Search

Complex adaptive learning is intelligent. It is adaptive, learns in feedback loops, and generates hidden patterns as many individuals, elements or particles interact in complex adaptive systems (CAS). CAS highlights adaptation in life and lifeless complex systems cutting across all traditional natural and social sciences disciplines. However, discovering a universal law in CAS and understanding the underlying mechanism of distribution formation, such as a non-Gauss distribution in complex quantum entanglement, remains highly challenging. Quantifying the uncertainty of CAS by probability wave functions, the authors explore the inherent logical relationship between Schr\"odinger's wave equation in quantum mechanics and Shi's trading volume-price wave equation in finance. Subsequently, the authors propose a non-localized wave equation in quantum mechanics if cumulative observable in a time interval represents momentum or momentum force in Skinner-Shi (reinforcement-frequency-interaction) coordinates. It reveals that the invariance of interaction as a universal law exists in quantum mechanics and finance. The theory shows that quantum entanglement is an interactively coherent state instead of a consequence of the superposition of coherent states. As a resource, quantum entanglement is non-separable, steerable, and energy-consumed. The entanglement state has opposite states subject to interaction conservation between the momentum and reversal forces. Keywords: complex adaptive systems, complex adaptive learning, complex quantum systems, non-localized wave equation, interaction conservation, interactively coherent entanglement PACS: 89.75.-k (Complex Systems); 89.65.Gh (Economics, Econophysics, Financial Markets, Business and Management); 03.65.Ud (Entanglement and Quantum Nonlocality), Comment: 19 pages in total, 8 figures, 1 table, and 59 references

Community detection is a significant and challenging task in network science. Nowadays, plenty of attention has been paid on local methods for community detection. Greedy expanding is a popular and efficient class of local algorithms, which typically starts from some selected central nodes and expands those nodes to obtain provisional communities by optimizing a certain quality function. In this paper, we propose a novel index, called local superiority index (LSI), to identify central nodes. In the process of expansion, we apply the fitness function to estimate the quality of provisional communities and ensure that all provisional communities must be weak communities. Evaluation based on the normalized mutual information suggests: (1) LSI is superior to the global maximal degree index and the local maximal degree index on most considered networks; (2) The greedy algorithm based on LSI is better than the classical fast algorithm on most considered networks.

In some systems, the connecting probability (and thus the percolation process) between two sites depends on the geometric distance between them. To understand such process, we propose gravitationally correlated percolation models for link-adding networks on the two-dimensional lattice $G$ with two strategies $S_{\rm max}$ and $S_{\rm min}$, to add a link $l_{i,j}$ to connect site $i$ and site $j$ with mass $m_i$ and $m_j$, respectively; $m_i$ and $m_j$ are sizes of the clusters which contain site $i$ and site $j$, respectively. The probability to add the link $l_{i,j}$ is related to the generalized gravity $g_{ij} \equiv m_i m_j/r_{ij}^d$, where $r_{ij}$ is the geometric distance between $i$ and $j$, and $d$ is an adjustable decaying exponent. In the beginning of the simulation, all sites of $G$ are occupied and there is no link. In the simulation process, two inter-cluster links $l_{i,j}$ and $l_{k,n}$ are randomly chosen and the generalized gravities $g_{ij}$ and $g_{kn}$ are computed. In the strategy $S_{\rm max}$, the link with larger generalized gravity is added. In the strategy $S_{\rm min}$, the link with smaller generalized gravity is added, which include percolation on the Erd\H os-R\'enyi random graph and the Achlioptas process of explosive percolation as the limiting cases, $d \to \infty$ and $d \to 0$, respectively. Adjustable strategies facilitate or inhibit the network percolation in a generic view. We calculate percolation thresholds $T_c$ and critical exponents $\beta$ by numerical simulations. We also obtain various finite-size scaling functions for the node fractions in percolating clusters or arrival of saturation length with different intervening strategies., Comment: 26 pages, 8 figures, Chinese Journal of Physics, accepted for publication

In this paper, we study the control of the traffic-driven epidemic spreading by immunization strategy. We consider the random, degree-based and betweeness-based immunization strategies, respectively. It is found that the betweeness-based immunization strategy can most effectively prevent the outbreak of traffic-driven epidemic. Besides, we find that the critical number of immune nodes above which epidemic dies out is increased with the enhancement of the spreading rate and the packet-generation rate., Comment: 5 pages, 6 figures. arXiv admin note: text overlap with arXiv:1802.01246, arXiv:1802.01258

