Natural language is a remarkable example of a complex dynamical system which combines variation and universal structure emerging from the interaction of millions of individuals. Understanding statistical properties of texts is not only crucial in applications of information retrieval and natural language processing, e.g. search engines, but also allow deeper insights into the organization of knowledge in the form of written text. In this thesis, we investigate the statistical and dynamical processes underlying the co-existence of universality and variability in word statistics. We combine a careful statistical analysis of large empirical databases on language usage with analytical and numerical studies of stochastic models. We find that the fat-tailed distribution of word frequencies is best described by a generalized Zipf’s law characterized by two scaling regimes, in which the values of the parameters are extremely robust with respect to time as well as the type and the size of the database under consideration depending only on the particular language. We provide an interpretation of the two regimes in terms of a distinction of words into a finite core vocabulary and a (virtually) infinite noncore vocabulary. Proposing a simple generative process of language usage, we can establish the connection to the problem of the vocabulary growth, i.e. how the number of different words scale with the database size, from which we obtain a unified perspective on different universal scaling laws simultaneously appearing in the statistics of natural language. On the one hand, our stochastic model accurately predicts the expected number of different items as measured in empirical data spanning hundreds of years and 9 orders of magnitude in size showing that the supposed vocabulary growth over time is mainly driven by database size and not by a change in vocabulary richness. On the other hand, analysis of the variation around the expected size of the vocabulary shows anomalous fluctuation scaling, i.e. the vocabulary is a nonself-averaging quantity, and therefore, fluctuations are much larger than expected. We derive how this results from topical variations in a collection of texts coming from different authors, disciplines, or times manifest in the form of correlations of frequencies of different words due to their semantic relation. We explore the consequences of topical variation in applications to language change and topic models emphasizing the difficulties (and presenting possible solutions) due to the fact that the statistics of word frequencies are characterized by a fat-tailed distribution. First, we propose an information-theoretic measure based on the Shannon-Gibbs entropy and suitable generalizations quantifying the similarity between different texts which allows us to determine how fast the vocabulary of a language changes over time. Second, we combine topic models from machine learning with concepts from community detection in complex networks in order to infer large-scale (mesoscopic) structures in a collection of texts. Finally, we study language change of individual words on historical time scales, i.e. how a linguistic innovation spreads through a community of speakers, providing a framework to quantitatively combine microscopic models of language change with empirical data that is only available on a macroscopic level (i.e. averaged over the population of speakers).

Many of the present problems that we are facing arise as unanticipated side-effects of our own actions. Moreover, the solutions implemented to solve important problems often create new problems. To avoid unintended consequences, understanding complex systems is essential in devising policy instruments and in improving environmental management. Thus, this thesis investigated systems modeling approaches to under- stand complex systems and monitor the environmental performance of management actions. The overall aim of the work was to investigate the usefulness of different systems modeling approaches in supporting environmental management. A driver- based, pressure-oriented approach was adopted to investigate systems modeling tools. Material/substance flow analysis, environmental footprinting, input-output analysis, process-based dynamic modeling, and systems dynamics modeling approaches were applied in different cases to investigate strengths and weaknesses of the tools in generating an understanding of complex systems. Three modeling and accounting approaches were also tested at different systems scales to support environmental mon- itoring. Static modeling approaches were identified as fundamental to map, account, and monitor physical resource metabolism in production and consumption systems, whereas dynamic modeling showed strengths in understanding complex systems. The results suggested that dynamic modeling approaches should be conducted on top of static analysis to understand the complexity of systems when devising and testing policy instruments. To achieve proactive monitoring, a pressure-based assessment was proposed instead of the mainstream impact/state-based approach. It was also concluded that the LCA community should shift the focus of its assessments to pressures instead of impacts. / Många