This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Solving the rr-pseudoforest Deletion Problem in Time Independent of rr

Bin Sheng Email: shengbinhello@nuaa.edu.cn College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Collaborative Innovation Center of Novel Software Technology and Industrialization, Nanjing, Jiangsu, 211106, PR China
Abstract

The feedback vertex set problem is one of the most studied parameterized problems. Several generalizations of the problem have been studied where one is to delete vertices to obtain graphs close to acyclic. In this paper, we give an FPT algorithm for the problem of deleting at most kk vertices to get an rr-pseudoforest. A graph is an rr-pseudoforest if we can delete at most rr edges from each component to get a forest. Philip et al. introduced this problem and gave an O(crk)O^{*}(c_{r}^{k}) algorithm for it, where crc_{r} depends on rr double exponentially. In comparison, our algorithm runs in time O((10k)k)O^{*}((10k)^{k}), independent of rr.

1 Preliminary

The Feedback Vertex Set problem is one of the 21 problems proved to be NP-hard by Karp [11]. It asks to delete a minimum number of vertices to make a given graph into a forest. The problem has important applications in artificial intelligence [1, 2], bio-computing [3, 10], operating system research [16], and so on.

The Feedback Vertex Set problem has attracted a lot of attention from the parameterized complexity community. Both its undirected and directed versions are fixed-parameter tractable when parameterized by the solution size [6, 7, 8, 9]. For the undirected case, the state-of-the-art algorithm runs in time O(3.619k)O^{*}(3.619^{k}) in the deterministic setting [6], and O(2.7k)O^{*}(2.7^{k}) in the randomized setting [12]. Here the OO^{*} notation hides polynomial factors in nn, and kk is the solution size.

Relaxing the acyclic requirement, researchers have defined several classes of almost acyclic graphs. A graph FF is an rr-pseudoforest if we can delete at most rr edges from each component in FF to get a forest. A pseudoforest is a 1-pseudoforest. An almost rr-forest is a graph from which we can delete rr edges to get a forest.

Philip et al.[14] introduced the problem of deleting vertices to get an almost acyclic graph. There are several results in this line of research. In [14], the authors gave an O(crk)O^{*}(c_{r}^{k}) algorithm for rr-pseudoforest deletion, which asks to delete at most kk vertices to get an rr-pseudoforest. They also gave an O(7.56k)O^{*}(7.56^{k}) time algorithm for the problem of pseudoforest deletion. Bodlaender et al. [4] gave an improved algorithm for pseudoforest deletion running in time O(3knkO(1))O(3^{k}nk^{O(1)}).

Rai and Saurabh [15] gave an O(5.0024(k+r))O^{*}(5.0024^{(k+r)}) algorithm for the Almost Forest Deletion problem, which asks to delete at most kk vertices to get an almost rr-forest. Lin et al.[13] gave an improved algorithm for this problem that runs in time O(5k4r)O^{*}(5^{k}4^{r}).

To the author’s knowledge, there has been no follow-up research for the problem of rr-pseudoforest deletion. In this paper, we provide a simple branching algorithm that runs in time O((10k)k)O^{*}((10k)^{k}). The crc_{r} in the algorithm by Philip et al. [14] is a large constant that depends on rr double exponentially. In contrast, the running time of our algorithm is independent of rr, which is somewhat surprising.

2 Notation and Terminology

Here we give a brief list of the graph theory concepts used in this paper; for other notation and terminology, we refer readers to [5].

For a graph G=(V(G),E(G))G=(V(G),E(G)), V(G)V(G) and E(G)E(G) are its vertex set and edge set respectively. A non-empty graph GG is connected if there is a path between every pair of vertices. Otherwise, we call it disconnected.

