A large number of physical problems can be modeled by means of a system of coupled non-linear parabolic equations. When the unknowns of such system represent quantities that must comply with a set of inequality constraints, it is crucial that numerical solution of the system remain within admissible domain at every step of the simulation, like, for example, a problem with unknowns that are concentrations, which have to remain non-negative. In this thesis we choose a relevant instance of the above class of problems, namely a model for tumor growth introduced by Perthame, Lorz and Lorenzi, and develop an efficient solution algorithm for its calculation that preserves the positivity of the unknown cell concentration through the simulations pro- cess. We adopt a first-order time-stepping procedure with automatic time- step selection. For the spatial discretization we used a finite element scheme that relies on a balanced quad-tree mesh. The computational domain is or- dered and partitioned by nodes. The resulting system of nonlinear algebraic equations is solved with a projected inexact quasi-Newton-Krylov method, globalized by compiling it with a gradient descent direction. The accuracy, robustness and scalability of the algorithm is observed by numerical experiments.
Un gran numero di problemi fisici possono essere modellati per mezzo di un sistema di equazioni accoppiate non-lineari paraboliche. Quando le incognite di tale sistema rappresentano quantitá che devono soddisfare un insieme di vincoli di disuguaglianza, é cruciale che la soluzione numerica del sistema rimanga in un dominio ammissibile ad ogni passo della simulazione come, per esempio, un problema con incognite che sono concentrazioni, che devono permanere non-negative. In questa tesi scegliamo un caso rilevante della suddetta classe di problemi, ossia un modello per la crescita dei tumori introdotto da Perthame, Lorz e Lorenzi, e sviluppiamo un efficiente algoritmo risolutivo per il suo calcolo, che preservi la positivitá della concentrazione incognita di cellule attraverso il processo di simulazione. Adottiamo una procedura di selezione automatica di passo temporale di primo ordine. Per la discretizzazione spaziale usiamo uno schema a elementi finiti che si basa su una mesh quad-tree. Il dominio computazionale é ordinato e partizionato rispetto ai nodi nodi. Il sistema risultante di equazioni algebriche non-lineari é risolto con un metodo proiet- tato inesatto di quasi-Newton-Krylov, globalizzato per mezzo dell’uso della direzione di discesa del gradiente. L’accuratezza, robustezza e scalabilitá dell’algoritmo sono osservate attraverso esperimenti numerici.
A scalable positivity preserving solution scheme for coupled nonlinear advection-diffusion-reaction problems
AVRAMOVA, TEMENUZHKA VALENTINOVA
2017/2018
Abstract
A large number of physical problems can be modeled by means of a system of coupled non-linear parabolic equations. When the unknowns of such system represent quantities that must comply with a set of inequality constraints, it is crucial that numerical solution of the system remain within admissible domain at every step of the simulation, like, for example, a problem with unknowns that are concentrations, which have to remain non-negative. In this thesis we choose a relevant instance of the above class of problems, namely a model for tumor growth introduced by Perthame, Lorz and Lorenzi, and develop an efficient solution algorithm for its calculation that preserves the positivity of the unknown cell concentration through the simulations pro- cess. We adopt a first-order time-stepping procedure with automatic time- step selection. For the spatial discretization we used a finite element scheme that relies on a balanced quad-tree mesh. The computational domain is or- dered and partitioned by nodes. The resulting system of nonlinear algebraic equations is solved with a projected inexact quasi-Newton-Krylov method, globalized by compiling it with a gradient descent direction. The accuracy, robustness and scalability of the algorithm is observed by numerical experiments.File | Dimensione | Formato | |
---|---|---|---|
2018_07_Avramova.pdf
non accessibile
Descrizione: Tesi di laurea magistrale
Dimensione
1.88 MB
Formato
Adobe PDF
|
1.88 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/141689