Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.
Updated
Source. Theorem 1c, statement on printed p. 150 and proof on p. 151 (published PDF).
Statement. Let , , be finite matroids on the same finite set , with ranks , where . There is an indexed partition
with empty parts permitted, if and only if
Proof. A partition gives , proving necessity. Assume (1). We show how to enlarge any family of pairwise disjoint independent sets whose union misses an element .
Whenever , at least one index satisfies
Otherwise the disjoint sets , together with the unassigned , would give , contrary to (1).
Start with . As long as , choose an index satisfying (2) for and set
By Lemma 2, this set has rank and is a proper subset of . Thus the chain ends at some with
We prove that a family with such a chain can be enlarged, by induction on its chain length . Put . If is independent, insert there and finish. This case always occurs when , by (3)–(4).
Otherwise let be the unique circuit in from Lemma 3. Because , is independent in . Let be the first index for which is independent. Then
The circuit containment follows because every dependent subset of contains its unique circuit. Choose . Since by , we have and .
Replace by , leaving the other sets unchanged. Lemma 3 makes independent. The family remains disjoint, has the same union size, and now leaves unassigned. We claim that the first steps of the old chain are a valid chain for this new family and .
For , both belong to . Proceed through the steps in increasing order, assuming the ambient is unchanged. If , nothing changes. If , put and . These are independent sets of the same size. The circuit shows that lies in the old span , so . The consequence in Lemma 2 gives . The strict rank inequality (2) also remains true because the independent-set size is unchanged.
This proves the claim, with no need to rebuild a potentially longer chain: and . Induction on the strictly smaller length now enlarges the new family. Hence the original family can also be enlarged, possibly after rearranging its elements.
Start with all and repeat this enlargement. Each successful enlargement increases the union size by one; finiteness of ends the process with a partition. For the initial family already is one.
The preservation of the shortened chain is the source's specific augmentation argument, expanded above. The earlier Edmonds papers mentioned on p. 151 are not being substituted for this proof.