Short Cycles make W-hard problems hard:

FPT algorithms for W-hard Problems in Graphs

with no short Cycles ?

Venkatesh Raman and Saket Saurabh

The Institute of Mathematical Sciences,
Chennai 600 113.
{vraman|saket}@imsc.res.in

Abstract. We show that several problems that are hard for various
parameterized complexity classes on general graphs, become fixed pa-
rameter tractable on graphs with no small cycles.
More specifically, we give fixed parameter tractable algorithms for Domi-
nating Set, t-Vertex Cover (where we need to cover at least t edges)
and several of their variants on graphs with girth at least five. These
problems are known to be W [i]-hard for some i ≥ 1 in general graphs.
We also show that the Dominating Set problem is W [2]-hard for bipartite
graphs and hence for triangle free graphs.
In the case of Independent Set and several of its variants, we show these
problems to be fixed parameter tractable even in triangle free graphs.
In contrast, we show that the Dense Subgraph problem where one is
interested in finding an induced subgraph on k vertices having at least
l edges, paramaterized by k, is W [1]-hard even on graphs with girth at
least six.
Finally, we give an O(log p) ratio approximation algorithm for the Dom-
inating Set problem for graphs with girth at least 5, where p is the size
of an optimum dominating set of the graph. This improves the previous
O(log n) factor approximation algorithm for the problem, where n is the
number of vertices of the input graph.

1 Introduction

Parameterized complexity is a practical approach to deal with intractable

computational problems having some small parameters. For decision prob-
lems with input size n, and a parameter k (which typically, and in all the
?
A preliminary version of this paper appeared as Triangles, 4-cycles and Parameter-
ized (In-)Tractability in the proceedings of 10th Scandinavian Workshop on Algo-
rithm Theory (SWAT 2006) [24].
2

problems we consider in this paper, is the solution size), the goal here
is to design an algorithm with runtime f (k)nO(1) where f is a function
of k alone, as contrasted with a trivial nk+O(1) algorithm. Problems hav-
ing such an algorithm is said to be fixed parameter tractable (FPT),
and such algorithms are practical when small parameters cover practical
ranges. The book by Downey and Fellows [8] provides a good introduc-
tion to the topic of parameterized complexity. For recent developments
see the books by Flum and Grohe [14] and Niedermeier [17].
There is a hierarchy of intractable parameterized problem classes
above FPT, the main ones are:

F P T ⊆ M [1] ⊆ W [1] ⊆ M [2] ⊆ W [2] ⊆ · · · ⊆ W [P ] ⊆ XP.

The principal analogue of the classical intractability class NP is W [1].

A convenient source of W [1]-hardness reductions is provided by the re-
sult that Independent Set is complete for W [1] [8]. Other highlights
of the theory include that Dominating Set, by contrast, is complete
for W [2] [8]. Surprisingly we show that these problems and several of
their variants that are known to be hard in the W-hierarchy, are fixed
parameter tractable on graphs that have no short cycles – more specif-
ically on graphs with girth at least five. These problems are known to
be NP-complete on such graphs as well [4, 5]. We also look at the Set
Cover problem where the size of the intersection of any pair of sets is
bounded by a fixed constant. While the general version of Set Cover is
known to be W [2]-complete, we prove this special version fixed parameter
tractable.
Most of our algorithms are based on the method of kernelization. The
main idea of kernelization is to replace a given instance (I, k) by a simpler
instance (I 0 , k 0 ) using some data reduction rules in polynomial time such
that (I, k) is a yes instance if and only if (I 0 , k 0 ) is a yes instance and |I 0 |
is bounded by a function of k alone. The reduced instance is called kernel
for the problem. For most of our problems we give polynomial sized kernel
in polynomial time.

1.1 Organization of the Rest of the Paper

In Section 2, we look at the Dominating Set problem and show that
the problem is W [2]-complete even in bipartite graphs and split graphs
(a graph in which the vertices can be partitioned into a clique and an
independent set). Though variations of Dominating Set like Red-Blue
Dominating Set [10] and Constrained Dominating Set [16] have
 3

been studied before and shown to be W[2]-complete, to the best of our

knowledge the standard Dominating Set problem (which we consider
here) in bipartite graphs has not been studied before. Our observation
means that the dominating set problem is W [2]-complete in triangle free
graphs. Then we show that the problem is FPT if the input graph has
girth at least 5. It turns out that this result can be generalized to several
variants of the Dominating Set problem on graphs with girth at least
five.
In Section 3, we look at the Set Cover problem for which Domi-
nating Set is a special instance. Set Cover problem is known to be
W [2]-complete [8]. Here we show that if the set cover instance satisfies
the property that the intersection of any pair of its sets is bounded by a
fixed constant then the problem is fixed parameter tractable.
In Section 4, we look at t-Vertex Cover and t-Dominating Set
problems. These are generalizations of Vertex Cover and Dominating
Set problems. In the t-Vertex Cover problem, we are interested in
finding a set of at most k vertices covering at least t edges and in the
t-Dominating Set problem the objective is to find a set of at most k
vertices that dominates at least t vertices. Both these problems have been
parameterized in two different ways: by k alone and by both k and t. Both
these problems are fixed parameter tractable when parameterized by both
k and t. Bläser [6] gave O(2O(t) nO(1) ) algorithm for both the problems
using color coding technique. Guo et. al. [18] have shown that t-Vertex
Cover is W [1]-complete when parameterized by k alone. It is easy to
see that the t-Dominating Set is W [2]-complete by a reduction from
Dominating set when parameterized by k alone. We show that both
these problems are fixed parameter tractable in graphs with girth at least
five, when parameterized by k alone.
In Section 5, we look at the Independent Set problem and several
of its variants. We show that these problems are fixed parameter tractable
in triangle free graphs while they are W[1]-complete in general graphs.
In contrast to our results in earlier sections, in Section 6, we exhibit
a problem that is W [1]-hard in graphs with no small cycles. This is the
Dense Subgraph problem [22]. Here, given a graph G = (V, E) and
positive integers k and l, the problem is to determine whether there exists
a set of at most k vertices C ⊆ V such that the induced subgraph on C
has at least l edges; here k is the parameter.
In Section 7, we deviate and look at the approximability result of
the Dominating Set problem. We conclude that the Dominating Set
problem is as hard to approximate in bipartite graphs as in general undi-
4

rected graphs. We also give an approximation algorithm of factor O(log p)

for the Dominating Set problem if the input graph has girth at least
5, where p is the size of an optimum dominating set of the input graph.
This improves the previously known approximation algorithm of factor
O(log n), where n is the number of vertices in the input graph.
Section 8 gives some concluding remarks and open problems.
We assume that all our graphs are simple and undirected. Given a
graph G = (V, E), n represents number of vertices, and m represents the
number of edges. For a subset V 0 ⊆ V , by G[V 0 ] we mean the subgraph
of G induced on V 0 . By N (u) we represent all vertices (excluding u) that
are adjacent to u, and by N [u], we refer to N (u) ∪ {u}. Similarly, for a
subset D ⊆ V , we define N [D] = ∪v∈D N [v]. By the girth of a graph, we
mean the length of the shortest cycle in the graph. We say that a graph
is a Gi graph if the girth of the graph is at least i. A vertex is said to
dominate all its neighbors.

2 Dominating Set and its Variants

In this section we look at the Dominating Set problem and its variants.

Dominating Set: Given a graph G = (V, E) and an integer k ≥ 0,

determine whether there exists a set D ⊆ V , of size at most k, such
that for every vertex u ∈ V , N [u] ∩ D 6= ∅.

We say that the set D “dominates” the vertices of G. We first show

that Dominating Set problem is W[2]-complete in bipartite graphs and
split graphs by a reduction from the same problem in general undirected
graphs. Then we give a fixed parameter tractable algorithm for the prob-
lem in graphs with girth at least 5.

