Contracting edges to destroy a pattern: A complexity study
Dipayan Chakraborty, R. B. Sandeep

TL;DR
This paper studies the computational complexity of contracting edges in a graph to eliminate a certain pattern, showing NP-completeness in most cases and analyzing fixed-parameter tractability for H-free contractions.
Contribution
It characterizes the complexity of H-free Contraction problems, proving NP-completeness for all but trivial cases and analyzing fixed-parameter tractability for tree patterns.
Findings
H-free Contraction is polynomial-time solvable only when H is a small complete graph.
Most H-free Contraction problems are NP-complete.
H-free Contraction is W[2]-hard for most trees, with a few exceptions.
Abstract
Given a graph G and an integer k, the objective of the -Contraction problem is to check whether there exists at most k edges in G such that contracting them in G results in a graph satisfying the property . We investigate the problem where is `H-free' (without any induced copies of H). It is trivial that H-free Contraction is polynomial-time solvable if H is a complete graph of at most two vertices. We prove that, in all other cases, the problem is NP-complete. We then investigate the fixed-parameter tractability of these problems. We prove that whenever H is a tree, except for seven trees, H-free Contraction is W[2]-hard. This result along with the known results leaves behind three unknown cases among trees.
| Problem | P | NPC | FPT | W-hard |
|---|---|---|---|---|
| Edge Editing | [trivial] | otherwise [1] | For all [6] | |
| Edge Deletion | [trivial] | otherwise [1] | For all [6] | |
| Edge Completion | [trivial] | otherwise [1] | For all [6] | |
| Vertex Deletion | [trivial] | otherwise [19] | For all [6] | |
| Edge Contraction | [trivial] | otherwise [Theorem 3.1] | () [15, 20], ( [8, 20], MSO1 expressibility) | W[2]-hard for 3-connected non-complete graphs, diamond [8, 15], () [8, 20], all trees except 7 trees [Theorem 4.1] |
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAdvanced Graph Theory Research · Limits and Structures in Graph Theory · semigroups and automata theory
\setlistdepth
9
11institutetext: LIMOS, Université Clermont Auvergne, France, 11email: [email protected]: Department of Mathematics and Applied Mathematics, University of Johannesburg, South Africa33institutetext: Department of Computer Science and Engineering, Indian Institute of Technology Dharwad, India, 33email: [email protected]
Contracting edges to destroy a pattern: A complexity study††thanks: This work is partly sponsored by SERB(India) grants “Complexity dichotomies for graph modification problems” (SRG/2019/002276), and “Algorithmic study on hereditary graph properties” (MTR/2022/000692), and a public grant overseen by the French National Research Agency as part of the “Investissements d’Avenir” through the IMobS3 Laboratory of Excellence (ANR-10-LABX-0016), and the IDEX-ISITE initiative CAP 20-25 (ANR-16-IDEX-0001). We also acknowledge support of the ANR project GRALMECO (ANR-21-CE48-0004).
Dipayan Chakraborty 1122
R. B. Sandeep 22
Abstract
Given a graph and an integer , the objective of the -Contraction problem is to check whether there exists at most edges in such that contracting them in results in a graph satisfying the property . We investigate the problem where is ‘-free’ (without any induced copies of ). It is trivial that -free Contraction is polynomial-time solvable if is a complete graph of at most two vertices. We prove that, in all other cases, the problem is NP-complete. We then investigate the fixed-parameter tractability of these problems. We prove that whenever is a tree, except for seven trees, -free Contraction is W[2]-hard. This result along with the known results leaves behind only three unknown cases among trees.
Keywords:
Edge contraction problem H-free NP-completeness W[2]-hardness Trees
1 Introduction
Let be any graph property. Given a graph and an integer , the objective of the -Contraction problem is to check whether contains at most edges so that contracting them results in a graph with property . This is a vertex partitioning problem in disguise: Find whether there is a partition of the vertices of such that each set in induces a connected subgraph of , (the graph obtained by contracting each set in into a vertex) has property , and . These problems, for various graph properties , have been studied for the last four decades. Asano and Hirata [2] proved that the problem is NP-complete if is any of the following classes - planar, series-parallel, outerplanar, chordal. When is a singleton set , then the problem is known as -Contraction. Brouwer and Veldman [5] proved that -Contraction is polynomial-time solvable if is a star, and NP-complete if is a connected triangle-free graph other than a star graph. Belmonte, Heggernes, and van ’t Hof [3] proved that it is polynomial-time solvable when is a split graph. Golovach, Kaminski, Paulusma, and Thilikos [13] studied the problem when is ‘minimum degree at least ’ and proved that the problem is NP-complete even for and W[1]-hard when parameterized by . Heggernes, van ’t Hof, Lokshtanov, and Paul [16] proved that the problem is fixed parameter tractable when is the class of bipartite graphs. Guillemot and Marx [14] obtained a faster FPT algorithm for the problem. Cai, Guo [8], and Lokshtanov, Misra, and Saurabh [20] proved that the problem is W[2]-hard when is the class of chordal graphs. Garey and Johnson [11] mentioned that, given two graphs and , the problem of checking whether can be obtained from by edge contractions is NP-complete. Edge contraction has applications in Graph minor theory (see [21]), Hamiltonian graph theory [17], and geometric model simplification [12].
We consider the -free Contraction problem: Given a graph and an integer , find whether can be transformed, by at most edge contractions, into a graph without any induced copies of . The parameter we consider is . Unlike graph contraction problems, other major graph modification problems are well-understood for these target graph classes. In particular, P versus NP-complete dichotomies are known for -free Edge Editing, -free Edge Deletion, -free Edge Completion [1], and -free Vertex Deletion [19] (here, the allowed operations are edge editing, edge deletion, edge completion, and vertex deletion respectively). It is also known that all these problems are in FPT for every graph [6]. The picture is far from complete for -free Contraction. See Table 1. It is trivial to note that -free Contraction is polynomial-time solvable if is a complete graph of at most 2 vertices. Cai, Guo [8, 15], and Lokshtanov, Misra, and Saurabh [20] proved the following results for -free Contraction.
- •
FPT when is a complete graph
- •
If is a path or a cycle, then the problem is FPT when has at most 3 edges, and W[2]-hard otherwise.
- •
W[2]-hard when is 3-connected but not complete, or a star graph on at least 5 vertices, or a diamond.
The W[2]-hardness results mentioned above also imply NP-completeness of the problems. Guo [15] proved that the problem is NP-complete when is a complete graph on vertices, for every . Eppstein [10] proved that the Hadwiger number problem (find whether the size of a largest clique minor of a graph is at least ) is NP-complete. This problem is essentially -free Contraction, if we ignore the parameter. This result implies that the problem is NP-complete when is a . We build on these results and prove the following.
