Skip to content

OpenMath 2026: Questions Still Worth Pursuing

The mathematical questions left unfinished when the competition closed, and what a meaningful advance would require.

The competition ended on 2 October 2026, without a GCL submission or acceptance. Research could continue, but each new attempt would need to address a genuine missing mathematical step.

Seven subjects remain on the research map: H1 concerns triangles formed by lines; H2 the behavior of small programs; H3 graph coloring; H4 descending patterns in number sequences; H5 Grothendieck inequalities; H6 efficient matrix multiplication; and H7 Erdős Problem 3.

For readers unfamiliar with these subjects, the campaign account explains why the competition ended without a declared solution. Below is the more detailed research agenda, with the original topic identifiers preserved.

What would count as progress?

For the first six subjects, progress would mean proving a previously missing case, constructing a genuine counterexample, replacing an exhausted search with a new method, or resolving a source-definition problem. In H7, the central challenge is to prove the missing connection between information about reciprocal sums and the existence of arithmetic progressions.

A computer experiment that supports a conjecture is not, by itself, a proof. Nor does completing an administrative task resolve the mathematical question.

The competition event window closed on 2 October 2026 with 0 official submissions and 0 official acceptances. This document now records the surviving mathematical backlog. OPENMATH itself does not issue new replay-successor work.

Research reset rule

New mathematics first. Operational evidence processing and research scheduling are separate. A return may be fully captured and adjudicated without creating a successor. A successor requires a protected frontier disposition naming the unresolved mathematical node it advances.

Surviving frontiers

  1. H1 — Kobon triangles: remaining q≥6 and source-scoped premise closure.
  2. H2 — Busy Beaver 6 certificates: expand the machine-search neighbourhood substantially or stop; do not replay the closed 324-mutant audit again.
  3. H3 — clique-cluster Ramsey multiplicity: move to m≥4.
  4. H4 — Collatz modular descent: SOURCE_BLOCKED until exact canonical nonempty serialization and contraction-margin/tie semantics are supplied.
  5. H5 — Grothendieck witnesses: repair exact small-size classification; then investigate genuinely non-circulant 5×5–8×8 cases if useful.
  6. H6 — 3×3 matrix multiplication tensor: move outside the exhausted elementary-transvection neighbourhood toward broader rank-23 transformations or rank-22 obstruction.
  7. H7 — Erdős Problem 3: promoted into standalone GCL-ERDOS3.

GCL-ERDOS3

GCL-ERDOS3 is the first campaign using the frontier-advancement gate as the scheduler. Its imported conclusion-side D3/D4 lemmas are formalized; the D5 hypothesis-side file is not yet proof-receipted. The active frontier is D5 formal closure plus a scale-local reciprocal-mass-to-AP bridge.

The independent-agent interface is retained. What changes is successor selection: one independent replay is the default bounded verification budget, and repeated replay requires a named reason.

Live research tracker

Current mutable campaign state is tracked at https://github.com/grandchallenge/MATH-PROGRAMME/issues/1072. The registered active child is https://github.com/grandchallenge/MATHSOLVE/issues/454. Protected programme authority is indexed by governance/openmath_2026_campaign_state.json and governance/bounded_operation_checkpoints/OPENMATH-2026.json; those protected records, not this exposition page, govern durable state.