1. 狄增如 北京师范大学管理学院系统科学系 北京师范大学复杂性研究中心 2011.4
有向网络中的同步能力
狄增如
北京师范大学管理学院系统科学系
北京师范大学复杂性研究中心
2011.4

2. 网络拓扑、权重对结构与功能的影响
以网络上的动力学同步为切入点

3. 网络同步问题 网络拓扑 + 作用强度 二元网 >>> 加权网
网络拓扑 + 作用强度
二元网 >>> 加权网
Physics Letters A 364 (2007) 488–493
Menghui Li, Ying Fan,Dahui Wang, Daqing Li, Jinshan Wu and Zengru Di
Small-world effect induced by weight randomization on regular networks

4. 加权网络上的混沌同步 Daqing Li, Menghui Li, J. Wu, Z. Di, Y.Fan,
Eur. Phys. J. B. Vol.57, (2007)

6. 更一般的情况下会如何？

8. Optimization of synchronizability
Finally,Nishikawa and Motter gave the weight distributing an extreme way by imposing the weight of some edges to 0. They get an oriented tree with normalized input strength and no directed loops. Moreover, they proved that the R of the tree is 1, meaning that the tree is the optimal structure for synchronizability.
T. Nishikawa, A.E. Motter, Synchronization is optimal in nondiagonalizable networks, Phys. Rev. E 73 (2006)

9. EPL, 87 (2009) An Zeng, Yanqing Hu and Zengru Di Optimal tree for both synchronizability and converging time

10. Center of the network and the shortest spanning tree
signaling process could be described by a simple but clear mathematical mechanism.
Suppose we have a network with N nodes; it can be represented mathematically by an adjacency matrix A. And the I is a matrix where all the diagonal elements are 1 and others are 0.
Then the column i of matrix will represent the effect of source node i to the whole network in m steps.
So we can get a N-dimensional vector that records each node's signal quantity which represents the effect of the source node. If all the elements of a column are nonzero, the signal of the source node i has affected the whole network. To find the center, we should simply find which column reaches totally nonzero at first.

12. Center of the network and the shortest spanning tree
The shortest spanning tree can be created by the signaling process too.
Supposing node i is the center of the network, we mark i as a used node. The second level includes all the unused nodes connected to the node i, then mark these nodes and the edges from node i to them as used. To create the third level, we consider each node in the second level as a sub-center, and the process is the same as the center i.
Specifically, if a node is marked, it as well as the edges connecting to it can not be marked by other nodes any more.
After all the nodes are marked, we can get the shortest spanning tree.

13. Center of the network and the shortest spanning tree
The topology of the original network is related to the depth of such effective directed trees. For instance, the center of the scale-free network usually has big degree. This kind of center can reduce the depth of the tree significantly. On the contrary, the homogenous random network does not have such kind of center, so the depth of the spanning tree from these networks will always be longer.

15. Seung-Woo Son, Beom Jun Kim, Hyunsuk Hong, and Hawoong Jeong PRL 103, (2009) Dynamics and Directionality in Complex Networks
RDG Method

16. synchronization Enhancement

17. The Problem

18. RDG method
Random exponential networks
Random scale-free networks

19. REBG Method
Random exponential networks
Random scale-free networks

20. Synchronization enhancement (REBD-RDG)
Random exponential networks
Random scale-free networks

21. 谢谢大家！