We study the traffic-driven epidemic spreading on scale-free networks with tunable degree distribution. The heterogeneity of networks is controlled by the exponent $\gamma$ of power-law degree distribution. It is found that the epidemic threshold is minimized at about $\gamma=2.2$. Moreover, we find that nodes with larger algorithmic betweenness are more likely to be infected. We expect our work to provide new insights into the effect of network structures on traffic-driven epidemic spreading., Comment: 8 pages, 5 figures. arXiv admin note: text overlap with arXiv:1503.00145, arXiv:1503.00149

The scattering of a positron by a muon in the presence of a linearly polarized laser field is investigated in the first Born approximation. The theoretical results reveals: 1) at large scattering angle, an amount of multiphoton processes take place in the course of scattering. The photon emission processes predominant the photon absorption ones. 2) Some nonlinear phenomena about oscillations, dark angular windows, and asymmetry can be observed in angular distributions. We analyze the reason giving rise to dark windows and geometric asymmetry initially noted in the potential scattering. 3) we also analyze the total differential cross section, the result shows that the larger the incident energy is, the smaller total differential cross section is. The reason of these new results are analysed.

An extended social force model with a dynamic navigation field is proposed to study bidirectional pedestrian movement. The dynamic navigation field is introduced to describe the desired direction of pedestrian motion resulting from the decision-making processes of pedestrians. The macroscopic fundamental diagrams obtained using the extended model are validated against camera-based observations. Numerical results show that this extended model can reproduce collective phenomena in pedestrian traffic, such as dynamic multilane flow and stable separate-lane flow. Pedestrians' path choice behavior significantly affects the probability of congestion and the number of self-organized lanes.

It is a longstanding debate on the absence of threshold for susceptible-infected-susceptible (SIS) model on networks with finite second order moment of degree distribution. The eigenvector localization of the adjacency matrix for a network gives rise to the inactive Griffiths phase featuring slow decay of the activity localized around highly connected nodes due to the dynamical fluctuation. We show how it dramatically changes our understanding of SIS model, opening up new possibilities for the debate. We derive the critical condition for Griffiths to active phase transition: on average, an infected node can further infect another one in the characteristic lifespan of the star subgraph composed of the node and its nearest neighbors. The system approaches the critical point of avoiding the irreversible dynamical fluctuation and the trap of absorbing state. As a signature of the phase transition, the infection density of a node is not only proportional to its degree, but also proportional to the exponentially growing lifespan of the star. And the divergence of the average lifespan of the stars is responsible for the vanishing threshold in the thermodynamic limit. The eigenvector localization exponentially reinforces the infection of highly connected nodes, while it inversely suppresses the infection of small-degree nodes.

To understand the multiple relations between developers and projects on GitHub as a whole, we model them as a multilayer bipartite network and analyze the degree distributions, the nearest neighbors' degree distributions and their correlations with degree, and the collaborative similarity distributions and their correlations with degree. Our results show that all degree distributions have a power-law form, especially, the degree distribution of projects in watching layer has double power-law form. Negative correlations between nearest neighbors' degree and degree for both developers and projects are observed in both layers, exhibiting a disassortative mixing pattern. The collaborative similarity of both developers and projects negatively correlates with degree in watching layer, while a positive correlations is observed for developers in forking layer and no obvious correlation is observed for projects in forking layer.

The intersecting pedestrian flow on the 2D lattice with random update rule is studied. Each pedestrian has three moving directions without the back step. Under periodic boundary conditions, an intermediate phase has been found at which some pedestrians could move along the border of jamming stripes. We have performed mean field analysis for the moving and intermediate phase respectively. The analytical results agree with the simulation results well. The empty site moves along the interface of jamming stripes when the system only has one empty site. The average movement of empty site in one Monte Carlo step (MCS) has been analyzed through the master equation. Under open boundary conditions, the system exhibits moving and jamming phases. The critical injection probability $\alpha_c$ shows nontrivially against the forward moving probability $q$. The analytical results of average velocity, the density and the flow rate against the injection probability in the moving phase also agree with simulation results well., Comment: 11 pages, 15 figures

We ascertain the modularity-like objective function whose optimization is equivalent to the maximum likelihood in annotated networks. We demonstrate that the modularity-like objective function is a linear combination of modularity and conditional entropy. In contrast with statistical inference methods, in our method, the influence of the metadata is adjustable; when its influence is strong enough, the metadata can be recovered. Conversely, when it is weak, the detection may correspond to another partition. Between the two, there is a transition. This paper provides a concept for expanding the scope of modularity methods.

