Fund Project:Project supported by the National Key Research and Development Program of China (Grant Nos. 2017YFB0803204, 2018YFB0804002), the National Natural Science Foundation of China (Grant Nos. 61872382, 61802429), and the Research and Development Program in Key Areas of Guangdong Province, China (Grant No. 2018B010113001)
Received Date:19 May 2019
Accepted Date:09 July 2019
Available Online:01 September 2019
Published Online:20 September 2019
Abstract:The robustness of complex networks plays an important role in human society. By further observing the networks on our planet, researchers find that many real systems are interdependent. For example, power networks rely on the Internet to transfer operation information, predators have to hunt for herbivores to refuel themselves, etc. Previous theoretical studies indicate that removing a small fraction of nodes in interdependent networks leads to a thorough disruption of the interdependent networks. However, due to the heterogeneous weak inter-layer links, interdependent networks in real world are not so fragile as the theoretical predictions. For example, an electronic components factory needs raw materials which are produced by a chemical factory. When the chemical factory collapses, the electronic components factory will suffer substantial drop in the production, however, it can still survive because it can produce some other raw materials by itself to sustain its production of some products. What is more, because of the heterogeneity on real industry chains, different electronic components factories produce different kinds of products, which still guarantees the diversity of electronic goods on the whole. In this paper, we develop a framework to help understand the robustness of interdependent networks with heterogeneous weak inter-layer links. More specifically, in the beginning, a fraction of 1–p nodes are removed from network A and their dependency nodes in network B are removed simultaneously, then the percolation process begins. Each connectivity link of a node with weak inter-layer dependency is removed with a probability γ after the failure of its counterpart node. The γ values for different nodes are various because of heterogeneity. At the end, the nodes can survive as long as one of the remaining connectivity links reaches the giant component. We present an analytical solution for solving the giant component size and analyzing the crossing point of the phase transition of arbitrary interdependent random networks. For homogeneous symmetric Erd?s-Rényi networks, we solve the continuous transition point and the critical point of γ. The simulation results are in good agreement with our exact solutions. Furthermore, we introduce two kinds of γ distributions to analyze the influence of heterogeneous weak inter-layer links on the robustness of interdependent networks. The results of both distributions show that with the increase of heterogeneity, the transition point pc decreases and the networks become more robust. For the first simple γ distribution, we also find the percolation transition changes from discontinuous one to continuous one by improving the heterogeneity. For the second Gaussian γ distribution, a higher variance makes the interdependent networks more difficult to collapse. Our work explains the robustness of real world interdependent networks from a new perspective, and offers a useful strategy to enhance the robustness by increasing the heterogeneity of weak inter-layer links of interdependent networks. Keywords:interdependent networks/ cascading failures/ percolation/ phase transition
3.仿真与讨论首先考虑同质对称弱相依网络, 图1和图2分别为ER和scale-free (SF)[26]对称弱相依网络的逾渗仿真结果. 图中给出了巨分量${\mu _\infty }$和级联失效迭代次数(number of iterative, NOI)与初始保留节点比例p的关系. 从图1和图2可以看出仿真结果与理论分析拟合. 随着$\gamma $值的减小, 相变点${p_{\rm{c}}}$逐渐减小的同时网络鲁棒性有所提升. 对于非连续相变, NOI曲线在相变点处存在尖峰, 连续相变的NOI曲线一直比较平缓. 图 1 同质对称ER弱相依网络对于不同γ值的巨分量${\mu _\infty }$与p对应关系(网络节点数为200000, 平均度为4) (a) 巨分量大小${\mu _\infty }$与p对应关系, 空心标记表示仿真结果, 实线是根据(17)和(18)式得到的理论值; (b) 级联失效迭代次数 Figure1. Simulation results of ${\mu _\infty }$ versus p for homogeneous symmetric interdependent ER networks for different γ (each network has 200000 nodes, average degree is 4). (a) The size of the giant component ${\mu _\infty }$ versus p. The symbols represent the simulation results, and the solid lines show the corresponding analytical predictions of Eqs. (17) and (18). (b) Number of iterative failures.
