Wiki
Wiki

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

Updated


Source. Theorem 2c, printed pp. 150–152 (published PDF).

Statement. For finite matroids Mi=(E,Fi)M_i=(E,\mathcal F_i) on one finite set, 1≤i≤k1\le i\le k, there are pairwise disjoint bases BiB_i of the respective matroids if and only if

∣A∣≥∑i=1k(ri(E)−ri(E∖A))(A⊆E).(1)|A|\ge\sum_{i=1}^k\bigl(r_i(E)-r_i(E\setminus A)\bigr) \qquad(A\subseteq E). \tag{1}

Proof. If the bases exist, then

∣Bi∩A∣=ri(E)−∣Bi∖A∣≥ri(E)−ri(E∖A).|B_i\cap A|=r_i(E)-|B_i\setminus A| \ge r_i(E)-r_i(E\setminus A).

Disjointness permits summing these lower bounds inside AA, proving (1).

Conversely, (1) at A=EA=E gives the nonnegative integer N=∣E∣−∑iri(E)N=|E|-\sum_i r_i(E). It is at most ∣E∣|E|. Let M0M_0 be the uniform matroid consisting of all subsets of EE of size at most NN; it is a truncation of the free matroid and has rank r0(A)=min⁡(N,∣A∣)r_0(A)=\min(N,|A|).

A family of disjoint bases BiB_i leaves exactly NN elements, and so extends to a partition into independent sets of M0,M1,…,MkM_0,M_1,\ldots,M_k. Conversely, in any such partition,

∣E∣=∣I0∣+∑i∣Ii∣≤N+∑iri(E)=∣E∣.|E|=|I_0|+\sum_i|I_i| \le N+\sum_i r_i(E)=|E|.

Equality forces ∣I0∣=N|I_0|=N and ∣Ii∣=ri(E)|I_i|=r_i(E) for each ii. Thus the IiI_i are the required bases.

By Theorem 1c, the partition exists exactly when

∣A∣≤min⁡(N,∣A∣)+∑iri(A)(A⊆E).|A|\le\min(N,|A|)+\sum_i r_i(A)\qquad(A\subseteq E).

For ∣A∣≤N|A|\le N this is automatic, as is ∣A∣≤N+∑iri(A)|A|\le N+\sum_i r_i(A); for ∣A∣>N|A|>N the two tests are identical. Hence the criterion is equivalent to

∣A∣≤∣E∣−∑iri(E)+∑iri(A).|A|\le |E|-\sum_i r_i(E)+\sum_i r_i(A).

Replacing AA by its complement gives exactly (1). The nonnegativity check preceded the construction of M0M_0, so it also covers zero ranks and an empty ground set. □\square