return a non-zero flag the driver keeps the six-parameter
solution and reports the non-grav parameters as NaN, which is right, but said so
only at DEBUG. The caller sees flag 0, a chi-square belonging to the gravity fit,
and NaN parameters, with nothing to explain them. Warn instead, and name the flag
the joint fit returned: 6 is a weakly-constrained solution, not a failure to
converge, and 2 is a fit rejected on chi-square.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

@kjnapier kjnapier left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Looks good

@matthewholman
matthewholman merged commit 58ca94b into main Sep 3, 2026
7 checks passed
@matthewholman
matthewholman deleted the fix/nongrav-reporting branch September 3, 2026 21:58
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.

A failed non-grav refinement is invisible, and its placeholder amplitude is not negligible

2 participants