图 2 同质对称SF弱相依网络对于不同$\gamma $的巨分量${\mu _\infty }$与p对应关系(网络节点数为20000, 平均度为4, $\lambda = 2.6$) (a) 巨分量大小${\mu _\infty }$与p对应关系, 空心标记表示仿真结果, 实线是根据(17)式和(18)式得到的理论值; (b) 级联失效迭代次数 Figure2. Simulation results of ${\mu _\infty }$ versus p for homogeneous symmetric interdependent SF networks for different γ (each network has 200000 nodes, average degree is 4, $\lambda = 2.6$). (a) The size of the giant component ${\mu _\infty }$ versus p. The symbols represent the simulation results, and the solid lines show the corresponding analytical predictions of Eqs. (17) and (18). (b) Number of iterative failures.
下面通过图形示意方法讨论逾渗相变点数值解, 令$D(f) = {\rm{rhs}} - f$, 其中${\rm{rhs}}$表示(18)式等号右边的部分, 显然$D(0) = 0$, 即$D(f)$在0点处与f轴相交. 随着初始保留节点比例p的增大, $D(f)$曲线会逐渐上升, 直到p增大到相变点, $D(f)$曲线会第一次与f轴相切, 此时的p即为相变点pc. 图3和图4分别为同质ER, SF弱相依网络的逾渗数值解示意图, 图中给出了$D(f)$与f的对应关系. 从图3和图4可以看出, 当$p = {p_{\rm{c}}}$时, $D(f)$曲线会与图f轴相切, 并且连续相变的切点为0 (图3(a)、图4(a)), 非连续相变的切点大于0 (图3(b)、图4(b)). 图 3 同质对称ER弱相依网络对于不同γ值的数值解示意图(网络平均度为4; 在相变点pc处, $D(f)$曲线与f轴相切) (a) γ = 0.1; (b) γ = 0.4 Figure3. Graphical solutions of homogeneous symmetric interdependent ER networks percolation transition for different γ: (a) γ = 0.1; (b) γ = 0.4. The average degree is 4. At the transition point pc, the curve of D(f) tangents to f axis.
图 4 同质对称SF弱相依网络对于不同γ值的数值解示意图(网络平均度为4, $\lambda = 2.6$; 在相变点pc处, $D(f)$曲线与f轴相切) (a) γ = 0.5; (b) γ = 0.9 Figure4. Graphical solutions of homogeneous symmetric interdependent SF networks percolation transition for different γ: (a) γ = 0.5; (b) γ = 0.9. The average degree is 4, $\lambda = 2.6$. At the transition point pc, the curve of D(f) tangents to f axis.
图5给出了根据(21)和(23)式求出的不同平均度同质对称ER弱相依网络相变点pc与γ对应关系. 从图5可以看出, 对于同质对称ER相依网络, ${\gamma _{\rm{c}}}$取值唯一且与网络度分布无关, 这与前文理论分析结果一致. 此外, 随着平均度的增加, ${p_{\rm{c}}}$随之减少, 意味着网络中更多的连接边可使网络更加鲁棒. 图 5 不同平均度同质对称ER弱相依网络相变点pc与γ对应关系, 其中网络的节点数为200000; 空心标记为仿真结果; 理论值分别通过实线和短划线表示, 其中实线为连续相变, 短划线为非连续相变; 垂直的点状线为连续相变和非连续相变的边界 Figure5. Simulation results of pc versus γ for homogeneous symmetric interdependent ER networks with different $\left\langle k \right\rangle $, each network has 200000 nodes. The symbols represent the simulation results. The corresponding analytical predictions are shown by lines, solid lines and dashed lines represent continuous and discontinuous phase transitions, respectively. The vertical dotted line is the boundary of continuous and discontinuous regions.
