    # On the Approximability and Hardness of the Minimum Connected Dominating Set with Routing Cost Constraint

In the problem of minimum connected dominating set with routing cost constraint, we are given a graph G=(V,E), and the goal is to find the smallest connected dominating set D of G such that, for any two non-adjacent vertices u and v in G, the cost of routing between u and v through D (the number of internal nodes on the shortest path between u and v in G[D ∪{u,v}]) is at most α times that through V. For general graphs, the only known previous approximability result is an O( n)-approximation algorithm (n=|V|) for the case α = 1 by Ding et al. For 1 < α < 5, no non-trivial approximation algorithm was previously known even on special graphs like unit disk graphs. When α > 1, we give an O(n^1-1/α( n)^1/α)-approximation algorithm. When α≥ 5, we give an O(√(n) n)-approximation algorithm. Finally, we prove that, when α =2, unless NP ⊆ DTIME(n^poly n), the problem admits no polynomial-time 2^^1-ϵn-approximation algorithm, for any constant ϵ > 0, improving upon the Ω( n) bound by Du et al. (albeit under a stronger hardness assumption).

## Authors

05/28/2020

### Liar's Domination in Unit Disk Graphs

In this article, we study a variant of the minimum dominating set proble...
12/20/2019

### An Õ(log^2 n)-approximation algorithm for 2-edge-connected dominating set

In the Connected Dominating Set problem we are given a graph G=(V,E) and...
07/21/2020

### A 3/2-Approximation for the Metric Many-visits Path TSP

In the Many-visits Path TSP, we are given a set of n cities along with t...
02/22/2018

### The Clever Shopper Problem

We investigate a variant of the so-called "Internet Shopping Problem" in...
04/19/2021

### It's Good to Relax: Fast Profit Approximation for Virtual Networks with Latency Constraints

This paper proposes a new approximation algorithm for the offline Virtua...
06/25/2018

### Maximum Rooted Connected Expansion

Prefetching constitutes a valuable tool toward efficient Web surfing. As...
01/26/2019

### Star Routing: Between Vehicle Routing and Vertex Cover

We consider an optimization problem posed by an actual newspaper company...
##### This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

## 1 Introduction

### 1.1 Motivation

In wireless network routing, a common approach is to select a set of nodes as the virtual backbone. The virtual backbone is responsible for relaying packets. Specifically, when a node generates a packet destined to , the packet is routed through path , where every internal node belongs to the virtual backbone. To realize this idea, we can model the wireless network as a graph , where is the set of nodes in the wireless network, and if and only if and can communicate with each other directly. Thus, a connected dominating set of is a virtual backbone for the wireless network.111A set is a dominating set of if every vertex in is adjacent to . Furthermore, if induces a connected subgraph of , then is called a connected dominating set of . One of the concerns in constructing the virtual backbone is the routing cost. Specifically, the routing cost of sending a packet from the source to the destination is the number of internal nodes (relays) in the routing path from to . For example, the routing cost is if the routing path is . The routing cost should not be too high even if packets are only allowed to be routed through the virtual backbone. Next, we give the formal definition of the problem.

### 1.2 Problem Definition

Let be the subgraph of induced by . Let be the number of internal vertices on the shortest path between and in . For example, if and are adjacent, then . If and are not adjacent and have a common neighbor, then . Furthermore, given a vertex subset of , is defined as , i.e., the number of internal vertices on the shortest path between and through . We use to denote the number of vertices in graph . When the graph we are referring to is clear from the context, we simply write , , and instead of , , and , respectively.

###### Definition 1.

Given a connected graph and a positive integer , the Connected Dominating set problem with Routing cost constraint (CDR-) asks for the smallest connected dominating set of , such that, for every two vertices and , if and are not adjacent in , then .

### 1.3 Preliminary

#### 1.3.1 An Equivalent Problem

In the CDR- problem, we need to consider all the pairs of non-adjacent nodes. Ding et al. discovered that to solve the CDR- problem, it suffices to consider only vertex pairs such that , i.e., and are not adjacent but have a common neighbor . We call the corresponding problem the 1-DR- problem.

###### Definition 2.

Given a connected graph and a positive integer , the 1-DR- problem asks for the smallest dominating set of , such that, for every two vertices and , if , then .

