Skip to content

Fix two C6 failures behind the MatC6 long orbit - #555

Open
fingolfin wants to merge 1 commit into
masterfrom
mh-ai/matc6-long-orbit
Open

fingolfin wants to merge 1 commit into
masterfrom
mh-ai/matc6-long-orbit

Conversation

@fingolfin

@fingolfin fingolfin commented Nov 15, 2026 •

Copy link
Copy Markdown
Member

On the 27-dimensional projective node of MatC6.tst, 3^(1+6):Sp(6,3), the first C6 call returned TemporaryFailure in 3 of 100 seeds. Method selection then ran StabilizerChainProj on PG(26,25) before retrying C6: over 10^7 points in seed 6, taking 477 s in CI. There were two causes.

First, there was a bug: RECOG.basis2 advanced k before unbinding a scalar gens[k], so the generator after each scalar was dropped unseen. With 13 scalars among 19 conjugates (seed 48) it found 2 of 3 hyperbolic pairs.

Second problem: C6 takes the conjugates of an element of the r-core R from TestAbelian as generators of R. The conjugators are the node's pseudo-random elements, which can be badly mixed: in our example, 6 of 8 generators lie in R, and in seeds 6 and 90, we find that 27 and 32 of the first 40 random elements map to elements of order at most 2 in Sp(6,3). The conjugates then span a subspace of R/Z(R) that the group does not normalise, and some generator fails to rewrite. Now C6 adds a conjugate of a random subproduct under each failing generator and retries, up to 4 rounds per dimension of R/Z(R). The successful path is unchanged.

With both fixes the first C6 call succeeds for seeds 1-100.

Assisted-by: Generative AI


To review this, I recommend to turn on "Hide whitespace diffs"

@fingolfin
fingolfin added this pull request to stack #556 November 15, 2026 08:57
@codecov

codecov Bot commented Nov 15, 2026 •

Copy link
Copy Markdown

Codecov Report

❌ Patch coverage is 95.83333% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 86.39%. Comparing base (6b55721) to head (0c82d97).

Files with missing lines Patch % Lines
gap/projective/c6.gi 95.83% 1 Missing ⚠️
Additional details and impacted files
@@            Coverage Diff             @@
##           master     #555      +/-   ##
==========================================
- Coverage   86.46%   86.39%   -0.08%     
==========================================
  Files          49       49              
  Lines       19382    19382              
==========================================
- Hits        16759    16745      -14     
- Misses       2623     2637      +14     
Files with missing lines Coverage Δ
gap/projective/c6.gi 98.07% <95.83%> (-0.25%) ⬇️

... and 5 files with indirect coverage changes

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@fingolfin
fingolfin force-pushed the mh-ai/matc6-long-orbit branch from ad86424 to 246ece8 Compare November 15, 2026 10:06
Base automatically changed from mh-ai/use-isscalarmat to master November 15, 2026 16:09
On the 27-dimensional projective node of MatC6.tst, 3^(1+6):Sp(6,3),
the first C6 call returned TemporaryFailure in 3 of 100 seeds. Method
selection then ran StabilizerChainProj on PG(26,25) before retrying C6:
over 10^7 points in seed 6. There were two causes.

RECOG.basis2 advanced k before unbinding a scalar gens[k], so the
generator after each scalar was dropped unseen. With 13 scalars among
19 conjugates (seed 48) it found 2 of 3 hyperbolic pairs.

C6 takes the conjugates of an element of the r-core R from TestAbelian
as generators of R. The conjugators are the node's pseudo-random
elements, which can be badly mixed: 6 of 8 generators lie in R, and in
seeds 6 and 90, 27 and 32 of the first 40 random elements map to
elements of order at most 2 in Sp(6,3). The conjugates then span a
subspace of R/Z(R) that the group does not normalise, and some
generator fails to rewrite. Now C6 adds a conjugate of a random
subproduct under each failing generator and retries, up to 4 rounds
per dimension of R/Z(R). The successful path is unchanged.

With both fixes the first C6 call succeeds for seeds 1-100.

Assisted-by: Generative AI
@fingolfin
fingolfin force-pushed the mh-ai/matc6-long-orbit branch from 246ece8 to 0c82d97 Compare November 15, 2026 16:09
Comment thread gap/projective/c6.gi
else
k:=k+1;
fi;
Unbind(gens[k]);

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

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

so here we advanced k before the unbind...

@fingolfin
fingolfin marked this pull request as ready for review November 15, 2026 23:41

This branch has not been deployed

No deployments
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.

1 participant