接下来考虑异质对称弱相依网络. 为了便于仿真分析, 首先考虑一种较为简单的$\gamma $分布情况, 即$\gamma = \left\{ {\bar \gamma - \Delta \gamma ,\bar \gamma + \Delta \gamma } \right\}$且$p(\gamma ) \sim \left\{ {q,1 - q} \right\}$, 表示网路中任取节点的弱相依节点失效后, 其连接边失效概率为$\bar \gamma - \Delta \gamma $的概率为q, 连接边失效概率为$\bar \gamma + \Delta \gamma $的概率为$1 - q$. 图6给出了不同简单$\gamma $分布的同质($\Delta \gamma = 0$)和异质($\Delta \gamma \ne 0$) ER相依网络的仿真结果. 从图6可以看出, 在相同$\gamma $概率均值的情况下, 异质性的引入会减小${p_{\rm{c}}}$并提高网络的鲁棒性. 图 6 不同简单γ分布的同质和异质ER弱相依网络的仿真结果, 其中网络节点数为200000, 平均度是4, $q = 0.5$; 空心标记表示仿真结果, 实线是理论分析值 Figure6. Simulation results of heterogeneous and homogeneous symmetric ER interdependent networks with different γ distributions, each network has 200000 nodes, average degree is 4, $q = 0.5$. The symbols represent the simulation results, and the solid lines show the corresponding analytical predictions.
图7是简单γ分布异质对称ER弱相依网络相变点与$\Delta \gamma $值对应关系的仿真结果, 图中连续相变与非连续相变分界线与(23)式求解方法类似, 通过对(19)式两边求f的偏导, 再将${f_{\rm{c}}} = 0$代入得到 图 7 简单γ分布异质对称ER弱相依网络相变点${p_{\rm{c}}}$与$\Delta \gamma $值的仿真结果, 其中网络节点数为200000, 平均度是4, $q = 0.5$; 空心标记表示仿真值, 短划线和实线分别表示非连续相变与连续相变的理论值, 点状线是连续相变和非连续相变的分界线 Figure7. Simulation results of critical point ${p_{\rm{c}}}$ of simple heterogeneous symmetric ER interdependent networks versus $\Delta \gamma $, each network has 200000 nodes, average degree is 4, $q = 0.5$. The symbols represent the simulation results. The corresponding analytical predictions are shown by lines, solid lines and dashed lines represent continuous and discontinuous phase transitions, respectively. The dotted line is the boundary of continuous and discontinuous regions.
其中$\bar \gamma $是均值, ${\sigma ^2}$是方差, $\alpha $为归一化调节参数. 当$\sigma = 0$时, 异质网络退化为同质网络, 所有节点的$p(\gamma )$相同. 图8给出了连接边失效概率服从高斯分布的异质ER相依网络仿真结果. 从图8可以看出, 随着高斯分布方差增大, 网络相变点${p_{\rm{c}}}$逐渐减小, 网络鲁棒性相应提高. 方差越大意味着网络异质程度越高, 因此, 网络异质程度与网络鲁棒性是正相关的, 这也从一定程度上解释了真实复杂网络抗毁性能优于理论结果的事实, 该结论与图7的仿真结果一致. 图 8 连接边失效概率服从高斯分布的异质对称ER弱相依网络巨分量${\mu _\infty }$与p的对应关系, 其中高斯分布的均值$\bar \gamma $为0.7, $\sigma $分别为0.9, 0.4, 0.2; 根据(25)式, $\sigma = 0$时$p(\gamma ) = \delta (\gamma - \bar \gamma )$; 空心标记为仿真结果, 实线为理论分析值 Figure8. Simulation results of ${\mu _\infty }$ versus p for heterogeneous symmetric ER interdependent networks with Gaussian distributions of connectivity link failure probability. The average value $\bar \gamma $ are set as 0.7, $\sigma $ are set as 0.9, 0.4, 0.2, respectively. According to Eq. (25), $p(\gamma ) = \delta (\gamma - \bar \gamma )$ when $\sigma = 0$. The symbols represent the simulation results, and the solid lines show the corresponding analytical predictions.