The multiplicity of an edge is the appearance number of it in the multigraph. An edge uvuv is a loop if u=vu=v, and we call it a loop at uu. The degree of a vertex is the number of its appearances as end-vertex of some edge. Bypassing a vertex vv of degree 2 means to delete vv and add an edge between its two neighbors uu and ww(even if u=wu=w or there is already an edge between uu and ww). A forest is a graph in which there is no cycle and a tree is a connected forest. A graph H=(V(H),E(H))H=(V(H),E(H)) is a subgraph of a graph G=(V(G),E(G))G=(V(G),E(G)), if V(H)V(G)V(H)\subseteq V(G) and E(H)E(G)E(H)\subseteq E(G). A subgraph HH of GG is an induced subgraph of GG if for every u,vV(G)u,v\in V(G), edge uvE(H)uv\in E(H) if and only if uvE(G)uv\in E(G). For XV(G)X\subseteq V(G), G[X]G[X] denotes the subgraph of GG induced by XX.

For a positive integer rr, a graph is an r-pseudoforest if we can make each component into a forest by deleting at most rr edges.

3 FPT Algorithm for rr-Pseudoforest Deletion

In this section, we give an FPT algorithm for rr-pseudoforest deletion. Let us give the formal problem definition first.

rr-Pseudoforest Deletion

Instance: An undirected graph GG, two integers rr and kk.

Parameter: kk.

Output: Decide if there is a set XV(G)X\in V(G) with |X|k|X|\leq k such that GXG-X is an rr-pseudoforest.

Let (G,k)(G,k) be an instance of rr-pseudoforest deletion, where GG contains mm edges and nn vertices.

Now we give some reduction rules to simplify the given instance.

Reduction Rule 1: If there is a component CC in GG that is an rr-pseudoforest, then delete CC and get a new instance (GC,k)(G-C,k).

Reduction Rule 1 is safe since there is no need to delete any vertex from a component that is an rr-pseudoforest.

Reduction Rule 2: If there are at least r+1r+1 loops at a vertex vv, then delete vv and decrease kk by 1.

Reduction Rule 2 is safe since the graph G[v]G[v] is not an rr-pseudoforest and every solution must contain vv.

Reduction Rule 3: If GG contains an edge uvuv of multiplicity greater than r+2r+2, then reduce its multiplicity to r+2r+2.

Every solution intersects {u,v}\{u,v\} if edge uvuv has multiplicity at least r+2r+2. Reducing the multiplicity of uvuv to r+2r+2 does not affect the set of solutions. Thus, Reduction Rule 3 is safe.

Reduction Rule 4: If GG contains a vertex vv of degree at most 1, then delete vv and get a new instance (Gv,k)(G-v,k).

On the one hand, every solution of (G,k)(G,k) is a solution of (Gv,k)(G-v,k). On the other hand, attaching a leaf to any component of an rr-pseudoforest still gives an rr-pseudoforest. And so every solution of (Gv,k)(G-v,k) is also a solution of (G,k)(G,k) since dG(v)1d_{G}(v)\leq 1. Thus, Reduction Rule 4 is safe.

Reduction Rule 5: If GG contains a vertex vv of degree 2, then bypass vv.

Lemma 1.

Reduction Rule 5 is safe.

Proof.

Let HH be the graph obtained from GG by bypassing vv. Let NG(v)={u,w}N_{G}(v)=\{u,w\}. We show that (H,k)(H,k) is a yes-instance if and only if (G,k)(G,k) is a yes-instance.

On the one hand, assume (H,k)(H,k) is a yes-instance, and XX is a solution of it. Suppose X{u,w}X\cap\{u,w\}\neq\emptyset, then dGX(v)1d_{G-X}(v)\leq 1. By adding vv as an isolated vertex or a leaf to HXH-X, we get GXG-X. It follows that GXG-X is an rr-pseudoforest since HXH-X is an rr-pseudoforest. Thus, XX is also a solution of (G,k)(G,k). Otherwise, X{u,w}=X\cap\{u,w\}=\emptyset. In this case, the edge uwuw is in some component of the rr-pseudoforest HXH-X. By subdividing uwuw in HXH-X, we get GXG-X, which is still an rr-pseudoforest. And so XX is a solution of (G,k)(G,k). In both cases, (G,k)(G,k) is a yes-instance, and XX is a solution of it.