In real networks, the dependency between nodes is ubiquitous; however, the dependency is not always complete and homogeneous. In this paper, we propose a percolation model with weak and heterogeneous dependency; i.e., dependency strengths could be different between different nodes. We find that the heterogeneous dependency strength will make the system more robust, and for various distributions of dependency strengths both continuous and discontinuous percolation transitions can be found. For Erd\H{o}s-R\'{e}nyi networks, we prove that the crossing point of the continuous and discontinuous percolation transitions is dependent on the first five moments of the dependency strength distribution. This indicates that the discontinuous percolation transition on networks with dependency is determined not only by the dependency strength but also by its distribution. Furthermore, in the area of the continuous percolation transition, we also find that the critical point depends on the first and second moments of the dependency strength distribution. To validate the theoretical analysis, cases with two different dependency strengths and Gaussian distribution of dependency strengths are presented as examples., Comment: 8 pages, 8 figures

The communication networks in real world often couple with each other to save costs, which results in any network does not have a stand-alone function and efficiency. To investigate this, in this paper we propose a transportation model on two coupled networks with bandwidth sharing. We find that the free-flow state and the congestion state can coexist in the two coupled networks, and the free-flow path and congestion path can coexist in each network. Considering three bandwidth-sharing mechanisms, random, assortative and disassortative couplings, we also find that the transportation capacity of the network only depends on the coupling mechanism, and the fraction of coupled links only affects the performance of the system in the congestion state, such as the traveling time. In addition, with assortative coupling, the transportation capacity of the system will decrease significantly. However, the disassortative coupling has little influence on the transportation capacity of the system, which provides a good strategy to save bandwidth. Furthermore, a theoretical method is developed to obtain the bandwidth usage of each link, based on which we can obtain the congestion transition point exactly., Comment: 9 pages, 5 figures

Comparing with single networks, the multiplex networks bring two main effects on the spreading process among individuals. First, the pathogen or information can be transmitted to more individuals through different layers at one time, which enlarges the spreading scope. Second, through different layers, an individual can also transmit the pathogen or information to the same individuals more than once at one time, which makes the spreading more effective. To understand the different roles of the spreading scope and effectiveness, we propose an epidemic model on multiplex networks with link overlapping, where the spreading effectiveness of each interaction as well as the variety of channels (spreading scope) can be controlled by the number of overlapping links. We find that for Poisson degree distribution, increasing the epidemic scope (the first effect) is more efficient than enhancing epidemic probability (the second effect) to facilitate the spreading process. However, for power-law degree distribution, the effects of the two factors on the spreading dynamics become complicated. Enhancing epidemic probability makes pathogen or rumor easier to outbreak in a finite system. But after that increasing epidemic scopes is still more effective for a wide spreading. Theoretical results along with reasonable explanation for these phenomena are all given in this paper, which indicates that the epidemic scope could play an important role in the spreading dynamics., Comment: 12 pages, 5 figures

We study the percolation in coupled networks with both inner-dependency and inter-dependency links, where the inner- and inter-dependency links represent the dependencies between nodes in the same or different networks, respectively. We find that when most of dependency links are inner- or inter-ones, the coupled networks system is fragile and makes a discontinuous percolation transition. However, when the numbers of two types of dependency links are close to each other, the system is robust and makes a continuous percolation transition. This indicates that the high density of dependency links could not always lead to a discontinuous percolation transition as the previous studies. More interestingly, although the robustness of the system can be optimized by adjusting the ratio of the two types of dependency links, there exists a critical average degree of the networks for coupled random networks, below which the crossover of the two types of percolation transitions disappears, and the system will always demonstrate a discontinuous percolation transition. We also develop an approach to analyze this model, which is agreement with the simulation results well., Comment: 9 pages, 4 figures

As a generation of the classical percolation, clique percolation focuses on the connection of cliques in a graph, where the connection of two $k$-cliques means that they share at least $l1$ makes a step-function-like discontinuous phase transition in the thermodynamic limit and a continuous phase transition for $l=1$. More interesting, our analysis shows that at the critical point, the order parameter $\phi_c$ for $l>1$ is neither $0$ nor $1$, but a constant depending on $k$ and $l$. All these theoretical findings are in agreement with the simulation results, which give theoretical support and clarification for previous simulation studies of clique percolation., Comment: 6 pages, 5 figures