nuvarande miljö- och utvecklingsproblem har uppstått som oförutsedda biverkningar av människans egna handlingar. De lösningar som prövats har i sin tur ofta skapat  nya problem. Det därför viktigt att förstå hur komplexa system fungerar och att utforma styrmedel och ledningssystem som minimerar risken för oönskade bieffekter. Den här avhandling har använt olika modelleringsmetoder för att öka förståelsen för komplexa system och bidra med kunskaper om hur miljöprestanda och förvaltningsåtgärder kan följas upp på ett mer effektivt sätt. Det övergripande syftet med arbetet var att undersöka användbarheten av olika modelleringsmetoder för att effektivisera den fysiska resurshanteringen i samhället. I arbetet har ett flödesbaserat och aktörsinriktat arbetssätt (pressure based and driver oriented approach) använts i modelleringen.  Material- och substansflödesanalys, miljöfotavtryck, input-output analys, processbaserad dynamisk modellering och systemdynamiska modelleringsmetoder studerades för att undersöka styrkor och svagheter hos de olika metoderna/verktygen.  Tre olika modellerings- och redovisningsmetoder för att stödja miljöövervakning testades också i olika systemskalor. Statiska modelleringsmetoder (räkenskaper) identifierades som grundläggande för att kartlägga, kontoföra och övervaka den fysiska resursmetabolismen i produktions- och konsumtionssystem, medan dynamisk modellering visade sin styrka i att skapa förståelse för komplexa system. Resultaten pekar på att dynamiska modelleringsmetoder bör användas som ett komplement till statiska analyser för att förstå komplexiteten i systemen när man utformar och testar styrmedel. För att uppnå proaktiv övervakning bör flödesbaserade räkenskaper utnyttjas i större utsträckning i stället för den vanliga tillstånds- och påverkansövervakningen (state/impact monitoring). En viktig slutsats är därför att LCA-samfundet bör flytta fokus i sina bedömningar från påverkan till flöden. / <p>QC 20160830</p>

L'objectif de ce thèse est l'étude des systèmes dynamiques avec interaction à longue portée. La complexité de leur dynamique met en évidence des propriétés contre-intuitives et inattendues, comme l'existence d'états stationnaires hors-équilibre (QSS). Dans le QSS on peut observer des propriétés particulières: chaleur spécifique négative, inéquivalence des ensembles statistiques et phénomènes d'auto-organisation. Les théories des interactions LR ont été appliquées pour décrire la dynamique des systèmes auto-gravitants, de tourbillons bidimensionnels, de systèmes avec interactions onde-particule et des plasmas chargés. Mon travail s'est tout d'abord consacré à l'extension de la solution de Lynden-Bell pour le modèle HMF, en généralisant l'analyse à des conditions initiales de «water-bag" à plusieurs niveaux, qui approchent des conditions initiales continues. En suite je me suis intéressé à la caractérisation formelle de la thermodynamique des QSS dans l'ensemble statistique canonique. En appliquant la théorie standard, il est possible de mesurer une chaleur spécifique "cinétique'' négative. Cette propriété inattendue amène à la violation du second principe de la thermodynamique. Un tel résultat nous pousse à reconsidérer l'applicabilité de la théorie thermodynamique actuelle aux systèmes LR. En suite j'ai étudié, pour le modèle α-HMF, la persistance des caractéristiques typiques du régime LR, dans le limite dynamique à courte portée. Les résultats suggèrent une généralisation de la définition des systèmes LR. Le dernier chapitre est consacré à la caractérisation d'un nouveau modèle LR, extension naturelle du précédent α-HMF et d'intérêt potentiel applicatif. / The scope of this thesis is the study of systems with long-range interactions (LR). The complexity of their dynamics evidences counter-intuitive and unexpected properties, as for instance the existence of out-of-equilibrium stationary states (QSS). Considering a system in the QSS, one may observe peculiar properties, as negative specific heat, statistical ensemble inequivalence and phenomena of self-organizations. The main theories of long-range interactions have been applied to describing self-gravitating systems, two-dimensional vortices, systems with wave-particle interactions and charged plasmas. My work has been initially dedicated to extending the Lynden-Bell solution for the HMF model, generalizing the analysis to multi-level water-bag initial condition that could approximate continuous distributions. Then I concentrated to the formal characterization of the thermodynamics of QSS in the canonical statistical ensemble. By applying the standard theory, it is possible to measure negative “kinetic” specific heat. This latter unexpected property leads to a violation of the second principle of thermodynamics. Such result forces us to reconsider the applicability of the accepted thermodynamic theory to LR systems. Afterwards I studied, in the context of the α-HMF model, the persistence of the typical characteristics of the LR regime in the limit of short-range dynamics. The results obtained suggests a generalization of the definition of LR systems. The last chapter is dedicated to the characterization of a novel LR model, a natural extension of α-HMF and of potential applicability.

