Excluded Minor
Excluded Minor
66 (2008)
Abstract
The domination number of a graph G = (V, E) is the minimum size of a dominating set
U ⊆ V , which satisfies that every vertex in V \ U is adjacent to at least one vertex in U . The
notion of a problem kernel refers to a polynomial time algorithm that achieves some provable
reduction of the input size. Given a graph G whose domination number is k, the objective is
to design a polynomial time algorithm that produces a graph G0 whose size depends only on
k, and also has domination number equal to k. Note that the graph G0 is constructed without
knowing the value of k. Problem kernels can be used to obtain efficient approximation and exact
algorithms for the domination number, and are also useful in practical settings.
In this paper, we present the first nontrivial result for the general case of graphs with an
excluded minor, as follows. For every fixed h, given a graph G that does not contain Kh as a
topological minor, our polynomial time algorithm constructs a subgraph G0 of G, such that if
the domination number of G is k, then the domination number of G0 is also k and G0 has at most
k c vertices, where c is a constant that depends only on h. This result is improved for graphs
that do not contain K3,h as a topological minor, using a simpler algorithm that constructs a
subgraph with at most ck vertices, where c is a constant that depends only on h.
Our results imply that there is a problem kernel of polynomial size for graphs with an
excluded minor and a linear kernel for graphs that are K3,h -minor-free. The only previous
kernel results known for the dominating set problem are the existence of a linear kernel for the
planar case as well as for graphs of bounded genus.
Key words: H-minor-free graphs, degenerated graphs, dominating set problem, fixed-
parameter tractable algorithms, problem kernel.
∗
Schools of Mathematics and Computer Science, Tel-Aviv University, Tel-Aviv, 69978, Israel. Research supported
in part by a grant from the Israel Science Foundation, and by the Hermann Minkowski Minerva Center for Geometry
at Tel Aviv University. E-mail: [email protected]
†
School of Computer Science, Tel-Aviv University, Tel-Aviv, 69978, Israel. This paper forms part of a Ph.D. thesis
written by the author under the supervision of Prof. N. Alon and Prof. Y. Azar in Tel Aviv University. E-mail:
[email protected]
ISSN 1433-8092
1 Introduction
The notion of a kernel for the dominating set problem refers to a polynomial time algorithm that
given a graph G whose domination number is k, constructs a graph G0 whose size depends only on
k, and also has domination number equal to k. It is easy and known that a parameterized problem
is kernelizable if and only if it is fixed-parameter tractable. Thus, a fixed-parameter algorithm for
the dominating set problem gives a trivial kernel whose size is some function of k, not necessarily
a polynomial. Problem kernels can be used to obtain efficient approximation and exact algorithms
for the domination number, and are also useful in practical settings.
Our main result is a polynomial problem kernel for the case of graphs with an excluded minor.
This is the most general class of graphs for which a polynomial problem kernel has been established.
To the best of our knowledge, the only previous results are a linear kernel for the planar case as
well as for graphs of bounded genus. Our algorithms generalize and simplify the known results
for the planar case [4, 8]. For a general introduction to the field of parameterized complexity, the
reader is referred to [12] and [14].
Fixed-Parameter Algorithms for the Dominating Set Problem. The dominating set
problem on general graphs is known to be W [2]-complete [12]. This means that most likely there
is no f (k) · nc -algorithm for finding a dominating set of size at most k in a graph of size n for any
computable function f : N → N and constant c. This suggests the exploration of specific families
of graphs for which this problem is fixed-parameter tractable.
The method of bounded search trees has been used to give an O(8k n) time algorithm for
the dominating set problem in planar graphs [3] and an O((4g + 40)k n2 ) time algorithm for the
problem
√
in graphs of bounded√genus g ≥ 1 [13]. The algorithms √
for planar graphs were improved to
O(4 6 34k n) [1], then to O(227 k n) [17], and finally to O(215.13 k k + n3 + k4 ) [15]. Fixed-parameter
algorithms are now known also for map graphs [9] and for constant powers of H-minor-free graphs
[10]. The running time given√
in [10] for finding a dominating set of size k in an H-minor-free graph
G with n vertices is 2O( k) nc , where c is a constant depending only on H. In a previous paper,
we proved that the dominating set problem is fixed-parameter tractable for degenerated graphs,
by establishing an algorithm with running time kO(dk) n for finding a dominating set of size k in a
d-degenerated graph with n vertices [5].
Kernels for the Dominating Set Problem. The reduction rules introduced by Alber,
Fellows, and Niedermeier were the first to establish a linear problem kernel for planar graphs [4].
The kernel obtained was of size 335k, where k is the domination number of the graph. Fomin and
Thilikos proved that the same rules of Alber et al. provide a linear kernel of size O(k + g) for
graphs of genus g [16]. Chen et al. improved the upper bound for the planar case to 67k [8]. They
also gave the first lower bound, by proving that for any > 0, there is no (2 − )k kernel for the
planar dominating set problem, unless P = N P . It is interesting to note that Alber, Dorn, and
Niedermeier introduced a reduction rule that explores the joint neighborhood of l distinct vertices
[2], but this general rule has been applied only for l = 1 and l = 2, in order to prove that the
directed dominating set problem on planar graphs has a linear size kernel. Their reduction rule
generates a constraint, which is encoded by a corresponding gadget in the graph.
