This repository contains a proposed complete solution of Erdős Problem 81, together with a publication manuscript, a Lean 4 formalization, and reproducible finite certificates.
For all sufficiently large n, the proposed result determines the exact
extremal value
max cp(G) = floor(n(n + 1) / 6),
where the maximum is over all n-vertex chordal graphs and cp(G) is the
minimum number of cliques whose edge sets partition E(G). In particular,
every chordal graph satisfies cp(G) <= n²/6 + O(n).
The proof and its formalization are complete within the trust boundary below. Because the result would resolve a long-standing open problem, it should be treated as a proposed proof until it has received independent review by specialists in extremal or probabilistic graph theory.
The proof combines a global fractional argument with a local integral one.
- Pack triangles and copies of
K₄fractionally, with gains2and5. TheK₄term pushes complete graphs below the extremal quadratic scale, leaving the complete-split family as the relevant terminal branch. - A vertex-copy inequality and discrete convexity produce a monotone path of single-vertex copies from any chordal graph to a complete-split graph.
- Near the extremal complete-split family, root regularization and balanced
edge-colouring construct an actual
K₂/K₃partition with a strict saving for every missing spoke or outside edge. A first-entry argument transfers this local stability back along the copy path. - Away from the extremal family, a fixed quadratic deficit absorbs the subquadratic loss in fractional-to-integral packing transfer. A signed edge count on complete-split graphs supplies the matching lower bound.
The manuscript gives the conventional proof and explains the relation to earlier complete-split and clone-copy work.
manuscript/main.texis the paper source;manuscript/main.pdfis the built review copy.lean/contains the pinned Lean 4 project and its verification gate.preparations/erdos81_stability_closure/contains the main exact finite certificates.preparations/erdos81_rigidity_addendum/contains a separate rigidity addendum and its replayable certificates; it is not part of the Lean theorem perimeter.references/SOURCES.mdrecords primary sources and immutable reference snapshots.docs/VERIFICATION_STATUS.mdgives the detailed proof and formalization ledger.
The principal checked declarations are
eventualSharpUpperBound_of_inputs :
ExternalInputs.Inputs -> EventualSharpUpperBound
eventualSharpEquality_of_inputs :
ExternalInputs.Inputs -> EventualSharpEquality
erdos81_of_inputs :
ExternalInputs.Inputs -> Erdos81Statement
The complete-split lower bound and the passage from the eventual result to
n²/6 + O(n) are proved inside Lean. The conditional upper bound uses three
explicit interfaces:
- Vizing's edge-colouring theorem;
- the Häggkvist--Janssen list-edge-colouring bound for complete graphs;
- the uniform finite-family packing transfer specialized to triangles and
K₄.
The third interface packages the transfer conclusion with attained rational primal and dual witnesses. These assumptions are theorem arguments, not Lean axioms or hidden placeholders. The project therefore checks the complete conditional reduction, but does not claim a self-contained kernel proof of the three external interfaces.
lean/check.sh builds the project and performs a fail-closed environment
audit of every public Erdos81.* declaration. The only permitted transitive
kernel assumptions are propext, Classical.choice, and Quot.sound.
The dependency-free certificate replays need only Python 3:
make replayFor a first Lean build, install elan and fetch the manifest-pinned Mathlib cache:
cd lean
lake exe cache get
./check.shBuild the manuscript and reject unresolved references, citations, and box overflow diagnostics with:
make -C manuscript clean checkFrom a prepared checkout, make check runs both certificate replays, the Lean
gate, and the manuscript check. The larger regeneration audits use the pinned
packages in requirements-audit.txt and can be run with make audit after
creating a virtual environment.
The mathematical proof was developed by Morluto (X), with GPT assistance; the shared proof-development conversation records that derivation. Jacobian, a research tool developed at Preference Labs, played a substantial role in the derivation and verification workflow.
The manuscript and Lean 4 formalization were written by N0zoM1z0 (X).
This is a Preference Labs research project.
Before submission, the anonymous author placeholder and affiliations should be finalized, appropriate licences selected, the artifact archived at a stable version, and independent specialist review obtained. Any public claim about the Lean development should retain the conditional boundary stated above.