This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal noncommercial researchand education use, including for instruction at the authors institutionand sharing with colleagues.Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third partywebsites are prohibited.In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further informationregarding Elsevier’s archiving and manuscript policies areencouraged to visit:http://www.elsevier.com/authorsrights
Author's personal copy
Theoretical Computer Science 511 (2013) 85–108
Contents lists available at ScienceDirect
Theoretical Computer Science
journal homepage: www.elsevier.com/locate/tcs
A novel parameterised approximation algorithm for
minimum vertex cover
✩
Ljiljana Brankovic
a
, Henning Fernau
b,
∗
a
School of Electrical Engineering and Computer Science, The University of Newcastle, Callaghan, NSW 2308, Australia
b
Fachbereich 4, Abteilung Informatik, Universität Trier, D54286 Trier, Germany
a r t i c l e i n f o
Keywords:
Parameterised algorithmsApproximation algorithmsVertex cover
a b s t r a c t
Parameterised approximation is a relatively new but growing field of interest. Itmerges two ways of dealing with NPhard optimisation problems, namely polynomialapproximationandexactparameterised(exponentialtime)algorithms.Weexemplifythisidea by designing and analysing parameterised approximation algorithms for
minimumvertex cover
. More specifically, we provide a simple algorithm that works on anyapproximation ratio of the form
2
l
+
1
l
+
1
,
l
=
1
,
2
,...
, and has complexity that outperformspreviously published algorithms based on sophisticated exact parameterised algorithms.Inparticular,for
l
=
1(factor1
.
5approximation)ouralgorithmrunsintime
O
∗
(
1
.
0883
k
)
,where parameter
k
≤
23
τ
, and
τ
is the size of a minimum vertex cover. Additionally,we present an improved polynomialtime approximation algorithm for graphs of averagedegree at most four and a limited number of vertices with degree less than two.Crown Copyright
©
2012 Published by Elsevier B.V. All rights reserved.
1. Introduction
1.1.
minimum vertex cover
: A hard problem and how to deal with it
Given a graph
G
=
(
V
,
E
)
, where
V
is the set of vertices and
E
the set of edges of
G
, a
vertex cover C
of
G
is a subset of
V
whose removal leaves only isolated vertices. The problem of finding a minimum vertex cover in an arbitrary graph, called
minimum vertex cover
or
minvc
for short, has long attracted the attention of an army of theoretical computer scientistsaroundtheglobe.
minvc
isamongthefirstgraphproblemseverproventobeNPhard,beingonthefamouslistofKarp[39],and it has remained one of the most studied NPhard graph problems in complexity theory ever since. In particular, it hasbeen the paradigmatic testbed problem for the development of parameterised algorithms, being among the first problemspresented in any introduction to that field. Minimum vertex cover can be formulated as a decision problem as follows.
Problem name:
vertex cover (VC)
Given:
A graph
G
=
(
V
,
E
)
, a nonnegative integer
k
Parameter:
k
Output:
Is there a
vertex cover C
⊆
V
such that

C
 ≤
k
?
✩
Anextendedabstractofthispaperwaspresentedatthe21st‘‘InternationalSymposiumAlgorithmsandComputation’’(ISAAC2010),JejuIsland,Korea(Brankovic and Fernau (2010) [12]).
∗
Corresponding author. Tel.: +49 651 201 2827.
Email addresses:
Ljiljana.Brankovic@newcastle.edu.au (L. Brankovic), fernau@unitrier.de (H. Fernau).
03043975/$ – see front matter Crown Copyright
©
2012 Published by Elsevier B.V. All rights reserved.doi:10.1016/j.tcs.2012.12.003
Author's personal copy
86
L. Brankovic, H. Fernau / Theoretical Computer Science 511 (2013) 85–108
Both the polynomialtime approximation approach and the fixedparameter approach to
VC
have their advantagesand limitations, as detailed in Section 1.2 below. It is therefore of interest to combine both approaches, aiming at betterapproximation factors by allowing FPTtime. We refer the reader to the survey by D. Marx [42] and to the conceptualframework introduced by L. Cai and X. Huang [13] for more details on how notions from parameterised complexity can beextendedtoapproximationalgorithms.ThemostrecentsurveyonparameterisedapproximationcanbefoundontheslidesofD.Marxwww.cs.bme.hu/
∼
dmarx/papers/marxdagstuhl11091approx.pdf .ThisseemstobetheonlyreferencewheretheW[1]hard famous
(Topological) Bandwidth
problem was shown to be constantfactor fixedparameter approximable.Notice that
Minimum Bandwidth
is unlikely to be constantfactor approximable in polynomialtime; for more detailssee [24].In this paper, we follow the definition by Y. Chen et al.
1
[19], which in our context can be phrased as follows. Givena graph
G
and a parameter
k
, a
parameterised approximation algorithm with (constant) approximation ratio
ρ
for
minimumvertex cover
produces a vertex cover
C
such that