We say that and form a target couple, denoted by , if . We say that a set covers a target couple if . Hence, the 1-DR- problem asks for the smallest dominating set that covers all the target couples. Notice that any feasible solution of the 1-DR- problem must induce a connected subgraph of . The equivalence between the CDR- problem and the 1-DR- problem is stated in the following theorem.

###### Theorem 1 (Ding et al.).

is a feasible solution of the CDR- problem with input graph if and only if is a feasible solution of the 1-DR- problem with input graph .

###### Corollary 1.

Any -approximation algorithm of the 1-DR- problem is an -approximation algorithm of the CDR- problem.

In this paper, we thus focus on the 1-DR- problem.

#### 1.3.2 Feasibility of the 1-DR-α Problem for α≥5

Next, we give the basic idea of finding a feasible solution of the 1-DR- problem for used in previous researches, e.g., in . One of our algorithms still uses this idea. First, find a dominating set . Thus, for any target couple , there exist and in , such that and dominate and , respectively.222 dominates if or and are adjacent. Let . For any two vertices and in , if , then we add the internal vertices of the shortest path between and on to . Observe that . Hence, and is a feasible solution of the 1-DR- problem for .

###### Lemma 1.

Let be a dominating set of . Let be a vertex subset of such that, for any two vertices and in , if , then . Then, is a feasible solution of the 1-DR- problem with input and .

### 1.4 Previous Result

##### Previous result on general graphs

When , the 1-DR- problem can be transformed to the set cover problem, i.e., cover all the vertices (to form a dominating set) and cover all the target couples. Observe that each target couple can be covered by a single vertex. The resulting approximation ratio is  . When is sufficiently large, e.g., , any connected dominating set is feasible for the CDR- problem. Note that, for any , the size of the minimum connected dominating set is a lower bound of the CDR- problem. Since the connected dominating set can be approximated within a factor of  [12, 21], the CDR- problem can be approximated within a factor of . If falls between these two extremes, e.g., , the only known previous result is the trivial -approximation algorithm. On the hardness side, it has been proved that, unless , there is no polynomial-time algorithm that can approximate the CDR- problem within a factor of () for   and  [8, 10], where is the maximum degree of .

###### Open Question 1 (Du and Wan ).

Is there a polynomial-time -approximation algorithm for the CDR- problem for ?

##### Previous result on Unit Disk Graph (UDG)

Most of the studies on the CDR- problem focused on UDG [5, 8, 10, 11, 19]. UDG exhibits many nice properties that enable constant factor approximation algorithms (or PTAS) in many problems where only -approximation algorithms (or worse) are known in general graphs, e.g., the minimum (connected) dominating set problem and the maximum independent set problem [3, 4, 20]. All the previous research on the CDR- problem on UDG leveraged constant bounds of the maximum independent set or the minimum dominating set. However, all the previous research only solved the case where (by Lemma 1), and the best result so far is a PTAS by Du et al. . When , the only known previous result is the trivial -approximation algorithm.

### 1.5 Our Result and Basic Ideas

In this paper, we first give an approximation algorithm of the 1-DR- problem on general graphs for any constant . A critical observation is that the 1-DR- problem is a special case of the Set Cover with Pairs (SCP) problem . Hassin and Segev proposed an -approximation algorithm for the SCP problem, where is the number of targets to be covered. However, since there are target couples to be covered, directly applying the -approximation bound yields a trivial upper bound for the 1-DR-2 problem. We re-examine the analysis in  and find that, when applying the algorithm to the 1-DR-2 problem, the approximation ratio can also be expressed as . Nevertheless, in this paper, we give a slightly simplified algorithm with an easier analysis for the SCP problem. The algorithm and analysis also make it easy to solve the generalized SCP problem. We obtain the following result, which is the first non-trivial result of the CDR- problem for on general graphs and for on UDG.

###### Theorem 2.

For any constant , there is an -approximation algorithm for the 1-DR- problem.

Apparently, the above performance guarantee deteriorates quickly as increases. In our second algorithm, we apply the aforementioned idea of finding a feasible solution when , i.e., Lemma 1. We have the following result.

###### Theorem 3.

When , there is an -approximation algorithm for the 1-DR- problem.

Finally, we answer Open Question 1 negatively. We improve upon the hardness result for the 1-DR-2 problem (albeit under a stronger hardness assumption) [8, 10]. In this paper, we give a reduction from the MIN-REP problem .