On the other hand, assume (G,k)(G,k) is a yes-instance, and XX is a solution of it. Suppose vXv\in X. Consider X=X{u}{v}X^{\prime}=X\cup\{u\}-\{v\}. Note that |X||X||X^{\prime}|\leq|X|. If uXu\in X, then HX=GXH-X^{\prime}=G-X. If uXu\not\in X, then HX=GX{u}H-X^{\prime}=G-X-\{u\}. In both cases, HXH-X^{\prime} is an rr-pseudoforest. Thus XX^{\prime} is a solution for (H,k)(H,k), and (H,k)(H,k) is a yes-instance.

Otherwise, vXv\not\in X. If dGX(v)=2d_{G-X}(v)=2, then we get HXH-X from GXG-X by bypassing vv. If dGX(v)=1d_{G-X}(v)=1, then we get HXH-X from GXG-X by deleting vv. In both cases, the number of edges and vertices decrease by the same amount, thus HXH-X is also an rr-pseudoforest. If dGX(v)=0d_{G-X}(v)=0, then we get HXH-X from GG by deleting the isolated vertex vv. Above all, we know that (H,X)(H,X) is a yes-instance, and XX is a solution of it. ∎

Reduction Rule 6: If k<0k<0, terminate and conclude that (G,k)(G,k) is a no-instance.

We call (G,k)(G,k) a reduced instance if none of Reduction Rules 1-6 applies to it. Observe that if (G,k)(G,k) is a reduced instance, then GG satisfies the following conditions:

  1. (P1)

    each edge has multiplicity at most r+2r+2;

  2. (P2)

    the minimum degree is at least 3;

  3. (P3)

    there are at most rr loops at any vertex.

Lemma 2.

If a connected graph GG is an rr-pseudoforest, then |E(G)||V(G)|+r1|E(G)|\leq|V(G)|+r-1.

Proof.

We can make each connected component of an rr-pseudoforest into a forest by deleting at most rr edges. Since GG is a connected rr-pseudoforest, |E(G)||V(G)|+r1|E(G)|\leq|V(G)|+r-1. ∎

Definition 1.

An rr-pseudoforest with one vertex and rr loops is called an rr-loop. An rr-pseudoforest with two vertices and r+1r+1 edges is called an (r+1)(r+1)-edge.

Lemma 3.

Let (G,k)(G,k) be a reduced instance of rr-pseudoforest deletion. Let XV(G)X\subseteq V(G) be a subset of vertices such that F=GXF=G-X is an rr-pseudoforest. Suppose FF contains t1t_{1} rr-loops, t1t^{\prime}_{1} components consists of one vertex and at least (r+2)/3(r+2)/3 loops, t2t_{2} (r+1)(r+1)-edges, t2t^{\prime}_{2} components consists of two vertices and at least 2(r+2)/32(r+2)/3 edges. If |V(F)|=t1+t1+2t2+2t2+s|V(F)|=t_{1}+t^{\prime}_{1}+2t_{2}+2t^{\prime}_{2}+s, then the following statements hold.

  1. 1.

    |E(F)|rt1+rt1+(r+1)t2+(r+1)t2+(r+2)s/3|E(F)|\leq rt_{1}+rt^{\prime}_{1}+(r+1)t_{2}+(r+1)t^{\prime}_{2}+(r+2)s/3.

  2. 2.

    2m3n+(2r+1)t1+(2(r+2)/3+1)t1+(r+2)t2+(2(r+2)/3+1)t22m\geq 3n+(2r+1)t_{1}+(2(r+2)/3+1)t^{\prime}_{1}+(r+2)t_{2}+(2(r+2)/3+1)t^{\prime}_{2}.

  3. 3.

    For any c>0c>0, vX(d(v)c)mcn[(rc)t1+(rc)t1+(r+12c)t2+(r+12c)t2+(r+23c)s/3]\sum_{v\in X}(d(v)-c)\geq m-cn-[(r-c)t_{1}+(r-c)t^{\prime}_{1}+(r+1-2c)t_{2}+(r+1-2c)t^{\prime}_{2}+(r+2-3c)s/3].

