Full-class non-tightness at every finite degree

Prove that the rank-dependent primal–dual protocol is not tight at every finite degree d≥2, even when optimized over the full admissible class of measurable nonnegative rank-only gain-sharing functions; specifically, establish Γ_A(d)<γ(d) for online-degree-bounded instances and \tilde{Γ}_A(d)<\tilde{γ}(d) for d-regular instances.

Background

The paper proves strict certifying gaps at degree two for both online-degree-bounded and 2-regular graph classes, including over the full measurable admissible function class. It also proves that the corresponding gaps vanish asymptotically as the degree tends to infinity. These results motivate the conjecture that strict non-tightness persists at every finite degree d≥2, despite the asymptotic disappearance of the gap.

The conjecture concerns the fixed rank-dependent splitting rule and its expected edgewise certificate, not randomized primal–dual analysis in general. Establishing it would extend the degree-two protocol-barrier results to all finite degrees and clarify whether the observed gap is a universal finite-degree phenomenon.

References

Our degree-two results motivate the conjecture that the rank-dependent protocol is not tight at any finite degree $d\geq2$, even over the full admissible class:

\Gamma_{\cA}(d)<\gamma(d),\qquad \tilde\Gamma_{\cA}(d)<\tilde\gamma(d) \qquad\text{for every integer }d\text{ with }2\leq d<\infty.

— Certifying Capacity and Gaps of the Rank-Dependent Gain-Sharing Framework for Ranking  (2610.17545 - Xu, 13 Oct 2026) in Section 7, Conclusion and Future Directions, subsection “Full-class gaps at every finite degree”

For fixed $\mathfrak P$, $\cS$, and \rank{}, can tightness be characterized through structural properties of the instance class $\cG$?

— Certifying Capacity and Gaps of the Rank-Dependent Gain-Sharing Framework for Ranking  (2610.17545 - Xu, 13 Oct 2026) in Section 7, Conclusion and Future Directions, subsection “Structural conditions for tightness”