###### Theorem 4.

Unless , for any constant , the 1-DR-2 problem admits no polynomial-time -approximation algorithm, even if the graph is triangle-free333If the graph is triangle-free, then any two vertices with a common neighbor form a target couple. or the constraint that the feasible solution must be a dominating set is ignored444One may drop the constraint that the solution must be a dominating set, and focuses on minimizing the number of vertices to cover all the target couples. This theorem also applies to such a problem..

### 1.6 Relation with the Basic k-Spanner Problem

When we ignore the constraint that any feasible solution must be a connected dominating set, the CDR- problem is similar to the basic -spanner problem. For completeness, we give the formal definition of the basic -spanner problem. Given a graph , a -spanner of is a subgraph of such that for all and in , where is the number of edges in the shortest path between and in . The basic -spanner problem asks for the -spanner that has the fewest edges. The CDR- problem differs with the basic -spanner problem in the following three aspects: First, in the CDR- problem, we find a set of vertices , and all the edges in the subgraph induced by can be used for routing; while in the basic -spanner problem, only edges in can be used. Second, in the CDR- problem, the objective is to minimize the number of chosen vertices; while in the basic -spanner problem, the objective is to minimize the number of chosen edges. Finally, in the basic -spanner problem, the distance is measured by the number of edges; while in the CDR- problem, the distance is measured by the number of internal nodes. Despite the above differences, these two problems share similar approximability and hardness results. Althöfer et al. proved that every graph has a -spanner of at most edges, and such a -spanner can be constructed in polynomial time [1, 7]. Since the number of edges in any -spanner is at least , this yields an -approximation algorithm for the basic -spanner problem. For , there is an -approximation algorithm due to Kortsarz and Peleg , and this is the best possible . For , Berman et al. proposed an -approximation algorithm . For , Dinitz and Zhang proposed an -approximation algorithm . On the hardness side, it has been proved that for any constant and for , unless , there is no polynomial-time algorithm that approximates the basic -spanner problem to a factor better than  .

## 2 Two Algorithms for the 1-DR-α Problem

### 2.1 The First Algorithm

We first give the formal definition of the Set Cover with Pairs (SCP) problem.

###### Definition 3.

Let be a set of targets. Let be a set of elements. For every pair of elements , denotes the set of targets covered by . The Set Cover with Pairs (SCP) problem asks for the smallest subset of such that .

Let be the number of elements in the optimal solution. We only need to consider the case where and .

#### 2.1.1 Approximating the SCP Problem

Our algorithm is a simple greedy algorithm: in each round, we choose at most two elements and that maximize the number of covered targets. Specifically, is an empty set initially. In each round, we select a set such that and increases the number of covered targets the most, i.e., , where

 g(P′)=|⋃{v1,v2}⊆S∪PC({v1,v2})|−|⋃{v1,v2}⊆SC({v1,v2})|.

We then add to and repeat the above process until all the targets are covered. The algorithm terminates once all targets are covered.555In , in each round, a set is added to , where .

###### Theorem 5.

The above algorithm is an -approximation algorithm for the SCP problem.

###### Proof.

Let be the number of uncovered targets after round . In the first round, some pair of elements in the optimal solution can cover at least targets. Since we choose a pair of elements greedily in each round, . In the second round, there exists a pair of elements in the optimal solution that can cover at least targets among the uncovered targets. Again, we choose the pair of elements that increases the number of covered targets the most. Hence, . In general, . After rounds, the number of uncovered targets is at most . Hence, after rounds, all targets are covered. Let be the number of elements chosen by the algorithm. Since we choose at most two elements in each round, . Finally, since , . ∎

Note that, in Theorem 5, we can replace with any upper bound of the size of solutions obtained by any polynomial-time algorithm for the SCP problem. This is achieved by executing both and our algorithm. Choosing the best between the two outputs yields the desired approximation ratio. An example is replacing with .

#### 2.1.2 Approximating the 1-DR-2 Problem

To transform the 1-DR-2 problem to the SCP problem, we treat each target couple as a target. Moreover, we treat each vertex as a target so that the output is a dominating set. The set of elements in the SCP problem is the vertex set of . consists of all the vertices that are dominated by in and all the target couples that are covered by in . In this SCP instance, and . It is easy to verify the following result.

###### Theorem 6.

There is an -approximation algorithm for the 1-DR-2 problem.