C
 ≤
ρ

C
∗

at least if

C
∗
 ≤
k
. More precisely, our algorithms willbehave as follows.
•
If our algorithms answers NO, then

C
∗

>
k
, that is, graph
G
does not have a vertex cover of size
k
.
•
If our algorithm produces a vertex cover
C
, then

C
 ≤
ρ
min
{
k
,

C
∗
}
.We give an alternative description in terms of the size

C
∗

of some minimum vertex cover
C
∗
:
•
If

C
∗
 ≤
k
, our algorithm produces a vertex cover
C
such that

C
 ≤
ρ

C
∗

.
•
If
k
<

C
∗
 ≤
ρ
k
, our algorithm can either answer NO, or produce a vertex cover
C
of size

C
 ≤
ρ
k
.
•
If

C
∗

> ρ
k
, our algorithm answers NO.In all cases, our algorithm runs in time
O
∗
(
f
(
k
))
for some function
f
, where
O
∗
notation suppresses polynomial factors,
i.e.
,
O
∗
(
f
(
k
))
is a shorthand for
O
(
f
(
k
)
poly
(
n
))
.Generally speaking, parameterised approximation is an emerging research area that appears very promising, especiallyfor those problems where under certain complexitytheoretic assumptions no further advancements in approximationand/or parameterised algorithms alone are to be expected. For example, it has been shown [13] that MAX SNP completeproblems (such as
minimum vertex cover
restricted to subcubic graphs) admit a fixedparameter approximation scheme,while no polynomialtime approximation scheme (PTAS) can be expected for such problems. On the other hand, W[P]hardminimisation problems do not admit constantfactor fixedparameter approximation algorithms unless FPT equals W[P][25], which is considered to be quite unlikely.
1.2. Earlier results on
VC
We report on the main results in each of the three approaches to the vertex cover problem: polynomialtime approximation, exact parameterised algorithms and parameterised (exponentialtime) approximation.1.
Polynomialtimeapproximation.
Despiteallefforts,thebestknownconstantfactorapproximationalgorithmforgeneralgraphs is still (basically) a factor 2 approximation. More specifically, G. Karakostas [38] derived an approximation factorof 2
−
Θ
(
1
√
log
n
)
. I. Dinur and S. Safra [23] showed that there is no polynomialtime approximation algorithm for
minvc
achievinganapproximationratiobetterthan10
√
5
−
21
≈
1
.
36067,unlessP
=
NP.Moreover,assumingthattheUniqueGames Conjecture is true, no approximation factor of the form 2
−
ε
is possible for any
ε >
0, as shown by S. Khot andO. Regev [40]. There are better results for some specific classes of graphs, most notably graphs with bounded degree(2
−
2
∆
[34] and 2
−
5
∆
+
3
+
ε
,
ε >
0 [8]), and 4colourable graphs (1
.
5) [34]. A good (although somewhat outdated)survey is contained in the third chapter of D. Hochbaum’s book [35]. The above mentioned results for degreeboundedgraphswereimprovedbyE. Halperin[33]towardsaperformancefactorof
(
2
−
(
1
−
o
(
1
)))
·
2lnln
∆
ln
∆
.Evenforgraphswithmaximumdegree3noapproximationfactorarbitrarilyclosetoone(
i.e.
,noPTAS)canbeachievedinpolynomialtime,asshown by P. Alimonti and V. Kann [3]. Another interesting result is due to T. Imamura and K. Iwama [36] who showed
that graphs
G
=
(
V
,
E
)
of maximum degree
∆
∈
Ω

V
·
loglog
(

V

)
log
(

V

)
and average degree
d
a
v
g
can be approximatedwith a factor of
21
+
da
v
g
2
∆
provided that

