Determine boundary-regime asymptotics for cycle-cover probability

Determine the precise asymptotic behavior of \(\Pr(\mathcal N)\), equivalently the probability that a directed graph sampled from a directed step-graphon admits cycle covers of all required sizes, when the concentration vector \(x^*\) lies on the boundary of the cycle polytope \(\vec{\mathcal X}(S)\) while satisfying feasibility but not strict feasibility.

Background

The main theorems characterize convergence of Pr⁡(N)\Pr(\mathcal N) to one away from the boundary of the cycle polytope. When x∗∈X⃗(S)∖relint⁡X⃗(S)x^*\in\vec{\mathcal X}(S)\setminus\operatorname{relint}\vec{\mathcal X}(S), the paper proves only an upper bound of one-half on the limiting superior of the probability that the sampled graph has a cycle cover of all vertices. The exact asymptotics may depend on the face containing x∗x^*, as in the previously studied symmetric case.

References

A second direction concerns the gap between $\mathcal N$, $\mathcal S$, and stability itself: even deterministically, no graph-theoretic characterization of stable digraphs is known, and our results show that $\mathcal N$ and $\mathcal S$ separate on the directed step-graphon model (Example~\ref{ex:NS_separation}), mirroring the deterministic gap between them.

— Structural stability of systems and cycle covers in random graphs  (2610.01607 - Belabbas, 1 Oct 2026) in Section 1, subsection “Main results”; Section 10, “Conclusion and open problems”