#### 2.1.3 The Set Cover with α-Tuples (SCT-α) Problem

In the 1-DR-2 problem, every target couple can be covered by no more than two vertices. In the 1-DR- problem, every target couple can be covered by no more than vertices. Hence, we consider the following generalization of the SCP problem.

###### Definition 4.

Let be a set of targets. Let be a set of elements. Let be a positive integer constant greater than one. For every -tuple , denotes the set of targets covered by . The Set Cover with -Tuples (SCT-) problem asks for the smallest subset of such that .

We only need to consider the case where and ( is a constant).

#### 2.1.4 Approximating the SCT-α Problem and the 1-DR-α Problem

The algorithm for the SCT- problem is a straightforward generalization of the algorithm for the SCP problem. The difference is that, in each round, we choose a set of at most elements that increases the number of covered targets the most. The transformation from the 1-DR- problem to the SCT- problem is also similar to the previous transformation. The value of in the constructed SCT- instance is equal to that in the 1-DR- instance. Again, and in the constructed SCT- instance. Theorem 2 is a direct result of the following theorem.

###### Theorem 7.

There is an -approximation algorithm for the SCT- problem.

We have the following claim, whose proof can be found in the appendix.

###### Claim 1.

When , .

Proof of Theorem 7: Let be the number of uncovered targets after round . By a similar argument in the proof of Theorem 5, we get that . After rounds, the number of uncovered targets is at most one. Hence, after rounds, all targets are covered. Let be the number of elements chosen by the algorithm. Since we choose at most elements in each round, . Since , .

Let . When , the approximation ratio is . When , . The proof then follows from Claim 1 and . ∎

### 2.2 The Second Algorithm

The second algorithm is designed for the 1-DR- problem when . It has a better approximation ratio than that of the previous algorithm when . The algorithm is suggested in Lemma 1: We first find a dominating set by any -approximation algorithm. Let . For any two vertices and in , if , we then add at most three vertices to so that .

Proof of Theorem 3: Let be the size of the minimum dominating set in . Let be the size of the optimum of the 1-DR- problem. Since any feasible solution of the 1-DR- problem must be a dominating set, . . Since , we have . ∎

## 3 Inapproximability Result

### 3.1 The MIN-REP Problem

We prove Theorem 4 by a reduction from the MIN-REP problem . The input of the MIN-REP problem consists of a bipartite graph , a partition of , , and a partition of , , such that and . Every (respectively, ) has size (respectively, ). and are called super nodes, and two super nodes and are adjacent if some vertex in and some vertex in are adjacent in . If and are adjacent, then and form a super edge. In the MIN-REP problem, our task is to choose representatives for super nodes so that if and form a super edge, then some representative for and some representative for are adjacent in . Note that a super node may have multiple representatives. Specifically, the goal of the MIN-REP problem is to find the smallest subset such that if and form a super edge, then must contain two vertices and such that , and . In this case, we say that covers the super edge . The inapproximability result of the MIN-REP problem is stated as the following theorem.

###### Theorem 8 (Kortsarz et al.).

For any constant , unless , there is no polynomial-time algorithm that can distinguish between instances of the MIN-REP problem with a solution of size and instances where every solution is of size at least , where is the number of vertices in the input graph of the MIN-REP problem.

### 3.2 The Reduction

Given inputs , , and of the MIN-REP problem, we construct a corresponding graph of the 1-DR-2 problem. When , , and are clear from the context, we simply write instead of . Initially, . Hence, contains , , and . For each super node (respectively, ), we create two corresponding vertices and (respectively, and ) in . If is in super node (respectively, is in super node ), then we add two edges and (respectively, and ) in . If and form a super edge, then we add two vertices and to , and we add four edges , , , to . (respectively, ) is called the relay of and (respectively, and ).

Before we complete the construction of , we briefly explain the idea behind the construction so far. If two super nodes and form a super edge, then and () have a common neighbor in , i.e., the relay . Because and are not adjacent, and form a target couple. To transform a solution of the 1-DR-2 problem to a solution of the MIN-REP problem, we need to transform to another feasible solution for the 1-DR-2 problem so that none of the relays is chosen, and only vertices in are used to connect and . This is the reason that we have two corresponding vertices for each super node (and thus two relays for each super edge). Under this setting, to connect to and to , choosing two vertices in is no worse than choosing the relays. Figure 1: An example of the reduction from the MIN-REP problem to the 1-DR-2 problem.