Proof.

Proof of statement 1: Because FF is an rr-pseudoforest, we know that |E(C)||V(C)|+r1|E(C)|\leq|V(C)|+r-1 for each component CC in FF. Since the ratio |V(C)|+r1|V(C)|\frac{|V(C)|+r-1}{|V(C)|} decreases when |V(C)||V(C)| increases, we have |E(C)||V(C)|r+23\frac{|E(C)|}{|V(C)|}\leq\frac{r+2}{3} for each component CC in FF with at least 3 vertices. It follows that |E(F)|rt1+rt1+(r+1)t2+(r+1)t2+(r+2)s/3|E(F)|\leq rt_{1}+rt^{\prime}_{1}+(r+1)t_{2}+(r+1)t^{\prime}_{2}+(r+2)s/3.

Proof of statement 2: Vertices in each (r+1)(r+1)-edge have average degree r+1r+1 and the vertex in an rr-loop has degree 2r2r. After exhaustive applications of Reduction Rule 1, there is no rr-pseudoforest component in GG. Note that there are t1t_{1} rr-loops, t1t^{\prime}_{1} components consists of one vertex and at least (r+2)/3(r+2)/3 loops, t2t_{2} (r+1)(r+1)-edges, t2t^{\prime}_{2} components consists of two vertices and at least 2(r+2)/32(r+2)/3 edges in FF. Thus GG contains at least t1t_{1} vertices of degree at least 2r+12r+1, t1t^{\prime}_{1} vertices of degree at least 2(r+2)/3+12(r+2)/3+1 and t2t_{2} vertices of degree at least r+2r+2, t2t^{\prime}_{2} vertices of degree at least 2(r+2)/3+12(r+2)/3+1. Combining with the fact that δ(G)3\delta(G)\geq 3, we have

2m=vV(G)d(v)3n+(2r+1)t1+(2(r+2)/3+1)t1+(r+2)t2+(2(r+2)/3+1)t2.2m=\sum_{v\in V(G)}d(v)\geq 3n+(2r+1)t_{1}+(2(r+2)/3+1)t^{\prime}_{1}+(r+2)t_{2}+(2(r+2)/3+1)t^{\prime}_{2}.

Proof of statement 3: Since every edge in E(G)E(F)E(G)\setminus E(F) is incident to at least one vertex in XX, we have |E(G)||E(F)|+vXd(v).|E(G)|\leq|E(F)|+\sum_{v\in X}d(v). It follows that

mrt1+rt1+(r+1)t2+(r+1)t2+(r+2)s/3+vXd(v).m\leq rt_{1}+rt^{\prime}_{1}+(r+1)t_{2}+(r+1)t^{\prime}_{2}+(r+2)s/3+\sum_{v\in X}d(v). (1)

Since |V(G)|=|V(F)|+|X||V(G)|=|V(F)|+|X|, denoting |X||X| by xx, we have

n=t1+t1+2t2+2t2+s+x.n=t_{1}+t^{\prime}_{1}+2t_{2}+2t^{\prime}_{2}+s+x. (2)

Combining (1) and (2), for any positive constant cc, we have vX(d(v)c)mcn[(rc)t1+(rc)t1+(r+12c)t2+(r+12c)t2+(r+23c)s/3].\sum_{v\in X}(d(v)-c)\geq m-cn-[(r-c)t_{1}+(r-c)t^{\prime}_{1}+(r+1-2c)t_{2}+(r+1-2c)t^{\prime}_{2}+(r+2-3c)s/3].

Let Φ=(v1,v2,,vn)\Phi=(v_{1},v_{2},\ldots,v_{n}) be an ordering of V(G)V(G) that satisfies d(v1)d(v2)d(vn)d(v_{1})\geq d(v_{2})\geq\dots\geq d(v_{n}). Let V10k={vi|1i10k}V_{10k}=\{v_{i}|1\leq i\leq 10k\} be the set of 10k10k vertices with the largest degrees. The following lemma shows that every solution of a reduced instance intersects V10kV_{10k}.

Lemma 4.

