Dominik Farhan

Matroids: Rank function and submodularity

Rank function

We define rank function r:2XNr: 2^X \rightarrow \mathbb{N} as follows r(A)=maxBABSBr(A) = \max_{B \subseteq A \land B \in S} |B|

Theorem (Equivalent definition of rr). Function r:2XNr: 2^X \rightarrow \mathbb{N} is the rank of a matroid iff it satisfies the following properties
  1. r()=0r(\emptyset) = 0
  2. r(Y)r(Y{y})r(Y)+1r(Y)\leq r(Y \cup \{ y\}) \leq r(Y) + 1
  3. r(Y{y})=r(Y{z})=r(Y)    r(Y{y,z})=r(Y)r(Y \cup \{y\}) = r(Y \cup \{z\}) = r(Y) \implies r(Y \cup \{ y, z\}) = r(Y)
Solution (    \implies)
Solution (\Longleftarrow)

Submodularity

Theorem (Submodularity). r:2XNr: 2^X \rightarrow \mathbb{N} is the rank of a matroid iff the following holds
  1. YX:0r(Y)Y\forall Y \subseteq X: 0 \leq r(Y) \leq |Y|
  2. ZY    r(Z)r(Y)Z \subseteq Y \implies r(Z) \leq r(Y)
  3. r(YZ)+r(YZ)r(Y)+r(Z)r(Y \cup Z) + r(Y \cap Z) \leq r(Y) + r(Z).
Solution (    \implies)
Solution (\Longleftarrow)

The previous theorem tells us that matroids are the systems where rank is monotone and submodular.

Submodular functions !

They are good for modeling value in combinatorial auctions. Let there be XX and f:2XNf: 2^X \rightarrow \mathbb{N}. Then for all xXx\in X we define Δfx:2XZ:TX:Δfx(T)=f(T{x})f(T).\Delta f_x : 2^X \rightarrow \mathbb{Z}: \forall T \subseteq X: \Delta f_x (T) = f(T \cup \{x\}) - f(T). In essence how much we value adding xx to TT.