Let be the set of vertices in corresponding to the super nodes in . Similarly, let . Let be the set of all relays. To complete the construction, we add four vertices (hubs) , , , and to . In , all the vertices in , , , and are adjacent to , , , and , respectively. Moreover, every relay is adjacent to and . These four hubs induce a 4-cycle in . Finally, for each hub , we create two dummy nodes and , and add two edges and to . This completes the construction of . Fig. 1 shows an example of the reduction. Let and be the set of hubs and the set of dummy nodes, respectively. Hence, the vertex set of is . Let be the set of neighbors of in . We then have

 N(px)⊆X∪R∪{hPX} if px∈PX. N(py)⊆Y∪R∪{hPY} if py∈PY. N(x)⊆PX∪Y∪{hX,R} if x∈X. N(y)⊆PY∪X∪{hY,R} if y∈Y. N(hX,R)∖M=X∪R∪{hPX,hPY}. N(hY,R)∖M=Y∪R∪{hPX,hPY}. N(hPX)∖M=PX∪{hX,R,hY,R}. N(hPY)∖M=PY∪{hX,R,hY,R}. N(m)⊆H if m∈M. N(r)⊆PX∪PY∪{hX,R,hY,R} if r∈R.

Observe that . We have the following lemma.

###### Lemma 2.

.

It is easy to check that, for any two adjacent vertices and in , and have no common neighbor. Hence, we have the following lemma.

###### Lemma 3.

is triangle-free.

We say that a target couple is in if and . It is easy to verify the following two lemmas.

###### Lemma 4.

Only can cover the target couples in .

###### Lemma 5.

is a dominating set of .

The proof of the following lemma can be found in the appendix.

###### Lemma 6.

covers all the target couples except those in .

Let and be vertices in and , respectively. Observe that, if is a path in , then and . We then have the following lemma.

###### Lemma 7.

covers target couples and if and only if at least one of the following conditions is satisfied.

1. There exist and such that and are paths in and .

2. .

### 3.3 The Analysis

Let be an instance of the MIN-REP problem with inputs , , and . Let be the instance of the 1-DR-2 problem with input . To prove the inapproximability result, we use the following two lemmas.

###### Lemma 8.

If has a solution of size , then has a solution of size .

###### Lemma 9.

If every solution of has size at least , then every solution of has size at least .

Proof of Theorem 4: By Theorem 8, for any constant , unless , there is no polynomial-time algorithm that can distinguish between instances of the MIN-REP problem with a solution of size and instances where every solution is of size at least . By the above two lemmas, it is hard to distinguish between instances of the 1-DR-2 problem with a solution of size and instances in which every solution is of size at least . Therefore, for any constant , unless , there is no polynomial-time algorithm that can approximate the 1-DR-2 problem by a factor better than . Lemma 2 implies that, for any constant , unless , there is no -approximation algorithm for the 1-DR-2 problem. By considering sufficiently large instances and a small enough , we have the hardness result claimed in Theorem 4. On the other hand, let 1-DR- be the problem obtained by removing the constraint that any feasible solution must be a dominating set from the 1-DR-2 problem. Thus, in the 1-DR- problem, we only focus on covering target couples. By Lemmas 4 and 5, a solution is feasible for the 1-DR- problem with input if and only if is a feasible solution of . Thus, the inapproximability result also applies to the 1-DR- problem. Finally, the proof follows from Lemma 3. ∎

Lemma 8 is a direct result of the following claim.

###### Claim 2.

If is a feasible solution of , then is a feasible solution of .

###### Proof.

Since is a dominating set, by Lemma 6, it suffices to prove that every target couple in is covered by . Note that cannot be a target couple if . This is because and do not have a common neighbor if . If , then the common neighbor must be . By the construction of , this implies that and form a super edge. Since is a feasible solution of , there exists and such that and are adjacent in and . Again, by the construction of , is a path in . Hence, covers . ∎

To prove Lemma 9, we use the following claim.

###### Claim 3.

has an optimal solution , such that is a feasible solution of .

Proof of Lemma 9: Let be the optimal solution of . By the assumption, we have . It suffices to prove that is an optimal solution for , which implies that every feasible solution of has size at least . The feasibility of follows from Claim 2. For the sake of contradiction, assume that the optimal solution of has size smaller than . Claim 3 and Lemma 4 then imply that is not an optimal solution of , which is a contradiction. ∎

Proof of Claim 3: Let be any optimal solution of . By Lemmas 4, 6, and 7, . If , by Lemma 7, each target couple is covered by some and some . By the construction of , such and also cover the super edge in . Because each super edge in has a corresponding target couple in , is a feasible solution of .

