Wiki
Wiki

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

Updated


Source interface. Section 1, printed p. 147, and the discussion of graph decompositions in Sections 2 and 4 (published PDF). This expands the elementary graphic example used by the source, not a separate numbered theorem or a reconstruction of its cited graph-theoretic papers.

Statement. Let GG be a finite loopless graph with fixed vertex set VV and edge set EE. The acyclic edge sets form a matroid on EE. If c(A)c(A) is the number of connected components of (V,A)(V,A), counting isolated vertices, its rank is

r(A)=∣V∣−c(A).r(A)=|V|-c(A).

Consequently, for k≥1k\ge1:

  1. EE partitions into kk forests, with empty parts allowed, exactly when ∣A∣≤k(∣V∣−c(A))|A|\le k(|V|-c(A)) for all A⊆EA\subseteq E.
  2. There are kk edge-disjoint spanning forests, each spanning every connected component of GG, exactly when
∣E∖A∣≥k(c(A)−c(E))(A⊆E).|E\setminus A|\ge k(c(A)-c(E)) \qquad(A\subseteq E).

For connected GG, these bases are spanning trees.

Proof. A subset of a forest is a forest. If F⊆AF\subseteq A is a maximal forest, its connected components are exactly those of (V,A)(V,A). Otherwise a path in (V,A)(V,A) between two different components of (V,F)(V,F) supplies an edge joining different forest components, which could be added without creating a cycle.

A finite forest with vertex set VV and cc components has ∣V∣−c|V|-c edges. Each nontrivial finite tree has a leaf: an endpoint of a longest path cannot have an additional neighbor without either extending the path or creating a cycle. Thus one can remove a leaf and its incident edge repeatedly from every nontrivial tree, leaving one vertex per component; each removal lowers both the edge and vertex counts by one. The formula also holds for the empty vertex set. Thus every maximal forest in AA has ∣V∣−c(A)|V|-c(A) edges, proving the matroid axiom and rank.

Apply Theorems 1 and 2 with this rank. The first criterion is immediate. In the second, r(E)−r(A)=c(A)−c(E)r(E)-r(A)=c(A)-c(E), giving the displayed inequality. A base is exactly a forest with the same components as GG, as the maximality argument showed. □\square

No numbered Erdős-problem implication or current graph bound is inferred here. The exact equivalence to other vertex-partition formulations in the cited Tutte and Edmonds papers remains outside this source compilation.