Definition
Rank (Matroid)
The rank of a matroid is defined as the cardinality of any basis of , i.e.,
Submodularity
Submodularity
Let be a finite matroid with rank function . For all ,
Submodularity
Fix . We construct one independent set in whose intersection with already has the largest possible size.
Fill the intersection, then extend to the union
Choose an independent set with . Extend by adding elements of while preserving independence, until no further element can be added. Call the resulting set .
By exchangeability, a maximal independent subset of has maximum size: a larger independent subset would supply another element to add. Hence
Why the intersection stays exactly
Suppose extending to added some . Since , we would have
By heredity, deleting elements from the independent set leaves an independent set. In particular, delete everything except and :
Schematic: and are outside . The orange region contains the retained elements , not necessarily all of . Deleting exposes an independent subset that is too large; the same argument works for any size of .
The remaining set would therefore satisfy all three conditions:
But an independent subset of has size at most . This would force
a contradiction. Adding elements outside cannot repair the dependent subset . Thus extending to adds no further element of :
Bound the two portions
Both and are independent, since they are subsets of . By the definition of rank,
Count the overlap twice
The two portions and cover and overlap exactly in . Adding their sizes counts every element of once, and every element of once more:
Shown for and : each column is one element, and each dot is one occurrence in the sum. The same count gives, for arbitrary and ,
This is the required inequality.
Function
Definition
Link to originalRank Function (Matroid)
The notion of ranks extends to arbitrary subsets. The rank function is
the size of the largest independent set inside . The rank is .