Let XX be any solution of rr-pseudoforest deletion for a reduced instance (G,k)(G,k), then XV10kX\cap V_{10k}\neq\emptyset.

Proof.

Suppose there is a solution XX of rr-pseudoforest deletion on (G,k)(G,k), which satisfies XV10k=X\cap V_{10k}=\emptyset and |X|k|X|\leq k. We show a contradiction by counting the number of edges in GG. Let F=GXF=G-X be the resulting rr-pseudoforest after deleting XX from GG. Suppose FF contains t1t_{1} rr-loops, t1t^{\prime}_{1} components consisting of one vertex and at least (r+2)/3(r+2)/3 loops, t2t_{2} (r+1)(r+1)-edges, and t2t^{\prime}_{2} components consisting of two vertices and at least 2(r+2)/32(r+2)/3 edges. Denote |V(F)|=t1+t1+2t2+2t2+s|V(F)|=t_{1}+t^{\prime}_{1}+2t_{2}+2t^{\prime}_{2}+s. In the following, we assume that n=t1+t1+2t2+2t2+s+x51kn=t_{1}+t^{\prime}_{1}+2t_{2}+2t^{\prime}_{2}+s+x\geq 51k, as otherwise, we may solve the problem via brute force. The assumption also implies that t1+t1+2t2+2t2+s50kt_{1}+t^{\prime}_{1}+2t_{2}+2t^{\prime}_{2}+s\geq 50k.

By the choice of V10kV_{10k}, the degree of each vertex in XX is at most d(v10k)d(v_{10k}). Since |X|k|X|\leq k, it follows that

i=110k(d(vi)c)10vX(d(v)c)\sum_{i=1}^{10k}(d(v_{i})-c)\geq 10\sum_{v\in X}(d(v)-c) (3)

By the definition of V10kV_{10k} and the assumption that XV10k=X\cap V_{10k}=\emptyset, we have X{v10k+1,v10k+2,,vn}X\subseteq\{v_{10k+1},v_{10k+2},...,v_{n}\} and so

i>10kn(d(vi)c)vX(d(v)c).\sum_{i>10k}^{n}(d(v_{i})-c)\geq\sum_{v\in X}(d(v)-c). (4)

Combining inequalities (3) and (4), we have

vV(G)(d(v)c)11vX(d(v)c).\sum_{v\in V(G)}(d(v)-c)\geq 11\sum_{v\in X}(d(v)-c). (5)

Since vV(G)d(v)=2m\sum_{v\in V(G)}d(v)=2m, it follows that

2mcn11vX(d(v)c).2m-cn\geq 11\sum_{v\in X}(d(v)-c). (6)

Thus 2mcn2m-cn\geq 11{mcn[(rc)t1+(rc)t1+(r+12c)t2+(r+12c)t2+(r+23c)s/3]}.11\{m-cn-[(r-c)t_{1}+(r-c)t^{\prime}_{1}+(r+1-2c)t_{2}+(r+1-2c)t^{\prime}_{2}+(r+2-3c)s/3]\}.

According to Statement 2 in Lemma 3, we have 10cn4.5[3n+(2r+1)t1+(2(r+2)/3+1)t1+(r+2)t2+(2(r+2)/3+1)t2]11[(rc)t1+(rc)t1+(r+12c)t2+(r+12c)t2+(r+23c)s/3].10cn\geq 4.5[3n+(2r+1)t_{1}+(2(r+2)/3+1)t^{\prime}_{1}+(r+2)t_{2}+(2(r+2)/3+1)t^{\prime}_{2}]-11[(r-c)t_{1}+(r-c)t^{\prime}_{1}+(r+1-2c)t_{2}+(r+1-2c)t^{\prime}_{2}+(r+2-3c)s/3].

Therefore, (10c13.5)n[4.5(2r+1)11(rc)]t1+[(3(r+2)+4.5)11(rc)]t1+[4.5(r+2)11(r+12c)]t2+[(3(r+2)+4.5)11(r+12c)]t211(r+23c)s/3.(10c-13.5)n\geq[4.5(2r+1)-11(r-c)]t_{1}+[(3(r+2)+4.5)-11(r-c)]t^{\prime}_{1}+[4.5(r+2)-11(r+1-2c)]t_{2}+[(3(r+2)+4.5)-11(r+1-2c)]t^{\prime}_{2}-11(r+2-3c)s/3.