Our Results. By introducing a novel reduction rule, we prove that the dominating set problem
on graphs with an excluded minor admits a polynomial problem kernel. For graphs that are K3,h -
minor-free, the reduction rules of Alber, Fellows, and Niedermeier [4] are shown to give a linear
problem kernel. All the reduction rules described in this paper have the property that the only
modifications made to an input graph are the removal of vertices and edges. This implies that the
graph obtained, as a result of applying the rules, is a subgraph of the input graph. The advantages
1
of this approach are its simplicity and the fact that it preserves monotone properties, like planarity,
being H-minor-free, and degeneracy. We show that the rules of Alber et al. can also be described
in such a way.
Techniques. Our new reduction rule uses a succinct representation of all subsets of some
bounded size that dominate a given set of vertices. Interestingly, this is done by applying a fixed-
parameter algorithm for finding dominating sets in degenerated graphs. A challenging part of the
combinatorial proofs is to show that given a graph with an excluded minor and a dominating set
D of size k, there exists a subset of vertices U whose size is linear in k, such that all vertices not
in D ∪ U belong to the ”inner neighborhood” of a constant number of vertices from D ∪ U .
2 Preliminaries
The paper deals with undirected and simple graphs. Generally speaking, we will follow the notation
used in [7] and [11]. For a graph G = (V, E) and a vertex v ∈ V , N (v) denotes the set of all vertices
adjacent to v (not including v itself), whereas N [v] denotesS ∪ {v}. This is generalized
N (v) S to the
neighborhood of arbitrary sets by defining N (A) := v∈A N (v) \ A and N [A] := v∈A N [v]. The
graph obtained from G by deleting a vertex v is denoted G − v. The subgraph of G induced by
some set V 0 ⊆ V is denoted by G[V 0 ].
A dominating set of a graph G = (V, E) is a subset of vertices U ⊆ V , such that every vertex
in V \ U is adjacent to at least one vertex in U . The domination number of a graph G, denoted by
γ(G), is the minimum size of a dominating set. For a set of vertices A, if U ⊆ N [A], then we say
that A dominates U .
A graph G is d-degenerated if every induced subgraph of G has a vertex of degree at most d. A
d-degenerated graph with n vertices has less than dn edges. An edge is said to be subdivided when
it is deleted and replaced by a path of length two connecting its ends, the internal vertex of this
path being a new vertex. A subdivision of a graph G is a graph that can be obtained from G by a
sequence of edge subdivisions. If a subdivision of a graph H is the subgraph of another graph G,
then H is a topological minor of G. A graph H is called a minor of a graph G if it can be obtained
from a subgraph of G by a series of edge contractions.
In this paper, we consider only simple paths, that is, paths of the form x0 − x1 − · · · − xk , where
the xi are all distinct. The vertices x1 , . . . , xk−1 are the inner vertices of the path. The number of
edges of a path is its length. Suppose that D is a dominating set of G = (V, E) and n and l are
two constants. We denote by D b n,l the set of all vertices v ∈ V \ D for which there are n vertex
disjoint paths of length at most l from v to n different vertices of D. To avoid confusion, we stress
the fact that v is the starting vertex of all the paths, but any other vertex belongs to at most one
of the paths. The vertices of D b n,l are called central vertices, and when the values of n and l are
clear from the context, the simpler notation D b will be used.
Proposition 3.1. [6, 18] There exists a constant c such that, for every h, every graph that does
not contain Kh as a topological minor is ch2 -degenerated.
[19, 20, 21] There exists a constant c such that, for every h, every graph with
Proposition 3.2. √
no Kh minor is ch log h-degenerated.
2
The following Lemma from [5] gives an upper bound on the number of cliques of a prescribed
fixed size in a degenerated graph.
Lemma 3.3. If a graph G with n vertices is d-degenerated, then for every k ≥ 1, G contains at
d
most k−1 n copies of Kk .
The following results that bound the number of central vertices are stated for graphs with
no topological Kh , but they obviously apply also to graphs that are Kh -minor-free with better
constants.
Lemma 3.4. For a fixed h, suppose that G = (V, E) does not contain Kh as a topological minor,
and D is a dominating set of size k. For every fixed l, there exists a constant c that depends only
b h−1,l | ≤ ck.
on l and h, such that |D
Proof. Denote d = sh2 , where s is the constant from Proposition 3.1. Initially we set B to be
equal to D. Consider the vertices of V \ B in some arbitrary order. For each vertex w ∈ / B, if
there exist two vertex disjoint paths of length at most l from w to two vertices b1 , b2 ∈ B, such
that b1 and b2 are not connected, add the edge {b1 , b2 } and remove the vertex w from the graph
together with the two paths (the vertices b1 and b2 remain in the graph). Denote the resulting
graph by G0 . Obviously, G0 [B] does not contain Kh as a topological minor and therefore has at
most d|B| = dk edges. The number of edges in the induced subgraph G0 [B] is at least the number
of deleted vertices divided by (2l − 1), which means that at most dk(2l − 1) vertices were deleted so
far. We now return all the removed vertices back to the graph, add them to the set B, and declare
that this is the end of the first phase.