Dans cette thèse, conjointement au développement d'un modèle de négociation bilatérale automatisée entre agents, dans un contexte à informations incomplètes, intégrant les effets de la personnalité du comportement humain sur le processus et sur les résultats de la négociation, nous proposons une architecture de tels agents (″vendeur ″ ou ″acheteur″). Pour cela, une nouvelle approche de génération d'offres a été présentée en proposant notamment trois familles de tactiques adaptatives (par rapport aux intervalles de réservation et aux dates limites), à savoir : les tactiques dépendantes du temps (supposé continu), les tactiques dépendantes du comportement et les tactiques indépendantes du temps. Cette thèse s'intéresse aussi à la prise en considération des effets de la personnalité (de l'agent négociateur) sur les issues de la négociation. En fait, en recourant à modèle appelé ‘‘cinq grands facteurs de la personnalité'' et en introduisant des orientations cognitives, nous avons développé une architecture d'agent négociant basé sur la personnalité. Notre architecture s'inspire principalement de la théorie des jeux. En effet, la connaissance de l'agent artificiel en termes de la négociation est considérée comme une certaine orientation mentale du négociateur favorisant les concessions de ce dernier vers l'un des trois équilibres (au sens de la théorie des jeux) suivants : Gagnant-Perdant, Perdant-Gagnant, ou Gagnant-Gagnant. Selon l'orientation privilégiée et la personnalité du négociateur, un tel agent négociateur décide de la combinaison adéquate des tactiques (modèles etc.) afin de moduler, en conséquence, les issues escomptées de la négociation / In this thesis, besides the developing a bilateral automated negotiation model between agents, in incomplete information state, integrating the personality effects of human on the negotiation process and outcomes, we proposed an architecture of such agents (“buyer” or “seller”). To do so, a new offer generation approach of three adaptive families of tactics has been proposed as follows : the time dependent tactics (time supposed as continuous), behavior dependent tactics, and time independent tactics.This thesis takes into consideration also the personality effects (of negotiator agent) on negotiation process and outcome. In fact, with regard to “Big five” personality model and introducing the cognitive orientations, we have developed a negotiator agent's architecture based on personality. This architecture is, mainly, inspired from the game theory. In fact, the artificial agent's cognition in terms of negotiation is considered as a certain negotiator's mental orientation favorising the concession of the negotiator agent towards one of following three equilibria (based on game theory) : Win-Lose, Lose-Win, and Win-Win According to the privileged orientation and the personality of negotiator, such a negotiator agent decides the adequate combination of tactics (models, etc) in order to modulate, consequently, the expected outcomes of negotiation

Ces dernières années, la puissance croissante des ordinateurs a permis à la fois de rassembler une quantité sans précédent de données décrivant la société moderne et d'envisager des outils numériques capables de s'attaquer à l'analyse et la modélisation les processus dynamiques qui se déroulent dans cette réalité complexe. Dans cette perspective, l'approche quantitative de la physique est un des catalyseurs de la croissance de nouveaux domaines interdisciplinaires visant à la compréhension des systèmes complexes techno-sociaux. Dans cette thèse, nous présentons dans cette thèse un cadre théorique et numérique pour simuler des épidémies de maladies infectieuses émergentes dans des contextes réalistes. Dans ce but, nous utilisons le rôle crucial de la mobilité des agents dans la diffusion des maladies infectieuses et nous nous appuyons sur l'étude des réseaux complexes pour gérer les ensembles de données à grande échelle décrivant les interconnexions de la population mondiale. En particulier, nous abordons deux différents problèmes de santé publique. Tout d'abord, nous considérons la propagation d’une épidémie au niveau mondial, et présentons un modèle de mobilité (GLEAM) conçu pour simuler la propagation d'une maladie de type grippal à l'échelle globale, en intégrant des données réelles de mobilité dans le monde entier. La dernière pandémie de grippe H1N1 2009 a démontré la nécessité de modèles mathématiques pour fournir des prévisions épidémiques et évaluer l'efficacité des politiques d'interventions. Dans cette perspective, nous présentons les résultats obtenus en temps réel pendant le déroulement de l'épidémie, ainsi qu'une analyse a posteriori portant sur les stratégies de lutte et sur la validation du modèle. Le deuxième problème que nous abordons est lié à la propagation de l'épidémie sur des systèmes en réseau dépendant du temps. En particulier, nous analysons des données décrivant les mouvements du bétail en Italie afin de caractériser les corrélations temporelles et les propriétés statistiques qui régissent ce système. Nous étudions ensuite la propagation d'une maladie infectieuse, en vue de caractériser la vulnérabilité du système et de concevoir des stratégies de contrôle. Ce travail est une approche