- •
-free Contraction is NP-complete if is not a complete graph on at most 2 vertices.
- •
-free Contraction is W[2]-hard if is a tree which is neither a star on at most 4 vertices ( , , , ) nor a bistar in , , .
Our W[2]-hardness results, along with known positive results, leaves behind only three open cases among trees -
,
,
.
2 Preliminaries
Graphs.
All graphs considered in this paper are simple and undirected. A complete graph and a path on vertices are denoted by and respectively. A universal vertex of a graph is a vertex adjacent to every other vertex of the graph. An isolated vertex is a vertex with degree 0. For an integer , a star on vertices, denoted by , is a tree with a single universal vertex and degree-1 vertices. The universal vertex in is also called the center of the star. For integers such that , a bistar on vertices, denoted by , is a tree with two adjacent vertices and , where degree-1 vertices are attached to and degree-1 vertices are attached to . The bistar is the star and the bistar is . We say that is the central edge of the bistar. By we denote the disjoint union of the graphs and . A graph is -free if it does not contain any induced copies of . In a graph , replacing a vertex with a graph is the graph obtained from by removing , introducing a copy of , and adding edges between every vertex of the and every neighbor of in . A separator of a connected graph is a subset of its vertices such that (the graph obtained from by removing the vertices in ) is disconnected. A separator is universal if every vertex of the separator is adjacent to every vertex outside the separator. We will be using the term ‘universal (resp. ) separator’ to denote a universal separator which induces a (resp. ). Let be a subset of vertices of a graph . By we denote the graph induced by in . For a graph and two subsets and of vertices of , by we denote the set of edges in , where each edge in the set is having one end point in and the other end point in .
Contraction.
Contracting an edge in a graph is the operation in which the vertices and are identified to be a new vertex such that is adjacent to every vertex adjacent to either or . Given a graph and a subset of edges of , the graph obtained by contracting the edges in does not depend on the order in which the edges are contracted. Every vertex in represents a subset of vertices (which are contracted to ) of such that induces a connected graph in . Let be the subgraph of containing all vertices of and the edges in . There is a partition of vertices of implied by : Every set in corresponds to the vertices of a connected component in . We note that many subsets of edges may imply the same partition - it does not matter which all edges of a connected subgraph are contracted to get a single vertex. The graph is nothing but the graph in which there is a vertex corresponding to every set in and two vertices in are adjacent if and only if there is at least one edge in between the corresponding sets in . The graph is equivalently denoted by . Assume that is a partition of a subset of vertices of . Then by we denote the graph obtained from by contracting each set in into a single vertex. The cost of a set in is the number , which is equal to the minimum number of edges required to form the set . The cost of , denoted by cost, is the sum of costs of the sets in . We observe that cost , where is the number of vertices of . We say that touches a subset of vertices of , if there is at least one edge in such that either or is in . Let be two non-adjacent vertices of . Identifying and in is the operation of removing and , adding a new vertex , and making adjacent to every vertex adjacent to either or .
Fixed parameter (in)tractability.
A parameterized problem is fixed-parameter tractable (FPT) if it can be solved in time -time, where is a computable function and is the input. Parameterized problems fall into different levels of complexities which are captured by the W-hierarchy. A parameterized reduction from a parameterized problem to a parameterized problem is an algorithm which takes as input an instance of and outputs an instance of such that the algorithm runs in time (where is a computable function), and is a yes-instance of if and only if is a yes-instance of , and (for a computable function ). We use parameterized reductions to transfer fixed-parameter intractability. For more details on these topics, we refer to the textbook [9]. The problem that we deal with in this paper is defined as follows.
-free Contraction: Given a graph and an integer , can be modified into an -free graph by at most edge contractions?
3 NP-completeness
In this section we prove that -free Contraction is NP-complete whenever is not a complete graph of at most two vertices. In Section 3.1 we obtain reductions for the cases when is connected but does not have any universal separator and universal separator. In Section 3.2 we deal with non-star graphs with universal separator or universal separator. In Section 3.3 we resolve the stars. The cases when is a or a are handled in Section 3.4. These come as base cases in the inductive proof of the main result of the section given in Section 3.5. We crucially use the following results.
Proposition 1 ([15, 10])
-free Contraction* is NP-complete when is a , or a , or a , for any .*
3.1 A general reduction
A vertex cover of a graph is a subset of its vertices such that for every edge of , either or is in . Vertex Cover is the decision problem in which the objective is to check whether the given graph has a vertex cover of size at most . The problem is very well-known to be an NP-complete problem. Inspired by a reduction by Asano and Hirata [2], we introduce the following reduction from Vertex Cover which handles connected graphs without any universal separator and universal separator.
Construction 1
Let be a graph without any isolated vertices, and be a connected non-complete graph. We obtain a graph from and as follows.
- •
Subdivide each edge of once, i.e., for every edge , introduce a new vertex and make it adjacent to both and , and delete the edge .
- •
Replace each new vertex by a copy .
- •
Let be a non-universal vertex of (the existence of is guaranteed as is not a complete graph). Identify of every copy of (introduced in the previous step) to be a single vertex named .
Let the resultant graph be . The vertices in copied from form the set , which forms an independent set in . For each edge in , the vertices, except , of the copy of is denoted by . By we denote any such set. We note that is adjacent to every vertex in , as does not have any isolated vertices. This completes the construction. An example is shown in Figure 1.
Let be a connected non-complete graph with vertices and without any universal or separator. Let be a graph without any isolated vertices. Let be obtained from by Construction 1. Assume that is a yes-instance of Vertex Cover and let be any vertex cover of size at most of . Let . Let itself denote the vertex obtained by contracting the edges in . The following can be observed directly from the fact that is a vertex cover.
Observation 1
The vertex is a universal vertex in .
Before, proving that is -free, we prove the following lemma.
Lemma 1
Let be a vertex in . Let be any neighbor of in . Then none of the subsets of induces an in .
Proof
For a contradiction, assume that a subset of induces an in . Since is not a universal vertex in , and both (by Observation 1) and (by the construction) are universal to in , we obtain that neither nor induces in . Therefore, , for some vertex in . Let be the number of edges in . Let be the number of edges in the graph induced by in . Clearly, , where and denote the degree of and degree of respectively in , and is 1 if there is an edge between and in , and 0 otherwise. Since induces an in , we obtain that (note that both and are universal in the induced by ). Therefore, we obtain that . Since is not universal in , we obtain that . Therefore, . If is universal in , then we get that , and if is not universal in , then we obtain that . Therefore, we get a contradiction.
Lemma 2
If is a yes-instance of Vertex Cover, then is a yes-instance of -free Contraction.
