Proofs for acyclic concentrators #
Let G be an acyclic (n, m)-concentrator whose inputs are sources (in-degree zero), with
2 ≤ m < n.
- Elimination (
exists_eliminate). Removing at least two edges,Gturns into an acyclic(n - 1, m)-concentrator on the same vertices, with sources as inputs. Every input has an edge leaving it. If some input has two, remove its edges and drop it: the other walks never visit a source. Otherwise every inputxhas one edge, to its head, and distinct inputs have distinct heads, since two inputs with a common head cannot be routed together. If some headuofxis entered by that edge only, remove it and makeuan input in place ofx:uis neither an input nor an output, as every output is reached from inputs other thanx. Otherwise letwbe a first vertex, in the topological order, among the non-input vertices that meet an edge. The edges enteringwcome from inputs whose head isw, so there is at most one, and thenwwould be a head entered only by it; so no edge entersw. Thenwis no output, and its edges can be removed. The last two steps keepnand remove an edge, so induction on the number of edges applies. - Eliminating
n - 2 minputs leaves a(2 m, m)-concentrator with at least2 (n - 2 m)fewer edges, and the concentrator bound applies to it. Forh = min(m, n - m)this givesh ≤ (A + η) (M - 2 n + h)⁺ + 3 log₂ (2 M) + Cand the baselineM ≥ 2 n - h - 1.
Removing the edges at a vertex #
The multigraph with the edges at x removed.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Walks avoiding x survive the removal of the edges at x.
Removing edges does not increase in-degrees.
Removing edges keeps a multigraph acyclic.
The edges kept and the edges at x partition the edges.
Walks and sources #
Every vertex of a chain after its first is entered by an edge.
A vertex entered by an edge has positive in-degree.
Walks avoid sources. A directed walk visits a vertex of in-degree zero only at its start.
A directed walk from a vertex of in-degree zero to another vertex continues along an edge leaving it, and the rest of the walk avoids it.
The three elimination steps #
Removing the edges at a vertex keeps the inputs sources.
Dropping an input. Removing the edges at a source input leaves a concentrator on the other inputs.
Contracting an input into its head. Let the input input i₀ be a source whose only
edge e₀ leads to a vertex u entered by no other edge, and let every other input be a
source. Removing e₀ and making u an input in place of input i₀ leaves a concentrator,
provided m inputs other than input i₀ exist.
Removing a dead vertex. Removing the edges at a non-input vertex entered by no edge leaves a concentrator with the same inputs and outputs.
Elimination #
The head of the only edge leaving a source lies on every walk from it to another vertex.
An edge leaves a source input that reaches an output.
Elimination. An acyclic (n + 1, m)-concentrator whose inputs are sources, with
2 ≤ m ≤ n, turns into an acyclic (n, m)-concentrator on the same vertices and outputs, whose
inputs are sources, with at least two fewer edges.
Repeated elimination. Eliminating j inputs from an acyclic (n, m)-concentrator whose
inputs are sources, with 2 ≤ m and m + j ≤ n, removes at least 2 j edges.
Acyclicity #
Ranks increase along directed walks of positive length.
Acyclicity is the absence of directed cycles. A finite multigraph has a vertex numbering increasing along every edge exactly when no vertex reaches itself along a directed walk of positive length.
Bounds #
The edge baseline for acyclic concentrators. For h = min(m, n - m), an acyclic
(n, m)-concentrator with M edges whose inputs are sources, with 2 ≤ m < n, has
2 n ≤ M + h + 1.
The finite bound for acyclic concentrators. For h = min(m, n - m), an acyclic
(n, m)-concentrator with M edges whose inputs are sources, with 2 ≤ m < n, satisfies
h ≤ (A + η) (M - (2 n - h))⁺ + 3 log₂ (2 M) + C.
The asymptotic edge bound for acyclic concentrators, for a general ordering coefficient.