How to identify the influential spreaders in social networks is crucial for accelerating/hindering information diffusion, increasing product exposure, controlling diseases and rumors, and so on. In this paper, by viewing the k-shell value of each node as its mass and the shortest path distance between two nodes as their distance, then inspired by the idea of the gravity formula, we propose a gravity centrality index to identify the influential spreaders in complex networks. The comparison between the gravity centrality index and some well-known centralities, such as degree centrality, betweenness centrality, closeness centrality, and k-shell centrality, and so forth, indicates that our method can effectively identify the influential spreaders in real networks as well as synthetic networks. We also use the classical Susceptible-Infected-Recovered (SIR) epidemic model to verify the good performance of our method., Comment: 4 tables and 4 figures, accepted by Physica A

In this paper, we study the role of degree mixing in the naming game. It is found that consensus can be accelerated on disassortative networks. We provide a qualitative explanation of this phenomenon based on clusters statistics. Compared with assortative mixing, disassortative mixing can promote the merging of different clusters, thus resulting in a shorter convergence time. Other quantities, including the evolutions of the success rate, the number of total words and the number of different words, are also studied.

The interplay between traffic dynamics and epidemic spreading on complex networks has received increasing attention in recent years. However, the control of traffic-driven epidemic spreading remains to be a challenging problem. In this Brief Report, we propose a method to suppress traffic-driven epidemic outbreak by properly removing some edges in a network. We find that the epidemic threshold can be enhanced by the targeted cutting of links among large-degree nodes or edges with the largest algorithmic betweeness. In contrast, the epidemic threshold will be reduced by the random edge removal. These findings are robust with respect to traffic-flow conditions, network structures and routing strategies. Moreover, we find that the shutdown of targeted edges can effectively release traffic load passing through large-degree nodes, rendering a relatively low probability of infection to these nodes.

The bidirectional selection between two classes widely emerges in various social lives, such as commercial trading and mate choosing. Until now, the discussions on bidirectional selection in structured human society are quite limited. We demonstrated theoretically that the rate of successfully matching is affected greatly by individuals neighborhoods in social networks, regardless of the type of networks. Furthermore, it is found that the high average degree of networks contributes to increasing rates of successful matches. The matching performance in different types of networks has been quantitatively investigated, revealing that the small-world networks reinforces the matching rate more than scale-free networks at given average degree. In addition, our analysis is consistent with the modeling result, which provides the theoretical understanding of underlying mechanisms of matching in complex networks.

A minimal model based on individual interactions is proposed to study the non-Poisson statistical properties of human behavior: individuals in the system interact with their neighbors, the probability of an individual acting correlates to its activity, and all individuals involved in action will change their activities randomly. The model creates rich non-Poisson spatial-temporal properties in the activities of individuals, in agreement with the patterns of human communication behaviors. Our findings provide insight into various human activities, embracing a range of realistic social interacting systems, particularly, intriguing bimodal phenomenons. This model bridges priority queues and punctuated equilibrium, and our modeling and analysis is likely to shed light on non-Poisson phenomena in many complex systems., Comment: 17 pages, 7 figures

It is known that the stationary distribution of the random walk process is dependent on the structure of the network. This could provide us a solution of the network reconstruction. However, the stationary distribution of the random walk process can only reflect the relative size of node degrees directly, how to infer the real connection is still a problem. In this paper, we will propose a method to reconstruct network by the random walk process, which can reconstruct the total number of links, degree sequence and links sequentially. In our method, only the stationary distribution is used, and no data of the evolution process is needed, such as the first passage time. We perform our method on some network models and real-world network, the results indicate our method can reconstruct networks accurately, even when we can not get the exact stationary distribution., Comment: 5 pages, 5 figures

We find there is relationship between the associated bigraph and the cluster (or community) detecting on network. By imbedding the associated bigraph of some network (suppose it has cluster structures) into some space, we can identify the clusters on this network, which is a new method for network cluster detecting. And this method, of which the physical meaning is clear and the time complexity is acceptable, may provide us a new point to understand the structure and character of networks. In this paper, We test the methods on serval computer-generated networks and real networks. A computer-generated network with 128 vertexes and the Zachary Network, which presents the structure of a karate club, can be partitioned correctly by these methods. And the Dolphin Network, which presents the relationship between 62 dolphins on the coast of New Zealand, is partitioned reasonably., Comment: 6 pagers,6 figures