Proof
Let be a vertex cover of size at most of and . We claim that is -free. For a contradiction, assume that there is a copy of induced by a set in . By Observation 1, is a universal vertex in . Since the vertices in along with induces a star graph in , and is a not a subgraph of a star graph (a non-complete star graph has a universal separator), cannot be a subset of . Therefore, must include some vertices from at least one copy of , say . Since is universal to and is not a universal vertex in , we obtain that does not induce an . Since is a vertex cover, either or is in . Assume that both are in . Then must contain at least one vertex not in and then is a universal separator in the induced by . Assume that and . By Lemma 1, none of the subsets of induces an . Let denote the union of all (for every edge in ). If contains at least one vertex not in , then is a universal separator in the induced by . If does not contain any such vertex, then must contain a vertex from for some . Then either or is a universal separator or is a universal separator in the induced by . Thus, in all these cases, we obtain contradictions.
Lemma 3
Let be a yes-instance of -free Contraction. Then is a yes-instance of Vertex Cover.
Proof
Let be a set of edges of such that and is -free. We create a vertex cover of as follows.
- •
For every vertex of , if is in , then include in .
- •
For every edge of , if there is an edge , where , then include in .
- •
For every edge of , if there is an edge , or an edge , where , include either or in arbitrarily.
Clearly, . Now we prove that is a vertex cover of . We note that induces an in for every edge in . Therefore, a vertex in must be touched by (the only neighbors of are , and ). Therefore, by the construction of , either or is in . Hence is a vertex cover of .
NP-completeness of Vertex Cover and Lemmas 2, 3 prove Lemma 4.
Lemma 4
Let be a connected non-complete graph with neither a universal separator nor a universal separator. Then -free Contraction is NP-complete.
3.2 Graphs with universal clique separators
In this section, we handle the graphs with either a universal separator (except stars) or with a universal separator. We note that cannot have both a universal separator and a universal separator. Further cannot have more than one such separator.
Construction 2
Let be any graphs and let be any subset of vertices of . Let be positive integers. We obtain a graph from as follows. For every set of vertices of , where induces a clique on vertices in , do the following: Introduce copies of and make every vertex of the copies adjacent to every vertex of . Let denote the set of new vertices introduced for , and let be the set of all new vertices.
Let be a graph with a universal separator or a universal separator. Let the set of vertices of the separator be denoted by . Assume that there exists at least two non-isomorphic components in (therefore, cannot be a star). Let be a component in with minimum number of vertices. Let be the number of times appears (as a component) in . Let be the set of vertices of a copy of . Let be the graph obtained from by removing the vertices of every component isomorphic to in . Let be an instance of -free Contraction. Let be obtained from by Construction 2. An example of the construction is shown in Figure 2.
Lemma 5
Let be a yes-instance of -free Contraction. Then is a yes-instance of -free Contraction.
Proof
Let be a partition of vertices of such that each set induces a connected graph, cost(\mathcal{P}^{\prime})$$\leq k, and is -free. Let be obtained from by adding the singleton sets corresponding to vertices in . Let be the singleton sets corresponding to vertices in , i.e., . Clearly, cost(\mathcal{P})$$=cost(\mathcal{P}^{\prime})$$\leq k. We claim that is -free. For a contradiction, assume that induces an in . Since there are at least two components in , we obtain that degree of a vertex in is strictly less than that of a vertex in in . Therefore, None of the sets in can act as a universal vertex (a vertex in ) in the induced by . Assume that . Let be the universal vertex of the induced . Since is a component in with minimum number of vertices, if a singleton set corresponding to a vertex in (for any vertex ) is in , then each of the singleton sets corresponding to vertices in (forming a copy of ) is in . Among the sets in , the set is adjacent to only the singleton sets corresponding to vertices in (for ). Therefore, the singleton sets corresponding to , for , are not in . Therefore, there must be an induced copy of in , which is a contradiction. The case when can be proved in a similar way.
Lemma 6
Let be a yes-instance of -free Contraction. Then is a yes-instance of -free Contraction.
Proof
Let be a subset of edges of such that is -free and . Let be the partition of vertices of corresponding to . We create a partition from as follows. For every , add the set to . Clearly, is a partition of vertices of and each set in induces a connected subgraph in . Further, cost cost . Now, it is sufficient to prove that is -free. For a contradiction, assume that has an induced by .
Let . Let be a set corresponding to the universal separator (of ) which is part of the induced . Consider any vertex . Since there are copies of attached to , at least of them are not touched by (i.e., each vertex of those copies of forms a singleton set in ). Therefore, the singleton sets in corresponding the copies of along with the sets in containing sets in induce an in , which is a contradiction.
With similar arguments we can handle the case of . Let . Let be two sets corresponding to the two vertices in the universal separator (of ), which is part of the induced . Since and are adjacent, there exists a vertex and a vertex such that is an edge in . Since there are copies of attached to , at least of them are not touched by (i.e., each vertex of those copies of forms a singleton set in ). Therefore, the singleton sets in corresponding the copies of along with the sets in containing sets in induce an in , which is a contradiction.
We obtain Lemma 7 from Lemmas 5 and 6.
Lemma 7
Let be a graph with a universal separator or a universal separator, denoted by . Assume that has at least two components which are not isomorphic. Let be a component in with minimum number of vertices. Let be obtained from by removing all components of isomorphic to . Then there is a polynomial-time reduction from -free Contraction to -free Contraction.
What remains to handle is the case when has a universal separator or a universal separator such that is a disjoint union of a graph . The diamond graph is an example. For this we need the concept of an enforcer - a structure to forbid contraction of certain edges. Enforcers are used widely in connection with proving hardness results for edge modification problems (see [7, 15, 22]).
We define an enforcer of a graph as a graph with two specified adjacent vertices such that the following hold true for every instance of -free Contraction and for every edge of : is a yes-instance of -free Contraction if and only if there is a partition of vertices of (where each set in induces a connected subgraph of ) such that is -free, cost(\mathcal{P})$$\leq k, and and are in different sets in . Here, is the graph obtained from by attaching copies of by identifying with , and with for each copy of .
Now, we will handle the case in which is the disjoint union of copies of a connected graph . Assume that is not a star graph. Let be a vertex in and be a vertex in . Clearly, is an edge. Subdivide by introducing a new vertex , making it adjacent to both and , and deleting the edge . We note that contracting the edge in creates . We will prove that along with is an enforcer for .
Let be obtained from as described above. An example is shown in Figure 3. Let denote the set of vertices of copies of , except . Let be the set of all new vertices, i.e., . Let the number of vertices in be .
Lemma 8
Let be a partition of vertices of (where each set in induces a connected subgraph in ) such that cost(\mathcal{\hat{P}})$$\leq k, and is -free. Let and denote the sets in containing and respectively. Then .