E
 ≤
∆
·
(

V
−
∆
)
. This yields factor
43
approximations for dense regular graphs(
i.e.
,
d
a
v
g
=
∆
) and an approximation factor of
32
provided that
d
a
v
g
≥
23
∆
for dense graphs.For certain graph classes like planar graphs, polynomialtime approximation schemes are known [6]. Some classesof graphs even have polynomialtime solutions,
e.g.
, bipartite graphs, where
minimum vertex cover
can be solved bymatching techniques due to the Theorem of König and Egerváry. One of the oldest online sources of this theorem seemsto be a paper of T. Gallai [31].
1A long version can be obtained from the homepage of M. Grohe in Berlin http://www2.informatik.huberlin.de/
∼
grohe/pub/chegrogru06.pdf .
Author's personal copy
L. Brankovic, H. Fernau / Theoretical Computer Science 511 (2013) 85–108
87
For the ease of reference, we make explicit the following three results that are the most relevant to this paper:
Lemma 1.
[8] There exists a polynomialtime factor
7
/
6
+
ε
approximation algorithm for
minimum vertex cover
for any graph with maximum degree
3
and any
ε >
0
.
Lemma 2.
There exist polynomialtime approximation algorithms for
minimum vertex cover
for any graph with maximumdegree
4
that guarantee factors of
3
/
2
[34] or even of
97
+
ε
, for any
ε >
0
[8].
Lemma 3.
[32] For an arbitrary graph with average degree d
a
v
g
, there exists a factor
(
4
d
a
v
g
+
1
)/(
2
d
a
v
g
+
3
)
approximationalgorithm for
minimum vertex cover
, providing that the size of a minimum vertex cover is at least

