Teoria de Valores Extremos Aplicada a Redes Complexas

• Ano de defesa: 2013
• Autor(a) principal: Borges, Rafael Ribaski
• Orientador(a): Pinto, Sandro Ely de Souza
• Banca de defesa: Viana, Ricardo Luiz , Rodrigues Junior, Pedro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: UNIVERSIDADE ESTADUAL DE PONTA GROSSA
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciências
• Departamento: Fisica
• País: BR
• Palavras-chave em Português: eventos extremos; caminhada aleatória; redes complexas
• Palavras-chave em Inglês: extreme events; random walk; complex networks
• Área do conhecimento CNPq: CNPQ::CIENCIAS EXATAS E DA TERRA::FISICA
• Link de acesso: http://tede2.uepg.br/jspui/handle/prefix/905

The extreme value theory is a branch of statistics and probability. It deals with the asymptotic distributions of extreme values (maximum or minimum) temporal series. The events which takes the average values removed are classified as extreme events. Examples include natural disasters such as goods, earthquakes or an event that causes a strong impact on society. Considering the scenario of complex networks, some examples of extreme events are congestion in networks of roads, power outages in power transmission networks and web servers congested. Thus, understanding the mechanisms that occur in such events is of great interest, because the prediction of these occurrences can minimize its efects, or even avoid them. Thus, the objectives of this study were: 1) to describe the asymptotic behavior of exceedances of a threshold specified by the generalized extreme value distribution, 2) extend the study to the probability of extreme events in complex networks with random topology, small world and scale free. This work was carried out by simulations of random walk pattern and shorter paths. The results shows that for the nodes, also called vertices or sites with low connectivity (lesser degree) in the networks analyzed, the distribution of excesses is not of exponential type. This implies that this distribution is bounded above. The results for the nodes with higher degree were similar, but only for the scale-free network this behavior does not occur. This is due to the fact that the number of exceedances observed in this case is signicantly smaller than the other. It was checked analytically and numerically simulated by random walk pattern, the probability of extreme event is larger and the average time between them is smaller for nodes with lower degree when compared with nodes with higher degree. The spectrum of eigenvalues of the adjacency matrix of the network, which describes the links between nodes, provides conditions for a good agreement between the analytical results and the simulations. For simulations of random walk for shorter paths it was found that nodes with lower betweenness centralities are more likely to have extreme events.