Set c=10rc=10r and simplify the above inequality, we get (100r13.5)(t1+t1+2t2+2t2+s+x)(108r+4.5)t1+(102r+10.5)t1+(213.5r2)t2+(212r0.5)t2+11(29r2)s/3.(100r-13.5)(t_{1}+t^{\prime}_{1}+2t_{2}+2t^{\prime}_{2}+s+x)\geq(108r+4.5)t_{1}+(102r+10.5)t^{\prime}_{1}+(213.5r-2)t_{2}+(212r-0.5)t^{\prime}_{2}+11(29r-2)s/3.

It follows that (100r13.5)x(18+8r)t1+(2r+24)t1+(13.5r+25)t2+(12r+26.5)t2+(19r/3+37/6)s(8t1+2t1+13.5t2+12t2+19/3s)r2(nx)r2(511)kr100kr(100r-13.5)x\geq(18+8r)t_{1}+(2r+24)t^{\prime}_{1}+(13.5r+25)t_{2}+(12r+26.5)t^{\prime}_{2}+(19r/3+37/6)s\geq(8t_{1}+2t^{\prime}_{1}+13.5t_{2}+12t^{\prime}_{2}+19/3s)r\geq 2(n-x)r\geq 2(51-1)kr\geq 100kr, a contradiction to the fact that xkx\leq k.

The above analysis shows that the assumption XV10k=X\cap V_{10k}=\emptyset is not correct, thus every solution of (G,k)(G,k) must intersect with V10kV_{10k}. ∎

Lemma 4 enables us to design the following algorithm for rr-pseudoforest deletion.

Input: An undirected graph GG, positive integers rr and kk
Output: yes, if GG has an rr-pseudoforest deletion set of size at most kk; no, otherwise.
1 If GG is an rr-pseudoforest, then return yes. Else if k0k\leq 0, then return no.
2Exhaustively apply Reduction Rules 1-6 on (G,k)(G,k). Either get a reduced instance (G,k)(G^{\prime},k^{\prime}) or return no.
3If |V(G)|51k|V(G^{\prime})|\leq 51k, then apply a trivial branching algorithm to solve (G,k)(G^{\prime},k^{\prime}) in time 251knO(1){2}^{51k}n^{O(1)}.
4Otherwise, |V(G)|>51k|V(G^{\prime})|>51k. Order the vertices in V(G)V(G^{\prime}) in non-increasing order according to the vertex degree. Let V10kV^{\prime}_{10k} be the set of 10k10k vertices with the largest degrees. For each vertex uV10ku\in V^{\prime}_{10k}, solve (Gu,k1)(G^{\prime}-u,k-1) recursively.
If any branch gives a solution XuX^{\prime}_{u} of rr-pseudoforest deletion on (Gu,k1)(G^{\prime}-u,k-1), then Xu{u}X^{\prime}_{u}\cup\{u\} is a solution on (G,k)(G^{\prime},k). Otherwise, return no.
Algorithm 1 FPT algorithm for rr-pseudoforest deletion
Theorem 1.

The rr-pseudoforest deletion problem parameterized by the solution size kk can be solved in time O((10k)k)O^{*}((10k)^{k}).

Proof.

Let (G,k)(G,k) be a given instance of the rr-pseudoforest deletion problem, where GG is an undirected graph and kk is an integer. Each recursive call in Step 4 of Algorithm 1 decreases the parameter by 1, and thus the height of the search tree is at most kk^{\prime}. At each step, the problem branches into at most 10k10k^{\prime} subproblems. Hence, the number of leaves in the search tree is at most (10k)k(10k)k(10k^{\prime})^{k^{\prime}}\leq(10k)^{k}. It follows that Algorithm 1 solves the problem of rr-pseudoforest deletion and runs in time O((10k)k)O^{*}((10k)^{k}). ∎

4 Conclusion

In this paper, we design an FPT algorithm for the rr-pseudoforest deletion problem. The running time of our algorithm is independent of rr, and so our algorithm improves over the result in [14], when rr is large compared with kk.

