The aim of the present manuscript is to investigate a novel perspective in the modeling and control of complex system both in the deterministic and stochastic case, with particular attention to numerical methods, control methodologies and uncertainty quantification. In Chapter 1 we construct a Boltzmann--type control of a consensus dynamics. Based on a microscopic model we design a Boltzmann-type optimal control thanks to the model predictive control (MPC) approach in the case of an instantaneous control. In Chapter 2 we apply the MPC control approach in the context of complex networks where a control acts over a minimum set of nodes/agents influencing the dynamics of the whole network. We observed how the introduction of a suitable selective control depending on the connection degree of each node is capable of driving the overall opinion towards consensus. Chapter 3 regards the derivation of performance bounds for mean--field MPC control with varying horizon. Here we have established a computable and provable bound on the difference in the cost functional for MPC-controlled and optimally controlled dynamics in the case of a large number of agents. In Chapter 4 we developed numerical schemes that conserve structural property of mean--field equations: non--negativity of the solution, entropy dissipation and large time behavior. The methods here developed are second order accurate, they do not require any restriction on the mesh size and are capable to capture the asymptotic steady states with arbitrary accuracy. These properties are essential for a correct description of the underlying physical problem. Applications of the schemes to several nonlinear Fokker-Planck equations describing emerging collective behavior in socio-economic and life sciences are presented. We further applied in Chapter 5 the developed numerical techniques to a multivariate non--local model for opinion dynamics over complex network, where interactions depend on the opinion itself and on the discrete number of connections. It has been proven that for different choices of weights functions the introduced generator of the network may produce stationary scale-free degree distributions as well as uniform random graphs. Further, we observed that the presence of a small portion of highly connected agents may drive the overall dynamics towards their position. In Chapter 6 we further investigate multivariate models with a direct application to decision science. In particular, we introduce and discuss multivariate kinetic models describing the influence of competence in the evolution of decisions in a multiagent system. The exchange mechanism includes the role of the agents’ tendency to behave in the same way as if they were as good, or as bad, as their partner: the so-called equality bias. The presented modeling is inspired by real experiments and reproduces the empirical findings. The role of stochastic quantities in the dynamics is investigated in Chapter 7. Here we analyze the effect of the uncertainty in the interaction parameter in a second-order alignment model of the Cucker-Smale type using a generalized polynomial chaos approach. We observed that the presence of negative tails in the distribution of the random inputs lead to the divergence of the expected velocities of the system of agents, even in the regimes of unconditional flocking. We formalized a selective MPC approach to stabilize the dynamics and to steer the expected velocities toward the desired one, even in the divergence regimes. Chapter 8 is part of an ongoing work. Here we derived mean-field equations from microscopic Cucker-Smale type model dependent on random input in several cases. Further, a novel structure preserving numerical technique based on both Monte Carlo and stochastic Galerkin-gPC methods is addressed. Numerical experiments show that the scheme maintain the spectral accuracy for the statistical quantities of interest, like mean and variance.
Scopo del presente lavoro è proporre un approccio inedito per i modelli di sistemi complessi, sia in caso deterministico che stocastico, con particolare attenzione a metodi numerici, metodologie di controllo e di quantificazione dell’incertezza. Nel Capitolo 1 introduciamo un controllo di tipo Boltzmann per dinamiche di consenso. Tale metodo è basato su equazioni microscopiche a sulla tecnica del model predictive control (MPC) per il controllo instantaneo. Nel Capitolo 2 applichiamo un approccio di tipo MPC nel contesto di reti complesse, dove il controllo agisce sui nodi/agenti influenzando la dinamica dell’intera rete. Osserviamo come l’introduzione di un appropriato termine di selezione, dipendente dal grado di connessione di ogni nodo, riesca a portare l’opinione del sistema di agenti verso il consenso. Il Capitolo 3 riguarda la derivazione di limiti di performace per il controllo MPC nel caso di equazioni di campo medio a orizzonte di controllo variabile. Abbiamo qui dimostrato l’esistenza di tali limiti sulla differenza nel funzionale di costo MPC e di dinamiche con controllo ottimo. Nel Capitolo 4 abbiamo sviluppato schemi numerici per la conservazione delle proprietà strutturali di equazioni di campo medio come: non-negatività, dissipazione dell’entropia e stati di equilibrio. I metodi sono accurati al second’ordine, non richiedono alcuna restrizione di discretizzazione e catturano lo stato stazionario con accuratezza arbitraria. Tali proprietà sono essenziali per una descrizione fisicamente corretta del problema. Abbiamo proposto qui alcune applicazioni a equazioni di tipo Fokker-Planck non-lineari che emergono in modelli di comportamento collettivo tipici delle scienze socio-economiche e della vita. Nel Capitolo 5 gli schemi del capitolo precedente sono applicate a un modello multivariato per dinamiche di opinione su reti complesse, dove le interazioni dipendono dall’opinione stessa e dal numero di connessioni di ogni agente. La generazione di network a partire da un master equation è stata trattata analiticamente dimostrando l’emergenza di distribuzioni stazionarie scale-free o di grafi