interdisciplinaire qui combine les techniques de la physique statistique et de l'analyse des systèmes complexes dans le contexte de la mobilité des agents et de l'épidémiologie numérique. / In recent years, the increasing availability of computer power has enabled both to gather an unprecedented amount of data depicting the global interconnections of the modern society and to envision computational tools able to tackle the analysis and the modeling of dynamical processes unfolding on such a complex reality. In this perspective, the quantitative approach of Physics is catalyzing the growth of new interdisciplinary fields aimed at the understanding of complex techno-socio-ecological systems. By recognizing the crucial role of host mobility in the dissemination of infectious diseases and by leveraging on a network science approach to handle the large scale datasets describing the global interconnectivity, in this thesis we present a theoretical and computational framework to simulate epidemics of emerging infectious diseases in real settings. In particular we will tackle two different public health related issues. First, we present a Global Epidemic and Mobility model (GLEaM) that is designed to simulate the spreading of an influenza-like illness at the global scale integrating real world-wide mobility data. The 2009 H1N1 pandemic demonstrated the need of mathematical models to provide epidemic forecasts and to assess the effectiveness of different intervention policies. In this perspective we present the results achieved in real time during the unfolding of the epidemic and a posteriori analysis on travel related mitigation strategies and model validation. The second problem that we address is related to the epidemic spreading on evolving networked systems. In particular we analyze a detailed dataset of livestock movements in order to characterize the temporal correlations and the statistical properties governing the system. We then study an infectious disease spreading, in order to characterize the vulnerability of the system and to design novel control strategies. This work is an interdisciplinary approach that merges statistical physics techniques, complex and multiscale system analysis in the context of hosts mobility and computational epidemiology.

La principale limitation du modèle connexionniste (réseau neuronal artificiel) CMAC (Cerebellar Model Articulation Controller) et son applicabilité à la résolution des problèmes inhérents au systèmes automatisés complexes (robots, véhicules autonomes, etc.) est liée à la taille de mémoire requise par ce type de modèle connexionniste. Il est pertinent de rappeler que la capacité de mémoire exigée par le CMAC dépend, en premier lieu, de la précision de la quantification des signaux d'entrée, puis, de la dimension de l'espace des entrées (espace caractéristique) du système modélisé. Dans le cas des applications requérant une exécution en temps-réel, la tendance est à la réduction de l'espace caractéristique (aussi petit que possible) et la précision (de la quantification : aussi faible que possible). Cependant, souvent, les systèmes complexes impliquent plusieurs entrées. Pour résoudre le problème inhérent à cet antagonisme et la taille de la mémoire, nous nous sommes intéressés, dans la présente thèse, à l'influence des paramètres intervenant dans la précision de la quantification et dans la capacité de la généralisation sur la qualité d'approximation du modèle CMAC. L'objectif escompté était d'arriver à des structures optimales du CAMC pour le contrôle des systèmes dynamiques complexes. Les robots bipèdes (humanoïdes) et des véhicules volants hypersoniques sont deux domaines d'applications très actuelles impliquant des systèmes complexes. Nous avons appliqué des concepts étudiés aux problèmes soulevés par les deux domaines précités. Des résultats obtenus à partir de la simulation ont montré que des structures optimales ou quasi-optimales conduisant à une réduction sensible d'erreur de modélisation peuvent être obtenues. Ces résultats ont montré que les choix effectués dans l'optimisation de la structure permet une réduction de la taille de la mémoire requise (par le CMAC) et une réduction du temps d'exécution à la fois / The main limitation of the CMAC (Cerebellar Model Articulation Controller) network in realistic applications for complex automated systems (robots, automated vehicles, etc…) is related to the required memory size. It is pertinent to remind that the memory used by CMAC depends firstly on the input signal quantification step and secondly on the input space dimension. For real CMAC based control applications, on the one hand, in order to increase the accuracy of the control the chosen quantification step must be as small as possible; on the other hand, generally the input space dimension is greater than two. In order to overcome the problem relating the memory size, how both the generalization and step quantization parameters may influence the CMAC's approximation quality has been discussed. Our goal is to find an optimal CMAC structure for complex dynamic systems' control. Biped robots and Flight control design for airbreathing hypersonic vehicles are two actual areas of such systems. We have applied the investigated concepts on these two quite different areas. The presented simulation results show that an optimal or sub-optimal structure carrying out a minimal modeling error could be achieved. The choice of an optimal structure allows decreasing the memory size and reducing the computing time as well

