Skip to content

docs(paper): drop the negative-results section and repair what cited it - #54

Merged
tschm merged 2 commits into
mainfrom
paper_drop_negatives
Aug 8, 2026
Merged

docs(paper): drop the negative-results section and repair what cited it#54
tschm merged 2 commits into
mainfrom
paper_drop_negatives

Conversation

@tschm

@tschm tschm commented Aug 8, 2026

Copy link
Copy Markdown
Member

Summary

Removes §8, "Two negative results" — the implicit-Q experiment (geqrf/ormqr) and the
Numba port. 126 lines out; the paper goes from fifteen pages to twelve.

The section was not self-contained

Seven passages referred to it and five leaned on it for a quantitative claim, so
deleting it alone would have left those claims citing evidence the paper no longer carries.
Each is rewritten to stand on its own:

where was now
Abstract "We also report two negative results in full…" with figures removed
Contributions list fourth item citing §8 removed
Updating vs recomputing "we quantify it in §8" states the O(n³) vs O(n⁴) consequence directly
Least-squares remark "§8 shows why that costs a factor of 57" says plainly the quantities are already available in factored form
Folding remark "§8 shows what happens when this folding is undone" reference dropped
Interpreter floor "…and §8 confirms it by removing it" stops at the claim
Limitations "§8 establishes that closing that gap requires leaving pure NumPy" "…appears to require…"

That last hedge is deliberate. It is the one claim whose only support was the removed
section, and asserting it without the measurement would be worse than weakening it. If the
optional Numba accelerator lands, that sentence should become a forward reference to it
rather than a hedge.

Note

numba2015 is now uncited. I left the entry in the bibliography rather than deleting it,
since adopting Numba as an optional extra would need it back.

Testing

  • make paper-tectonic — compiles clean, 0 unresolved references
  • make fmt

docs/paper/quadprog.tex only.

🤖 Generated with Claude Code

Section 8 reported two experiments that were pursued and abandoned: leaving the
orthogonal factor implicit through geqrf/ormqr, and a Numba port. Removed --
126 lines, and the paper goes from fifteen pages to twelve.

The section was not self-contained. Seven passages referred to it, and five of
those leaned on it for a quantitative claim, so each has been rewritten to stand
on its own rather than to cite evidence the paper no longer carries:

* the abstract's promise of "two negative results in full", with the figures it
  quoted, is gone;
* the contributions list loses its fourth item;
* "we quantify it in S8", on updating versus recomputing the factor, now states
  the O(n^3) against O(n^4) consequence directly;
* "S8 shows why that costs a factor of 57", after the least-squares remark, now
  says plainly that the quantities are already available in factored form and
  that recomputing them each iteration is the more expensive route;
* the remark on folding R_G^{-1} and Q into J loses its forward reference;
* "a floor set by the language implementation ... and S8 confirms it by removing
  it" now stops at the claim itself;
* Limitations said S8 "establishes" that closing the small-n gap requires leaving
  pure NumPy; it now says this "appears to" be so, since the experiment that
  established it is no longer reported.

That last hedge is deliberate rather than cosmetic. It is the one claim in the
paper whose only support was the removed section, and overstating it without the
measurement would be worse than weakening it.

The numba2015 entry is left in the bibliography, now uncited. It is not orphaned
for long if the optional accelerator lands.

Compiles clean under tectonic with no unresolved references.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Copilot AI lite review requested due to automatic review settings August 8, 2026 20:01
@tschm
tschm merged commit 373e7f0 into main Aug 8, 2026
43 checks passed

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

Pull request overview

This PR updates the LaTeX paper to remove the “Two negative results” section and revises all passages that previously relied on that section so the paper remains internally consistent (including removing/rewriting claims that previously cited §8).

Changes:

  • Deletes the “Two negative results” section entirely.
  • Removes or rewrites references in the abstract, contributions list, and multiple sections that previously cited the removed section for support.
  • Softens one previously-measured claim in Limitations (“establishes” → “appears to require”) to avoid overstating without the removed evidence.

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

Comment thread docs/paper/quadprog.tex
Comment on lines 719 to 721
method. Small problems remain slower than the compiled reference by up to
13 times, and \S\ref{sec:negative} establishes that closing that gap
requires leaving pure NumPy. The structure detection of \S\ref{sec:structure}
13 times, and closing that gap appears to require leaving pure NumPy. The structure detection of \S\ref{sec:structure}
recognises single-nonzero columns and general sparsity in the slack product but
@tschm
tschm deleted the paper_drop_negatives branch August 8, 2026 20:05
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