SCCS Colloquium - Sep 4, 2019

From Sccswiki
Jump to navigation Jump to search
Date: September 4, 2019
Room: 02.07.023
Time: 15:00 - 16:00

Christian Menges: Optimization and Evaluation of the Linked-Cell Algorithm

This is a Bachelor's thesis submission talk. Christian is advised by Fabio Gratl.

The Linked-Cell algorithm is used as an underlying data structure in molecular dynamics simulations to store and organize particles. It allows us to quickly find neighbors of a particle within a given radius without checking all stored particles, which is necessary to effectively simulate short-range interactions. In this thesis, four different approaches for faster neighbor search and improved SIMD performance are presented and evaluated. First of all, the variation of the cell size is discussed, continuing with two schemes to combine multiple cells to maximize their performance using SIMD operations. Further, the reduction of search space for possible neighbors using sorted cells is discussed. The last chapter deals with an outlook on adaptive approaches which allows a combination of different cell sizes and sorted cells. All these techniques show a performance improvement depending on the characteristics of the simulated experiment, especially on the density of particles and their distribution inside the domain.

Keywords: AutoPas, molecular dynamics, linked-cell algorithm

Language: English

Alec Gliga: Simulation of Multivariate Distributions with various Univariate Marginals

This is a Bachelor's thesis submission talk. Alec is advised by Tobias Neckel.

The ultimate purpose of copula simulations is the generation of samples from any multivariate or joint distribution, through the Monte Carlo sampling method. Simulations of such multivariate distributions have proven to be useful, when the number of marginals contained in a multivariate distribution is high, and analytic approaches get highly expensive in terms of computational effort or do not even exist. A second important advantage of copulas is their ability to decouple the dependence structure and marginals from one another, such that those can be modeled separately. Copulas are mainly used inquantitative finance, but have found applications in various other fields such as climate andweather research, civil engineering and even medical research.

Keywords: Multivariate Distributions, Copulas

Language: English