Consider a vertex v ∈ D b at the beginning of the phase. There are h − 1 vertex disjoint paths of
length at most l from v to a set H of h− 1 different vertices of D. Assume that when v is considered
in the arbitrary order, all the vertices of these h − 1 paths are still in the graph. We claim that the
h − 1 vertices of H cannot all be adjacent to each other, since otherwise they form a topological
Kh together with v. This means that if v was not removed from the graph during the phase, then
this can only happen in case there exists a vertex u on one the h − 1 vertex disjoint paths, which
was removed from the graph before v was considered. This vertex u was later added to B at the
end of the phase.
The set B is updated, but its size always remains O(k). We continue to perform additional
phases in which the vertices not in B are considered in some arbitrary order. For each vertex v ∈ D, b
there are h − 1 vertex disjoint paths of length at most l from v to h − 1 different vertices of D, and
these paths contain at most (h − 1)(l − 1) inner vertices. Thus, after 1 + (h − 1)(l − 1) phases, all
the vertices of Db will be added to B. This proves that |D| b = O(k).
Lemma 3.5. Suppose that G = (V, E) does not contain Km,h as a topological minor, and D is a
dominating set of size k. For every fixed l, there exists a constant c that depends only on l, m, and
b m,l | ≤ ck
h, such that |D
Proof. The proof is similar to that of Lemma 3.4, so we highlight only the modifications needed.
Initially we set B to be equal to D. During a phase, as long as there is still a vertex w ∈ / B for
which there are two vertex disjoint paths of length at most l from w to two vertices b1 , b2 ∈ B, such
that b1 and b2 are not connected, add the edge {b1 , b2 } and remove the vertex w from the graph
together with the two paths. Denote the resulting graph by G0 .
Consider a vertex v ∈ D b at the beginning of a phase. There are m vertex disjoint paths of
length at most l from v to a set M of m different vertices. Assume that none of the vertices on
these m paths were removed during the phase. This means that if v was not removed either, then
3
this can only happen in case G0 [M ] is a clique of size m. Since G0 does not contain Km,h as a
topological minor, there can be at most h − 1 vertices with m vertex disjoint paths to M . It follows
from Lemma 3.3 that there are at most O(k) cliques of size m in G0 [B], which means that only
b were not accounted for.
O(k) vertices of D
Lemma 4.1. Let G = (V, E) be a d-degenerated graph, and assume that B ⊆ V . If |B| > (4d+ 2)k,
then there are at most (4d + 2)k vertices in G that dominate at least |B|/k vertices of B.
Given a d-degenerated graph G = (V, E) and a set B ⊆ V that needs to be dominated, the
algorithm uses the method of bounded search trees. If |B| > (4d + 2)k, then denote by R the set
of all vertices that dominate at least |B|/k vertices of B. Every dominating set of size at most k
must contain a vertex from R. It follows from lemma 4.1 that |R| ≤ (4d + 2)k, so we can build our
search tree, by checking all possible options of adding one of the vertices of R to the dominating
set. This gives the following useful characterization of dominating sets in degenerated graphs.
Theorem 4.2. Suppose that G = (V, E) is d-degenerated and B ⊆ V . There is an a kO(dk) n time
algorithm for finding a family F of t ≤ (4d + 2)k k! pairs (Di , Bi ) of subsets of V , such that |Di | ≤ k
and |Bi | ≤ (4d + 2)k for every 1 ≤ i ≤ t, for which the following holds. If D ⊆ V is a subset of size
at most k that dominates B, then some i, 1 ≤ i ≤ t, satisfies that Di ⊆ D and Bi = B \ N [Di ].
Definition 5.1. Consider a subset of vertices A ⊆ V of the given graph G = (V, E). The neigh-
borhood of A is partitioned into four disjoint sets N1 (A), N2 (A), N3 (A), and N4 (A).
Note that in the original definitions from [4], which are described in section 6, the neighborhood
is partitioned into only three parts. Here, the definition of N3 (A) is modified and N4 (A) takes the
role of the ”inner neighborhood” of A. Here is a simple observation that follows immediately from
the previous definition.
Proposition 5.2. Let D be a dominating set of a graph G = (V, E). If v ∈ / N4 (A), then there is a
path of length at most 4 from v to a vertex of D, and the path does not contain any vertices of A.
4
Proof. Since v ∈
/ N4 (A), there is a path of length at most 3 from v to a vertex w ∈/ N [A], and the
path does not contain any vertices of A. Since D is a dominating set, this vertex w is adjacent to
some vertex d ∈ D. Since w ∈ / N [A], then obviously d ∈
/ A (it could be that d ∈ N (A)). This gives
a path of length at most 4 from v to d, as needed.
We now define our two reduction rules. Rule 2 applies Rule 1 as a subroutine. The main goal
of this section will be to analyze graphs for which Rule 2 cannot be applied anymore.
Rule 1: Let A ⊆ V be an independent set of the graph G = (V, E) and assume that N (v) 6= ∅
for every v ∈ A.
• For every 1 ≤ i ≤ t for which |Ai | > 2, let v, w ∈ Ai be two arbitrary distinct vertices.