Proof
For a contradiction, assume that . There are copies of in where all the copies are attached to . This implies that all of them cannot be destroyed by . Then there exists a set such that every vertex in is a singleton set in . Then those sets and induce an in , which is a contradiction.
Lemma 9
Let be a partition of vertices of (where each set in induces a connected subgraph in ) such that and are not in the same set in , and is -free. Let be obtained from by adding singleton sets corresponding to vertices in . Then is -free.
Proof
For a contradiction, assume that induces an in . Clearly, none of the subsets of induces an in . Hence at least one set in is part of , let it be from . Let be the subset of singleton sets in corresponding to . Also assume that and be the sets in containing and respectively. Clearly, induces an and not an . Therefore, contains sets corresponding to () or from . Therefore, either or acts as a universal vertex in the induced by . Recall that is identified with , a degree two vertex in , whose neighbors are not adjacent to each other (one of them is which is identified with ). Therefore, if acts as a universal vertex, then can have only one vertex. This is a contradiction as is not a star graph. Recall that is identified with a vertex of which is adjacent to at most vertices including , where is not adjacent to any neighbors of . Therefore, if acts as a universal vertex, can have only one vertex, which is a contradiction as is not a star graph.
Lemmas 8 and 9 imply that is an enforcer for .
Lemma 10
* is an enforcer for .*
Proof
We need to prove that is a yes-instance of -free Contraction if and only if there is a partition of vertices of (where each set in induces a connected subgraph of ) such that is -free, cost(\mathcal{P})$$\leq k, and and are in different sets in . Lemma 9 proves the backward direction, i.e., if there is a partition as specified, then is a yes-instance. For the forward direction, assume that is a yes-instance. Let be a partition of vertices of such that cost(\mathcal{\hat{P}})$$\leq k and is -free. Let and be the sets in containing and respectively. By Lemma 8, . Create from as follows: For every , include in . It is straight-forward to verify that is -free. Clearly, cost(\mathcal{P})$$\leqcost(\mathcal{\hat{P}})$$\leq k. Therefore, is an enforcer for .
Now we will come up with a reduction from -free Contraction to -free Contraction, where is , i.e., is isomorphic to .
Construction 3
Let be graphs and be an integer. Let be two vertices in . We obtain a graph from by introducing two new adjacent vertices and making them adjacent to all vertices of . Then we obtain a graph from as follows:
- •
For every edge (for ), attach copies of by identifying with and with .
- •
For every edge (for ), attach copies of by identifying with and with .
- •
Attach copies of by identifying with and with .
Recall that is a non-star graph with a universal separator or a universal separator, where the universal separator is denoted by . Let be an instance of -free Contraction, where . Let be the enforcer we created for (Lemma 10). Let and be the graphs obtained from by Construction 3.
Lemma 11
Let be a yes-instance of -free Contraction. Then is a yes-instance of -free Contraction.
Proof
To prove that is a yes-instance of -free Contraction, by Lemma 10, it is sufficient to prove that is a yes-instance with a solution such that and are not in the same set in , and are not in the same set in (for every ), and and are not in the same set. Let be a partition of vertices of such that is -free and cost(\mathcal{P}^{\prime})$$\leq k. We obtain , a partition of vertices of , from by including two singleton sets and . We observe that if has an induced , then and can act only as universal vertices in the and hence must contain an induced , which is a contradiction. It is straight-forward to verify that satisfies the rest of the required properties.
Lemma 12
Let be a yes-instance of -free Contraction. Then is a yes-instance of -free Contraction.
Proof
Let be a yes-instance of -free Contraction. Then Lemma 10 implies that there exists a partition of vertices of such that , , and , for every vertex , where denote the sets containing respectively. Further, is -free and cost(\mathcal{P})$$\leq k. This implies that and . Let . Clearly, cost(\mathcal{P}^{\prime})$$=cost(\mathcal{P})$$\leq k. We claim that is -free. For a contradiction, assume that induces an in . Then, we obtain that there is an induced in by , if , or by , if . This gives us a contradiction.
Lemmas 11 and 12 prove Lemma 13.
Lemma 13
Let be a graph with a universal separator or a universal separator, denoted by . Assume that is a disjoint union of copies of a graph , for some . Let be not a star and let be . Then there is a polynomial-time reduction from -free Contraction to -free Contraction.
3.3 Stars
Here, we prove that -free Contraction is NP-complete whenever is a star graph of at least 4 vertices.
Construction 4
Let be a graph and be integers. We obtain a graph from as follows.
- •
Introduce four cliques having vertices each. Make every vertex of adjacent to . Similarly, make every vertex of adjacent to every vertex of , and every vertex of adjacent to every vertex of .
- •
*For every vertex , introduce cliques of vertices each, and make adjacent to every vertex of . Let denote the union of all *s.
An example is given in Figure 4. For two integers , by we denote the graph obtained by introducing a vertex adjacent to all vertices of copies of s. The following is a trivial observation.
Observation 2
Let be a yes-instance of -free Contraction for any integer . Then does not have as an induced subgraph.
We give a reduction from -free Contraction to -free Contraction, for every . For an instance of -free Contraction, we obtain an instance of -free Contraction by applying Construction 4 on .
Lemma 14
Let be a yes-instance of -free Contraction. Then is a yes-instance of -free Contraction.
Proof
Let be a partition of vertices of such that cost(\mathcal{P}^{\prime})$$\leq k, is a clique (i.e., -free), and every set in induces a connected graph in . Construct from by introducing singleton sets corresponding to the vertices . Clearly, cost(\mathcal{P})$$=cost(\mathcal{P}^{\prime})$$\leq k. It is straight-forward to verify that a maximum sized induced star graph in is . Therefore, is -free.
Lemma 15
Let be a yes-instance of -free Contraction. Then is a yes-instance of -free Contraction.
Proof
Let be a partition of vertices of such that is -free and cost(\mathcal{P})$$\leq k. For a vertex in , let denote the set containing in . First we claim that , where . If, on the contrary, for some , we observe that, in , there is a induced by and the vertices in . Therefore, by Observation 2, the claim is true. By symmetry, we obtain that , , and for vertices . This implies that there is no set in containing vertices from both and . Therefore, there is a subset of such that is a partition of vertices of and every set in induces a connected subgraph in . Now, it is sufficient to prove that that is a clique. For a contradiction assume that there is a induced in by . We observe that for every clique attached to , there is a partition in such that (all the vertices in cannot be contracted to as the budget is only and has vertices). Therefore, for every vertex , is a center of a in , where the leaves are from the cliques attached to . This along with the sets induce a in , which is a contradiction.
Lemmas 14 and 15, and the NP-completeness of -free Contraction (Proposition 1) prove Lemma 16.