If , then some . We can further assume that both and are in ; otherwise, by Lemma 7, we can remove from , the resulting solution is smaller and is still feasible. We then replace and with some and some satisfying the first condition in Lemma 7. By Lemma 7, the resulting solution is still feasible, and the size remains the same. Repeat the above replacing process until the resulting solution does not contain any relay. The proof then follows from the argument of the case where . ∎

## 4 Transforming the 1-DR-α Problem to Other Related Problems

Submodular Cost Set Cover Problem: The 1-DR- problem can also be considered as a special case of the submodular cost set cover problem [9, 14, 23]. In the set cover problem, we are given a set of targets and a set of objects . Each object in can cover a subset of (specified in the input). The goal is to choose the smallest subset of that covers . In the submodular cost set cover problem, there is a non-negative submodular function that maps each subset of to a cost, and the goal is to find the set cover with the minimum cost. To transform the 1-DR- problem with input to the submodular cost set cover problem, let be the union of and the set of all target couples, and let be the set of all subsets of with size at most . Hence, each object in is a subset of . An object can cover a vertex if is adjacent to some vertex in or . An object can cover a target couple if . The cost of a subset of is simply the size of the union of objects in , i.e., the number of distinct vertices specified in .

Iwata and Nagano proposed a -approximation algorithm and an -approximation algorithm, where is the maximum frequency,  . Koufogiannakis and Young also proposed an -approximation algorithm when the cost function is non-decreasing . It is easy to see that these algorithms give trivial bounds for the 1-DR- problem. When the cost function is integer-valued, non-decreasing, and satisfies , Wan et al. proposed a -approximation algorithm, where , is the largest number of targets that can be covered by an object in , and is the -th Harmonic number . Du et al. applied this algorithm to the 1-DR- problem on UDG for and obtained a constant factor approximation algorithm . It is unclear whether or not can be upper bounded by for some when applied to the 1-DR- problem on general graphs.

Minimum Rainbow Subgraph Problem on Multigraphs: Given a set of colors and a multigraph , where each edge is colored with one of the colors, the Minimum Rainbow Subgraph (MRS) problem asks for the smallest vertex subset of , such that each of the colors appears in some edge induced by . The 1-DR-2 problem can be transformed to the MRS problem as follows. Let be the input graph of the 1-DR-2 problem. Let be the union of and the set of all target couples. The set of colors for the MRS problem is . The input multigraph of the MRS problem has the same vertex set as . To form a dominating set, for each , is incident to loops in , where is the degree of in . Each of these loops receives a different color in . For each target couple in , if is a common neighbor of and in , we add a loop with color to . Finally, for each target couple in , if is a path in , we add an edge with color to . The MRS problem can be transformed to the SCP problem. When the input graph is simple, Tirodkar and Vishwanathan proposed an -approximation algorithm .