Remove all the vertices of Ai \ {v, w} from the graph.
• Let F be a family of t ≤ (4d + 2)k k! pairs (Di , Bi ) of subsets of V , such that |Di | ≤ k and
|Bi | ≤ (4d + 2)k for every 1 ≤ i ≤ t for which the following holds. If D ⊆ V is a subset of
size at most k that dominates N3 (A) ∪ N4 (A), then some i, 1 ≤ i ≤ t, satisfies that Di ⊆ D
and Bi = (N3 (A) ∪ N4 (A)) \ N [Di ].
S
• Denote W := A ∪ ti=1 (Di ∪ Bi ). Remove all edges between vertices of (N3 (A) ∪ N4 (A)) \ W .
• Apply Rule 1 to the resulting graph and the independent set N4 (A) \ W .
Lemma 5.3. Let A ⊆ V be an independent set of the graph G = (V, E). Applying Rule 1 to G and
A does not change the domination number.
Proof. It is enough to prove that if a graph G = (V, E) contains an independent set {x, y, z}, such
that N (x) = N (y) = N (z) 6= ∅, then γ(G − z) = γ(G). We first prove that γ(G) ≤ γ(G − z). Let
D be a dominating set of G − z. If D ∩ N (x) 6= ∅, then D is also a dominating set of G. Otherwise,
{x, y} ⊆ D, so we can add one of the vertices of N (x) to D \ {y} and get a dominating set of G of
size |D|.
Now we prove the other direction γ(G − z) ≤ γ(G). Let D be a minimum dominating set of
G. It cannot be the case that {x, y, z} ⊆ D, since adding one of the vertices of N (x) to D \ {y, z}
results in a smaller dominating set. Thus, we can assume, without loss of generality, that z ∈ / D,
and therefore D is a dominating set of G − z.
Proof. Using the notations of Rule 2, denote by G0 the graph obtained from G by removing all
edges between vertices of (N3 (A)∪N4 (A))\W , just before Rule 1 is applied. It follows from Lemma
5.3 that in order to verify that Rule 2 does not change the domination number, it is enough to
prove that γ(G0 ) = γ(G). It is obvious that γ(G0 ) ≥ γ(G), since removing edges cannot decrease
5
the domination number. We now prove that γ(G0 ) ≤ γ(G). Let D be a minimum dominating
set of G, and let D 0 ⊆ D be a set of minimum size that dominates N3 (A) ∪ N4 (A). Obviously
|D 0 | ≤ k, since otherwise (D ∪ A) \ D 0 would be a smaller dominating set of G. Thus, from Theorem
4.2, some i, 1 ≤ i ≤ t, satisfies that Di ⊆ D 0 and Bi = (N3 (A) ∪ N4 (A)) \ N [Di ]. To prove that
D is also a dominating of G0 , we need to show that the vertices of (N3 (A) ∪ N4 (A)) \ W are
dominated by D in G0 , since the neighborhood of all other vertices remained the same. Assume
that v ∈ (N3 (A) ∪ N4 (A)) \ W . Since Bi ⊆ W , it follows that v ∈ / Bi , and therefore v is dominated
in G by some vertex d ∈ Di . This means that v is still dominated by d in G0 , since Di ⊆ W . This
completes the proof that Rule 2 does not change the domination number.
We now prove that when Rule 2 is applied, at least one vertex of N4 (A)\W is removed from the
graph G0 . First, note that (N3 (A)∪N4 (A))\W is an independent set, and therefore N4 (A)\W is also
independent. Given a vertex v ∈ N4 (A) \ W , obviously N (v) ⊆ A ∪ N3 (A) ∪ N4 (A) and N (v) 6= ∅,
adjacent in G0
since it is adjacent to at least one vertex of A. The important property of v is that it is S
only to vertices of W , since all other edges incident at v were removed. Since W = A∪ ti=1 (Di ∪Bi ),
it follows that |W | ≤ k + (4d + 2)k k!(k + (4d + 2)k)) = (4d + 3)k(4d + 2)k k! + k. It is easy to verify
k+1
that 2 · 2|W | + |W | ≤ 2|W |+2 ≤ 2(4dk+3k) < N4 (A). Thus, |N4 (A) \ W | ≥ |N4 (A)| − |W | > 2 · 2|W | .
By the pigeonhole principle, we conclude that there are three distinct vertices x, y, z ∈ N4 (A) \ W ,
such that N (x) = N (y) = N (z) 6= ∅. One of these three vertices will be removed by Rule 1.
The next Lemma is useful for showing that most of the vertices of a graph belong to an ”inner
neighborhood” of a set of vertices of constant size.
Lemma 5.5. Let D be a dominating set of the graph G = (V, E). If n ≥ 1 and v ∈ b n+1,4 ,
/ D∪D
b n+1,4 of size at most 40n , such that v ∈ N4 (A).
then there exists a subset A ⊆ D ∪ D 5
Proof. Let q be the maximum number of disjoint paths of length 4 from v to q different vertices of
D. Since v ∈ / D ∪ D,b it follows from the definition of D b that q ≤ n. Construct q such paths, whose
total length is the minimum possible. Denote by B the set of all vertices that appear in these q
paths and call the inner vertices of these paths B 0 := B \ (D ∪ {v}). Assign t := 3n(n + n2 + n4 ) + 1,
and assume, by contradiction, that v ∈ / N4 (A) for all subsets A ⊆ D ∪ Db of size at most 4(n + t − 1).