Lemma 16
For , -free Contraction is NP-complete.
3.4 Small graphs
In this section, we prove that -free Contraction is NP-complete when is either a or a . First we handle the case of with a simple reduction from -free Contraction.
Construction 5
Let be a graph and be an integer. We obtain a graph by attaching pendant vertices, denoted by a set , to every vertex in . Let denote the set of all newly added vertices.
Let be an instance of -free Contraction. We obtain an instance of -free Contraction by applying Construction 5 on .
Lemma 17
Let be a yes-instance of -free Contraction. Then is a yes-instance of -free Contraction.
Proof
Let be a partition of vertices of such that cost(\mathcal{P}^{\prime})$$\leq k and is -free. We obtain a partition of vertices of from by introducing singleton sets corresponding to the vertices in . Clearly, cost(\mathcal{P})$$=cost(\mathcal{P}^{\prime})$$\leq k. Since there is no edge induced by the sets corresponds to vertices in , we obtain that if is not -free, then there is an induced in , which is a contradiction.
Lemma 18
Let be a yes-instance of -free Contraction. Then is a yes-instance of -free Contraction.
Proof
Let be a partition of vertices of such that is -free and cost(\mathcal{P})$$\leq k. We obtain a partition of vertices of as follows: For every set , include in . Since induces a connected graph in , induces a connected graph in . Assume that there is a induced by . Let and . Since there is a set of pendant vertices attached to and a set of pendant vertices attached to , at least one vertex from and at least one vertex from form singleton sets in . Then, those two sets along with the sets containing and the set containing in induces a in , which is a contradiction.
Now, the NP-completeness of -free Contraction follows from that of -free Contraction (Proposition 1) and Lemmas 17 and 18.
Lemma 19
-free Contraction* is NP-complete.*
Next we prove the hardness for -free Contraction by a reduction from Domatic Number. Domatic number of a graph is the size of a largest set of disjoint dominating sets of the graph, which partitions the vertices of the graph. For example, the domatic number of a complete graph is the number of vertices of it, and that of a star graph is 2. The Domatic Number problem is to find whether the domatic number of the input graph is at least or not. It is known that Domatic Number is NP-complete [11] even for various classes of graphs [4, 18]. Recall that -free graphs are exactly the class of complete multipartite graphs. The reduction that we use is exactly the same as the reduction for the NP-completeness of Hadwiger number problem (which is equivalent to -free Contraction) described by Eppstein [10]. The proof requires some adaptation.
Let be an instance of Domatic Number such that has vertices. Let be the vertices of . We can safely assume that there are no universal vertices in (otherwise, remove it and decrease by 1 to obtain an equivalent instance). We obtain a graph as follows:
- •
Introduce a clique of size , called the upper vertices.
- •
Introduce an independent set of size , called the middle vertices.
- •
Introduce a clique of vertices, where is a clique of vertices, for . These vertices are called the lower vertices.
- •
Make every vertex in adjacent to every vertex in .
- •
The vertex is adjacent to (every vertex of) if and only if dominates , i.e., either or is adjacent to in .
Let be the number of vertices in , i.e., . Let . This completes the reduction. An example is given in Figure 5.
The idea behind Eppstein’s reduction is the following: If has disjoint dominating sets, then a partition of vertices of can be obtained by the subsets of corresponding to the dominating sets of (subsets of is an independent set, so take one vertex from to make sure that each set induces a connected subgraph). On the other hand, if is a yes-instance of -free Contraction, then there exists a partition (which is a solution) such that each subset of contained in (the set containing a vertex in the partition) implies a dominating set in . It turns out that, the largest complete multipartite graph that can be obtained from by contractions is a complete graph.
Lemma 20
If the domatic number of is at least , then there is a partition of vertices of such that every set in induces a connected graph in and is a complete multipartite graph of at least vertices.
Proof
Assume that is a set of mutually disjoint dominating sets of . We define a partition of vertices of as follows:
- •
Let be the subset of vertices of corresponding to , i.e., . Form a set , for .
- •
Every vertex in forms a singleton set in .
It is straight-forward to verify that there are sets in , each set induces a connected graph in , and is a complete graph.
Let be a partition of vertices of such that each set in forms a connected subgraph and is a complete multipartite graph. Further assume that . Observations 3, 4, 5, and Lemma 21 are essentially taken from [10]. Observation 3 follows from the fact that cost(\mathcal{P})$$+|\mathcal{P}|=N.
Observation 3
cost .
Observation 4
For each , there is a set such that .
Proof
Assume that for every vertex in , the set in containing contains at least one vertex from outside . Then cost is at least as there are vertices in , which contradicts Observation 3.
Observations 5, 6, and Lemma 21 assumes that is a clique.
Observation 5
* cannot contain a singleton subset of .*
Proof
Since there are no universal vertices in , the degree of a vertex in is at most . Since is a clique of size at least , we obtain that cannot contain a singleton subset of .
Observation 6 follows from Observation 4 and the assumption that is a clique.
Observation 6
If contains a vertex in , then it contains a vertex from .
Lemma 21
* has a set of at least mutually disjoint dominating sets.*
Proof
By Observation 3, cost . Let be the subset of such that each set in contains at least one vertex from . By Observation 5, each set contains at least two vertices. Since is an independent set and induces a connected subgraph of , we obtain that is not a subset of . Therefore, cost(). This implies that contains exactly one vertex outside . By Observation 6, every set containing a vertex has at least one vertex from . As a set containing a vertex from contains exactly one vertex outside , we obtain that does not contain any vertex from . Now we obtain mutually disjoint dominating sets of as follows. For every , is the vertices in corresponding to , i.e., . It is sufficient to prove that is a dominating set of . Assume that a vertex is not dominated by in . By Observation 4, there is a set such that . Then, is not adjacent to that set, which is a contradiction.
Lemma 22
Assume that there is a partition of vertices of into sets inducing connected subgraphs such that is a complete multipartite graph on at least vertices. Then is a clique.
Proof
For a contradiction, assume that is not a clique. Let be two nonadjacent vertices of . Let be the sets in containing respectively.
Claim 1: cannot be a subset of .
Proof of Claim 1: Assume that . Clearly, cannot contain any vertex from . Therefore, . Without loss of generality, assume that contains at least one vertex, say , from . Then every other vertex in contributes towards the cost of . For example, if () is part of , then it contributes toward the cost of . If is not part of , then the set containing must contain some other vertices from outside to make sure that it is adjacent to (otherwise there is a induced by and the set containing ). Since cost is at most (Observation 3), this implies that sets in containing vertices from are the only sets containing at least two vertices and all other sets are singleton sets. Further, each set containing at least one vertex from contains exactly one vertex outside and that vertex must be from (to make sure that is adjacent to - otherwise there is a induced by , and ). This implies that every vertex in forms a singleton set in . Then any such set (recall that ) along with and induces a , which is a contradiction.
