Documentary state: DRAFT_UNACCEPTED. This draft has NOT been editorially admitted. Mathematical claims require independent checking.
OAI-26 family 096: the Gaussian propeller conjecture
Editorial status: DRAFT_UNACCEPTED. NO_CERTIFICATION. This article reports the pinned publisher's claims and does not independently verify the proof.
The question in plain language
Imagine dividing a high-dimensional cloud of Gaussian-distributed points into finitely many groups. Each group has a first moment, describing its weighted displacement from the origin. The Gaussian propeller conjecture asks how large the sum of the squared lengths of these moments can be.
The pinned OAI-26 overview claims one sharp universal upper bound, attained by a simple three-part arrangement resembling a propeller. It also connects this inequality, together with a separate Unique Games theorem, to a claimed computational hardness threshold for a form of kernel clustering.
A small example
In the plane, divide space into three equal angular sectors, each with angle \(2\pi/3\). The sectors meet at the origin and point in three directions, like a three-bladed propeller. If a standard Gaussian point falls in each sector, its weighted first moment points toward the middle of that sector.
The mathematical quantity is the sum of the squared Euclidean norms of these unnormalized Gaussian first moments. A partition may have more than three pieces, and the surrounding Gaussian space may have arbitrarily many coordinates. The asserted theorem says that these extra freedoms never improve on the three-sector example.
What is known
The protected Family 096 catalogue identifies one manuscript, The Gaussian propeller bound in every dimension (September 24, 2026).
The pinned publisher overview claims that every finite measurable partition of a Gaussian space satisfies the first-moment inequality
where \(\gamma\) denotes standard Gaussian measure and the \(A_i\) form the partition. In dimension at least two, three equal planar sectors extended orthogonally into the remaining dimensions reportedly attain equality.
The overview further describes a conditional combination: together with a separate Unique Games theorem, the propeller inequality yields NP-hardness of improving the stated loss factor
for identity-target kernel clustering at fixed \(k\geq3\), on rational centered positive-semidefinite inputs. This computational implication must not be attributed to the Gaussian inequality alone, nor generalized to arbitrary kernel-clustering formulations.
GCL verification state
The source is pinned to openai/math@1581d4cbdd69a0faf34b2089614fe2a814316a6a. The protected catalogue and overview provide the manuscript identity and source-reported bound.
This editorial pass does not independently verify the partition inequality, the equality case, or the reduction connecting it to kernel-clustering hardness. The separate Unique Games result requires independent source identification and audit. No Family 096-specific Lean proof, Comparator replay, proof-semantic correspondence, or MATHCERT certificate has been established.
What remains
Read the pinned manuscript and verify the Gaussian normalization, the definition of unnormalized first moments, any restrictions on finite measurable partitions, and the treatment of null sets. Recalculate the three-sector equality value and determine whether all equality cases are classified or only an extremizer is exhibited.
For the computational implication, locate the separate Unique Games theorem and check all reduction hypotheses: fixed \(k\), rational inputs, centered positive-semidefinite matrices, identity target, and the precise interpretation of the approximation loss factor. Do not identify conditional hardness with an unconditional claim.
Complete mathematical Markdown and generated-compendium validation, source-specific critical review, and protected merge/readback. Documentary admission does not independently certify the mathematics.
Sources and claim boundaries
Boundary: NO_CERTIFICATION. The geometric bound and its stated computational consequence are publisher-reported claims, not independently certified results.
Indexed primary manuscript paths:
- The Gaussian propeller bound in every dimension —
preprints/The-Gaussian-Propeller-Bound-in-Every-Dimension-September-24-2026/main.pdf