Skip to content

Products-of-linears start systems do not homogenize (ExpandToFunctionTree variable count mismatch) #376

Description

@ofloveandhate

Symptom

Homogenizing a system whose functions include a product of linears (the two-step sphere-cap regeneration's L1 * L2 start, and linear-product start structures generally) fails with

ExpandToFunctionTree: products-of-linears variable count mismatch

Why it matters

The cellular-decomposition reliability work established that complex-gamma moves must track projectively (homogenize + patch) so paths only go to infinity at the end, never mid-path (mid-path affine poles were diagnosed as pre:GoingToInfinity divergence). That fix was applied to the regeneration witness moves and collapsed mid-path divergence — but the product-to-sphere deformation cannot be projectivized because its products-of-linears start system will not homogenize, so that one move family still tracks affinely.

The measurable consequence: the residual preimage-curve cap losses in the whitney gauntlet (1–2 truncated paths per dirty run, SecurityMaxNormReached/pre:GoingToInfinity during regen-slicemove/regen-linprod: sphere-cap), each of which costs a critpointset a genuine point and produces mis-assembled faces downstream. Raising the security ceiling (max_norm/path_truncation 1e14) recovers most, but the principled fix is projective tracking for this move family, which is blocked here.

Ask

Make homogenization handle products-of-linears function structures (or provide an expansion path so the homogenizer sees plain polynomials), so the sphere-cap deformation can join the projective-move policy.

🤖 Generated with Claude Code

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions