Abstract
A new method is used to construct the claw network. The weight of edge in next generation on claw network is related to the strength of the node in former generation. Then we calculate the exact expression of the average receiving time (ART) on claw network after dividing the whole network into four blocks. The result of ART indicates that ART grows exponentially with the network order as a power-law function. Moreover, the accurate relation of average shortest path (ASP) on claw network is determined by a novel method. The key of this method is to divide the whole claw network into four blocks, and then divide each block network into some smaller blocks to calculate ASP. The result of ASP implies that ASP increases sublinearly with the size of the claw network.
Export citation and abstract BibTeX RIS
1. Introduction
Complex networks are a new research field in recent years and many scholars have also obtained many research results. Complex networks have been applied to many interdisciplinary fields since they show many superiority, such as communication network, airport network, biological network and so on. As a part of complex networks, deterministic networks are widely studied including the structure of network [1–3], spectra of the network[4–7], the applications of spectra [8–10] and so on.Furthermore, some dynamical processes are also hot fields in complex networks, like random walks [11–13], coherence problem [14, 15], epidemic spreading [16], Lotka-Volterra model [17], evolutionary game [18, 19], and so on.
Meanwhile, the trapping problem on network has also attracted the attention of many scholars. Usually scholars will fix a node as a trap to study the trapping problem [20, 21]. Two important quantities about trapping problem are the mean first-passage time (MFPT) which is the time from a node to the trap node [22, 23] and the average receiving time (ART) which is an average of the mean first-passage times. The hot topic issue in the study of random walks is how the ART scales with the size of network.
In addition, the average shortest path (ASP) is also a basic measure of complex networks. It is related to many other structural characteristics of networks, for example, degree distribution [24, 25], centrality [26], fractal [27–29]. Besides, ASP has a strong efficiency on various dynamical processes, such as synchronization [30]. Because of its importance and practicability, ASP has attracted considerable attention. However, the most of networks they considered are special networks, which the whole network conforms to some certain laws and the edges weights of most networks they considered are related to the generation of the network.
Motivated by the previous study, we construct the claw network with a new method. The claw network is a network proposed by the authors and related to the weight of the edge in weighted network. It is different from the previous deterministic network generated by the substitution rules [31, 32] and valuable to study the dynamic process of such network. The claw network has two features, one is the global feature which is a common feature of the networks, another is the local feature which is a unique feature of claw network. However, local feature may make it impossible to study the degree distribution and strength distribution of the claw network. In addition, the weight of edge in next generation is related to the strength of the node in former generation. Then, we calculate the exact expression of ART after dividing claw network into four blocks. Because of the particularity of network, the previous methods of calculating ASP are not suitable for claw network, we propose a novel method to calculate ASP. The step of this method is to divide the whole claw network into four blocks, and then divide each block network into smaller blocks to calculate ASP.
The rest of the paper is arranged as follows: in section 2, claw network is constructed using a new method, which has a special topological structure different from previous network. In section 3, we divide the claw network into four block, then we work out the sum of MFPTs, finally we obtain the accurate relation of ART, which implies that ART grows exponentially with the network order as a power-law function. In section 4, we first divide every block network into small blocks after dividing the whole claw network into four blocks. Based on this operation, we find the sums of shortest paths of every block network and the whole claw network. Eventually, we obtain the result which shows that ASP increases sublinearly with the size of claw network. In section 5, we give the conclusion about ART and ASP after analyzing the expressions of ART and ASP respectively.
2. The claw network
The process of network construction is described in this section. Let Gn denote the n-th graph and Nn be the set of nodes in Gn. The strength si of node i represents the sum of the weights of all edges connected to node i and the initial value of si depends on the sum. Our initial graphs are G0 and h. G0 is a non-clique connected graph consisting of three nodes and two edges of unit weight. h is a weighted graph with two nodes and a unit weighted edge. Given the n-th network Gn, Gn+1 is obtained by the following two steps.
Step 1: For every node i with strength si in Gn, we take si copies of h, and link them to node i by unit weighted edge.