uniformi. Inoltre, si osserva come la presenza di una porzione di nodi altamente connessi nel grafo forzi la dinamica verso l’opinione di tali agenti. Il Capitolo 6 è dedicato alla costruzione di un modello multivariato per le scienze decisionali. In particolare, introduciamo modelli di tipo cinetico per l’influenza della competenza nell’evoluzione di decisioni in sistemi multi-agente. Il meccanismo di scambio include la tendenza degli agenti a comportarsi come se fossero capaci, o incapaci, al pari del loro partner: fenomeno chiamato equality-bias. Il modello è ispirato da misurazioni empiriche ed è in grado di riprodurre i risultati sperimentali. L’azione di quantità stocastiche sulla dinamica è oggetto del Capitolo 7. Qui consideriamo interazioni soggette a incertezza per il modello di Cucker-Smale utilizzando tecniche di caos polinomiale. Osserviamo come la presenza di code negative nella distribuzione del parametro casuale porti alla divergenza in tempo finito del valore atteso delle velocità del sistema anche in regimi di flocking incondizionato. Un controllo selettivo MPC è poi introdotto per stabilizzare la dinamica e guidare le velocità attese verso una quantità desiderata anche in regimi di divergenza. Il Capitolo 8 è parte di un lavoro in corso di svolgimento. Abbiamo derivato equazioni di tipo campo medio dal modello microscopico di Cucker-Smale con stocasticità. Proponiamo inoltre un nuovo schema numerico che conserva le proprietà strutturali, basato sia su tecniche Monte Carlo che di tipo Galerkin-gPC stocastiche. Test numerici mostrano come tale schema mantenga l’accuratezza spettrale per le quantità statistiche come media e varianza.
Boltzmann-type and mean-field modeling of social dynamics: numerics, control, uncertainty quantification
ZANELLA, Mattia
2017
Abstract
The aim of the present manuscript is to investigate a novel perspective in the modeling and control of complex system both in the deterministic and stochastic case, with particular attention to numerical methods, control methodologies and uncertainty quantification. In Chapter 1 we construct a Boltzmann--type control of a consensus dynamics. Based on a microscopic model we design a Boltzmann-type optimal control thanks to the model predictive control (MPC) approach in the case of an instantaneous control. In Chapter 2 we apply the MPC control approach in the context of complex networks where a control acts over a minimum set of nodes/agents influencing the dynamics of the whole network. We observed how the introduction of a suitable selective control depending on the connection degree of each node is capable of driving the overall opinion towards consensus. Chapter 3 regards the derivation of performance bounds for mean--field MPC control with varying horizon. Here we have established a computable and provable bound on the difference in the cost functional for MPC-controlled and optimally controlled dynamics in the case of a large number of agents. In Chapter 4 we developed numerical schemes that conserve structural property of mean--field equations: non--negativity of the solution, entropy dissipation and large time behavior. The methods here developed are second order accurate, they do not require any restriction on the mesh size and are capable to capture the asymptotic steady states with arbitrary accuracy. These properties are essential for a correct description of the underlying physical problem. Applications of the schemes to several nonlinear Fokker-Planck equations describing emerging collective behavior in socio-economic and life sciences are presented. We further applied in Chapter 5 the developed numerical techniques to a multivariate non--local model for opinion dynamics over complex network, where interactions depend on the opinion itself and on the discrete number of connections. It has been proven that for different choices of weights functions the introduced generator of the network may produce stationary scale-free degree distributions as well as uniform random graphs. Further, we observed that the presence of a small portion of highly connected agents may drive the overall dynamics towards their position. In Chapter 6 we further investigate multivariate models with a direct application to decision science. In particular, we introduce and discuss multivariate kinetic models describing the influence of competence in the evolution of decisions in a multiagent system. The exchange mechanism includes the role of the agents’ tendency to behave in the same way as if they were as good, or as bad, as their partner: the so-called equality bias. The presented modeling is inspired by real experiments and reproduces the empirical findings. The role of stochastic quantities in the dynamics is investigated in Chapter 7. Here we analyze the effect of the uncertainty in the interaction parameter in a second-order alignment model of the Cucker-Smale type using a generalized polynomial chaos approach. We observed that the presence of negative tails in the distribution of the random inputs lead to the divergence of the expected velocities of the system of agents, even in the regimes of unconditional flocking. We formalized a selective MPC approach to stabilize the dynamics and to steer the expected velocities toward the desired one, even in the divergence regimes. Chapter 8 is part of an ongoing work. Here we derived mean-field equations from microscopic Cucker-Smale type model dependent on random input in several cases. Further, a novel structure preserving numerical technique based on both Monte Carlo and stochastic Galerkin-gPC methods is addressed. Numerical experiments show that the scheme maintain the spectral accuracy for the statistical quantities of interest, like mean and variance.File | Dimensione | Formato | |
---|---|---|---|
Fronte_firma.pdf
Open Access dal 05/04/2018
Dimensione
1.84 MB
Formato
Adobe PDF
|
1.84 MB | Adobe PDF | Visualizza/Apri |
Tesi.pdf
Open Access dal 05/04/2018
Dimensione
28.19 MB
Formato
Adobe PDF
|
28.19 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/74469
URN:NBN:IT:UNIFE-74469