Menger's Theorem
Edge version: For non-adjacent vertices in a graph , the maximum number of edge-disjoint - paths equals the minimum size of an - edge cut (minimum number of edges whose removal disconnects from ).
Vertex version: For non-adjacent vertices in a graph , the maximum number of internally vertex-disjoint - paths equals the minimum size of an - vertex separator (minimum number of vertices, excluding and , whose removal disconnects from ).
Proof
Edge version (from max-flow min-cut). Direct each edge of as two antiparallel arcs, each with capacity 1. The max - flow equals the max number of edge-disjoint paths (by flow decomposition and integrality). The min cut in this network corresponds to a minimum edge cut in . By max-flow min-cut: max edge-disjoint paths = min edge cut.
Vertex version (by vertex splitting). Create a directed graph : replace each vertex by two copies and connected by arc of capacity 1. Each original edge becomes arcs and of capacity (or ). Keep and .
In : flow through vertex in the original graph is limited to 1 by the arc . An integer max flow of value decomposes into paths that are internally vertex-disjoint in (since each vertex arc has capacity 1).
A minimum cut in : cutting vertex arcs corresponds to a vertex separator in (since cutting the -capacity edges is never optimal). The min cut capacity equals the min vertex separator size.
By max-flow min-cut in : max vertex-disjoint paths = min vertex separator.
Whitney's theorem: is -connected (i.e., ) if and only if every pair of vertices has internally vertex-disjoint paths. Proof: by Menger's theorem, vertex-disjoint paths exist iff every vertex separator has size , which is the definition of -connectivity. Similarly, is -edge-connected iff every pair has edge-disjoint paths.
Computing the edge connectivity requires max-flow computations (fix , compute max-flow to each , take minimum). The Stoer-Wagner algorithm computes for undirected graphs in without max-flow, using minimum cut phases based on maximum adjacency ordering.