Step 2: The weight of every edge in Gn is increased from w to .
Figure 1 shows the construction processes of claw network from n = 0 to n = 2.
In order to describe the topological structure of the network better and study conveniently, the claw network is divided into four blocks: two blocks ( and ) are connected to node 1, and two blocks ( and ) are connected to nodes 2 and 3 respectively. It is obvious that the four blocks of Gn have the same structure. For G1, let node 1 be the hub node of and , nodes 2 and 3 be hub nodes of and respectively, nodes 4, 5, 6, 7, 8, 9, 10 and 11 in G1 be sub-hub nodes, which are defined as the nodes whose strength are second only to the hub node in each block network. Since the four blocks of Gn have the same structure, we take as an example to show the process of network generation. For n > 0, Gn+1 is obtained from the four blocks of Gn when Gn and G1 are given. Firstly, four copies of are taken and connected to node 1 of G1, nextly two copies of are taken and connected to node 4 of G1, and they are also taken and connected to node 8 of G1 similarly, finally is got. By repeating above operation, we can obtain , and , then Gn+1 is got.
Take n = 1 as an example, G2 is obtained from the four blocks of G1 when G1 is given. Firstly, we take four copies of and connect these copies to node 1 , nextly we take two copies of and connect these copies to node 4, then take two copies of and connect these copies to node 8, finally we get . By repeating the above operation, , and can be obtained, eventually we can get G2. Figure 2 shows the construction process of claw network.
Download figure:
Standard image High-resolution imageFrom the construction of claw network, we can easily get
3. Average receiving time on claw network
This section gives the exact expression of average receiving time and shows how it scales with the size of network. On claw network, we consider that the method of walking is weight-dependent walk, it means that every node moves from its current location to any of its neighbors according to probability proportional to the weight of the edge. The transition probability is
Where h(i) represents the neighbors of node i, and si is the strength of node i.
In this paper, we consider the trap node is node 1. and Fi(n) are defined as the mean first-passage time (MFPT) for an agent walking from node i to node j and the MFPT from node i to the trap respectively, where n represents the n-th generation network. Let be the average receiving time (ART), which is an average of Fi(n) of all nodes except the trap. Based on the definition of ART, we have
Let , in order to determine , we should determine Ttot(n) firstly. Combining the division of Gn, we can get
Because and , and are symmetrical, we can simplify equation (4).
Then we need to determine and . Based on the special structure of claw network, we have the following relationships.
For :
Where is MFPT from node 4 to node 1 in Gn.
For :
Where is MFPT from node 2 to the trap in Gn, is MFPT from node 6 to node 2 in Gn.
Note that , thus we need to find the and . Let X be the MFPT from node i to any of its old neighbors in Gn, and let Y be the MFPT from a new neighbor of node j generated in Gn+1 to an old neighbor of node i in Gn. Let S be the strength of node i in Gn, then the strength of node i in Gn+1 is . According to the method of walking and the construction of claw network, we have the following equations:
It is apparent that X = 4 holds, so we know that if the MFPT from node i to node j in Gn is T, then the MFPT from node i to node j in Gn+1 is . Combining initial condition , we obtain
For , we have
From above equations, we can get , so we know
By introducing equations (1), (8) and (9) into equations (6) and (7) and combining the initial conditions and , we can obtain
Based on equations (5), (10) and (11), we can get
Based on equations (1), (3) and (12), we have
From equation (1), we note that
Considering equations (13) and (14), we can get
When the size of network is large enough, we obtain
The numerical formula equation (13) and the analytical formula equation (16) are compared for (see figure 3).
Download figure:
Standard image High-resolution imageAs shown in figure 3, the numerical result is slightly greater than the analysis result when and the same as the analysis result when n > 3 which shows that the analysis result is accurate when the network size tends to infinity. Equation (16) illustrates that ART increases exponentially with the network order as a power-law function of , which implies that ART grows sublinearly with the size of network.
4. Average weighted shortest path
This section we find the accurate expression of average shortest path (ASP). Based on the definition of ASP [31, 32] , we have the following equations.
Where is the shortest path between node i and node j in Gn.
Applying the previous generation method (substitution rule) of the network, the exact value of ASP can be obtained by dividing the network once. However, the network generated by the new method is related to the weight of the edge, so the exact solution of ASP cannot be obtained in previous method. Encouraged by the previous research, we find the exact solution of ASP by proposing a new method of dividing the network twice. Because claw network is different from the previous network, we need divide each into smaller block when we find the exact expression of ASP. Figure 4 shows the partitioning of network taking as an example.
Download figure:
Standard image High-resolution imageAs can be seen from the figure 4, has the following recursive relationship.
Where is the sum of all shortest paths of two nodes which belong to different blocks. Detailed derivation of is shown in appendix A. The result is as follows.
Substituting equation (20) into equation (19) and combining the initial condition , we can get
By calculating equation (21), we can obtain
Next, we continue to calculate Dtot(n). Based on partitioning of the claw network, Dtot(n) can be expressed as the following equation.
Where Ωn is the sum of all shortest paths of two nodes which are in different . See appendix B for detailed deducing. The conclusion of Ωn is
Considering equations (22), (23) and (24), we can get
Combining equations (1), (17) and (25), we have
Substituting equation (14) into (26), we can obtain
When Nn is large enough, we have
The numerical formula equation (26) and the analytical formula equation (28) are checked for 1 ≤ n ≤ 8 (see figure 5).
Download figure:
Standard image High-resolution imageAs can be seen from figure 5, the numerical result is slightly less than the analysis result when 0 ≤ n ≤ 4 and the same as the analysis result when n > 4, which shows that the analysis result equation (28) is desirable when the network tends to infinity. As shown in figure 5 and equation (28), ASP grows as a power-law function of the network order with the exponent, denoted by .
5. Conclusion
In this paper, a new method of generating network is introduced to construct the claw network. Moreover, we determine the exact solution of ART and ASP on claw network. The exact expression of ART is obtained after finding out the results of MFPTS by block method, then we test the analysis result and the numerical result of ART. From figure 3, we can see that before the fourth generation of claw network, the numerical result is slightly larger than the analysis result, after the fourth generation of the network, the analysis result is the same as the numerical result. This proves that our analysis about ART is reasonable, and the analysis result of ART implies that ART increases exponentially with the size of the network.
The accurate expression of ASP is obtained by twice-blocking method. Similarly, we examine the analysis result and the numerical result of ASP. From figure 5, we can see that before the fourth generation of the claw network, the numerical result is slightly smaller than the analysis result, and after the fourth generation of the claw network, the analysis result is the same as the numerical result. This shows that our analysis about ASP is also appropriate, and the analysis result of ASP displays that ASP increases sublinearly with the size of the network.
Acknowledgments
Research is supported by National Natural Science Foundation of China (Nos. 11 501 255, 11 671 172, 11 801 225), University Science Research Project of Jiangsu Province (No.18KJB110005) and Foundation for advanced talents of Jiangsu University (5503000039).
Appendix A.: The specific derivation of the
Let be the shortest path of two nodes which are in different blocks. From figure 4, we can get the following equation.
In order to obtain , we define
Where di1 is the shortest path of any node in to node 1. According to the topological structure of claw network, we have the recursive relation about .
The second term of RHS (right-hand side) of equation is the sum of shortest paths from nodes connected to sub-hub nodes (nodes 4 and 8) to node 1, The third term of RHS (right-hand side) of equation is the sum of shortest paths from nodes 4 and 8 to node 1. Simplified the equation, we have
Combining the initial condition , we can get
After getting , we have
and
By calculating,we can get
Appendix B.: The specific derivation of the Ωn
Let be the shortest path of two nodes which belong to different blocks, for example, denotes the shortest path of two nodes belonging to and respectively. From figure 2, we can get the following equation.
Because of the particularity of the initial graph, the expressions of Ωn and are not similar.
and
By calculating, we can get