V

/
2
.
2.
Exact parameterised algorithms.
Asmentioned,
VC
istheparadigmatictestbedproblemforthistypeofalgorithms.Forgeneral graphs, several parameterised algorithms offer running times of about
O
∗
(
1
.
28
k
)
[15,16,18,37,44]. For special
classesofgraphs,betterrunningtimeboundsareknown.Forexample,forcubicgraphswehave
O
∗
(
1
.
194
k
)
[17]andevenbetter claims [50]. For graphs of bounded genus,
O
∗
(
c
√
k
)
can be obtained [21]. Such running times cannot be obtainedfor general graphs unless the Exponential Time Hypothesis fails [14].3.
Exact moderately exponentialtime algorithms.
Here, the task is to solve
minimum vertex cover
, or equivalently, dueto Gallai’s identity [31],
Maximum Independent Set
, in running times
O
∗
(
c
n
)
on graphs of order
n
. A sequence of paperssuggested branching algorithms with
c
≈
1
.
2; the reader is referred to [29,30,41,46,47] and the references therein.
4.
Parameterised and exponentialtime approximation.
More recently, the idea of using parameterised algorithms forobtaining better approximation factors at the cost of exponentialtime (best confined to a small entity called theparameter) was proposed in several independent papers in 2006, as surveyed by D. Marx in [42]. The focus of thesepapers is either on nonapproximability or on problems where possibly new qualities of approximation can be achievedby investing exponentialtime. Since
minvc
turns out to be quite a nice problem already, both from the parameterisedcomplexity point of view and from the standpoint of approximation, a combination of both lines of attack on thisparticularNPhardproblemwasnotventuredthere.However,asnoconstantfactorimprovementsoverthewellknownfactor2approximationalgorithmsaretobeexpectedfor
minvc
usingpolynomialtimeonly,investingFPTtimeappearswarranted.Thebenefitsofallowingexponentialtimetoimprovetheapproximationfactorshavebeenstudiedbyseveralauthors.For example, E. Dantsin et al. [20] showed how to transform an existing approximation algorithm for MAX SAT intoan algorithm with a better approximation factor by allowing exponential time. In relation to our paper, the article byN. Bourgeois et al. [11] is the most relevant one, as it shows that if there exists an exact exponential time algorithmfor computing minimum vertex cover that runs in time
O
∗
(γ
n
)
, then an approximation factor of 2
−
ρ
, for
ρ
∈
(
0
,
1
]
can be obtained with running time
O
∗
(γ
ρ
n
)
. Furthermore, any parameterised algorithm for
VC
running in time
O
∗
(δ
k
)
that produces an associated feasible solution can be used to obtain a parameterised approximation algorithm withapproximationratio
ρ
andrunningtime
O
∗
(δ
ρ
k
)
.AsimilarresulthasbeenindependentlyobtainedbyM.Fellows,A.Kulik,F. Rosamond and H. Shachnai [27] via socalled fidelity preserving transformations.An interesting alternative view on the matter is offered by V. Vassilevska and R. Williams and S. L. M. Woo [49] whosuggest algorithms that either provide better approximation ratios than any polynomialtime approximation algorithmor better running times than any exact algorithm can do (under certain complexity assumptions). We do not follow thisapproach here, but recommend it for future research.
1.3. Main contributions and organisation of the paper
In this paper, we present the first genuine parameterised approximation algorithm for
minimum vertex cover
. Our
O
∗
(
1
.
09
k
)
factor
32
approximation algorithm is simple, yet faster than the
O
∗
(δ
.
5
k
)
=
O
∗
(
1
.
13
k
)
algorithm by N. Bourgeoiset al. that incorporates a quite sophisticated exact parameterised
O
∗
(
1
.
28
k
)
algorithm for
VC
. The main difference betweenour algorithm and other approaches (both the one of N. Bourgeois et al. and the one of M. Fellows et al.) lies in the fact thatweexploitthetargetapproximationfactorineachbranchingstep,asopposedtotakingtheexact(parameterised)algorithmas a black box. This way, we not only obtain far better running times, but our branching algorithms are distinguishedby their simplicity. It is also the first such algorithm that introduces localratio techniques known from polynomialtimeapproximation, see [7], into parameterised algorithmics, combining them with the search tree analysis itself, althoughsimilar ideas are the basis of the mentioned results due to M. Fellows et al. We generalise this factor
32
approximationalgorithm to any constant factor of the form
2
l
+
1
l
+
1
. Actually, it is also possible to obtain any other constant factor less than2, and we elaborate on this in the Conclusion. We also obtain an improved polynomialtime approximation algorithms forgraphs of average degree at most four. This outset also explains the organisation of the paper.The considerable improvements that we obtain in comparison with other approaches are due to the following mainingredients of our algorithms:1. We develop new and powerful approximationpreserving reduction rules that may be of independent interest, also inthe context of polynomialtime approximation. Actually, the announced polynomialtime approximation result that wepresent in this paper profits greatly from this idea.
Author's personal copy
88
L. Brankovic, H. Fernau / Theoretical Computer Science 511 (2013) 85–108
2. Our algorithm branches on a pair of vertices, rather than on a single vertex. The vertices are selected in such a way so asto ensure that together they have at least two neighbours more than any one of them separately, which adds greatly tothe efficiency of the algorithm. We note that this novel strategy is not applicable to the exact parameterised case.3. We interleave socalled worsening steps with exact branching steps in order to speed up the branching. In this way,we can exploit these worsening steps to create better branching situations. By way of contrast, the fidelity preservingtransformations of Fellows et al. put these worsening steps at the very beginning of a parameterised approximationalgorithm, using any exact parameterised in a second step as a black box. Conversely, Dantsin et al. first do some (exact)branchingandthenapplytheknownpolynomialtimeapproximationalgorithmsineachoftheleavesofthesearchtree.Bourgeois et al. (independently) suggested introducing a worsening step (by selecting an edge every time a vertex isput into the independent set) to obtain a factor0
.
5 approximation for
Maximum Independent Set
. However, their mainresult related to ours (
i.e.
, to
Minimum Vertex Cover
) relies on a different technique, splitting the graph into pieces andthen to solving the problem on these pieces.4. Thevalidityoftheapproachisjustifiedbyargumentsreminiscentofthelocalratiotechniqueknownfromapproximationtheory.
1.4. Some bits of graph theory
We consider simple undirected graphs and we use
G
=
(
V
,
E
)
to denote a graph
G
with the vertex set
V
and the edgeset
E
, where
E
can be seen as a set consisting of twoelement vertex sets, modelling the symmetric
adjacency relation
on
V
.Two edges
e
,
f
with
e
∩
f
̸ = ∅
are called
incident
. We use
∆
(
G
)
and
δ(
G
)
to denote the maximum and minimum degree in
G
,respectively,and
d
G
(v)
todenotethedegreeofthevertex
v
∈
V
within
G
,suppressingthesubscript
G
ifunderstoodfromthe context. The neighbourhood of the vertex
v
is denoted by
N
G
(v)
and contains all the vertices at distance exactly 1 from
v
, so that
d
G
(v)
= 
N
G
(v)