2.1 Dominating Set in Bipartite and Split Graphs

Theorem 1. Dominating Set problem is W [2]-complete in bipartite

graphs.

Proof. We prove this by giving a reduction from the Dominating Set

problem in general undirected graphs. Given an instance (G = (V, E), k)
of Dominating Set, we construct a bipartite graph H = (V 0 , E 0 ). Let
z1 and z2 be two new vertices (not in V ). Now V 0 = V1 ∪ V2 where
V1 = {u1 | u ∈ V } ∪ {z1 } and V2 = {u2 | u ∈ V } ∪ {z2 }. If there is an edge
(u, v) in E then we draw the edges (u1 , v2 ) and (v1 , u2 ). We also draw
 5

edges of the form (u1 , u2 ) for every u ∈ V . Finally, we add an edge from
every vertex in V1 to z2 . This completes the construction of H.
We show that G has a dominating set of size k if and only if H has
a dominating set of size k + 1. Let D be a dominating set of size k in G.
Then clearly D0 = {u1 | u ∈ D} ∪ {z2 } is a dominating set of size k + 1
in H. Conversely, let K be a dominating set in H of size k + 1. Observe
that either z1 or z2 must be part of K as z2 is the unique neighbor of z1 .
Without loss of generality, we can assume that z2 ∈ K, as otherwise we
could delete z1 and include z2 in K and still have a dominating set of size
at most k + 1 in H. Now take D = {u | u ∈ V, u1 or u2 ∈ K}. Clearly
D is of size k. We show that D is a dominating set in G. For any u ∈ / D,
u2 ∈
/ K and hence there exists some v1 ∈ K such that v1 dominates u2 in
H. But this implies v ∈ D and (v, u) ∈ E, which shows that v dominates
u. This proves that D is a dominating set of size k for G and establishes
the theorem. t
u

Since every bipartite graph is also triangle free, we have the following
corollary.
Corollary 1. Dominating Set problem is W [2]-complete in triangle
free graphs.

Theorem 2. Dominating Set problem is W [2]-complete in split graphs.

Proof. We again prove this by giving a reduction from the Dominat-

ing Set problem in general undirected graphs. Given an instance (G =
(V, E), k) of Dominating Set, we construct a split graph H = (V 0 , E 0 ).
We create two copies of V namely V1 = {u1 | u ∈ V } and V2 = {u2 | u ∈
V }. If there is an edge (u, v) in E then we draw the edges (u1 , v2 ) and
(v1 , u2 ). We also draw edges of the form (u1 , u2 ) for every u ∈ V . Now we
make H[V1 ] a complete graph by adding all arcs of the form (u1 , v1 ) for
every pair of vertex u1 , v1 ∈ V1 . This completes the construction of H.
It is easy to see that H is a split graph with H[V1 ] as a clique and H[V2 ]
as an independent set.
As in the proof of Theorem 1, it is easy to see that G has a dominating
set of size k if and only if H has a dominating set of size k. t
u

An undirected graph is chordal if every cycle of length greater than three

possesses a chord, that is, an edge joining two nonconsecutive vertices of
the cycle. It is well known that every split graph is also a chordal graph
and hence Theorem 2 implies that the Dominating Set is W [2]-complete
in chordal graphs. As a corollary of Theorem 2 we get the following.
6

Corollary 2. Dominating Set problem is W [2]-complete in chordal

graphs.

2.2 FPT Algorithm for Dominating Set in G5 Graphs

We give a fixed parameter tractable algorithm for the Dominating Set
problem in graphs with girth at least 5 (G5 graphs) and also observe that
various other W -hard problems become tractable for G5 graphs.
Our algorithm follows a branching strategy where at every iteration
we find a vertex that needs to be included in the Dominating Set which
we are trying to construct. Once a vertex is included, we can at best
delete that vertex. Though the neighbors of the vertex are dominated, we
can not remove these vertices from further consideration as they can be
useful to dominate other vertices.
Hence we resort to a coloring scheme for the vertices, similar to the one
suggested by Alber et al. in [2, 3]. At any point of time of the algorithm,
the vertices are colored as below:

1. Red - The vertex is included in the dominating set D which we are

trying to construct.
2. White - The vertex is not in the set D, but it is dominated by some
vertex in D.
3. Black - The vertex is not dominated by any vertex of D.

Now we define the dominating set problem on the graph with vertices
colored with White, Black or Red as above. We call a graph colored red,
white and black as above, as a rwb-graph.
RWB-Dominating Set: Let G be a G5 graph (graph with girth
at least 5) with vertices colored with Red, White or Black satisfying
the following conditions, and let k be a positive integer parameter.
Let R, W and B be the set of vertices colored red, white and black
respectively.
1. Every white vertex is a neighbor of a red vertex.
2. Black vertices have no red neighbors.
3. |R| ≤ k
Does G have at most k − |R| vertices that dominate all the black
vertices?
It is easy to verify that if we start with a general G5 graph with all
vertices colored black, and color all vertices we want to include in the
dominating set as red, and their neighbors as white, the graph we obtain
 7

at every intermediate step is a rwb-graph, and the problem we will have

at the intermediate steps is the RWB-Dominating Set problem.
The following lemma essentially shows that if the rwb-graph has a
black or white vertex dominating more than k black vertices, then such
a vertex must be part of every solution of size at most k, if one exists.
Lemma 1. Let (G = (R ∪ W ∪ B, E), k) be an instance of the RWB-
Dominating Set problem where G is a G5 graph and k a positive integer.
Let v be a black or white vertex with more than k − |R| black neighbors.
Then if G has a set of size at most k−|R| that dominates all black vertices,
then v must be part of every such set.

Proof. Let D be a set of size k − |R| that dominates all black vertices
in G, and suppose v ∈ / D. Let X be the set of black neighbors of v
which are not in D and Y be the set of black neighbors of v in D. So
|X| + |Y | > k − |R|. Observe that for every vx ∈ X we have a neighbor
ux ∈ D which is not in Y (otherwise v, vx , ux is a 3 length cycle). Similarly,
for x, y ∈ X, x 6= y ⇒ ux 6= uy . Otherwise v, x, ux , y will form a cycle
of length 4. This means that |D| ≥ |X| + |Y | > k − |R| which is a
contradiction. t
u

Given a rwb-graph, Lemma 1 suggests the following simple reduction

rule.

(R1) If there is a white or a black vertex v having more than k − |R|

black neighbors, then color v red and color its neighbors white.

Our goal now is to pick enough white or black vertices to dominate the
black vertices. So the following reduction rules are obviously justified.

(R2) If a white vertex is not adjacent to a black vertex, delete the white
vertex.
(R3) If there is an edge between two white vertices, delete the edge.
(R4) If |R| > k, then stop and return NO.

The following lemma follows from Lemma 1.

Lemma 2. Let G = (R∪W ∪B, E) be an instance of RWB-Dominating
Set and let G0 = (R0 ∪ W 0 ∪ B 0 , E 0 ) be the reduced instance after applying
rules (R1) to (R4) once. Let k be an integer parameter. Then G is a yes
instance if and only if G0 is a yes instance. That is G has a set of size at
most k − |R| dominating all vertices in B if and only G0 has a set of size
at most k − |R0 | dominating all vertices in B 0 .
8

Let G be an instance of RWB-Dominating Set and let G0 be the

reduced instance after applying the reduction rules (R1) − (R4) until no
longer possible. Then we show that if G0 is a yes instance (and hence G
is a yes instance), the number of vertices in G0 is bounded by polynomial
in k. More precisely we show the following lemma.
Lemma 3. Let (G, k) be a yes instance of RWB-Dominating Set and
(G0 , k 0 ) be the reduced instance of (G, k) after applying the rules (R1) −
R(4) until no longer possible. Then, the number of vertices in G0 is
O(k 3 ), that is, a kernel of size at most O(k 3 ) can be obtained for RWB-
Dominating Set.

Proof. Let R0 , B 0 and W 0 be the set of vertices colored red, black and
white respectively in G0 . We argue that each of |R0 |, |B 0 | and |W 0 | is
bounded by a function of k.
Because of (R4) (and the fact that G0 is a yes instance), |R0 | ≤ k.
Because of (R1), every vertex colored white or black has at most
k − |R0 | black neighbors. Also we know that no red vertex has a black
neighbor. Since G0 is a yes instance, there are at most k (k − |R0 | to be
more precise) black or white vertices dominating all black vertices. Since
each of them can dominate at most k black vertices, we conclude that
|B 0 | can be at most k 2 .
We argue that |W 0 | ≤ k 3 . Towards this end, we just show that every
black vertex has at most k white neighbors. Since |B 0 | ≤ k 2 , and every
white vertex is adjacent to some black neighbor (because of (R2) and
(R3)), the conclusion will follow.
Note that every white vertex has a red neighbor. Observe that the
white neighbors of any black vertex (any vertex for that matter) will
have all distinct red neighbors. I.e. if w1 and w2 are white neighbors of a
black vertex b, then there is no overlap between the red neighbors of w1
and the red neighbors of w2 . This is because if w1 and w2 have a common
red neighbor r, then we will have a 4-cycle b, w1 , r, w2 , b. Since |R0 | ≤ k,
it follows that a black vertex can have at most k white neighbors.
This proves the required claim. t
u

Thus we have the following theorem.

Theorem 3. The RWB-Dominating Set problem can be solved in
O(k k+O(1) + nO(1) ) time for G5 graphs.

Proof. It is easy to see that the reduction rules (R1) to (R4) take poly-
nomial time to execute. When none of these rules can be executed, by
 9

Lemma 3, we have that the number of vertices in the resulting graph is

O(k 3 ), and each vertex has at most k black neighbors. We can just try all
possible subsets of size at most k of the vertex set of the reduced graph,
to see whether that subset dominates all the black vertices. If any of them
does, then we say YES and NO otherwise. This will take O(k 3k+O(1) ) time.
Alternatively, we can apply a branching technique on the black ver-
tices, by selecting a black vertex or any of its neighbors in the dominating
set. More precisely, let v be a black vertex. Then we branch on N [v] by
including w ∈ N [v] in the possible dominating set D we are constructing
and look for a solution of size k − 1 in G − {w} where w is colored red
and all its neighbors are colored white for every w ∈ N [v]. This results in
an O((k + 1)k+O(1) ) time algorithm. t
u

Now to solve the general Dominating Set problem in G5 graphs, simply

color all vertices black and solve the resulting RWB-Dominating Set
problem using Theorem 3. Thus we have

Theorem 4. Parameterized Dominating Set problem can be solved in

O(k k+O(1) + nO(1) ) time for G5 graphs.

Parameterized version of Connected Dominating Set (where one

is interested in dominating set which is connected) or Independent
Dominating Set (where one is interested in dominating set which is
independent) are also known to be W[2]-complete [8]. Since the reduction
rules (R1)-(R4) apply for any dominating set, using Lemma 3 we can
obtain a kernel of size at most O(k 3 ) for both these problems. For the
Independent Dominating Set problem we also check that R remains
an independent set when we add a vertex to it while applying reduction
rule (R1), else we return NO. Furthermore in the proof of the Theorem 3,
we try all possible subsets of size at most k and look for a connected
or independent dominating set, as required. This results in the following
corollary.

Corollary 3. Parameterized Connected Dominating Set and Inde-

pendent Dominating Set problems can be solved in O(k 3k+O(1) +nO(1) )
time for G5 graphs.

A number of other variants of dominating set problem which are W[2]-

hard can be shown to be fixed parameter tractable in a similar way for
G5 graphs though not using kernelization. We give necessary details for
a few of them in the next subsections.
10

2.3 Red-Blue Dominating Set and Constraint Bipartite

Dominating Set

In this section we give FPT algorithms for Red-Blue Dominating

Set and Constrained Bipartite Dominating Set problems for G5
graphs. We first give an algorithm for Red-Blue Dominating Set prob-
lem which is defined as follows.

Red-Blue Dominating Set [10]: Given a bipartite graph G =

(V, E) with V bipartitioned as Vred ∪Vblue and a positive integer k.
Does there exist a subset D ⊆ Vred with |D| ≤ k and Vblue ⊆ N (D).

Theorem 5. Red-Blue Dominating Set is FPT for G5 graphs.

Proof. Any two vertices in Vred have at most one common neighbor in
Vblue as otherwise there will be a four cycle in G. Hence, the following
reduction rule is justified.

(R10 ) if x ∈ Vred has degree more than k then include x ∈ D.

The correctness of (R10 ) follows from the fact that if we do not select x
in D then we need more than k vertices from Vred to dominate N (x) as
any vertex y ∈ Vred , y 6= x, can dominate at most one vertex of N (x).
Hence after exhaustively applying reduction rule (R10 ) if the size of D is
more than k we answer NO.
Remove N [D] from G, i. e., set Vred = Vred \D and Vblue = Vblue \N (D).
Now the degree of every vertex in Vred is at most k and we are looking
for a set of size at most k − |D| in Vred such that it dominates all the
vertices of Vblue . Since every vertex in Vred has degree at most k, the size
of the set Vblue is bounded above by k 2 ((k − |S|)k to be precise) else the
answer is NO. We can not bound the size of the set Vred anymore, as we
do not have any bound on the degree of the vertices in Vblue . So to find
the desired dominating set in Vred (dominating all the vertices in Vblue )
we do as follows:

– For all partitions P of Vblue into at most k − |D| parts, say P =

{P1 , P2 , · · · , Pj }, 1 ≤ j ≤ k − |D|, for each Pi , 1 ≤ i ≤ j check
whether there exists a vertex ui ∈ Vred such that Pi ⊆ N (ui ). Call
the partition P valid if for all 1 ≤ i ≤ j, there exists ui ∈ Vblue such
that Pi ⊆ N (ui ) and the set {ui | 1 ≤ i ≤ j} is called the witness set.
If any partition P is valid then return YES with the corresponding
witness set else return NO.
 11

It is easy to see that there exits a subset of Vred of size at most k − |D|
dominating all vertices of Vblue if and only if there exists a valid partition.
Number of ways in which n indistinguishable objects can be partitioned
n+r−1
into r ways is r−1 [20]. Hence the total number of partitions P con-

sidered for our case is upper bounded by

k−|D|  2 
X k +i−1
.
i−1
i=1

Since the total number of partitions is upper bounded by O(k 2k+O(1) ), the
result that Red-Blue Dominating Set is FPT for G5 graphs follows.
t
u

Next we study Constraint Bipartite Dominating Set problem which

is defined as follows.

Constraint Bipartite Dominating Set (CBDS) [16]: Given

a bipartite graph G = (V, E) with V partitioned as V1 ∪ V2 and
positive integers k1 and k2 . Does there exist subsets D1 ⊆ V1 and
D2 ⊆ V2 with |D1 | ≤ k1 and |D2 | ≤ k2 such that V2 ⊆ N (D1 ) and
V1 ⊆ N (D2 ).

Theorem 6. Parameterized Constraint Bipartite Dominating Set

is FPT for G5 graphs.

Proof. To solve this problem we just need to solve two instances of Red-
Blue Dominating Set problem. The instances of Red-Blue Domi-
nating Set problem we solve are:

1. Vred = V1 , Vblue = V2 and parameter is k1 ; and

2. Vred = V2 , Vblue = V1 and parameter is k2 .

We return YES for CBDS problem if both the instances return YES and
as D1 the red-blue dominating set returned by instance 1 and as D2 the
red-blue dominating set returned by instance 2. If either of the instances
of Red-Blue Dominating Set problem returns NO, then we return NO
for the CBDS problem. t
u

2.4 Threshold Dominating Set

This problem generalizes Dominating Set and is formally defined as

follows.
12

Threshold Dominating Set (TDS) [7]: Given a graph G =

(V, E) and positive integers k and r. Is there a set of at most k
vertices V 0 ⊆ V such that for every vertex u ∈ V , N [u] contains
at least r elements of V 0 ?

Theorem 7. Threshold Dominating Set parameterized by k is FPT

for G5 graphs.

Proof. First we observe that if k < r, then the answer is NO. We assume
that r ≤ log n, as otherwise k ≥ log n and we have a kernel of size at
most 2k . Now we can solve the problem by checking all subsets of size at
most k for the desired threshold dominating set.
Our algorithm is again based on the following simple reduction rule
whose correctness follows from Lemma 1.
(R100 ) if x ∈ V has degree more than k then include x ∈ V 0 .
So basically we select all the vertices of degree more than k of V in V 0
and hence if the size of V 0 is more than k then we answer NO.
Next we assign a color to all the vertices. We assign white color to
all the vertices (including vertices in V 0 ) which have enough (at least r)
neighbors in V 0 and black to the rest. Let B and W , as usual, represent
the set of black and white vertices respectively and set B 0 = B \ V 0
and W 0 = W \ V 0 . Apply reduction rules (R2) and (R3) of Lemma 2
exhaustively. The rule (R3) makes G[W ] an independent set. Now the
problem reduces to finding a set S 0 of size at most k − |V 0 | in V \ V 0 such
that V 0 ∪ S 0 is a desired threshold dominating set for G, in particular for
the vertices of B. Since every vertex in V \ V 0 has degree at most k and
we are looking for S 0 of size at most k in V \ V 0 , the size of |B| is bounded
above by k 2 , as otherwise we answer NO.
Now what we have is a generalized version of Threshold Domi-
nating Set problem where we have a set of j ≤ k 2 black vertices
B = {u1 , · · · , uj }, each with a positive integer ri (ri = r − |N [vi ] ∩ V 0 |),
1 ≤ i ≤ j. We are looking for a set S 0 ⊆ (W 0 ∪ B 0 ) of size at most
k − |V 0 | such that for every ui ∈ B, |N (ui ) ∩ S 0 | ≥ ri in G0 where
the vertex set of G0 is V (G0 ) = B ∪ W 0 and the edge set of G0 is
E(G0 ) = {(u, v) ∈ E | u ∈ W 0 , v ∈ B or u ∈ B, v ∈ B}.
To solve this generalized version of Threshold Dominating Set
problem, we need to generalize our partition arguments used in the The-
orem 5 suitably. The major differences are that G0 is no more bipartite
and that there are vertices which need more than 1 (possibly r) vertices
in the desired threshold dominating set. To overcome this difficulty, we
 13

make a multiset M from B by having ri copies for each vertex ui ∈ B.

Clearly the size of |M | is bounded above by rk 2 . Now if we apply the
partition idea of Theorem 5 it is possible that the same vertex may domi-
nate multiple copies of the same vertex. To deal with this call a partition
P = {P1 , P2 , · · · , Pα } valid if (a) there exists a subset S 0 ⊆ B 0 ∪W 0 form-
ing a system of distinct representatives; that is for all 1 ≤ i ≤ α, there
exists a distinct ui ∈ S 0 such that Pi ⊆ N (ui ) and (b) each Pi contains at
most one copy of any vertex of B. The set S 0 is a witness set. So to find
the desired threshold dominating set in B 0 ∪ W 0 we proceed as follows.

– For all partitions P of M in at most k − |V 0 | parts, say P =

{P1 , P2 , · · · , Pα }, 1 ≤ α ≤ k − |V 0 |, we check whether P is a valid
partition. If any partition P is valid then return YES with the corre-
sponding witness set else return NO.

For a fixed partition P = {P1 , P2 , · · · , Pα }, we can do the validity testing

and find a corresponding witness set in polynomial time as follows. Testing
for duplicate copies in Pi ’s are easy. For the other part we first define the
set
Ii = {u ∈ (B 0 ∪ W 0 ) | Pi ⊆ NG0 (u)},
where NG0 (u) denotes the neighbors of u in G0 . Now we make the bipartite
incidence graph for the sets {I1 , · · · , Iα }, that is a bipartite graph G∗ =
(X ∪ Y, E 00 ), where X has a vertex xi for every set Ii and Y = ∪αl=1 Il
and there is an edge between (xi , u) if u ∈ Ii . Now finding a “valid”
system of distinct representatives reduces to finding a maximum bipartite
matching in G∗ saturating X, for which there is a classical polynomial
time algorithm of Edmonds [12].
The total number of partitions P considered for our case is upper
bounded by
k−|V 0 |  2
X rk + i − 1
,
i−1
i=1

which is at most O((rk 2 )k+O(1) ). Since r ≤ log n and (log n)k ≤ n +

(2k log k)k for all n and k ≤ n, we have the desired result that Threshold
Dominating Set problem is FPT for G5 graphs. t
u

3 Set Cover with Bounded Intersection among Sets

Dominating Set problem is well known to be a special instance of the

Set Cover problem defined below.
14

Set Cover: Given a base set (or universe) U = {s1 , s2 , · · · , sn }, a

collection S = {S1 , S2 , · · · , Sm } of subsets of U (Si ⊆ U, 1 ≤ i ≤ m)
such that ∪m i=1 Si = U and a positive integers k, does there exist a
sub-collection S 0 of S of size at most k such that ∪Sj ∈S 0 Sj = U.

Given an instance (G = (V, E), k) of the Dominating Set problem, we

can formulate it as an instance of the Set Cover problem by taking
U = V and S = {Sv = N [v] | v ∈ V }. It is easy to verify that G has
a dominating set of size k if and only if (U, S) has a set cover of size at
most k. Hence the parameterized version of the Set Cover problem is
W [2]-complete [8].
Here, we show that a special case of the Set Cover problem, that
generalizes the Dominating Set problem for G5 graphs to be fixed pa-
rameter tractable. More specifically, we show if the Set Cover instance
(U, S) satisfies the property that for any pair of sets Si and Sj in S,
|Si ∩ Sj | ≤ c, for a fixed constant c, then the problem is fixed param-
eter tractable. We call this variant of the Set Cover problem, where
every pair of sets in the given family intersect in at most c elements, as
Bounded Intersection Set Cover (BISC) problem.
Observe that if the input graph G of the dominating set problem is a
G5 graph then the sets in its corresponding set cover instance satisfies a
property that for any pair of sets Su and Sv in S, |Su ∩ Sv | ≤ 2.

Theorem 8. The BISC problem is fixed parameter tractable.

Proof. If there is a set Si ∈ S such that |Si | > ck then Si must be in

every k-sized set cover. Otherwise, we need more than k sets to cover all
the elements of U since every other set can cover at most c elements of
Si . So this gives us a following simple reduction rule:

Rule 1: Given a set cover instance, (U, S, k), if there exists Si ∈ S such
that |Si | > ck then obtain a new reduced instance of set cover as
following:
– U ← U − Si .
– S ← {S − Si | S ∈ S}. If there are multiple copies of some set,
then remove all but one copy of the same.
– k ←k−1

If k becomes 0 and U is non-empty then this is a no instance for the

problem and we stop. We apply the Rule 1 until all the sets in S is of
size at most ck 0 , where k 0 ≤ k. As k 0 sets of size ck 0 can only cover at
most ck 02 ≤ ck 2 elements of U, if |U| > ck 2 then it is a no instance of the
 15

problem. The reduction rule also ensures that every set in S is distinct.
But then the number of distinct sets of size at most ck in S can be at
most the number of distinct subsets of U. This gives us that if |U| ≥ 2ck
then
ck  ck
cek 2
   
X |U| |U|
|S| = ≤ ck ≤ ck = ceck k ck+1
i ck ck
i=1

and if |U| < 2ck then

ck  
X |U|
|S| = ≤ 22ck = 4ck .
i
i=1

Now it suffices to try each sub-collection S 0 ⊆ S of size k and return YES

if any of them covers the set U and NO otherwise. This has following time
complexity:
k
ceck k ck+1 ceck k ck+1 e
  
2
≤ = (ce)k (ek)ck .
k k

Since c is a fixed constant, it follows that the running time results in a

fixed parameter tractable algorithm. u
t

4 t-Vertex Cover and t-Dominating Set Problems

t-Vertex Cover and t-Dominating Set problems are respectively,

generalizations of classical Vertex Cover and Dominating Set prob-
lems. Here the objective is not to cover all the edges or to dominate all
the vertices but to cover at least t edges or to dominate at least t vertices
with at most k vertices. More precisely they are defined as follows:

t-Vertex Cover: Given a graph G = (V, E) and positive integers k

and t, does there exist a set of at most k vertices C ⊆ V such that
|{e = (u, v) ∈ E | C ∩ {u, v} =
6 ∅}| ≥ t.
t-Dominating Set: Given a graph G = (V, E) and positive integers
k and t, does there exist a set of at most k vertices D ⊆ V such that
|N [D]| ≥ t.

The t-Vertex Cover and t-Dominating Set problems have been

parameterized in two ways. They are either parameterized by k or by t
and k. Both these problems are FPT when parameterized by both k and
16

t [6] and are hard for different level of W -hierarchy when parameterized
by k alone. t-Vertex Cover is W [1]-complete [18] and t-Dominating
Set is W [2]-complete when parameterized by k alone.
Here, we first give a simple algorithm for the t-Vertex Cover when
parameterized by both t and k and then show that this problem is FPT
even when parameterized by k alone in G5 graphs. We then extend this
result to the t-Dominating Set problem for G5 graphs when parame-
terized by k alone.
Our algorithms for the t-Vertex Cover depend on the following
lemma.
Lemma 4. Let (G = (V, E), k, t) be a yes instance of the t-Vertex
Cover and v be a vertex of maximum degree in G. Then there ex-
ists a t-vertex cover C whose intersection with N [v] is nonempty, i.e.
N [v] ∩ C 6= ∅.

Proof. Since G is a yes instance of the problem there exists a t-vertex

cover C of size at most k and covering at least t edges. If N [v] ∩ C = ∅
then choose C 0 = C − {u} + {v} where u is any vertex in C. Since v is a
vertex of highest degree and none of its neighbors is in C, C 0 also covers
at least t edges and is of size at most k. t
u

Suppose that the given graph has maximum degree bounded by d.

Since there exists a t-vertex cover containing either a maximum degree
vertex u or one of the neighbors of u, we can branch on u and on each
of the (at most) d neighbors of u giving rise to a (d + 1)-way branching.
The following theorem is immediate from this.

Theorem 9. Let G be a graph with maximum degree d. Then t-Vertex

Cover can be solved in O((d + 1)k n) time.

Given a graph G = (V, E) and positive integer parameters t and k,

if there exists a vertex of degree at least t then we get a t-vertex cover
by choosing the vertex. So without loss of generality, we can assume that
every vertex has degree at most t − 1. Then from Theorem 9, we have
Corollary 4. t-Vertex Cover can be solved in O(tk n) in general graphs.
Suppose, instead of trying to cover at least t edges, we want to cover
all but t edges (where t is a parameter) using at most k vertices. That
is, we want an induced subgraph on n − k vertices with at most t edges.
We call it the (m − t)-Vertex Cover problem. Such a parameterization
is known as dual parameterization and dual problems are, in general,
 17

natural and equally interesting [8, 21]. For example Vertex Cover is
fixed parameter tractable whereas the dual of Vertex Cover is the
Independent Set problem (which is the same as choosing n − k vertices
to cover all edges) and is W[1] complete.
The (m−t)-Vertex Cover problem can also be parameterized in two
ways, by k alone and by k and t. When we have both t and k as parameters
then we solve this problem by branching on an edge e = (u, v). Here we
branch by choosing either the vertex u or the vertex v or e which means
that we are looking for a solution which contains either u or v or does
not cover e. So we get the following branching recurrence:

T (k, t) ≤ 2T (k − 1, t) + T (k, t − 1).

This immediately gives us the following theorem.

Theorem 10. (m − t)-Vertex Cover can be solved in O(3t+k (n + m))
time. Thus (m − t) Vertex Cover is fixed parameter tractable if param-
eterized by t and k.
When (m − t)-Vertex Cover problem is parameterized by k alone, we
can show the following theorem.
Theorem 11. The (m − t)-Vertex Cover problem is W [1]-hard when
parameterized by k alone.

Proof. We give a reduction from W[1]-complete t-Vertex Cover prob-

lem where we need at most k vertices to cover at least t edges. Given
(G = (V, E), k, t1 ), an instance of t-Vertex Cover problem, we map it
to (G = (V, E), k, t2 ) where t2 = |E|−t1 as an instance of (m−t)-Vertex
Cover problem. Now it is easy to see that (G = (V, E), k, t1 ) is a yes
instance of t-Vertex Cover problem if and only if (G = (V, E), k, t2 ) is
a yes instance of (m − t)-Vertex Cover problem. t
u

Now we show that the t-Vertex Cover problem is F P T for G5

graphs when parameterized by k alone. We will see that this result also
applies to (m − t)-Vertex Cover problem when parameterized by k
alone.
Theorem 12. t-Vertex Cover is fixed parameter tractable for G5 graphs
when parameterized by k alone. The algorithm runs in O((k + 1)k (n + m))
time.

Proof. Without loss of generality we can assume that the maximum de-
gree of this graph is not bounded by a function of k, otherwise the problem
18

is fixed parameter tractable by Theorem 9. Let v0 be a vertex of highest

degree and let v1 , v2 , . . . , vr be its neighbors. Further assume that

deg(v1 ) ≥ deg(v2 ) ≥ · · · deg(vk ) ≥ · · · ≥ deg(vr ).

Let A = {v0 , v1 , · · · , vk }. We show that if there exists any t-vertex cover

then there is one which contains either v0 or one of its k highest degree
neighbors. More precisely, we prove the following claim:
Claim : There exists a t-vertex cover C such that A ∩ C 6= ∅, if one
exists.
The claim says that if there exists any t-vertex cover then there exists a
t-vertex cover C containing at least one vertex of A. We then branch on
the vertices of the set A, and look for a solution of size k − 1, covering
t-deg(vi ) edges in G−{vi }, where 0 ≤ i ≤ k and recursively use this claim
on the respective subgraphs. Hence the claim proves that t-vertex cover
is fixed parameter tractable.
Now we are left with proving the claim. We show the claim by con-
tradiction. Assume to the contrary that no t-vertex cover intersects A.
By Lemma 4 we know that there exists a t-vertex cover C containing
one of v0 ’s neighbors. Let vl be a neighbor of v0 in C. Because of our
assumption l > k. Suppose for some vi , 1 ≤ i ≤ k, N (vi ) ∩ C = ∅. Then
we can obtain a t-vertex cover C 0 = C − {vl } + {vi } of size at most k and
covering at least t edges as deg(vi )≥ deg(vl ). So we now assume that for
each vi , 1 ≤ i ≤ k, N (vi ) ∩ C 6= ∅. Let Bi = N (vi ) ∩ C. Observe that
for each i, Bi does not contain vl otherwise that will imply v0 , vi , vl is a
triangle. Suppose for some i 6= j, u ∈ Bi ∩ Bj then v0 , vi , u, vj is a cycle
of length 4. Hence Bi ∩ Bj = ∅ for all i, j such that i 6= j. So this implies
that
Xk
|Bi | ≥ k.
i=1

So we have Bi 6= ∅, Bi ⊆ C − {vl } and their pairwise intersections are

empty. But this implies
k
X
|Bi | ≤ |C − {vl }| ≤ k − 1
i=1

which contradicts that ki=1 |Bi | ≥ k. This in turn proves the claim.
P
Since we branch on the vertices of A whose size is bounded by k + 1,
we get an algorithm of time complexity O((k + 1)k n). t
u
 19

Since the runtime in Theorem 12 was independent of t, we get

Theorem 13. (m − t)-Vertex Cover can be solved in O((k + 1)k (n +
m)) time for G5 graphs when parameterized by k only.
By arguments similar to those used in Theorem 12, we can show the
following.
Theorem 14. t-Dominating Set can be solved in O((k+1)k nO(1) ) time
for G5 graphs when parameterized by k only.

5 Independent Set and its Variants in G4 Graphs

Independent Set problem asks for an induced subgraph on k vertices

which only contains isolated vertices. More precisely:
Independent Set : Given a graph G = (V, E) and an integer k ≥ 0,
determine whether there exists a set of at most k vertices I ⊆ V such
that the subgraph induced by I does not contain any edges.
Independent Set problem is W[1]-complete for general graphs. We show
that the Independent Set and some of its variants are fixed parameter
tractable for triangle free graphs. We use Ramsey theory to get a kernel
of size O(k 2 ) for these problems.

Theorem 15. Parameterized Independent Set problem can be solved

in O(kn + k O(k) ) in G4 graphs (triangle free graphs).

Proof. Given any two integers p and q, there exists a number R(p, q) such
that any graph on at least R(p, q) vertices contains an independent set
of size p or a clique of size q. R(p, q), for various values of p and
q are
known as Ramsey Numbers. It is well known that R(p, q) ≤ p+q−2 q−1 [20].
And if n > R(p, q) then either an independent set of size p or a clique
of size q can be found in O((p + q)n) time by transforming the inductive
arguments used in the proof of Theorem 27.3 in [20] for the upper bound
of R(p, q) to a constructive algorithm.
If k ≤ 2, then we can check in linear time whether the graph has
an independent set of size 2 or not. So let us assume that k ≥ 3. If the
number of vertices n > k 2 ≥ R(k, 3) then we know that this graph has
either an independent set of size k or a clique of size 3. But since the input
graph is triangle free, we know that it must have an independent set of
size k and can be found in O(kn) time. Otherwise we know that n ≤ k 2 .
In this case, we try all possible subsets of size at most k to see whether
20

the graph has an independent set of size k or not. If any of them does,
then we answer YES and answer NO otherwise. This will take O(k O(k) )
time. This completes the proof. t
u

Theorem 15 can be extended to a larger class of problems where one is

interested in finding a subset inducing a “hereditary property”. A graph
property Π is a collection of graphs. A graph property Π is non-trivial if
Π has at least one graph and does not include all graphs. A non-trivial
property is said to be hereditary if a graph G is in property Π implies
that every induced subgraph of G is also in Π. Given any property Π, let
P (G, k, Π) be the problem defined below:
P(G, k, Π): Given a graph G = (V, E) and a positive integer k, deter-
mine whether there exists a set of k vertices V 0 ⊆ V such that G[V 0 ]
is in Π.
Khot and Raman [21] studied this problem and showed the following
theorem.
Theorem 16. (Khot and Raman [21]) Let Π be a hereditary property
that includes all independent sets but not all cliques (or vice versa). Then
the problem P (G, k, Π) is W [1] hard.
The proof of the following theorem is exactly as in the proof of The-
orem 15, by considering the Ramsey numbers R(k, c).
Theorem 17. Let Π be a hereditary property that includes all indepen-
dent sets. Then the problem P (G, k, Π) restricted to Gc graphs for any
fixed constant c ≥ 3 is fixed parameter tractable and can be solved in
O(kn + k O(k) nO(1) ) time.
Given a graph G = (V, E) and a positive integer k ≥ 0, Acyclic Sub-
graph, Bipartite Subgraph and Planar Subgraph problems ask
whether there exists a subset V 0 ⊆ V , such that |V 0 | ≥ k and G[V 0 ] is
acyclic, bipartite or planar respectively. All these problems are known to
be W[1]-hard [8, 21] in general graphs. As a corollary to Theorem 17 we
have following:
Corollary 5. Acyclic, Bipartite and Planar Subgraph problems
are fixed parameter tractable with time complexity O(kn + k O(k) nO(1) ) for
Gc graphs for any fixed constant c ≥ 3.
Corollary 5 shows that Acyclic and Planar Subgraph problems
are fixed parameter tractable for bipartite graphs. In fact we can easily
obtain much improved FPT algorithms for these problems for bipartite
 21

graphs. Observe that a bipartite graph has an independent set (and hence
planar or acyclic induced subgraph) on n/2 vertices. So, if k ≤ n/2 then
for both these problems the answer is YES and otherwise k > n/2 or
n < 2k and hence we get a kernel of size at most 2k for both the Acyclic
and Planar Subgraph problems for bipartite graphs. Now we check all
k sized subsets of the vertex set to see whether
the2ksubset induces an
n
acyclic subgraph or planar subgraph. Since k ≤ k ≤ 2 = 4k , we2k

get an O(4k nO(1) ) time algorithm for both these problems for bipartite
graphs.
Minimum feedback vertex set, which is a subset of vertices whose
removal makes the graph acyclic, is a complement of the vertex set of the
maximum Acyclic Subgraph problem. Fomin et al. [15] have shown
that the minimum feedback vertex set can be found in O(1.7548n ) time
in undirected graphs. So together with this result and the kernel of size
2k we get O(1.75482k nO(1) ) = O(3.0793k nO(1) ) time algorithm for the
Acyclic Subgraph problem. Putting together everything we get the
following theorem.
Theorem 18. The parameterized Acyclic Subgraph and Planar Sub-
graph problems can be solved in O(3.08k k O(1) + nO(1) ) and O(4k k O(1) +
nO(1) ) time, respectively, for bipartite graphs.
Another problem which can be shown to be FPT for Gc graphs for
any fixed constant c ≥ 3 is the Irredundant Set problem, which is
known to be W[1]-complete [9] in general graphs.
Irredundant Set: Given a graph G = (V, E) and a positive
integer k. Is there a set V 0 ⊆ V of cardinality at least k having
the property that each vertex u ∈ V 0 has a private neighbor ? A
private neighbor of a vertex u ∈ V 0 is a vertex u0 ∈ N [u] (possibly
u0 = u) with the property that for every vertex v ∈ V 0 \ {u},
u0 ∈
/ N [v].
This follows from a simple observation that every independent set is also
an irredundant set. Then the following theorem can be proved on the lines
of Theorem 15, by considering the Ramsey Numbers R(k, c).
Theorem 19. Irredundant Set is FPT for Gc graphs for any fixed
constant c ≥ 3.

6 Is everything easy on graphs with no small cycles ?

In contrast to the results presented in the previous sections, here we
show a problem to be W [1]-hard even in bipartite graphs with girth at
22

least 6 (G6 graphs). Observe that in graphs with large girth the Clique
problem is trivial. We look at Dense Subgraph problem [22] which is a
generalization of the Clique problem.

Dense Subgraph: Given a graph G = (V, E) and positive integers

k and l, determine whether there exists a set of at most k vertices
C ⊆ V such that G[C] has at least l edges, i.e. the induced subgraph
on C has at least l edges. (Note that l is at most k2 .)

It is easy to observe that Dense Subgraph problem is W[1]-hard when

parameterized by k, by a simple reduction from Clique. But we give a
reduction from Clique to Dense Subgraph problem parameterized by
k which shows that the problem is W[1]-hard even in bipartite graphs
with girth at least 6.

Theorem 20. Dense Subgraph is W [1]-hard for bipartite graphs with

girth at least 6 when parameterized by k.

Proof. We give a reduction from Clique. Let (G, k) be an instance of

Clique with k ≥ 3. We make the graph G = (V, E) bipartite by sub-
dividing every edge. Let G0 = (V 0 , E 0 ) be the resulting subgraph. Here,
V 0 = V ∪ W where W = {wuv | (u, v) ∈ E} and E 0 , the set of edges,

consists of (u, wuv ) and (v, wuv ) for every wuv ∈ W . Take k 0 = k + k2
and l = 2 k2 .

