===== A hybrid level-set method on the GPU =====

==== Facts ====
^ Type           | master project|
^ Place          | internal      |
^ Supervisors     | Andrei Jalba |
^ Student        | Laurens Timmermans            |
^ Thesis         | [[http://alexandria.tue.nl/extra1/afstversl/wsk-i/timmermans2011.pdf|download]]      |
^ Start/End date |  / 19-12-2011       |


==== Abstract ====
Historically, there have been two approaches to tracking evolving
interfaces: Lagrangian and Eulerian. While it is difficult to track an
interface that changes topology using a Lagrangian approach, the
Eulerian approach handles such cases with ease. Unless expensive
high-order finite differencing schemes are used however, the Eulerian
approach exhibits what is observed as a loss of mass. These orthogonal
strengths and weaknesses of the Lagrangian and Eulerian approaches have
lead to an interest in a third, hybrid approach.

A number of hybrid methods have been proposed in recent literature that
augment the level-set method, which take an Eulerian approach, with
Lagrangian-style particles. In this thesis, a comparison is made between
a number of these proposals. The Self-Adaptive Oriented Particles
Level-Set (SA-OPLS) method was found to provide excellent results with
respect to mass-preservation, while additionally containing a robust
self-adaptive mechanism.

The advent of programmable graphics processing units (GPUs) has provided
access to a highly parallel computing platform. Both the Lagrangian and
Eulerian approach, and therefore naturally also a hybrid approach, have
been recognized to exhibit a large amount of parallelism, allowing them
to benefit from such highly parallel architectures. One example is the
GPU-accelerated Sorted Tile List (GPU-STL) method, which provides a
highly efficient implementation of the level-set method.

In this thesis, a hybrid level-set implementation on the GPU is
proposed, based on the combination of GPU-STL and SA-OPLS. The final
results of this implementation are promising, comparing favourably
against similar work presented in recent related literature.