22.08.2013 Views

Edge-connectivity of undirected and directed hypergraphs

Edge-connectivity of undirected and directed hypergraphs

Edge-connectivity of undirected and directed hypergraphs

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Section 4.4. (k, l)-edge-<strong>connectivity</strong> <strong>of</strong> <strong>directed</strong> <strong>hypergraphs</strong> 79<br />

is k-rooted-connected, then it can be shrinked (meaning that we delete all but one tail-<br />

nodes <strong>of</strong> every hyperarc) to a k-rooted connected digraph. For (k, l)-edge-<strong>connectivity</strong> no<br />

similar statement is true; in fact, Bang-Jensen <strong>and</strong> Thomassé [5] observed the following:<br />

Proposition 4.10 ([5]). It is NP-complete to decide whether a strongly connected <strong>directed</strong><br />

hypergraph can be shrinked to a strongly connected digraph.<br />

Pro<strong>of</strong>. Let D = (V, A) be a strongly connected digraph. We define a <strong>directed</strong> hypergraph<br />

D ′ = (V, A) the following way: A contains one hyperarc av for every v ∈ V , namely<br />

h(av) = v, <strong>and</strong> t(av) = {u ∈ V : uv ∈ A}. Then D ′ is strongly connected, <strong>and</strong> it can be<br />

shrinked to a strongly connected digraph if <strong>and</strong> only if D contains a Hamiltonian cycle.<br />

4.4.2 (k, l)-edge-<strong>connectivity</strong> augmentation<br />

Let D0 = (V, A0) be a <strong>directed</strong> hypergraph, <strong>and</strong> let k, l be non-negative integers. We<br />

are interested in the (k, l)-edge-<strong>connectivity</strong> augmentation <strong>of</strong> D0, either with the objective<br />

<strong>of</strong> minimizing the total size <strong>of</strong> the new hyperarcs, or with the objective <strong>of</strong> adding the<br />

minimum number <strong>of</strong> (r, 1)-hyperarcs . To solve these problems using the results in Section<br />

4.3, we define the following set function:<br />

⎧<br />

⎪⎨<br />

(k − ϱD0(X))<br />

p(X) :=<br />

⎪⎩<br />

+ if ∅ = X ⊆ V − s,<br />

(l − ϱD0(X)) + 0<br />

if s ∈ X ⊂ V ,<br />

if X = ∅ or X = V .<br />

(4.10)<br />

The set function p is positively crossing supermodular, so we can apply Theorems 4.8<br />

<strong>and</strong> 4.9. If k ≥ l, then the conditions <strong>of</strong> the theorems can be simplified:<br />

Claim 4.11. Let k ≥ l, <strong>and</strong> let p be the set function defined in (4.10). If<br />

γ ≥ <br />

p(Z)<br />

for every partition F <strong>of</strong> V , then<br />

for every partition F.<br />

Z∈F<br />

(|F| − 1) γ ≥ <br />

p(V − Z)<br />

Pro<strong>of</strong>. Suppose that (t − 1)γ < <br />

Z∈F p(V − Z) for a partition F = {X1, . . . , Xt} <strong>of</strong> V<br />

with t ≥ 2. If p(V − Xt) = 0, then <br />

Z∈F p(V − Z) = t−1<br />

i=1 p(V − Xi) ≤ t−1<br />

i=1 (p(Xi) +<br />

p(V − Xi)) ≤ (t − 1)γ, by considering the two-member partitions {Xi, V − Xi}. So we can<br />

Z∈F<br />

assume that p(V −Xi) > 0 for every i, thus <br />

Z∈F p(V −Z) = k+(t−1)l− t<br />

i=1 δD0(Xi) ≤<br />

k + (t − 1)l − t<br />

i=1 ρD0(Xi) ≤ (t − 1)k + l − t<br />

i=1 ρD0(Xi) ≤ t<br />

i=1 p(Xi) ≤ γ.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!