Les problèmes de la classification ont reçu une attention considérable dans des différents champs d'ingénierie comme traitement des images biomédicales, identification a partir de la voix, reconnaissance d'empreinte digitale etc. Les techniques d'intelligence artificielles, incluant les réseaux de neurones artificiels, permettent de traiter des problèmes de ce type. En particulier, les problèmes rencontrés nécessitent la manipulation de bases de données de tailles très importantes. Des structures de traitement adaptatives et exploitant des ensembles de classificateurs sont utilisées. Dans cette thèse, nous décrivons principalement le développement et des améliorations apportées à un outil de classification désigné par le terme Tree-like Divide to Simplify ou T-DTS. Nos efforts se sont portés sur l'un des modules de cet outil, le module d'estimation de complexité. L'architecture de l'outil T-DTS est très flexible et nécessite le choix d'un nombre important de paramètres. Afin de simplifier l'exploitation de T-DTS, nous avons conçu et développé une procédure automatique d'optimisation d'un de ces plus importants paramètres, le seuil de décision associé à la mesure de complexité. La contribution principale de cette thèse concerne le développement de modules pouvant s'implanté sur une architecture de calcul matérielle parallèle. Ceci permet de se rapprocher d'une implantation purement matérielle de l'outil T-DTS / Classification problems deal with separating group of objects into sets of smaller classes; this set of problems have received considerable attention in diverse engineering fields such as biomedical imaging, speaker identification, fingerprint recognition, etc. Several effective approaches for automated classification were suggested based on artificial intelligence techniques, including neural networks. Still, one of the major challenges faced by these approaches is a large scale of data required for successful classification. In this thesis, we explore a possible solution to this problem based on a module-based Tree-like Divide to Simplify (T-DTS) classification model. We focus on enhancing the key module of this approach - complexity estimation module. Furthermore, we provide an automated procedure for optimizing key complexity estimation parameters of the T-DTS model; this considerably improves usability and allows for a more effective configuration of decomposition reasoning of the approach. Another major contribution of this work employs further development of T-DTS modules that could be implemented using parallel computer architecture, thereby allowing T-DTS to utilize an underlying hardware to the fullest extent

In public policies, agents are part of an emergent and complex context, reason for which their actions should not be examined in isolation. The state of an agent is influenced by the state of others, in an environment where feedback is continuous and full of interactions. These characteristics result in a system where the total is more unpredictable and dazzling than the mere sum of its parts. As a result, there are a growing number of studies that use typical methods of complex systems to analyze public policies in various areas, such as healthcare, education, crime prevention, energy resources and others. Moreover, such distinct approach allows for more accessible investigations of public policy models, including policies that were not evaluated ex ante from the traditional lenses. This research had two main objectives: to verify how complex systems apply to the context of public policies theoretically and to present a practical application of a model, which was built based upon a case study. Since there is not a clear comprehension on how complex systems could benefit policy makers, this study presents, in its first part, a systematic literature review including some existing applications and the benefits of complexity science in the policy arena. On the whole, it can be asserted that there is a strong consensus that complex systems can be highly beneficial for policy makers and, consequently, for the overall population. Researchers perceive different benefits, such as the opportunity of testing policies a priori, the possibility of comparing different policies for the same topic, and the contemplation of new ideas and insights for better policy formulation. Although there are several simulations and models proposed for public policies in several areas, it lacks an empirical demonstration that effectively proves the benefits of applying complex systems in public policies, i.e., apparently, there are obstacles that prevent such models from having effects in the real world. In this way, the second part of the research presents an agent-based model that can be applied empirically in a government agency: a regulatory body. Such model allows policy makers to compare different enforcement strategies and anticipate side effects that would be difficult to predict without the use of simulations. In this