Domination and fractional domination in digraphs

In this paper, we investigate the relation between the (fractional) domination number of a digraph $G$ and the independence number of its underlying graph, denoted by $\alpha(G)$. More precisely, we prove that every digraph $G$ has fractional domination number at most $2\alpha(G)$, and every directed triangle-free digraph $G$ has domination number at most $\alpha(G)\cdot \alpha(G)!$. The first bound is sharp.


Introduction
Every digraph in this paper is simple, loopless and finite, where a digraph G is simple if for every two vertices u and v of G, there is at most one arc with endpoints {u, v}. Given a digraph G, we denote by V (G) and E(G) the vertex set and arc set of G, respectively. The independence number α(G) of a digraph G is the independence number of the underlying (undirected) graph of G. A digraph with independence number 1 is called a tournament. An Eulerian tournament is a tournament that is in addition Eulerian (i.e., the indegree of each vertex equals its outdegree).
Given a digraph G = (V, E), we say that a vertex of G dominates itself and all of its out-neighbors. A set of vertices S ⊆ V is called a dominating set if every vertex v ∈ V is either an element of S or is an out-neighbor of some element of S. The domination number γ(G) is the cardinality of a minimum (by cardinality) dominating set of G. Domination in tournaments has been well-studied [1], [6], [13], while little is known for domination in general digraphs. Recently, it was proved [4], [8], [9] that the domination number is closely related to the dichromatic number when the digraph is a tournament or a dense digraph. The topic of domination in undirected graphs (where S ⊆ V is a dominating set if every vertex v ∈ V is either an element of S or is a neighbor of some element of S) has also been studied extensively, see for example the monograph [10]. It is a well-known fact that in an undirected graph, any maximal independent set is also a dominating set.
Suppose we are given a digraph G = (V, E), a subset S of V , and a function g : is the set of in-neighbors of v. Thus, a natural linear relaxation of domination in digraphs arises as follows. A fractional dominating function of G is a function g : The fractional domination number γ * (G) is the smallest value of v∈V g(v) over all fractional dominating functions g of G. 1 The fractional domination number of a tournament was the main tool to prove the long standing Erdös-Sands-Sauer-Woodrow conjecture in [2].
In this paper, we show that for any digraph, its fractional domination number is at most twice its independence number, and this bound is sharp.
In contrast to the fractional domination number, it is not possible to bound the domination number of a digraph in terms of its independence number. Indeed, it was shown in [12] that almost surely a random tournament has domination number on the order of log n, much larger than its independence number of 1. However, the upper bound of log n on the domination number of a tournament can be extended to general digraphs. Sometimes, it is in fact possible to bound the domination number of a digraph purely in terms of its independence number. We discuss this further in the last section.

Notation
Given two vertices u, v, if uv, vu / ∈ E, we say that u and v are independent. We denote by N o G (v) the set of vertices that are independent with v (i.e., N o ). When it is clear from the context, we may omit the subscript G. Given a subset X of V , we denote by G[X] the induced subgraph of G on X. Given a digraph G = (V, E) and a function g on V , we write g(X) := v∈X g(v) for short. We sometimes use n to denote |V (G)|. Finally, we mention a trivial but useful observation regarding the independence number of a digraph.
has independence number at most α − 1.