The network topology can be described by the number of nodes and the interconnections among them. The degree of a node in a network is the number of connections it has to other nodes and the degree distribution is the probability distribution of these degrees over the whole network. Therefore, the degree is very important structural parameter of network topology. However, given the number of nodes and the degree of each node in a network, the topology of the network cannot be determined. Therefore, we propose the degree-layer theory of network topology to describe deeply the network topology. First, we propose the concept of degree-tree with the breadth-first search tree. The degrees of all nodes are layered and have a hierarchical structure. Second,the degree-layer theory is described in detail. Two new concepts are defined in the theory. An index is proposed to quantitatively distinguish the two network topologies. It also can quantitatively measure the stability of network topology built by a model mechanism. One theorem is given and proved, furthermore, and one corollary is derived directly from the theorem. Third, the applications of the degree-layer theory are discussed in the ER random network, WS small world network and BA scale-free network, and the influences of the degree distribution on the stability of network topology are studied in the three networks. In conclusion, the degree-layer theory is helpful for accurately describing the network topology, and provides a new starting point for researching the similarity and isomorphism between two network topologies., Comment: 6 pages, 4 figures

We propose a universal analysis for static routings on networks and describe the congestion characteristics by the theory. The relation between average transmission time and transmission capacity is described by inequality T0Rc0<=1. For large scale sparse networks, the non-trivial upper bond of transmission capacity Rc0 is limited by Rc0<=1/< 1/k > in some approximate conditions. the theoretical results agree with simulations on BA Networks., Comment: 7pages,4figures

We propose a strategy for achieving maximum cooperation in evolutionary games on complex networks. Each individual is assigned a weight that is proportional to the power of its degree, where the exponent alpha is an adjustable parameter that controls the level of diversity among individuals in the network. During the evolution, every individual chooses one of its neighbors as a reference with a probability proportional to the weight of the neighbor, and updates its strategy depending on their payoff difference. It is found that there exists an optimal value of alpha, for which the level of cooperation reaches maximum. This phenomenon indicates that, although high-degree individuals play a prominent role in maintaining the cooperation, too strong influences from the hubs may counterintuitively inhibit the diffusion of cooperation. We provide a physical theory, aided by numerical computations, to explain the emergence of the optimal cooperation. Other pertinent quantities such as the payoff, the cooperator density as a function of the degree, and the payoff distribution, are also investigated. Our results suggest that, in order to achieve strong cooperation on a complex network, individuals should learn more frequently from neighbors with higher degrees, but only to certain extent., Comment: this work was finished in (Dated: April 22, 2011)

Uncovering the mechanism behind the scaling law in human trajectories is of fundamental significance in understanding many spatio-temporal phenomena. In combination of the exploration and the preferential returns, we propose a simple dynamical model mainly based on the cascading processes to capture the human mobility patterns. By the numerical simulations and analytical studies, we show more than five statistical characters that are well consistent with the empirical observations, including several type of scaling anomalies, and the ultraslow diffusion property, implying the cascading processes associated with the other two mechanisms are indeed a key in the understanding of human mobility activities. Moreover, both of the diverse individual mobility and aggregated scaling move-lengths, bridging the micro and macro patterns in human mobility. Our model provides deeper understandings on the emergence of human mobility patterns., Comment: 9 pages, 6 figures

Facing the threats of infectious diseases, we take various actions to protect ourselves, but few studies considered an evolving system with competing strategies. In view of that, we propose an evolutionary epidemic model coupled with human behaviors, where individuals have three strategies: vaccination, self-protection and laissez faire, and could adjust their strategies according to their neighbors' strategies and payoffs at the beginning of each new season of epidemic spreading. We found a counter-intuitive phenomenon analogous to the well-known \emph{Braess's Paradox}, namely a better condition may lead to worse performance. Specifically speaking, increasing the successful rate of self-protection does not necessarily reduce the epidemic size or improve the system payoff. This phenomenon is insensitive to the network topologies, and can be well explained by a mean-field approximation. Our study demonstrates an important fact that a better condition for individuals may yield a worse outcome for the society., Comment: 17 pages, 5 figures

Human mobility patterns deeply affect the dynamics of many social systems. In this paper, we empirically analyze the real-world human movements based GPS records, and observe rich scaling properties in the temporal-spatial patterns as well as an abnormal transition in the speed-displacement patterns. We notice that the displacements at the population level show significant positive correlation, indicating a cascade-like nature in human movements. Furthermore, our analysis at the individual level finds that the displacement distributions of users with strong correlation of displacements are closer to power laws, implying a relationship between the positive correlation of the series of displacements and the form of an individual's displacement distribution. These findings from our empirical analysis show a factor directly relevant to the origin of the scaling properties in human mobility., Comment: 10 pages, 9 figures

