Wiki
Wiki

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

Updated


Source. Erdős (1945), unnumbered lemma and its proof, printed pp. 900–901 (published scan). The increasing orientation needed by Theorem 5 is explicit below.

Statement. Let N≥0N\ge0 and let aa be an integer with 0≤a≤N/20\le a\le N/2. In the Boolean lattice on [N][N], there are (Na)\binom Na pairwise vertex-disjoint increasing paths from rank aa to rank N−aN-a. Each path adds one element at each step. If the two endpoint ranks agree, the paths consist of the individual vertices.

Proof. The equal-rank case is immediate, so suppose a<N/2a<N/2. Orient each cover edge toward the larger rank and retain ranks a,…,N−aa,\ldots,N-a. Write M=(Na)M=\binom Na.

There are

P=(Na) (N−a)!a!=N!(a!)2P=\binom Na\,\frac{(N-a)!}{a!} =\frac{N!}{(a!)^2}

increasing paths from the bottom to the top layer: choose the starting aa-set, then choose in order N−2aN-2a of the remaining elements. A vertex of rank kk lies on exactly

Pk=k!a! (N−k)!a!=k!(N−k)!(a!)2P_k=\frac{k!}{a!}\,\frac{(N-k)!}{a!} =\frac{k!(N-k)!}{(a!)^2}

such paths. The first factor counts the choices and order below the vertex, and the second counts those above it. For a≤k≤N−aa\le k\le N-a, symmetry and unimodality give (Nk)≥(Na)\binom Nk\ge\binom Na. Hence

Pk≤D:=(N−a)!a!.P_k\le D:=\frac{(N-a)!}{a!}.

If a collection WW of vertices meets every increasing path, counting path incidences with WW gives P≤∣W∣DP\le |W|D, so ∣W∣≥P/D=M|W|\ge P/D=M. Endpoint vertices are allowed in WW.

To obtain increasing paths from this separator bound, use the exact finite integral-flow input. Replace every lattice vertex vv by v−,v+v^-,v^+ with an arc v−→v+v^-\to v^+ of capacity one. Replace every increasing cover edge u→vu\to v by u+→v−u^+\to v^- with capacity M+1M+1. Add a new source ss with a capacity-one arc to v−v^- for each bottom vertex, and a capacity-one arc from v+v^+ to a new sink tt for each top vertex.

This network is finite and acyclic, with nonnegative integer capacities, no arc into ss and no arc out of tt. Suppose it had an outgoing cut of capacity less than MM. No capacity-M+1M+1 cover arc belongs to that cut. Associate each cut arc of capacity one with its lattice vertex: use vv for v−→v+v^-\to v^+, for s→v−s\to v^- and for v+→tv^+\to t. The resulting set WW has size at most the cut capacity. Every increasing lattice path lifts to a source–sink path, which crosses the outgoing cut and therefore meets a lattice vertex in WW. This contradicts ∣W∣≥M|W|\ge M.

Every cut consequently has capacity at least MM. The source has total outgoing capacity MM, so the finite integral max-flow/min-cut theorem gives an integral flow of value exactly MM. Decompose it into MM unit source–sink paths: as long as the value is positive, follow positive-flow arcs from ss. Conservation prevents a dead end at an intermediate vertex, and acyclicity forces arrival at tt. Subtract one unit along the path and repeat. Integrality and nonnegativity are preserved at each step.

The capacity-one arcs v−→v+v^-\to v^+ prevent two extracted paths from using the same lattice vertex. Contracting these arcs and removing the new terminals leaves MM vertex-disjoint increasing lattice paths. Their number equals the size of the bottom layer, so every bottom vertex starts one of them. □\square

Source precision and external scope. The source quotes an undirected form of Menger's theorem, but its displayed counts count increasing paths, and its later replacement argument needs chains. The explicit upward orientation and integral-flow reduction supply that interface. They do not assert that the quoted undirected theorem is false. The integral-flow theorem is proved in the linked Ford–Fulkerson (1957) source and is an external input here; the split-network and decomposition deductions are included above. This is a later implementation of the source's Menger method, not a historical attribution to Erdős.

Use. Theorem 5.