Note that 4(n + t − 1) ≤ 40n5 .
We will now construct t paths of length at most 4 and a series of t subsets A1 ⊆ A2 ⊆ · · · ⊆ At
of size at most 4(n + t − 1). Let A1 := B ∩ (D ∪ D). b For each i from 1 to t, we do the following.
According to our assumption v ∈ / N4 (Ai ), which means by Proposition 5.2 that there is a path of
length at most 4 from v to a vertex of D \ Ai , and this path does not contain any vertices from
Ai . Denote by Pi the vertices of a minimum length path, which satisfies these properties. Define
Ai+1 := Ai ∪ (Pi ∩ (D ∪ D)) b and proceed to the next iteration to construct Pi+1 .
Note that |A1 | ≤ 4n and |Ai+1 | ≤ |Ai |+4. Thus, all the sets Ai are of size at most 4n+4(t−1) =
4(n + t − 1). After completing this process, we get t paths of length 4 that start at v. Note that
a vertex u ∈ D b can participate in at most one of these paths, since once it appears in a path Pi ,
it is immediately added to Ai+1 . Because of the maximality of q, each path Pi must contains a
vertex of B 0 . From now on, we will consider the last appearance of a vertex from B 0 in a path Pi
as the starting point of the path. This means that all the paths Pi start at a vertex of B 0 and are
of length at most 3. Since |B 0 | ≤ 3q ≤ 3n and the number of paths is t = 3n(n + n2 + n4 ) + 1, by
the pigeonhole principle there must be a vertex b ∈ B 0 that is a starting point of n + n2 + n4 + 1
paths of length at most 3.
There are three possible cases.
Case 1: The vertex b is a starting point of at least n + 1 paths of length 1. This means that b
is adjacent to n + 1 vertices of D and therefore b ∈ D, b which means that b ∈ A1 . Thus, b cannot
6
belongs to any path Pi , and we get a contradiction.
Case 2: The vertex b is a starting point of at least n2 + 1 paths of length 2. It follows from the
construction that all these paths are from b to a different vertex of D. A vertex u cannot be the
middle vertex of more than n of these paths, since this would imply that u ∈ D, b but as mentioned
b
before, vertices of D can appear in at most one path. Thus, there are at least n + 1 middle vertices
that are part of n + 1 vertex disjoint paths of length 2 from b to D, which implies that b ∈ D. b This
is a contradiction.
Case 3: The vertex b is a starting point of at least n4 + 1 paths of length 3. The vertex b
is the first vertex of these paths, whereas the fourth vertex is always a different vertex from D.
Denote by U2 and U3 the vertices that appear as a second and third vertex on one of these paths,
respectively. Recall that when creating the paths Pi , we always chose a path of minimum length
that leads to a vertex of D. This implies that U2 ∩ U3 = ∅. As before, vertices of U2 and U3 can
belong to at most n2 and n paths, respectively. The total number of paths is n4 + 1, and therefore
|U2 | ≥ n2 + 1. Since a vertex of U3 belongs to at most n paths, we can find n + 1 vertices of U2 that
can be matched to n + 1 different vertices of U3 in a way which would give n + 1 vertex disjoint
paths of length 3 from b to n + 1 different vertices of D. Thus, b ∈ D,b and we get a contradiction.
We reached a contradiction in all three cases, and the claim is proved.
7
Proposition 6.3. Let D be a dominating set of a graph G = (V, E). If u ∈ / N3 (v), then there is a
path of length at most 3 from u to a vertex of D, and the path does not contain v. If u ∈
/ N3 (v, w),
then there is a path of length at most 3 from u to a vertex of D, and the path contains neither v
nor w.
The following is a simplified presentation of the two reduction rules from [4]. As proved there,
these reduction rules do not change the domination number of the graph. Unlike the original
rules, in which new vertices can be added to the graph, in our formulation the only modifications
made to the graph are the removal of vertices and edges. Another useful property of the following
formulation is that in case a rule is applied, at least one vertex is removed from the graph.
Rule 3: Given a graph G = (V, E) and a vertex v ∈ V , if |N3 (v)| > 1, then do the following.
Let v 0 be some arbitrary vertex of N3 (v). Remove all the vertices of N3 (v) \ {v 0 } and all the edges
incident at v 0 , except for {v, v 0 }.
Rule 4: Let v and w be two distinct vertices of the graph G = (V, E). If |N3 (v, w)| > 2 and
N3 (v, w) cannot be dominated by a single vertex from N2 (v, w) ∪ N3 (v, w), then do the following.
• If both v and w dominate N3 (v, w), then let z and z 0 be two arbitrary distinct vertices of
N3 (v, w). Remove all the vertices of N3 (v) \ {z, z 0 } and all the edges incident at z and z 0 ,
except for the edges {v, z}, {w, z}, {v, z 0 }, {w, z 0 }.