How to promote the innovative activities is an important problem for modern society. In this paper, combining with the evolutionary games and information spreading, we propose a lattice model to investigate dynamics of human innovative behaviors based on benefit-driven assumption. Simulations show several properties in agreement with peoples' daily cognition on innovative behaviors, such as slow diffusion of innovative behaviors, gathering of innovative strategy on "innovative centers", and quasi-localized dynamics. Furthermore, our model also emerges rich non-Poisson properties in the temporal-spacial patterns of the innovative status, including the scaling law in the interval time of innovation releases and the bimodal distributions on the spreading range of innovations, which would be universal in human innovative behaviors. Our model provide a basic framework on the study of the issue relevant to the evolution of human innovative behaviors and the promotion measurement of innovative activities., Comment: 6 pages, 4 figures

We propose a model of two-way selection system. It appears in the processes like choosing a mate between men and women, making contracts between job hunters and recruiters, and trading between buyers and sellers. In this paper, we propose a model of two-way selection system, and present its analytic solution for the expectation of successful matching total and the regular pattern that the matching rate trends toward an inverse proportion to either the ratio between the two sides or the ratio of the state total to the smaller people number. The proposed model is verified by empirical data of the matchmaking fairs. Results indicate that the model well predicts this typical real-world two- way selection behavior to the bounded error extent, thus it is helpful for understanding the dynamics mechanism of the real-world two-way selection system., Comment: 8 pages, 4 figures

Uncovering human mobility patterns is of fundamental importance to the understanding of epidemic spreading, urban transportation and other socioeconomic dynamics embodying spatiality and human travel. According to the direct travel diaries of volunteers, we show the absence of scaling properties in the displacement distribution at the individual level,while the aggregated displacement distribution follows a power law with an exponential cutoff. Given the constraint on total travelling cost, this aggregated scaling law can be analytically predicted by the mixture nature of human travel under the principle of maximum entropy. A direct corollary of such theory is that the displacement distribution of a single mode of transportation should follow an exponential law, which also gets supportive evidences in known data. We thus conclude that the travelling cost shapes the displacement distribution at the aggregated level., Comment: 4 figures, 5 pages

The financial market and turbulence have been broadly compared on account of the same quantitative methods and several common stylized facts they shared. In this paper, the She-Leveque (SL) hierarchy, proposed to explain the anomalous scaling exponents deviated from Kolmogorov monofractal scaling of the velocity fluctuation in fluid turbulence, is applied to study and quantify the hierarchical structure of stock price fluctuations in financial markets. We therefore observed certain interesting results: (i) The hierarchical structure related to multifractal scaling generally presents in all the stock price fluctuations we investigated. (ii) The quantitatively statistical parameters that describes SL hierarchy are different between developed financial markets and emerging ones, distinctively. (iii) For the high-frequency stock price fluctuation, the hierarchical structure varies with different time period. All these results provide a novelty analogy in turbulence and financial market dynamics and a insight to deeply understand the multifractality in financial markets., Comment: 10 pages, 6 Figures

Often, vaccination programs are carried out based on self-interest rather than being mandatory. Owing to the perceptions about risks associated with vaccines and the `herd immunity' effect, it may provide suboptimal vaccination coverage for the population as a whole. In this case, some subsidy policies may be offered by the government to promote vaccination coverage. But, not all subsidy policies are effective in controlling the transmission of infectious diseases. We address the question of which subsidy policy is best, and how to appropriately distribute the limited subsidies to maximize vaccine coverage. To answer these questions, we establish a model based on evolutionary game theory, where individuals try to maximize their personal payoffs when considering the voluntary vaccination mechanism. Our model shows that voluntary vaccination alone is insufficient to control an epidemic. Hence, two subsidy policies are systematically studied: (1) in the free subsidy policy the total amount of subsidies is distributed to some individuals and all the donees may vaccinate at no cost, and (2) in the part-offset subsidy policy each vaccinated person is offset by a certain proportion of the vaccination cost. Simulations suggest that, since the part-offset subsidy policy can encourage more individuals to be vaccinated, the performance of this policy is significantly better than that of the free subsidy policy.