Full text
| |
| Author(s): |
Fedor Pisnitchenko
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Instituto de Matemática, Estatística e Computação Científica |
| Defense date: | 2008-09-26 |
| Examining board members: |
José Mario Martínez Pérez;
Maria Cristina de Castro Cunha;
Haroldo Fraga de Campos Velho;
Jin Yun Yuan;
Ricardo Caetano Azevedo Biloti
|
| Advisor: | José Mario Martínez Pérez; Sandra Augusta Santos |
| Abstract | |
In science and engeneering there is a wide class of problems that consist in solving a system of partial differential equations to find variables (such as velocity, temperature, displacement, etc.), given the necessary decision information (such as domain, initial and boundary conditions, etc.). However,it is very common for real problems that the decision information is incomplete and contains errors. On the other hand, there is some additional information about state variables, which come from other simulation or some kind of observations (observed data). A natural way to solve this kind of problem, using all the decision information, is to interpret it as an optimization problem. That is, minimize an objective function chosen such as distance between the observed data and the state variables, subject to the system discretization. In this work, we propose a Quasi-Newton method to solve the PDE-constrained problem using as models the unidimensional Rossby-Obukhov and Korteweg-de Vries equations. A very importante aspect of the method is that there is no stability restriction for the stepsize in the differential equations discretization. Another aspect is to be able to use stepsizes larger than the ones used in traditional evolutive methods such as finite differences. A large number of computational test was performed. The results were promising and showed the robustness of the method and its ability to solve large scale problems. (AU) | |