Observe that G0 is a bipartite graph as every cycle is of even length

and the girth is at least 6 as the girth of G is at least 3. We claim that G
has a clique of size k if and only if G0 has a subgraph on k 0 vertices with
at least l edges. Also note that every vertex in W has degree 2 as they
represent edges in the original graph. Now suppose G has a clique of size
k on vertex set C = {v1 , v2 , · · · , vk }. Then C 0 = C ∪ {wuv | u, v ∈ C} is a
vertex set of dense subgraph in G0 having k 0 vertices and l edges as G[C]
k


has at least 2 edges.
Conversely, let C 0 be a set of k 0 vertices such that G0 [C 0 ] has at least l
edges. Let O = V ∩ C 0 . Clearly G0 [C 0 ] is bipartite with O and N = C 0 − O
as the two parts of the vertex set, and every vertex in
N has degree at
most 2. Since the number of edges in G0 [C 0 ] = l = 2 k2 , and since every
vertex in N has degree at most 2, |N | ≥ k2 and hence |O| ≤ k. Let

t = |O|. We claim that t = k. Suppose not. Then t ≤ k − 1. Also, since

k ≥ 3, t ≥ 1. Let n1 and n2 be the degree 1 and degree 2 vertices in
N respectively. Since G has no multiple edges, no pair of vertices in N
with degree 2 can be adjacent to the same pair of vertices in O and hence
 23

t
. Then the number of edges in G[C 0 ] is:


n2 ≤ 2
 
k
2 ≤ |E(G[C 0 ])| = 2n2 + n1
2
= k 0 − t + n2
 
k
=k+ − t + n2
2
   
k t t


≤k+ −t+ ( since n2 ≤ 2 ).
2 2
From the above it implies that
   
k t
t+ ≤k+ . (1)
2 2
If t = 1 then    
1 k
k+ =k <1+ ,
2 2
a contradiction to inequality (1). So let 2 ≤ t ≤ k − 1. But then
       
t k−1 k k
k+ ≤k+ = +1< + t,
2 2 2 2
again a contradiction
to inequality (1). This implies that |O| = k. As a
result of this, |N | = k2 and every vertex in N has degree 2. Every vertex
of degree 2 in N represents an edge in G[O]. This shows that the vertices
of O form a clique in the original graph. t
u

7 Approximation of Dominating Set

In this section we give some results concerning approximation of the Dom-

inating Set problem for bipartite and G5 graphs. We refer to [25] for
all the basic definitions regarding approximation algorithms.
Feige [13] showed that (1-o(1)) ln n is a threshold below which the
Dominating Set problem cannot be approximated efficiently unless NP
has slightly super-polynomial time algorithm. Here, ln n represents natu-
ral logarithm. Under the same hypothesis (1−o(1))2
ln n
is a threshold below
which the Dominating Set problem can not be approximated for bipar-
tite graphs. This result follows from the reduction in Theorem 1. Towards
this end, we just show that if we have factor α approximation algorithm
for the dominating set problem in bipartite graphs then it implies 2α fac-
tor approximation algorithm for the dominating set problem in general
24

undirected graphs. Given a graph G, we apply Theorem 1 to obtain the

bipartite graph H and apply the factor α approximation algorithm for
dominating set problem in bipartite graphs to get a dominating set D for
H. We obtain a dominating set D0 for G from the dominating set D for
H as in the proof of Theorem 1. Let OP TG denote the size of an optimum
dominating set for the graph G. Now note that

|D0 | < |D|

≤ α · OP TH
≤ α · (OP TH − 1) + α
≤ α · OP TG + α · OP TG (since OP TH − 1 = OP TG )
= (2α) · OP TG .

Furthermore the result of Feige [13] together with Theorem 2 im-

ply that the approxmability of Dominating Set problem has the same
threshold of (1-o(1)) ln n even for split graphs. The above discussion re-
sults in the following theorem.
Theorem 21. Dominating Set problem can not be approximated ef-
ficiently below (1-o(1)) ln n in bipartite and split graphs unless N P ⊂
DT IM E(nO(log log n) ) .
An approximation algorithm of factor O(log n) is known for the Dom-
inating Set problem using the reduction to the Set Cover problem
(see discussion before Theorem 8 in Section 3) and the following propo-
sition.
Proposition 1 ([19, 23]). Let (U, S) be a set cover instance such that
|U | = n. ThenPwe can find a set cover S 0 ⊆ S of size at most Hn · (OP T )
where Hn = ni=1 1/i and OP T is the size of the optimum solution of
the set cover instance. Hn ≤ ln n + 1.
Here we outline a slightly improved approximation algorithm for Dom-
inating Set problem in G5 graphs. This approximation algorithm has
a factor O(log l) where l is the size of the optimum dominating set. The
idea of the algorithm is to use the reduction rules developed in Section
2.2 and obtain an instance of size O(l3 ) with the property that maximum
degree of the graph is bounded by l and then use the following proposition
on the corresponding set cover instance of the problem.
Proposition 2 ([11]). Let (U, S) be a set cover instance such that |U | =
n and the size of each set Si ∈ S is bounded by q, that is |Si | ≤ q. Then
we can find a set cover S 0 ⊆ S of size at most (Hq − 1/2) · (OP T ) where
 25

Hq = qi=1 1/i and OP T is the size of the optimum solution of the set
P
cover instance.
Observe that the reduction rules (R1) − (R4) depend on k whereras here
we have an optimization problem. Hence apply reduction rules for all
values for k between 1 and n and if the reduced instance as viewed as the
Set Cover problem instance satisfies the hypothesis of Proposition 2
then we obtain a dominating set for G by applying Proposition 2. Finally
we return the dominating set of smallest size among the ones obtained
for different k. Our detailed algorithm is described below. We outline our
algorithm in terms of rwb-graphs described in Section 2.2.
Algo-Dom-SET(G=(V,E))
(Input: A G5 graph. Output: A dominating set of G.)
Step 1: Given an undirected graph G = (V, E). Make it a rwb-graph by
coloring all vertices of V black; that is R = ∅, W = ∅ and B = V .
I = ∅.
Step 2: for (j = 1 to n) do as follows:
Step 2a: Apply reduction rules (R1) − (R4) on (G = (R ∪ W ∪
B, E), j) until no longer possible and obtain an instance (Gj =
(Rj ∪ W j ∪ B j , E j ), j − |Rj |).
Step 2b: If (|W j | + |B j | ≤ 2j 3 ) and the maximum degree of Gj is at
most j then

I = I ∪ {(Gj = (Rj ∪ W j ∪ B j , E j ), j − |Rj |)}

(In this step we obtain a set of instances which could possibly lead to
an optimum dominating set. So we have

I = {(Gk = (Rk ∪ W k ∪ B k , E k ), k − |Rk |) | |W k | + |B k | ≤ 2k 3

and maximum degree of Gk is at most k})
Step 3: We obtain a set cover instance (Uk , Sk ) from the reduced graph
Gk by taking U = B k and having sets Su for u ∈ (W k ∪ B k ). Su =
N (u) ∩ B k if u ∈ W k and Su = N [u] ∩ B k if u ∈ B k . Obtain P, the
set of instances for the set cover problem, by changing every instance
in I to the set cover instance. That is:

P = {(Uk , Sk ) | (Gk = (Rk ∪ W k ∪ B k , E k ), k − |Rk |) ∈ I}.

Step 4: Apply Proposition 2 to every instance of the set cover problem
in P and obtain the following set of solutions

SOL = {Sk0 | Sk0 ⊆ Sk , (Uk , Sk ) ∈ P}.