Fractional domination in digraphs
In this section, we present two proofs of Theorem 1.1. The first proof uses the duality of linear programming, while the second proof is by induction. We first present some useful lemmas.
Proof. Suppose that the lemma is false. Then for the function p, we have a contradiction.
Proof. We prove the lemma by induction on |V |. The lemma clearly holds for |V | = 1.
For |V | > 1, fix some function p : V → [0, 1] and apply Lemma 2.1 to obtain a vertex v such that which proves the lemma.
We have the following remarks.
Thus, for the stable set S, we have This proves the lemma.
In the first proof of Theorem 1.1, we will use the following linear program. Let S be the set of all maximal stable sets in G, and let A be the matrix with |V | rows and |S| columns, where for every v ∈ V, S ∈ S, Let us consider the following linear program and its dual The value of an optimal solution for (P) is at most 2.
Proof. We prove that an optimal solution to (P) is at most 2, by proving that the optimal solution to (D) is at most 2. Then we apply the Strong Duality Theorem (see [3], Theorem 17.2 for example) to complete the proof.
Suppose for a contradiction that the optimal solution of (D) is greater than 2. Then there is a function w such that 1 T w > 2 and A T w ≤ 1. Then for every S ∈ S, v∈V A(v, S)w(v) ≤ 1, and so by (1), This proves the lemma.
We now restate Theorem 1.1. First proof of Theorem 2.4. Invoking Lemma 2.3, there is a z such that Az ≥ 1 and 1 T z ≤ 2. Note that z is a vector of length |S| and w is a vector of length |V |. Let g(v) := S:v∈S z(S) for every v ∈ V . Let α = α(G).
Note that for every S ∈ S, |S| ≤ α since S is stable. We have Fix v, since Az ≥ 1, we have S∈S A(v, S)z(S) ≥ 1. In other words, is a fractional dominating function of G). This proves the theorem.
In the second proof of Theorem 2.4, we will use the following consequence of Farkas' Lemma (we refer the reader to [11] for the proof of Lemma 2.5; see also Theorem 1 in [5]). Second Proof of Theorem 2.4. We prove the theorem by induction on α(G). If α(G) = 1, then G is a tournament. Let p be a function satisfying Lemma 2.5. Let g(v) = 2p(v) for every v ∈ V . Then g(V ) = 2 and for every v, we have Thus, g is a fractional domination function of G. We conclude that γ * (G) ≤ g(V ) = 2 = 2α(G) for the case α(G) = 1.
If α(G) > 1, let p be a function satisfying Lemma 2.5. By Observation 1.3, we have In the rest of the proof, we write G v := G[N o (v)] for short. For each vertex v, let g v be a minimum fractional dominating function of G v . Set g(x) = 2p(x) + y∈N o (x) g y (x)p(y) for every vertex x. We show that g is a fractional dominating function of G. Note that x ∈ N o (y) if and only if y ∈ N o (x), and for every v, y with v ∈ G y , since g y is a fractional dominating function of G y . Fix v, we have (we omit the subscript G if applicable) We conclude that g is a fractional dominating function of G.
Note that g v is a minimum fractional dominating function of g v , and so Thus, γ * (G) ≤ v∈V g(v) ≤ 2α(G). This completes the proof.
We can show that the bound in Theorem 2.4 is sharp.
Proposition 2.6. Given an arbitrarily small positive real number ε, for any positive integer k, there exists a digraph G such that α(G) = k and γ * (G) > 2k − ε.
Proof. Let r = ⌈1/ε⌉ + 1. For k = 1, let G be an Eulerian tournament with V (G) = {v 1 , ..., v 2r−1 } and N + (v i ) = {v j : 1 ≤ j − i mod (2r − 1) ≤ r − 1}. Let g be a minimum fractional dominating function of G. Suppose that g(v r ) = min i g(v i ). Note that i g(v i ) = γ * (G) ≤ 2, and r is chosen sufficiently large so that g(v r ) < 2/(2r − 1) < ε. We have For k > 1, let r = ⌈k/ε⌉ + 1. Let G be a disjoint union of k tournaments G 1 , . . . , G k , each is constructed as in the case k = 1. Since the k tournaments are disjoint, γ * (G) = We can also show that almost surely a random tournament has a fractional domination number close to the upper bound of 2. First, we need the following proposition.
Proof. Suppose that the statement was false, then there is a function g : V → [0, 1] such that g(V ) < n/(d + 1) and g(N − [x]) ≥ 1 for every x. Then We also need Chernoff's inequality (see for example [14]). Proposition 2.8 (Chernoff's Inequality). Let X be a binomial random variable consisting of n Bernoulli trials, each with probability of success p. Then, for all 0 < ǫ < 1, Proof. By Chernoff's bound, the probability that a given vertex has out-degree more than n/2 + 10 √ n log n is O(n −C ) for some constant C > 1. Thus, the probability that there is a vertex with out-degree more than n/2 + 10 √ n log n is, by the union bound, at most n · O(n −C ) = o(1). Thus, almost surely all the vertices of T n have out-degree at most n/2 + 10 √ n log n. It follows that almost surely γ * (T n ) ≥ n n/2+10 √ n log n .

Dominating sets in digraphs
In the previous section, we showed that the fractional domination number of a digraph can be bounded from above by twice its independence number. In general, we cannot bound the (integral) domination number of a digraph in terms of its independence number, as mentioned towards the end of the introduction section. Nevertheless, these two quantities can be related.
It is well known that a tournament has a dominating set of size at most log n [13]. Analogously, we can show that a digraph G = (V, E) has a dominating set of size at most α(G) · log n. For S ⊆ V , let χ(S) denote the chromatic number of the underlying (undirected) graph of digraph G[S]. Since each stable set S has cardinality at most α(G), Lemma 3.1 implies the following Theorem.
Theorem 3.2. Every digraph G = (V, E) with n = |V | has a dominating set of size at most α(G) · log n.
When is it possible to bound the domination number of a digraph purely in terms of its independence number? For example, Theorem 3.2 implies that this can be done when the independence number of a digraph is sufficiently large.
Another case in which the domination number of a digraph can be bounded in terms of its independence number is when the digraph is directed-triangle-free. For example, a directed-triangle-free digraph has independence number bounded by α(G) · α(G)! (see Theorem 3 in [7]). Moreover, when α(G) = 2, this bound can be improved to γ(G) ≤ 3 (see Theorem 4 in [7]).
In [8], we conjectured that the dichromatic number of a directed-triangle-free digraph can be bounded as a polynomial function of α(G). Let χ(G) denote the dichromatic number of a digraph. Then for any digraph G, γ(G) and χ(G) can be related as follows.
This follows from the fact that in a legal coloring, each color class forms an induced acyclic digraph and every acyclic digraph has a kernel (i.e., an independent dominating set). Thus, if the aforementioned conjecture holds, then the following must also hold.
Moreover, as pointed out in [7], the best theoretically possible upper bound on the domination number of directed-triangle-free digraph G cannot be better than γ(G) ≤