. Given a graph
G
and two nonadjacent vertices in
G
,
u
and
v
, those neighbours of
u
that are notalso neighbours of
v
are called
private neighbours of u with respect to
v
, referring to the set
N
(
u
)
\
N
(v)
.
C
⊆
V
is a
vertex cover
of
G
=
(
V
,
E
)
if every edge
e
∈
E
has at least one of its endpoints in
C
. The size of the smallestvertex cover of
G
is also known as the
vertex cover number
of
G
and is denoted by
τ(
G
)
. The complement of a vertex coveris known as an
independent set
,
i.e.
, a set
I
of vertices of
G
such that no two vertices
u
,v
∈
I
are connected with an edge.Likewise, a set
M
of edges is called
independent
or a
matching
if, for any pair of edges
e
,
f
∈
M
,
e
∩
f
= ∅
,
i.e.
, no two edgesin
M
share a vertex.If
G
=
(
V
,
E
)
is a graph and
U
⊆
V
, then
G
[
U
]
denotes the graph
induced by U
,
i.e.
,
G
[
U
] =
(
U
,
F
)
with
F
= {
e
∈
E

e
⊆
U
}
.Wewilloftenneedtheoperationof
removing
or
deleting
verticesor,moregenerally,asetofvertices
D
fromagraph(together with the incident edges), referring to
G
−
D
=
G
[
V
\
D
]
. Note that
C
is a vertex cover of
G
if and only if
G
−
C
hasno edges.A
path
is a sequence
v
0
,...,v
ℓ
of vertices such that each two consecutive vertices are connected with an edge,
i.e.
, foreach
i
=
1
,
2
,...,ℓ
there exists an edge
e
i
= {
v
i
−
1
,v
i
} ∈
E
; we will also refer to such a path as a path between
v
0
and
v
ℓ
. If
ℓ
= {
v
1
,...,v
ℓ
}
,thenthepathiscalled
simple
.If
v
0
=
v
ℓ
,thenasimplepathiscalleda
cycle
.Acycleoflengththreeisalsoknownasa
triangle
.Agraph
G
iscalled
trianglefree
ifthereisnotriple
(
x
,
y
,
z
)
ofverticessuchthat
x
,
y
,
z
,
x
isatriangle.Weuse dist
G
(
u
,v)
to denote the
distance
between two vertices
u
and
v
,
i.e.
, the length
ℓ
of the shortest path between
u
and
v
.The following two Lemmas are crucial for the algorithm analysis that we present in Sections 3 and 4.
Lemma 4.
Consider a connected graph G with maximum degree
∆
(
G
)
. Let G
′
be a graph obtained from G by deleting one or more vertices together with their incident edges. If G
′
contains any edges at all, then there exists a vertex y in G
′
such that
0
<
d
G
′
(
y
) <
∆
(
G
)
.
Proof.
Consider a connected graph
G
=
(
V
,
E
)
with maximum degree
∆
(
G
)
. Let
G
′
=
(
V
′
,
E
′
)
be a graph obtained from
G
by deleting
X
⊆
V
(together with the incident edges) and let
Y
be the set of neighbours of vertices from
X
in
G
that are notcontained in
X
. Then
d
G
′
(
y
) <
∆
(
G
)
for all
y
∈
Y
, as each vertex in
Y
has a neighbour in
X
in graph
G
. If
d
G
′
(
y
)
=
0 for all
y
∈
Y
, then the only neighbours of
Y
vertices in
G
were from
X
. Since
G
is connected, there are no other vertices in
G
apartfrom vertices in
X
and
Y
, and thus
V
′
=
Y
and
E
′
= ∅
.
Lemma 5.
In a trianglefree graph with minimum degree
δ
≥
2
, any vertex
v
has a vertex u at distance two.
Proof.
Let
G
=
(
V
,
E
)
be a trianglefree graph with minimum degree
δ
≥
2. Consider some vertex
v
∈
V
. Since
d
(v)
≥
2,we have that
N
(v)
̸ = ∅
. Consider any
x
∈
N
(v)
. As
d
(
x
)
≥
2, it follows that there exists some vertex
u
∈
N
(
x
)
\{
v
}
. Sincethere are no triangles in
G
, it follows that
u
/
∈
N
(v)
. Hence, dist
(v,
u
)
=
2.
We occasionally need the following generalisation of the previous lemma:
Lemma 6.
In a trianglefree graph with minimum degree
δ
≥
2
, to any vertex
v
we can associate a matching M
v
,

M
v
 ≥
δ
−
1
,such that each matching edge xy
∈
M
v
obeys
{
x
,
y
} ∩
N
(v)
 = {
x
,
y
} ∩
N
[
v
] =
1
. This matching can be greedily found in polynomialtime.