Claim 2: It cannot happen that contains a middle vertex and contains a lower vertex.
Proof of Claim 2: For a contradiction, assume that contains a middle vertex and contains a lower vertex . Then every other vertex in contributes toward the cost of (as obtained in the proof of Claim 1). Therefore, contributes toward the cost. This implies that the sets containing at least one vertex from are the only sets containing at least two vertices and all other sets in are singleton sets. Further, each such set can contain only one vertex outside - clearly, that extra vertex cannot be from . Then is a singleton set and every vertex in forms a singleton set in . Then there is a induced in by and one singleton set formed by a vertex in , which is a contradiction.
Claim 1 and Claim 2 implies that both and are subsets of . Since both and must induce connected subgraph of , we obtain that both and are singleton sets. Let . Let be nonadjacent to (this is guaranteed as there are no universal vertices in ). By Claim 2, a set containing at least one vertex from must be adjacent to . Then every vertex in contribute toward cost. This is a contradiction as and cost .
Lemmas 20, 21, 22 imply that there is a polynomial-time reduction from Domatic Number to -free Contraction.
Lemma 23
-free Contraction* is NP-complete.*
3.5 Putting them together
Recall that the reduction from Vertex Cover in Section 3.1 does not handle disconnected graph. This is the main ingredient that remains to be added to obtain the main result of the section. This turns out to be easy. Guo [15] has a reduction for transferring the hardness of -free Contraction to -free Contraction, where is any component of .
Proposition 2 ([15])
Let be a disconnected graph. Let be any component of it. Then there is a polynomial-time reduction from -free Contraction to -free Contraction.
Proposition 2 does not help us to prove the hardness when every component of is either a or a . But there are simple reductions to handle them.
Lemma 24
Let be a disconnected graph with an isolated vertex . There is a polynomial-time reduction from -free Contraction to -free Contraction.
Proof
It is straight-forward to verify that is a yes-instance of -free Contraction if and only if is a yes-instance of -free Contraction.
Repeated application of Lemma 24 implies that there is a polynomial-time reduction from -free Contraction to -free Contraction, for every . Then the NP-Completeness of -free Contraction (Proposition 1) implies Lemma 25.
Lemma 25
For every , -free Contraction is NP-complete.
Now, we handle the case when is a disjoint union of copies of .
Lemma 26
Let , for any integer and let be . There is a polynomial-time reduction from -free Contraction to -free Contraction.
Proof
It is straight-forward to verify that is a yes-instance of -free Contraction if and only if is a yes-instance of -free Contraction.
Repeated application of Lemma 26 and the NP-completeness of -free Contraction (Lemma 19) give us the following Lemma.
Lemma 27
For every , -free Contraction is NP-complete.
Now we are ready to prove the main result of this section.
Theorem 3.1
Let be any graph other than and . Then -free Contraction is NP-complete.
Proof
We prove this by induction on , the number of vertices of . The base cases are when and , i.e., when is or or triangle (Proposition 1), or (Lemma 25), or (Lemma 23). Assume that .
Let be a disconnected graph. Assume that has a component with at least three vertices. By Proposition 2, there is a polynomial-time reduction from -free Contraction to -free Contraction. Then we are done by induction hypothesis. Assume that every component of is either a or a . If has an isolated vertex, then we are done by Lemma 24. If there are no isolated vertex in , then is isomorphic to , for . Then we are done by Lemma 27.
Let be a connected graph. If is complete, then Proposition 1 is sufficient. Assume that is non-complete. Therefore, there is a non-universal vertex in . Assume that has neither a universal separator nor a universal separator. Then we are done by Lemma 4. Assume that has either a universal separator or a universal separator, denoted by . Further assume that is not a star. Let has at least two non-isomorphic components. Let be any component in with least number of vertices. Let be obtained from by removing all copies of in . Then by Lemma 7, there is a polynomial-time reduction from -free Contraction to -free Contraction, and we are done. Assume that is disjoint union of copies of a graph , for some . Then Lemma 13 gives us a polynomial-time reduction from -free Contraction to -free Contraction. For the last case, assume that is a star graph. Then we are done by Lemma 16.
4 W[2]-hardness
In this section, we prove that -free Contraction is W[2]-hard whenever is a tree, except for 7 trees. All our reductions are from Dominating Set, which is well-known to be a W[2]-hard problem. First we obtain a reduction for all trees which are neither stars nor bistars. Then we come up with a reduction for a subset of bistars, a corner case of the same proves the case of stars. Then we come up with a reduction which handles the remaining bistars.
Recall that a dominating set of a graph is a subset of vertices of such that every vertex of is either in or adjacent to a vertex in . The objective of the Dominating Set problem is to check whether a graph has a dominating set of size at most or not.
4.1 A general reduction for trees
Here we handle all trees which are neither stars nor bistars. The reduction that we use is an adapted version of a reduction used in [8] (to handle 3-connected graphs) and a reduction used in [20] (to handle cycles).
Construction 6
Let be graphs and be an integer. Let be the set of vertices of . We construct a graph from as follows.
- •
Introduce a clique .
- •
Introduce copies of denoted by . Let denote the set of vertices of , for .
- •
Let be any vertex in . Identify ’s of all copies of . Let the vertex obtained so be denoted by . Let the remaining vertices in each copy be denoted by , i.e., induces , for .
- •
Make adjacent to every vertex in .
- •
Make adjacent to every vertex in if and only if or is adjacent to in .
This completes the construction. An example is shown in Figure 6.
Let be an instance of Dominating Set. Let be a tree which is neither a star nor a bistar. We obtain a graph from by applying Construction 6.
Lemma 28
Let be a yes-instance of Dominating Set. Then is a yes-instance of -free Contraction.
Proof
Let be a dominating set of size at most of . Let . Clearly, . We claim that is -free. Let itself denote the vertex obtained by contracting the edges in . To get a contradiction, assume that there is an induced by a set of vertices of . We observe that is a universal vertex in , due to the fact that is a dominating set of . Since does not contain a universal vertex (a tree has a universal vertex if and only if it is a star), cannot be in . Since is triangle-free, can have at most two vertices from . Assume that has no vertex in . Then must be a subset of , which is a contradiction (observe that and are nonadjacent in for ). Assume that has exactly one vertex, say , from . Then the rest of the vertices are from s adjacent to . Recall that is adjacent to either all or none of vertices in . Since is triangle-free, the vertices in from s adjacent to form an independent set. Then is a star, which is a contradiction. If has exactly two vertices from , then with similar arguments, we obtain that is a bistar, which is a contradiction.