26

Let V(Sk0 ) represent the set of vertices in Gk corresponding to the sets

in the collection Sk0 . Obtain the following set

DOM = {V(Sk0 ) ∪ Rk | Sk0 ∈ SOL & Rk the red vertices of Gk }.

of possible dominating sets for G and return the one with the mini-
mum size in DOM as a dominating set for G.

Theorem 22. Let G = (V, E) be a G5 graph on n vertices. Then the

Pp set of size at most (Hp+1 −
algorithm Algo-Dom-SET outputs a dominating
1/2) · p in polynomial time where Hp = i=1 1/i and p is the size of the
optimum solution of a dominating set of G. That is, Algo-Dom-SET is an
approximation algorithm with performance ratio of ln(p + 2) + 1/2 for the
dominating set problem for G5 graphs.

Proof. It is clear that the algorithm Algo-Dom-SET takes polynomial time.

Proposition 2 ensures that the algorithm returns a dominating set for G.
Now we show that the algorithm is a factor of Hp+1 − 1/2 approxima-
tion algorithm for the dominating set problem for G5 graphs which will
complete the proof of the theorem.
Let l be the smallest positive integer in Step 2 of the algorithm such
that (Gl = (Rl ∪ W l ∪ B l , E l ), l − |Rl |) ∈ I. The reduction rules ensures
that (G = (R ∪ W ∪ B, E), k) is a no instance for 1 ≤ k ≤ l − 1 and hence
we have p ≥ l.
Consider the instance (Gp = (Rp ∪W p ∪B p , E p ), p−|Rp |) ∈ I. Observe
that the instance Gp has an optimum dominating set of size p − |Rp | and
the maximum degree of the graph is bounded by p. When we apply the
factor (Hq − 1/2) set cover approximation algorithm in Step 4 on the
instance (Up , Sp ), where each set in Sp is bounded by p + 1, we obtain
Sp0 ⊆ Sp of size at most |Sp0 | ≤ (Hp+1 − 1/2)(p − |Rp |). Now the size of
the dominating set Rp ∪ V(Sp0 ) corresponding to this instance for G is :

|Rp | + |V(Sp0 )| = |Rp | + |Sp0 |

≤ |Rp | + (Hp+1 − 1/2)(p − |Rp |)
≤ |Rp |(Hp+1 − 1/2) + (Hp+1 − 1/2)(p − |Rp |)
= (Hp+1 − 1/2)p.

Since we return a dominating set of minimum size among the sets in

DOM as a dominating set for G it is clear that its size is also bounded
by (Hp+1 − 1/2)p. This completes the proof. t
u
 27

8 Conclusion and Discussions

In this paper we showed that if the input graphs do not possess short
cycles then the neighborhood problems such as Dominating Set, Inde-
pendent Set and several of their variants are fixed parameter tractable.
We have also shown that the restriction on girth is optimal if we do not
put further restriction on the graph classes. This is the first time, to
our knowledge, that the parameterized complexity of graph problems are
classified by girth.
Most of the algorithms given here are just parameterized complexity
classification algorithms. We believe that more efficient FPT algorithms
should be possible. Obtaining a O(ck nO(1) ), c a constant, algorithm for
all these problems remains an open problem.
We also gave an improved approximation algorithm for Dominating
Set problem in graphs with girth at least 5. It would be interesting to
explore the possibility of improved approximation algorithms for other
problems on graphs with no small cycles.
Furthermore, it is worth exploring excluding structures as subgraphs
other than cycles to see whether some W-hard problems become FPT.

Acknowledgments. We thank the referees for several useful comments.

We would also like to thank T. Erlebach and R. E. Tarjan for suggesting
an investigation of approximation algorithms for the Dominating Set
problem for graphs with no small cycles.

References
1. J. Alber, H. L. Bodlaender, H. Fernau, T. Kloks and R. Niedermeier. Fixed pa-
rameter algorithms for Dominating Set and related problems on Planar Graph.
Algorithmica 33 (2002) 461-493.
2. J. Alber, H. Fan, M. R. Fellows, H. Fernau, R. Niedermeier, F. Rosamond and U.
Stege. A refined search tree technique for dominating set on planar graphs. Journal
of Computer and System Sciences 71 (2005) 385-405.
3. J. Alber, M. R. Fellows and R. Niedermeier. Polynomial time data reduction for
dominating Set. Journal of the ACM 51(3) (2004) 363-384.
4. V. E. Alekseev. On easy and hard hereditary classes of graphs with respect to the
Independent Set problem. Discrete Applied Mathematics 132(1-3) (2003) 17-26.
5. V. E. Alekseev, D. V. Korobitsyn and V. V. Lozin. Boundary classes of graphs for
the Dominating Set problem. Discrete Mathematics 285(1-3) (2004) 1-6.
6. M. Bläser. Computing small partial coverings. Information Processing Letters
85(6) (2003) 327-331.
7. R. G. Downey and M. R. Fellows. Threshold Dominating Sets and an improved
characterization of W[2]. Theoretical Computer Science 209(1-2) (1998) 123-140.
28

8. R. G. Downey and M. R. Fellows. Parameterized Complexity. Springer Verlag,

(1999).
9. R. G. Downey, M. R. Fellows and V. Raman. The complexity of irredundant sets
parameterized by size. Discrete Applied Mathematics 100(3) (2000) 155-167.
10. R. G. Downey, M. R. Fellows, A. Vardy and G. Whittle. The parametrized complex-
ity of some fundamental problems in coding theory. SIAM Journal on Computing
29(2) (1999) 545-570.
11. R. Duh and M. Fürer. Approximation of k-set cover by semi-local optimization. In
the Proceedings of the 29th Annual ACM Symposium on Theory of Computing
(STOC), (1997) 256-264.
12. J. Edmonds. Paths, trees and flowers. Canadian Journal of Mathematics 17 (1965)
449-467.
13. U. Feige. A threshold of ln n for approximating set cover. Journal of the ACM
45(4) (1998) 634-652.
14. J. Flum and M. Grohe. Parameterized Complexity Theory. Springer-Verlag, (2006).
15. F. V. Fomin, S. Gaspers and A. V. Pyatkin. Finding a Minimum Feedback Vertex
Set in time O(1.7548n ). In the Proceedings of the 2nd International Workshop on
Parameterized and Exact Computation (IWPEC), LNCS 4169 (2006) 184-191.
16. H. Fernau. Parameterized algorithms: A graph-theoretic approach. Habilitationss-
chrift (Universität Tübingen, Tübingen, Germany) (2005).
17. R. Niedermeier. Invitation to Fixed-Parameter Algorithms. Oxford Lecture Series
in Mathematics and Its Applications, Oxford University Press, (2006).
18. J. Guo, R. Niedermeier and S. Wernicke. Parameterized complexity of generalized
Vertex Cover problems. In the Proceeding of the 9th International Workshop Al-
gorithms and Data Structures (WADS), LNCS 3608 (2005) 36-48.
19. David S. Johnson. Approximation algorithms for combinatorial problems. Journal
of Computer and System Sciences 9(3) (1974) 256-278.
20. S. Jukna. Extremal Combinatorics. Springer-Verlag, (2001).
21. S. Khot and V. Raman. Parameterized complexity of finding Subgraphs with hered-
itary properties. Theoretical Computer Science 289(2) (2002) 997-1008.
22. G. Kortsarz and D. Peleg. On choosing a dense subgraph. In the Proceeding of
the 34th Annual Symposium on Foundations of Computer Science (FOCS), (1993)
692-701.
23. L. Lovàsz. On the ratio of optimal fractional and integral covers. Discrete Mathe-
matics 13 (1975) 383-390.
24. V. Raman and S. Saurabh. Triangles, 4-Cycles and parameterized (in-) tractabil-
ity. In the Proceeding of the 10th Scandinavian Workshop on Algorithm Theory
(SWAT), LNCS 4059 (2006) 304-315.
25. V. V. Vazirani. Approximation Algorithms. Springer-Verlag, (2001).