Chaos最新文献

With the widespread application of artificial intelligence, human-machine interaction has become an essential component of social systems. This study investigates human-machine cooperation from an evolutionary game perspective by constructing a mixed spatial prisoner's dilemma environment that integrates reinforcement learning-based machine strategies and traditional reactive human strategies. The results show that machines interacting with tolerant human strategies tend to converge toward stable cooperative patterns and, under certain conditions, significantly enhance group cooperation. The effect of machine proportion is context-dependent: in low-temptation settings, machines strengthen cooperative stability, whereas in high-temptation environments, cooperation relies more on human strategies. Furthermore, the analysis of average Q-values reveals that machine learning not only reproduces conditional cooperation logic but is also deeply shaped by human strategic patterns. These findings highlight the critical role of humans in shaping machine learning and cooperative tendencies, offering new theoretical insights into the evolution of human-machine cooperation and methodological implications for applications such as intelligent manufacturing and autonomous driving.

Understanding how opinion leader characteristics influence network reconstruction represents a critical challenge in computational social science. This study presents a novel framework integrating leader-driven opinion dynamics with compressive sensing, systematically investigating how node centrality, initial opinion, acceptance rate, and opinion homogeneity affect reconstruction accuracy. The extensive experimental results for three real-world networks and three synthetic models show that leaders with lower centrality consistently outperform highly central nodes in network reconstruction. This occurs because high centrality leaders create rapid opinion convergence, reducing the informational diversity essential for accurate reconstruction, while lower centrality leaders preserve richer signal content. Our analysis shows that extremely conservative leaders (o=0.0) with high stubbornness (α=1.0) achieve optimal performance in moderately tolerant communities (ε=0.5), challenging conventional centrality-based leader selection strategies. These findings indicate that effective opinion leadership for network reconstruction requires consideration of dynamics-specific factors beyond traditional structural importance, with significant implications for marketing, public health interventions, and crisis communication applications.

The wide community of researchers has embraced Artificial Neural Networks (ANNs) to optimize several activities that include approximation alongside regression models. The training efficiency of ANNs depends heavily on the methods used to initialize the weights. The suggested weight initialization system develops the Hilbert matrix method to accelerate training convergence. The implementation of Mutual Information (MI) enables feature selection through the MI score ranking of features. The ordered features are distributed across a scaled Hilbert matrix to assign higher weight to higher-ranked elements and lower weight to lower ones, which results in more rapid training efficiency. This work achieves its main innovation through the combination of Mutual Information-based feature ranking together with Hilbert-matrix-based weight initialization procedures. The combined approach produces an initialization technique that advances convergence speed and strengthens learning stability. The experimental evaluation across several datasets established the superiority of the proposed MI-Hilbert weight initialization approach, which offered a better convergence speed while maintaining training stability when using MSE and R2 metrics for assessment.

In contemporary social networks, information is often transmitted through asynchronous, multi-channel environments where individuals participate in both pairwise and group-based interactions. These processes exhibit strong directionality. For example, interactions may occur from influential users to followers or from dominant voices within group discussions, but most existing contagion models rely on undirected, pairwise interactions and overlook both higher-order structure and directional influence. To address this issue, we propose a SAR (susceptible-adopted-recovered) model for information diffusion on directed multiplex higher-order networks. Each layer incorporates both dyadic and group-level interactions, and diffusion proceeds via interlayer alternation across layers. Directionality is embedded in the higher-order structure via a tunable directionality weight that captures heterogeneous influence among group members. Simulation results reveal a non-monotonic dependence of the final diffusion size on the interlayer alternation probability, with suppression emerging under intermediate alternation regimes. Enhancing directional transmission within higher-order structures can mitigate this suppression and facilitate broader diffusion. Theoretical predictions are consistent with simulation outcomes, validating the proposed framework. Our findings highlight the importance of incorporating directional group interactions and interlayer alternation in models of information diffusion, offering new insights into how structural and temporal heterogeneities jointly regulate information diffusion in multilayer social systems.