## References

•  I. Althöfer, G. Das, D. Dobkin, D. Joseph, and J. Soares. On sparse spanners of weighted graphs. Discrete & Computational Geometry, 9(1):81–100, Jan 1993.
•  P. Berman, A. Bhattacharyya, K. Makarychev, S. Raskhodnikova, and G. Yaroslavtsev. Approximation algorithms for spanner problems and directed steiner forest. Information and Computation, 222(Supplement C):93 – 107, 2013.
•  X. Cheng, X. Huang, D. Li, W. Wu, and D.-Z. Du. A polynomial-time approximation scheme for the minimum-connected dominating set in ad hoc wireless networks. Networks, 42(4):202–208, 2003.
•  G. K. Das, M. De, S. Kolay, S. C. Nandy, and S. Sur-Kolay. Approximation algorithms for maximum independent set of a unit disk graph. Information Processing Letters, 115(3):439 – 446, 2015.
•  L. Ding, W. Wu, J. Willson, H. Du, W. Lee, and D. Z. Du. Efficient algorithms for topology control problem with routing cost constraints in wireless networks. IEEE Transactions on Parallel and Distributed Systems, 22(10):1601–1609, Oct 2011.
•  M. Dinitz, G. Kortsarz, and R. Raz. Label cover instances with large girth and the hardness of approximating basic k-spanner. ACM Transactions on Algorithms, 12(2):25:1–25:16, Dec. 2015.
•  M. Dinitz and Z. Zhang. Approximating low-stretch spanners. In Proceedings of the Twenty-seventh Annual ACM-SIAM Symposium on Discrete Algorithms, SODA ’16, pages 821–840, Philadelphia, PA, USA, 2016. Society for Industrial and Applied Mathematics.
•  D.-Z. Du and P.-J. Wan. Routing-Cost Constrained CDS, pages 119–131. Springer New York, New York, NY, 2013.
•  H. Du, W. Wu, W. Lee, Q. Liu, Z. Zhang, and D.-Z. Du. On minimum submodular cover with submodular cost. Journal of Global Optimization, 50(2):229–234, Jun 2011.
•  H. Du, W. Wu, Q. Ye, D. Li, W. Lee, and X. Xu. Cds-based virtual backbone construction with guaranteed routing cost in wireless sensor networks. IEEE Transactions on Parallel and Distributed Systems, 24(4):652–661, April 2013.
•  H. Du, Q. Ye, J. Zhong, Y. Wang, W. Lee, and H. Park. Polynomial-time approximation scheme for minimum connected dominating set under routing cost constraint in wireless sensor networks. Theoretical Computer Science, 447(Supplement C):38 – 43, 2012. Combinational Algorithms and Applications (COCOA 2010).
•  S. Guha and S. Khuller. Approximation algorithms for connected dominating sets. Algorithmica, 20(4):374–387, Apr 1998.
•  R. Hassin and D. Segev. The set cover with pairs problem. In S. Sarukkai and S. Sen, editors, FSTTCS 2005: Foundations of Software Technology and Theoretical Computer Science: 25th International Conference, Hyderabad, India, December 15-18, 2005. Proceedings, pages 164–176, Berlin, Heidelberg, 2005. Springer Berlin Heidelberg.
•  S. Iwata and K. Nagano. Submodular function minimization under covering constraints. In IEEE FOCS, pages 671–680, Oct 2009.
•  G. Kortsarz. On the hardness of approximating spanners. Algorithmica, 30(3):432–450, Jan 2001.
•  G. Kortsarz, R. Krauthgamer, and J. R. Lee. Hardness of approximation for vertex-connectivity network design problems. SIAM Journal on Computing, 33(3):704–720, 2004.
•  G. Kortsarz and D. Peleg. Generating sparse 2-spanners. Journal of Algorithms, 17(2):222 – 236, 1994.
•  C. Koufogiannakis and N. E. Young. Greedy -approximation algorithm for covering with arbitrary constraints and submodular cost. Algorithmica, 66(1):113–152, May 2013.
•  C. Liu, H. Huang, H. Du, and X. Jia. Performance-guaranteed strongly connected dominating sets in heterogeneous wireless sensor networks. In IEEE INFOCOM 2016 - The 35th Annual IEEE International Conference on Computer Communications, pages 1–9, April 2016.
•  T. Nieberg and J. Hurink. A ptas for the minimum dominating set problem in unit disk graphs. In T. Erlebach and G. Persinao, editors, Approximation and Online Algorithms: Third International Workshop, WAOA 2005, Palma de Mallorca, Spain, October 6-7, 2005, Revised Papers, pages 296–306, Berlin, Heidelberg, 2006. Springer Berlin Heidelberg.
•  L. Ruan, H. Du, X. Jia, W. Wu, Y. Li, and K.-I. Ko. A greedy approximation for minimum connected dominating sets. Theoretical Computer Science, 329(1):325 – 330, 2004.
•  S. Tirodkar and S. Vishwanathan. On the approximability of the minimum rainbow subgraph problem and other related problems. Algorithmica, 79(3):909–924, Nov 2017.
•  P.-J. Wan, D.-Z. Du, P. Pardalos, and W. Wu. Greedy approximations for minimum submodular cover with submodular cost. Computational Optimization and Applications, 45(2):463–474, Mar 2010.

## Appendix A Proof of Claim 1

 n1−c =√n⋅α(nc)α−2lnt ⇔n2−2c =n⋅α(nc)α−2lnt (both sides are non-% negative) ⇔n2−2c−(1+c(α−2)) =αlnt ⇔n1−cα =αlnt.

When ,

 n1−cα =n1−(1−lnln(tα)lnn) =nlnln(tα)lnn (1) =(nln(ln(tα)))1lnn (2) =((ln(tα))ln