Lemma 29
Let be a yes-instance of -free Contraction. Then is a yes-instance of Dominating Set.
Proof
Let be a subset of edges of such that is -free and . We construct a new solution from as follows. If contains an edge touching , then we replace that edge with the edge in . Clearly . Since an edge touching kills only the induced by , which is killed by , we obtain that is -free. Let be the partition of vertices of corresponds to . Let be the set in containing . Let . Clearly, cost(\mathcal{P})$$|\leq k. We claim that is a dominating set of . Assume that there is a vertex in not dominated by . Then induces an in , which is a contradiction.
Lemmas 29 and 28 imply that there is a parameterized reduction from Dominating Set to -free Contraction.
Lemma 30
Let be a tree which is neither a star nor a bistar. Then -free Contraction is W[2]-hard.
4.2 Stars and Bistars
First we generalize a reduction given in [15] for -free Contraction. This generalized reduction covers all bistars such that and . As a boundary case (when ) we obtain hardness result for all stars of at least 5 vertices.
Construction 7
Let be a graph with a vertex set and let be integers. We construct a graph from as follows.
- •
Create two cliques .
- •
Make adjacent to if and only if or is an edge in .
- •
Create a vertex and two cliques and of vertices each.
- •
Make adjacent to all vertices of . Make a clique.
- •
Introduce cliques with vertices each and make them adjacent to . Let denote the set of these vertices.
- •
Introduce cliques with vertices each and make them adjacent to . Let denote the set of these vertices..
- •
Introduce cliques with vertices each and make them adjacent to . Let denote the set of these vertices.
- •
Introduce a clique of vertices and make it adjacent to . Let denote the set of these vertices.
- •
For every vertex in , attach degree-1 vertices.
Let be for and . Let be an instance of Dominating Set. We obtain from by Construction 7. An example is shown in Figure 7.
Lemma 31
Let be a yes-instance of Dominating Set. Then is a yes-instance of -free Contraction.
Proof
Let be a dominating set of size at most of . Let . Let itself denote the vertex obtained by contracting edges in . Since is a dominating set, is adjacent to every vertex of in . We claim that is -free. For a contradiction, assume that there is an induced in with the central edge being (we do not distinguish between and , i.e., we do not assume that is the vertex attached to degree-1 vertices and is the vertex attached to degree-1 vertices in the induced ). Let and . Then the largest bistar graph rooted at and is which does not have a as an induced subgraph. The case when and can be proved similarly. Let and . Then the largest bistar graph rooted at and is which does not have an induced . Let and . Then the largest bistar with the central edge is , which does not have a as an induced subgraph. For a pair of adjacent vertices in let denote the size of a maximum independent set in the graph induced by in . It is sufficient to prove that there exists no adjacent pair such that . Clearly, none of the degree-1 vertex can be in the pair we are looking for. The rest of the cases are handled below, each of them is straight-forward to verify.
- •
Let or or or . Then .
- •
Let or . Then .
- •
Let or . Then .
- •
Let and . Then .
- •
Let and . Then .
- •
Let and , or and , and . Then .
- •
Let and . Then .
By extending the definition of , we define as follows. Create a vertex and make it adjacent to copies of s. Attach degree-1 vertices to each of the vertices of one of such s. Let be known as the root of . An example is shown in Figure 8. The following observation is the bistar counterpart of Observation 2.
Observation 7
Let denote a copy of . Let has as an induced subgraph such that the degree-1 vertices in the are degree-1 vertices in . Then is a no-instance of -free Contraction.
Observation 8
Let be a subset of edges of such that and is -free. Then does not contain any edge from .
Proof
Assume that for some vertex . Then there is a in induced by , and the degree-1 vertices attached to the vertices of any one of the cliques formed by . Further, it satisfies the condition in Observation 8. Then we get a contradiction by Observation 7. By symmetry, we obtain that does not contain any edge from . Assume that , where . Then there is a induced by , and the degree-1 vertices attached to the vertices of . Further, it satisfies the condition in Observation 8. This results in a contradiction.
Observation 9
Let be a partition of vertices of such that is -free. Further assume that every set in is either a singleton set, or a subset of or a subset of . Let denote the set in containing . Then, for every set such that , at least one vertex of is adjacent to at least one vertex of .
Proof
For a contradiction, assume that none of the vertices in is adjacent to none of the vertices of . Then there is a induced by and singleton sets corresponding to a vertex in and vertices (one from each clique) corresponding to , and vertices attached to one selected vertex from . Therefore, is not -free, which is a contradiction.
Lemma 32
Let be a yes-instance of -free Contraction. Then is a yes-instance of Dominating Set.
Proof
Let be a minimal subset of edges of such that is -free and . By Observation 8, the edges from cannot be in . It can be easily verified that, due to the limited budget and the minimality of , does not contain any edges of other than those from . Let be the partition of vertices of corresponding to . Let denote the set containing . We create a dominating set of size at most of as follows. Include the vertices corresponding to in , i.e., include in . By Observation 9, for each such that , at least one vertex of is adjacent to at least one vertex of . Let is adjacent to a vertex in . Then for every vertex , include in . We claim that is a dominating set in . For a contradiction, assume that there is a vertex not adjacent to any vertex in . Let be the set containing . By the construction of , for every vertex in either or a vertex adjacent to is included in . This gives us a contradiction. Now, it is sufficient to prove that cost(\mathcal{P})$$\leq k. This follows from the fact that for we included vertices in and for every set such that , we included at most vertices in .
Lemmas 31 and 32 imply that there is a parameterized reduction from Dominating Set to -free Contraction.
Lemma 33
Let and . Then -free Contraction is W[2]-hard.
Now, we are left with the bistars where . For this, we come up with a reduction that handles more than this case. The reduction is a parameterized reduction but not a polynomial-time reduction, as the main structure, that we define next, used in the reduction has size exponential in . For integers , a -canopy is a graph obtained as follows. The vertices are arranged in levels - level 0 to level . The set of vertices in th level is denoted by . The set contains a single vertex, which is called the root. For , contains cliques, denoted by , each of size . The single vertex in is adjacent to all vertices in the clique . For , the edges between cliques in and cliques in are as follows. If is odd, then for every clique there are unique cliques in such that edges of a perfect matching is added between and each such clique, i.e., for every vertex in , there is a unique neighbor in each such clique . If is even, then every vertex in a clique is adjacent to all vertices of a unique clique in . An example is shown in Figure 9.
Construction 8
Let be a graph with vertices and let be integers. A graph is constructed from as follows.
- •
Introduce a set of cliques each of size . These cliques are denoted by , and the set of all these vertices is denoted by .
- •
Introduce a clique of vertices.
- •
Introduce a vertex and make it adjacent to .
