Edge-connectivity of undirected and directed hypergraphs
Edge-connectivity of undirected and directed hypergraphs
Edge-connectivity of undirected and directed hypergraphs
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
30 Chapter 2. Submodular functions<br />
Pro<strong>of</strong>. This well-known result can be seen as a special case <strong>of</strong> the following claim (note<br />
that the claim does not hold for non-negative real numbers!):<br />
Claim 2.5. Let x1, . . . , xn be non-negative rational numbers. Suppose that we apply re-<br />
peatedly the following operation: for some indices i < j < k < l where xj <strong>and</strong> xk are<br />
positive, decrease xj <strong>and</strong> xk by min{xj, xk}, <strong>and</strong> increase xi <strong>and</strong> xl by min{xj, xk}. Then<br />
this operation can be repeated only a finite number <strong>of</strong> times.<br />
Pro<strong>of</strong>. By multiplying all xi values by a suitable integer, we can assume that every xi is<br />
integer. Now suppose that there is an infinite sequence <strong>of</strong> operations, <strong>and</strong> let m be the<br />
smallest index for which xm decreases infinitely many times. Then one <strong>of</strong> x1, . . . , xm−1<br />
increases infinitely many times by at least 1, but decreases finitely many times, which is<br />
impossible since n<br />
i=1 xi remains constant <strong>and</strong> xi ≥ 0 for every i.<br />
Let X1, . . . , Xt be an ordering <strong>of</strong> the subsets <strong>of</strong> V compatible with the st<strong>and</strong>ard partial<br />
order; let xi := y(Xi). Then it follows from the claim that after finitely many uncrossing<br />
steps uncrossing is impossible, therefore y is positive on a cross-free family.<br />
To illustrate the usefulness <strong>of</strong> the uncrossing technique, let p : 2 V → Z ∪ {−∞} be a set<br />
function for which p(X) + p(Y ) ≤ p(X ∩ Y ) + p(X ∪ Y ) whenever X <strong>and</strong> Y are crossing.<br />
Lemma 2.4 implies that for any c : V → Z+ <strong>and</strong> any c-composition F there is a cross-free<br />
c-composition F ′ such that <br />
X∈F ′ p(X) ≥ <br />
X∈F p(X).<br />
2.2 Submodular <strong>and</strong> supermodular set functions<br />
2.2.1 Basic properties<br />
A set function b : 2 V → Z ∪ {∞} is called fully submodular (or submodular for short) if<br />
b(X) + b(Y ) ≤ b(X ∩ Y ) + b(X ∪ Y ) (2.1)<br />
holds for every X, Y ⊆ V . It is clear from the definition that the sum <strong>of</strong> fully submodular set<br />
functions is fully submodular. A modular set function (obtained from a function m : V → Z<br />
by m(X) := <br />
v∈X m(v)) is always fully submodular. There is an alternative way to<br />
characterize submodularity:<br />
Proposition 2.6. A set function b : 2 V → Z ∪ {∞} is fully submodular if <strong>and</strong> only if<br />
for all X ⊆ Y ⊂ V <strong>and</strong> v ∈ V − Y .<br />
b(X + v) − b(X) ≥ b(Y + v) − b(Y )