National Technical University of Athens (NTUA)
Large-scale, nonlinear computations with parallel iterative methods: application in the analysis of transport phenomena
Abstract
dc:descriptionLarge-scale computations are often required, in order to ‘capture’ mechanisms pertaining to transport phenomena, which involve many continuum spatial scales, usually in complicated geometries. For instance, a fine spatial discretization - leading to large-scale problems - is often required for ‘introverted’ mechanisms to ‘reveal’ themselves. Large-scale computations could also come from the need to capture singular behaviors near rough edges of boundaries or abrupt changes of boundary conditions, which also demands fine spatial discretization near the singularity. Therefore, it is of primary importance that the numerical procedures, used to tackle problems in transport phenomena, are efficiently parallelizable. Also, the programming model that is implemented in the numerical procedures should exploit the modern computer architectures as much as possible. Realizing this, we strive to create a general framework for cost-effective parallel computations in modern computers, applied to the research in transport phenomena. To be fair, cost-effectiveness cannot only rely on the parallelization of the procedures. In this work, is also adopted a nothing-goes-to-waste philosophy in the computational method that improves the overall cost-effectiveness. Regarding this, two different methods are proposed for extracting useful information about the critical instabilities of physical systems by processing by-products of the solution procedure. Another approach in this work for improving cost-effectiveness lies in the reduction of the degrees of freedom by incorporating a hybrid semi-analytic method in Laplace-type problems that efficiently accounts for possible singularities in the solution. Thus, the excessive mesh refinement at the singularity to capture its behavior is not needed, which results in a radical decrease of the degrees of freedom. To sum up, the proposed framework is divided in three branches: the efficient parallelization through a hybrid programming model, the implementation of two parallel-efficient iterative procedures utilizing their by-products for stability analysis, and the reduction of the degrees of freedom by a hybrid semi-analytic method.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2009
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Pashos, Georgios
- Πάσχος, Γεώργιος
Subjects
dc:subject × 22- Μέθοδος γενικευμένων ελάχιστων υπολοίπων
- Μέθοδος αναδρομικής προβολής
- Πρόβλημα ιδιοτίμων
- Συνοριακά στοιχεία
- Πρόβλημα Bratu
- Πρόβλημα κοιλότητας με ολισθαίνον τοίχωμα
- Προγραμματισμός OPENMP-MPI
- Ιδιομορφία
- Υβριδική μέθοδος
- GMRES
- Recursive projection method
- Eigenvalue problems
- Boundary elemetns
- Bratu problem
- Lid-driven cavity problem
- OPENMP-MPI programming
- Singularity
- Hybrid method
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Χημικού Μηχανικού
- Engineering and Technology
- Chemical Engineering
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/22528