- •
Introduce cliques each of size . These cliques are denoted by and the set of all these vertices is denoted by . Let the vertices in be .
- •
The vertices is adjacent to if and only if or and are adjacent in .
- •
For to , the edges of a perfect matching is added between and , i.e., each vertex in is made adjacent to a unique vertex in .
- •
For each vertex in , introduce a -canopy and identify with the root of the canopy.
An example of the construction is given in Figure 10.
Let be an instance of Dominating Set. Let be , where and . Let be constructed from by Construction 8.
Lemma 34
Let be a yes-instance of Dominating Set. Then is a yes-instance of -free Contraction.
Proof
Let be a dominating set of size at most of . Let . Clearly . We claim that is -free. Let itself denote the vertex obtained by contracting the edges in . Since is a dominating set of , is adjacent to all vertices of in . For a contradiction assume that there is an induced in with the central edge . If and , then the largest bistar where is the central edge is which does not have an induced . If and , then the largest bistar where is the central edge is which does not have an induced . For a pair of adjacent vertices in let denote the size of a maximum independent set in the graph induced by in . It is sufficient to prove that there exists no adjacent pair such that . The remaining cases are handled below, each of them is easy to verify.
- •
Let where is a clique in an odd-numbered level in any -canopy attached to the vertices of . Then .
- •
Let where is a clique in an th level, for any even in any -canopy attached to the vertices of . Then .
- •
Let , for . Then .
- •
Let . Then .
- •
Let . Then .
- •
Let and , where for a -canopy attached to a vertex in . Then .
- •
Let and . Then .
- •
Let . Then .
The following observation is straight-forward to verify.
Observation 10
Let be a partition of vertices of such that each set induces a connected graph and is -free and cost(\mathcal{P})$$\leq k. Then for every clique introduced while constructing , there is a subset of such that .
Observation 11
Let be a partition of vertices of such that each set induces a connected graph, is -free, and cost(\mathcal{P})$$\leq k. Further assume that is such a partition with least cost. Let be the set containing . Then does not contain any vertex in any of the -canopies attached to vertices in .
Proof
For a contradiction, assume that contains a -canopy vertex and let be the largest index such that contains a vertex in among all -canopies attached to vertices of . Since the budget is only , we obtain that . Let be any vertex in . Assume that is an even number. Then there is a induced by , sets in subsets of cliques in (Observation 10), a set corresponding to a clique in adjacent to , and sets corresponding to the vertices in adjacent to . Let be an odd number. Then there is a induced by , sets in subsets of cliques in (Observation 10), a set corresponding to the clique in is part of (such that ), and sets corresponding to the vertices in adjacent to .
Lemma 35
Let be a yes-instance of -free Contraction. Then is a yes-instance of Dominating Set.
Proof
Let be a partition of vertices of into connected subgraphs such that is -free and cost(\mathcal{P})$$\leq k. Further assume that is having the least cost. Let denote the set containing . By Observation 9, does not contain any vertex in the -canopies introduced during the construction of . Now, as we have done in the proof of Lemma 32, we can construct a dominating set of : Add the vertices in corresponding to in . For every such that , we know that at least one vertex of must be adjacent to . For all other vertices in , add in . It is straight-forward to verify that is a dominating set of size at most of .
Lemmas 34 and 35 imply that there is a parameterized reduction from Dominating Set to -free Contraction.
Lemma 36
Let . Then -free Contraction is W[2]-hard.
Now, Lemmas 30, 33, and 36 imply the main result of this section.
Theorem 4.1
Let be a tree which is neither a star of at most 4 vertices () nor a bistar in . Then -free Contraction is W[2]-hard.
We believe that our W[2]-hardness result on trees will be a stepping stone for an eventual parameterized complexity classification of -free Contraction. The most challenging hurdle for such a complete classification can be the graphs where each component is of at most 2 vertices, and the case of claw, the usual trouble-maker for other graph modification problems to -free graphs.
We conclude with some folklore observations. As noted in a version of [20], the property that “there exists at most edges contracting which
results in an -free graph” can be expressed in MSO1. The length of the corresponding MSO1 formula will be a function of . Then, there exists FPT algorithms for -free Contraction, whenever -free graphs have bounded rankwidth (See Chapter 7 of the textbook [9]).
This, in particular, implies that -free Contraction can be solved in FPT time.
It is known that every component of a paw-free graph is either triangle-free or complete multipartite [23], where
where paw is the graph
. Then the existance of FPT algorithms for -free Contraction and -free Contraction imply that there exists an FPT algorithm for -free Contraction.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] Aravind, N.R., Sandeep, R.B., Sivadasan, N.: Dichotomy results on the hardness of H-free edge modification problems. SIAM J. Discrete Math. 31 (1), 542–561 (2017). https://doi.org/10.1137/16M 1055797, https://doi.org/10.1137/16M 1055797
- 2[2] Asano, T., Hirata, T.: Edge-contraction problems. Journal of Computer and System Sciences 26 (2), 197–208 (1983)
- 3[3] Belmonte, R., Heggernes, P., van ’t Hof, P.: Edge contractions in subclasses of chordal graphs. Discret. Appl. Math. 160 (7-8), 999–1010 (2012). https://doi.org/10.1016/j.dam.2011.12.012, https://doi.org/10.1016/j.dam.2011.12.012
- 4[4] Bonuccelli, M.A.: Dominating sets and domatic number of circular arc graphs. Discret. Appl. Math. 12 (3), 203–213 (1985). https://doi.org/10.1016/0166-218X(85)90025-3, https://doi.org/10.1016/0166-218X(85)90025-3
- 5[5] Brouwer, A.E., Veldman, H.J.: Contractibility and np-completeness. Journal of Graph Theory 11 (1), 71–79 (1987)
- 6[6] Cai, L.: Fixed-parameter tractability of graph modification problems for hereditary properties. Inf. Process. Lett. 58 (4), 171–176 (1996). https://doi.org/10.1016/0020-0190(96)00050-6, https://doi.org/10.1016/0020-0190(96)00050-6
- 7[7] Cai, L., Cai, Y.: Incompressibility of H-free edge modification problems. Algorithmica 71 (3), 731–757 (2015). https://doi.org/10.1007/s 00453-014-9937-x
- 8[8] Cai, L., Guo, C.: Contracting few edges to remove forbidden induced subgraphs. In: Gutin, G.Z., Szeider, S. (eds.) Parameterized and Exact Computation - 8th International Symposium, IPEC 2013, Sophia Antipolis, France, September 4-6, 2013, Revised Selected Papers. Lecture Notes in Computer Science, vol. 8246, pp. 97–109. Springer (2013). https://doi.org/10.1007/978-3-319-03898-8_10, https://doi.org/10.1007/978-3-319-03898-8\_10