• If v dominates N3 (v, w) but w does not dominate it, then let v 0 be some arbitrary vertex of
N3 (v, w). Remove all the vertices of N3 (v) \ {v 0 } and all the edges incident at v 0 , except for
the edge {v, v 0 }. The case that only w dominates N3 (v, w) is handled in a symmetric manner.
• If neither v nor w dominate N3 (v, w), then let v 0 and w0 be two arbitrary distinct vertices
of N3 (v, w) such that v 0 is adjacent to v and w0 is adjacent to w. Remove all the vertices of
N3 (v) \ {v 0 , w0 } and all the edges incident at v 0 and w0 , except for the edges {v, v 0 }, {w, w0 }.
A graph is called reduced in case Rules 3 and 4 cannot be applied to it anymore. The following
definitions are specific to this section.
Definition 6.4. Let D be a dominating set of the graph G = (V, E).
e the set of vertices in V \ D that have at least two neighbors from D.
• Denote by D
• Suppose that d1 , d2 ∈ D are two distinct vertices. Denote by Inner(d1 , d2 ) the set of all inner
b e
S 3 of the type d1 −x−y−d2 , such that x, y ∈ N3 (d1 , d2 )\(D∪D3,3 ∪D).
vertices of paths of length
Denote Inner(D) := d1 ,d2 ∈D,d1 6=d2 Inner(d1 , d2 )
Lemma 6.5. For a fixed h ≥ 2, suppose that G = (V, E) is a reduced graph that contains no K3,h
e = O(k).
as a topological minor. If D is a dominating set of size k, then |D|
Proof. Assume that v ∈ D. e This means that v is adjacent to at least 2 vertices of D, so we
distinguish between three cases.
Case 1: The vertex v is adjacent to at least 3 vertices of D. Thus, by definition v ∈ D b 3,3 , and
it follows from Lemma 3.5 that |D b 3,3 | = O(k).
Case 2: The vertex v is adjacent to exactly 2 vertices d1 , d2 ∈ D and v ∈ / N3 (d1 , d2 ). It follows
from proposition 6.3 that there is a path of a length at most 3 from v to a vertex of D, and the
path does not use the vertices d1 and d2 . This implies that v ∈ D b 3,3 and we proceed as in the
previous case.
Case 3: The vertex v is adjacent to exactly 2 vertices d1 , d2 ∈ D and v ∈ N3 (d1 , d2 ). The
number of pairs d1 , d2 ∈ D for which there is a vertex v ∈
/ D such that N (v) ∩ D = {d1 , d2 } is O(k).
8
To see this, just connect each such pair d1 , d2 , in case they were not connected before. Denote
the resulting graph by G0 . The number of edges in G0 [D] is at least the number of pairs we are
counting. Since G0 [D] does not contain K3,h as a topological minor, it has O(k) edges.
It is now enough to prove that |N3 (d1 , d2 )| ≤ h for every two distinct vertices d1 , d2 ∈ D.
By contradiction, assume that |N3 (d1 , d2 )| > h ≥ 2. Since the graph is reduced, there is a vertex
v ∈ N2 (d1 , d2 )∪N3 (d1 , d2 ) that dominates N3 (d1 , d2 ). Note that v can possibly belong to N3 (d1 , d2 ).
This implies that d1 ,d2 , and v together with N3 (d1 , d2 ) \ {v} form a K3,h . This is a contradiction,
and the claim is proved.
Corollary 6.6. For a fixed h ≥ 2, let D be a dominating set of size k of a reduced graph G = (V, E)
that contains no K3,h as a topological minor. If a subset U ⊆ V of size m satisfies that D ∩ U = ∅,
then |N [U ]| = O(k + m).
Proof. The set D ∪ U is obviously a dominating set. A vertex v ∈ N [U ] \ (D ∪ U ) is adjacent to a
vertex of U and also to a vertex of D, since D is a dominating set. This means that v is adjacent
to at least two vertices of D ∪ U . The result now follows from Lemma 6.5.
Lemma 6.7. Suppose that G = (V, E) is a reduced graph that contains no K3,h as a topologi-
cal minor. If D is a dominating set of size k, then there are O(k) pairs d1 , d2 ∈ D for which
Inner(d1 , d2 ) 6= ∅.
Proof. Consider the pairs d1 , d2 ∈ D for which Inner(d1 , d2 ) 6= ∅ in some arbitrary order. For each
b 3,3 ∪ D)
such pair d1 , d2 , there are two vertices x, y ∈ N3 (d1 , d2 ) \ (D ∪ D e that appear on the path
d1 − x − y − d2 . We claim that both x and y do not belong to any other pair Inner(d01 , d02 ). To see
this, suppose by contradiction that x ∈ Inner(d1 , d2 ) ∩ Inner(d01 , d02 ) for {d01 , d02 } =
6 {d1 , d2 }. Since
x∈ e
/ D, it has only one neighbor in D, so assume, without loss of generality, that x is adjacent to
d1 = d01 and x appears on the two paths d1 − x − y − d2 and d1 − x − z − d02 . This implies that
x∈D b 3,3 , a contradiction, and the claim is proved.