sense, the objective of the second part of this research was to build an agent-based model of a public policy and for which a practical implementation could be carried out. Therefore, a public policy from a professional regulatory board in the healthcare area was chosen, for which two different strategies were tested, with the objective of comparing their efficiency and effectiveness. Such strategies were modeled and simulated with the use of Netlogo software with different scenarios. Results indicate that agent based models can serve as predictive tools for comparing and improving inspection strategies, and also as source of insights for anticipating unintended consequences that would hardly be noticed ex ante without the use of simulation tools / Em políticas públicas, as ações dos agentes envolvidos não podem ser analisadas de forma isolada. O estado de um agente é influenciado pelo estado dos demais, num ambiente em que o feedback é contínuo e repleto de interações. Essas características resultam num sistema onde o total é mais imprevisível e deslumbrante do que a mera soma de suas partes. Com isso, há um crescente número de estudos que utilizam métodos típicos de sistemas complexos para analisar políticas públicas de diversas áreas, como saúde pública, educação, segurança, recursos energéticos e outros. Além disso, essa forma diferente de abordagem permite que alguns modelos de políticas públicas sejam investigados com mais facilidade, incluindo políticas que sequer eram analisadas pelo prisma tradicional. Esta pesquisa teve dois objetivos principais: verificar como os sistemas complexos se aplicam às políticas públicas no campo teórico e apresentar uma aplicação prática de modelagem dentro do contexto de um estudo de caso. Como ainda não há um entendimento sistematizado sobre como sistemas complexos podem ser úteis em políticas públicas, este estudo apresenta, em sua primeira parte, uma revisão sistemática de literatura para uma melhor compreensão de como essas aplicações ocorrem e de quais benefícios essa ciência, de fato, pode trazer. Em decorrência desse estudo, pode-se afirmar que há consenso, na literatura, de que a teoria da complexidade é benéfica para formuladores de políticas e, consequentemente, para a população em geral. Tais benefícios são vistos de diversas formas pelos pesquisadores, como, por exemplo, a possibilidade de se testar políticas a priori, a possibilidade de se comparar diversos tipos de políticas para um mesmo problema e a obtenção de novas perspectivas e ideias para formulação de políticas. Apesar de haver diversas simulações e modelos propostos para políticas públicas em diversas áreas, não foi constatada uma demonstração empírica que comprove efetivamente o benefício de se aplicar sistemas complexos em políticas públicas, ou seja, aparentemente há obstáculos que impedem esses modelos terem efeitos nas políticas de facto. Dessa maneira, o objetivo da segunda parte da pesquisa foi o de construir um modelo baseado em agentes relacionado a uma política pública e cuja implementação prática fosse factível. Assim, foi selecionada uma política relacionada a um órgão público de fiscalização do exercício profissional (conselho de classe), especificamente na área da saúde, para a qual foram traçadas duas estratégias diferentes, com o objetivo de compará-las em termos de eficácia e de efetividade. Essas estratégias foram modeladas e simuladas em software específico de modelos baseados em agentes para análise dos resultados considerando diversos cenários possíveis. Os resultados indicam que os modelos baseados em agentes podem auxiliar o formulador de políticas a comparar diferentes estratégias de fiscalização e antecipar efeitos colaterais que dificilmente seriam constatados ex ante sem a utilização de simulações

Neste trabalho apresentamos uma análise Bayesiana para confiabilidade de sistemas de redes usando métodos de simulação de Monte Carlo via Cadeias de Markov. Assumimos diferentes densidades a priori para as confiabilidades dos componentes individuais, com o objetivo de obtermos sumários de interesse. A metodologia é exemplificada condiderando um sistema de rede com sete componentes e um caso especial de sistema complexo composto por nove componentes. Consideramos ainda confiabilidade de redes tipo k-out--of-m com alguns exemplos numéricos / In this work we present a Bayesian approach for network reliability systems using Marov Chain Monte Carlo methods. We assume different prior densities for the individual component reliabilities th to get the posterior summaries of interest. The methodology is exemplified considering a network system with seven components and a special case of complex system with nine components. We also consider k-out-of-m system reliabiility with some numerical examples

