Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Statement

Setting as on the Theorem 1.2 page: r≥3r\ge3 fixed, r∣nr\mid n, Hn,M\mathcal H_{n,M} the uniform random MM-edge rr-graph on [n][n]. Φ(H)\Phi(\mathcal H) denotes the number of perfect matchings of H\mathcal H (p. 3).

Theorem 1.5 (p. 3). For fixed ε>0\varepsilon>0 and M>(1+ε)(n/r)log⁡nM>(1+\varepsilon)(n/r)\log n, w.h.p.

Φ(Hn,M)>[e−(r−1)rM/n]n/re−o(n).\Phi(\mathcal H_{n,M})>\Bigl[e^{-(r-1)}rM/n\Bigr]^{n/r}e^{-o(n)}.

This is display (4) of the paper. The paper remarks (p. 3) that the right-hand side is within a subexponential factor of E Φ(Hn,M)\mathbb E\,\Phi(\mathcal H_{n,M}). Since the right-hand side is positive, Theorem 1.5 contains Theorem 1.2. The paper notes (p. 4) that a major difference from the 2008 Johansson-Kahn-Vu argument is the error term e−o(n)e^{-o(n)}, which was e−O(n)e^{-O(n)} there.

Proof pointer

Section 2 (pp. 4 to 7) removes the edges of K\mathcal K one at a time in uniform random order until MM remain and tracks log⁡Φ\log\Phi along the way: each removal multiplies Φ\Phi by 1−ξt1-\xi_t, where ξt\xi_t is the fraction of current perfect matchings using the removed edge, and ξt\xi_t has conditional mean γt=(n/r)/((nr)−t+1)\gamma_t=(n/r)/(\binom nr-t+1). Theorem 1.5 comes down to showing that the martingale ∑(ξi−γi)\sum(\xi_i-\gamma_i) stays small, which a bounded-differences argument (Section 3) gives once the increments ξi\xi_i are O(γi)O(\gamma_i). That increment bound comes from a property saying no edge lies in much more than its share of perfect matchings, established in Sections 5 to 9 using the entropy bounds of Section 4 (in particular the Brégman-type Theorem 4.2); routine degree conditions are handled in the appendix.

Read depth

Claims checked: Theorem 1.5 and display (4) were read on the print (p. 3), and the Section 2 reduction (pp. 4 to 7) was followed in outline. Sections 3 to 9 and the appendix were not checked. Nothing here is independently reviewed.

Dependencies

None in the corpus. The paper's argument follows the method of Johansson, Kahn and Vu (Random Structures Algorithms 33 (2008)).

Source. J. Kahn, Asymptotics for Shamir's problem, Adv. Math. 422 (2023), Paper No. 109019, doi:10.1016/j.aim.2023.109019; labels and pages are those of the edition named on the source card.

Bears on