In each case as above, we delete the vertices x and y, and add an edge between d1 and d2 ,
assuming this edge does not exist. Denote the resulting graph by G0 . Obviously, G0 [D] does not
contain K3,h as a topological minor and therefore has at most O(k) edges. The number of edges
in the induced subgraph G0 [D] is at least the number of pairs for which Inner(d1 , d2 ) 6= ∅, as
claimed.
Lemma 6.8. Let D be a dominating set of a reduced graph G = (V, E) that contains no K3,h as a
topological minor. Every two distinct vertices d1 , d2 ∈ D satisfy |Inner(d1 , d2 )| ≤ 2h2 .
Proof. By contradiction, assume that |Inner(d1 , d2 )| ≥ 2h2 + 1. This implies that |N3 (d1 , d2 )| >
2, and since the graph is reduced, there is a vertex v ∈ N2 (d1 , d2 ) ∪ N3 (d1 , d2 ) that dominates
N3 (d1 , d2 ). Let q the maximum number of internally-disjoint paths of the type d1 − x − y − d2 , such
that x, y ∈ Inner(d1 , d2 ), and denote by W the 2q inner vertices of these paths. Note that v can
possibly belong to W . We must have that q ≤ h, since otherwise d1 ,d2 , and v would be part of a
topological K3,h . Since |W | = 2q ≤ 2h, there are at least 2h(h − 1) + 1 vertices of Inner(d1 , d2 ) \ W
that appear on a path of the type d1 − x− y − d2 together with one of the vertices of W . Thus, there
is a vertex w ∈ W that belongs to at least h of these paths. Assuming, without loss of generality,
that w is adjacent to d1 , there are h + 1 different paths of length 2 from w to d2 , and the inner
vertices of these paths are from Inner(d1 , d2 ). Thus, w,d2 , and v are part of a topological K3,h .
This is a contradiction, and the claim is proved.
Lemma 6.9. Suppose that the reduced graph G = (V, E) contains no K3,h as a topological minor.
If D is a dominating set of size k, then |Inner(D)| = O(k).
9
Proof. Follows immediately from Lemmas 6.7 and 6.8.
Lemma 6.10. Suppose that the reduced graph G = (V, E) contains no K3,h as a topological minor.
If D is a dominating set of size k, then the number of vertices that appear on a path of length 3
between two vertices of D is O(k).
Proof. We examine the inner vertices of paths of the form d1 − v − x − d2 , such that d1 , d2 ∈ D. It
follows from Lemmas 3.5 and 6.5 that |D b 3,3 ∪ D|
e = O(k), which means that it remains to count the
number of vertices not in D ∪ D b 3,3 ∪ D.e Assume that v ∈ / D∪D b 3,3 ∪ D.
e Since v ∈ e it is adjacent
/ D,
to exactly one vertex of D, and therefore x ∈ b e b
/ D. If x ∈ D3,3 ∪ D, then v ∈ N [D3,3 ∪ D],e but it
follows from Corollary 6.6 that |N [D b 3,3 ∪ D]|
e = O(k). If either v or x do not belong to N3 (d1 , d2 ),
then this implies that x ∈ D b 3,3 , but this case has already been addressed. The only remaining case
is that v, x ∈ N3 (d1 , d2 ) \ (D ∪ D b 3,3 ∪ D),
e which means that v ∈ Inner(D), and we know from
Lemma 6.9 that |Inner(D)| = O(k).
Theorem 6.11. For every fixed h, given a graph G that does not contain K3,h as a topological
minor, there is a polynomial time algorithm that constructs a subgraph G0 of G, such that if γ(G) =
k, then γ(G0 ) = k and G0 has at most ck vertices, where c is a constant that depends only on h.
Proof. Suppose that G contains no K3,h as a topological minor and γ(G) = k. As long as the
conditions of Rules 3 and 4 are satisfied, apply these rules to get a reduced subgraph G0 . Alber et.
al [4] proved that γ(G0 ) = k ,so let D be a dominating set of G0 of size k. It follows from Lemma
e = O(k), so we need to count the number of vertices not in D ∪ D.
6.5 that |D| e Assume v ∈ / D∪D e
is adjacent to d1 ∈ D. If v ∈ N3 (d1 ), then in a reduced graph |N3 (d1 )| ≤ 1, which means that
there could be at most k vertices of this type. Assume now that v ∈ / N3 (d1 ), so by Proposition 6.3
there is a path of length at most 3 from v to a vertex d2 ∈ D, and d1 is not part of this path. We
examine a shortest path p from d1 to d2 , in which v is the second vertex of the path. Since v ∈ e
/ D,
it is adjacent to only one vertex of D, so the path p can be of length either 3 or 4.
In case p is of length 3, then it follows from Lemma 6.10 that there are at most O(k) vertices
of this type. If p is of length 4, denote it by d1 − v − x − y − d2 , where x, y ∈ / D. The vertex x is
adjacent to some vertex of D. It cannot be adjacent to d2 , since a path p on minimum length was
chosen. If x is adjacent to a vertex of D \ {d1 , d2 }, then x ∈ Db 3,3 and v ∈ N [D b 3,3 ], but it follows
from Corollary 6.6 that |N [D b 3,3 ]| = O(k). The remaining case is that d1 is the only vertex in D
that is adjacent to x. Since x ∈ / D is on a path of length 3 from d1 to d2 , it follows from Lemma
6.10 and Corollary 6.6 that the number of vertices v of this type is also O(k).