Introdução: Grande parte das ações para promover a atividade física no lazer em populações tem apresentado tamanhos de efeito pequenos ou inexistentes, ou resultados inconsistentes. Abordar o problema a partir da perspectiva sistêmica pode ser uma das formas de superar esse descompasso. Objetivo: Desenvolver um modelo baseado em agentes para investigar a conformação e evolução de padrões populacionais de atividade física no lazer em adultos a partir da interação entre atributos psicológicos dos indivíduos e atributos dos ambientes físico construído e social em que vivem. Métodos: O processo de modelagem foi composto por três etapas: elaboração de um mapa conceitual, com base em revisão da literatura e consulta com especialistas; criação e verificação do algoritmo do modelo; e parametrização e análise de consistência e sensibilidade. Os resultados da revisão da literatura foram consolidados e relatados de acordo com os domínios da busca (aspectos psicológicos, ambiente social e ambiente físico construído). Os resultados quantitativos da consulta com os especialistas foram descritos por meio de frequências e o conteúdo das respostas questões abertas foi analisado e compilado pelo autor desta tese. O algoritmo do modelo foi criado no software NetLogo, versão 5.2.1., seguindo-se um protocolo de verificação para garantir que o algoritmo fosse implementado acuradamente. Nas análises de consistência e sensibilidade, utilizaram-se o Teste A de Vargha-Delaney, coeficiente de correlação de postos parcial, boxplots e gráficos de linha e de dispersão. Resultados: Definiram-se como elementos do mapa conceitual a intenção da pessoa, o comportamento de pessoas próximas e da comunidade, e a percepção da qualidade, do acesso e das atividades disponíveis nos locais em que atividade física no lazer pode ser praticada. O modelo representa uma comunidade hipotética contendo dois tipos de agentes: pessoas e locais em que atividade física no lazer pode ser praticada. As pessoas interagem entre si e com o ambiente construído, gerando tendências temporais populacionais de prática de atividade física no lazer e de intenção. As análises de sensibilidade indicaram que as tendências temporais de atividade física no lazer e de intenção são altamente sensíveis à influência do comportamento atual da pessoa sobre a sua intenção futura, ao tamanho do raio de percepção da pessoa e à proporção de locais em que a atividade física no lazer pode ser praticada. Considerações finais: O mapa conceitual e o modelo baseado em agentes se mostraram adequados para investigar a conformação e evolução de padrões populacionais de atividade física no lazer em adultos. A influência do comportamento da pessoa sobre a sua intenção, o tamanho do raio de percepção da pessoa e a proporção de locais em que a atividade física no lazer pode ser praticada são importantes determinantes da conformação e evolução dos padrões populacionais de atividade física no lazer entre adultos no modelo. / Introduction: Most of the actions to promote leisure-time physical activity at the population level has shown small or null effect sizes, or inconsistent results. Approaching the problem from a systemic perspective can be one way to overcome this gap. Objective: To develop an agent-based model to explore how the interaction between psychological traits and built and social environments leads to the formation and evolution of leisure-time physical activity population patterns in adults. Methods: The modeling process consisted of three stages: development of a conceptual map, based on literature review and expert-based consultation; creation and verification of the models algorithm; and parametrization and consistency and sensitivity analyses. The results of the literature review were consolidated and reported by search domain (psychological, social and built environment aspects). The quantitative results of the expert-based consultation were described using frequency and the content of the open ended questions were analyzed and compiled by the author. The models algorithm has been created using NetLogo, version 5.2.1., following a verification protocol to ensure that the algorithm has been implemented accurately. The Vargha-Delaney test, partial rank correlation coefficients, boxplots, and line and scatterplot graphs were used during the consistency and sensitivity analysis. Results: The elements of the conceptual map are the persons intention, the behavior of the persons social network and the community at large, and the perception of quality, access and available activities of places where leisure-time physical activity is practiced. The model is a hypothetical community with containing two types of agents: people and places where leisure-time physical activity is practiced. People interact with each other and with the built environment, generating population temporal trends of leisure-time physical activity and intention. Sensitivity analysis indicated that the temporal trends of leisure-time physical activity and intention are highly sensitive to the influence of the persons current behavior on her future intention, the persons perception radius size, and the proportion of places where leisure-time physical activity leisure is practiced. Final considerations: The conceptual map and the agent-based model are suitable to investigate the formation and evolution of leisure-time physical activity population patterns in adults. The influence of the persons behavior on her intention, the persons perception radius size, and the proportion of places where leisure-time physical activity leisure is practiced seem to be important determinants of the formation and evolution of population patterns of leisure-time physical activity in this model.