5 Acknowledgements

This research is supported by the National Natural Science Foundation of China (No. 61802178).

References

  • [1] Reuven Bar-Yehuda, Dan Geiger, Joseph Naor, and Ron M. Roth. Approximation algorithms for the feedback vertex set problem with applications to constraint satisfaction and bayesian inference. SIAM J. Comput., 27(4):942–959, 1998.
  • [2] Ann Becker, Reuven Bar-Yehuda, and Dan Geiger. Randomized algorithms for the loop cutset problem. J. Artif. Intell. Res., 12:219–234, 2000.
  • [3] Guillaume Blin, Florian Sikora, and Stéphane Vialette. Querying graphs in protein-protein interactions networks using feedback vertex set. IEEE/ACM Trans. Comput. Biology Bioinform., 7(4):628–635, 2010.
  • [4] Hans L. Bodlaender, Hirotaka Ono, and Yota Otachi. A faster parameterized algorithm for pseudoforest deletion. Discrete Applied Mathematics, 236:42–56, 2018.
  • [5] Béla Bollobás. Modern Graph Theory, volume 184 of Graduate Texts in Mathematics. Springer, 2002.
  • [6] Yixin Cao, Jianer Chen, and Yang Liu. On feedback vertex set: New measure and new structures. Algorithmica, 73(1):63–86, 2015.
  • [7] Jianer Chen, Fedor V. Fomin, Yang Liu, Songjian Lu, and Yngve Villanger. Improved algorithms for feedback vertex set problems. J. Comput. Syst. Sci., 74(7):1188–1198, 2008.
  • [8] Rajesh Hemant Chitnis, Marek Cygan, Mohammad Taghi Hajiaghayi, and Dániel Marx. Directed subset feedback vertex set is fixed-parameter tractable. ACM Trans. Algorithms, 11(4):28:1–28:28, 2015.
  • [9] Marek Cygan, Marcin Pilipczuk, Michal Pilipczuk, and Jakub Onufry Wojtaszczyk. Subset feedback vertex set is fixed-parameter tractable. SIAM J. Discrete Math., 27(1):290–309, 2013.
  • [10] Zheng Fu and Tao Jiang. Computing the breakpoint distance between partially ordered genomes. J. Bioinformatics and Computational Biology, 5(5):1087–1101, 2007.
  • [11] Richard M. Karp. Reducibility among combinatorial problems. In Proceedings of a symposium on the Complexity of Computer Computations, held March 20-22, 1972, at the IBM Thomas J. Watson Research Center, Yorktown Heights, New York, USA, pages 85–103, 1972.
  • [12] Jason Li and Jesper Nederlof. Detecting feedback vertex sets of size k in O*(2.7k{}^{\mbox{\emph{k}}}) time. In Shuchi Chawla, editor, Proceedings of the 2020 ACM-SIAM Symposium on Discrete Algorithms, SODA 2020, Salt Lake City, UT, USA, January 5-8, 2020, pages 971–989. SIAM, 2020.
  • [13] Mugang Lin, Qilong Feng, Jianxin Wang, Jianer Chen, Bin Fu, and Wenjun Li. An improved FPT algorithm for almost forest deletion problem. Inf. Process. Lett., 136:30–36, 2018.
  • [14] Geevarghese Philip, Ashutosh Rai, and Saket Saurabh. Generalized pseudoforest deletion: Algorithms and uniform kernel. SIAM J. Discrete Math., 32(2):882–901, 2018.
  • [15] Ashutosh Rai and Saket Saurabh. Bivariate complexity analysis of almost forest deletion. Theor. Comput. Sci., 708:18–33, 2018.
  • [16] Abraham Silberschatz and Peter Galvin. Operating System Concepts, 4th edition. Addison-Wesley, 1994.