• Another challenging question is to characterize the families of graphs for which the dominating
set problem admits a linear kernel. We cannot rule out the possibility that a linear kernel
can be obtained for graphs with any fixed excluded minor.
10
References
[1] J. Alber, H. L. Bodlaender, H. Fernau, T. Kloks, and R. Niedermeier. Fixed parameter
algorithms for DOMINATING SET and related problems on planar graphs. Algorithmica,
33(4):461–493, August 2002.
[2] Jochen Alber, Britta Dorn, and Rolf Niedermeier. A general data reduction scheme for dom-
ination in graphs. In Jirı́ Wiedermann, Gerard Tel, Jaroslav Pokorný, Mária Bieliková, and
Julius Stuller, editors, SOFSEM, volume 3831 of Lecture Notes in Computer Science, pages
137–147. Springer, 2006.
[3] Jochen Alber, Hongbing Fan, Michael R. Fellows, Henning Fernau, Rolf Niedermeier,
Frances A. Rosamond, and Ulrike Stege. A refined search tree technique for dominating set
on planar graphs. J. Comput. Syst. Sci, 71(4):385–405, 2005.
[4] Jochen Alber, Michael R. Fellows, and Rolf Niedermeier. Polynomial-time data reduction for
dominating set. Journal of the ACM, 51(3):363–384, May 2004.
[5] Noga Alon and Shai Gutner. Linear time algorithms for finding a dominating set of fixed size
in degenerated graphs. In Guohui Lin, editor, COCOON, volume 4598 of Lecture Notes in
Computer Science, pages 394–405. Springer, 2007.
[6] Béla Bollobás and Andrew Thomason. Proof of a conjecture of Mader, Erdös and Hajnal on
topological complete subgraphs. Eur. J. Comb, 19(8):883–887, 1998.
[7] J. A. Bondy and U. S. R. Murty. Graph theory with applications. American Elsevier Publishing
Co., Inc., New York, 1976.
[8] Jianer Chen, Henning Fernau, Iyad A. Kanj, and Ge Xia. Parametric duality and kernelization:
lower bounds and upper bounds on kernel size. SIAM J. Comput., 37(4):1077–1106, 2007.
[9] Erik D. Demaine, Fedor V. Fomin, Mohammadtaghi Hajiaghayi, and Dimitrios M. Thilikos.
Fixed-parameter algorithms for (k, r)-center in planar graphs and map graphs. ACM Trans-
actions on Algorithms, 1(1):33–47, July 2005.
[10] Erik D. Demaine, Fedor V. Fomin, Mohammadtaghi Hajiaghayi, and Dimitrios M. Thilikos.
Subexponential parameterized algorithms on bounded-genus graphs and H-minor-free graphs.
Journal of the ACM, 52(6):866–893, November 2005.
[11] Reinhard Diestel. Graph theory, volume 173 of Graduate Texts in Mathematics. Springer-
Verlag, Berlin, third edition, 2005.
[12] R. G. Downey and M. R. Fellows. Parameterized complexity. Monographs in Computer Science.
Springer-Verlag, New York, 1999.
[13] John A. Ellis, Hongbing Fan, and Michael R. Fellows. The dominating set problem is fixed
parameter tractable for graphs of bounded genus. J. Algorithms, 52(2):152–168, 2004.
[14] J. Flum and M. Grohe. Parameterized complexity theory. Texts in Theoretical Computer
Science. An EATCS Series. Springer-Verlag, Berlin, 2006.
[15] Fedor V. Fomin and Dimitrios M. Thilikos. Dominating sets in planar graphs: branch-width
and exponential speed-up. In Proceedings of the Fourteenth Annual ACM-SIAM Symposium
on Discrete Algorithms, pages 168–177, 2003.
11
[16] Fedor V. Fomin and Dimitrios M. Thilikos. Fast parameterized algorithms for graphs on
surfaces: Linear kernel and exponential speed-up. In Josep Dı́az, Juhani Karhumäki, Arto
Lepistö, and Donald Sannella, editors, ICALP, volume 3142 of Lecture Notes in Computer
Science, pages 581–592. Springer, 2004.
[17] Iyad A. Kanj and Ljubomir Perkovic. Improved parameterized algorithms for planar dominat-
ing set. In Krzysztof Diks and Wojciech Rytter, editors, MFCS, volume 2420 of Lecture Notes
in Computer Science, pages 399–410. Springer, 2002.
[18] János Komlós and Endre Szemerédi. Topological cliques in graphs II. Combinatorics, Proba-
bility & Computing, 5:79–90, 1996.
[19] Alexandr V. Kostochka. Lower bound of the Hadwiger number of graphs by their average
degree. Combinatorica, 4(4):307–316, 1984.
[20] Andrew Thomason. An extremal function for contractions of graphs. Math. Proc. Cambridge
Philos. Soc., 95(2):261–265, 1984.
[21] Andrew Thomason. The extremal function for complete minors. J. Comb. Theory, Ser. B,
81(2):318–338, 2001.
12
http://eccc.hpi-web.de/