Understanding how adaptive mechanisms influence species coexistence remains a central issue in evolutionary ecology. In this study, we propose a spatial rock-paper-scissors model that incorporates fitness-driven adaptive competition, where the intensity of interspecific interactions dynamically adjusts according to local environmental fitness. Using extensive Monte Carlo simulations, we systematically explore how the sensitivity parameter (K) and migration rate (M) jointly shape spatial patterns, extinction probabilities, and long-term biodiversity. The results demonstrate that moderate fitness heterogeneity and intermediate dispersal rates favor the persistence of coexistence by stabilizing oscillatory dynamics and maintaining spiral-wave structures. In contrast, excessive sensitivity or mobility leads to spatial homogenization and increased extinction risks. These findings highlight the dual role of adaptability as both a stabilizing and destabilizing force in cyclic competition, offering new theoretical insights into the ecological mechanisms underlying biodiversity maintenance and informing conservation strategies that balance migration and environmental adaptation.

In this paper, we present explicit examples of Denjoy minimal sets that exhibit two- to one-hole transitions at some parameter values 0<ϵ<1 and cantorus to circle transitions as ϵ→1 and collapse to finite sets as ϵ→0. The limit ϵ→0 is an anti-integrable limit in the sense of Aubry. We describe all the transitions in terms of multi-hole Sturmian symbolic systems.

Extending a framework originally proposed by Cencetti et al. [Eur. Phys. J. B 90, 1 (2017)], we investigate a topological control mechanism for mediating Turing patterns on unweighted, undirected networks. Our primary contribution is proving boundedness results involved in targeted destabilization, in which we map Laplacian mode shifts to structural interventions. Through numerical simulations, we show the efficacy of this control scheme on a range of graph models, establishing theoretical expectations for the special case of the ring lattice. This work stands against a backdrop of real-world applications, moving the needle toward better understanding and engineering of network-driven pattern formation.

This paper innovatively constructs a class of Caputo hetero-order fractional predator-prey systems incorporating cannibalism, fear effect, and double time delays. By differentially setting the fractional orders of prey and predators, the heterogeneous memory effects exhibited by both species during their evolutionary processes are characterized. The model simultaneously integrates the cannibalistic behavior of prey and the fear effect induced by predators, and introduces double time delays (fear effect delay and gestation delay), breaking through the limitation of traditional same-order fractional models in describing the memory differences among species. The research adopts a progressive analysis approach: First, for the non-time-delay system, the uniqueness and boundedness of the system's solutions are proved, the existence conditions of the positive equilibrium point are given, and the local stability criterion is established based on the characteristic equation. Furthermore, with the cannibalism rate and fear parameter as bifurcation parameters, the Hopf bifurcation mechanism is analyzed. Second, for the time-delay system, the focus is on analyzing various combinations of time delays; with time delay as the bifurcation parameter, the stability of the equilibrium point and the conditions for Hopf bifurcation are derived. Finally, the correctness of the theoretical results is verified through multiple sets of numerical simulations.

In voluntary vaccination, adaptive adjustments in government subsidy policies play a crucial role in influencing vaccination levels. The Bush-Mosteller model, a type of reinforcement learning, offers an excellent framework to study the decision-making process of the government. In this work, we study how the government adaptively adjusts to different subsidy policies that affect the vaccination level. Here, we incorporate the per capita treatment cost for infections and the per capita subsidy into a Bush-Mosteller model where the former serves as the payoff and the latter defines the aspiration level, and the gap between the payoff and the aspiration level determines whether the stimulus is positive, resulting in continuation of the current policy, or negative, prompting policy adjustment. Our results reveal that while increasing the total subsidy amount can enhance vaccination levels, reducing the relative vaccination costs fails to increase vaccination levels under a fixed subsidy budget. Provided the total subsidy exactly covers vaccination costs, vaccination levels depend on the dominant strategy: dominance of the partial-offset policy results in a decline, dominance of the free subsidy policy leads to an increase, and the coexistence of both policies maintains the initial level. This study sheds light on the role of adaptive subsidy policies driven by reinforcement learning in shaping vaccination dynamics.

Networked systems-from smart grids and autonomous fleets to social networks-are ubiquitous yet complex, with agents interacting amid topological dependencies and challenges like dynamic environments or malicious attacks. Game theory, control theory, and optimization offer tools to model these systems, but bridging theory with real-world complexity remains a key gap. This Chaos Focus Issue tackles this by exploring intelligent game theory in networked systems, featuring 26 papers across four themes: cooperation promotion, distributed systems, complex structures, and game applications. It links theoretical insights (e.g., cooperative dynamics in structured populations) to practical solutions (e.g., epidemic control, infrastructure protection), advancing resilient, efficient networked system design.