You are here: NA&AM › Agenda NA&AM seminars › NATW Seminar by Koen Poppe: Gaussian distributed quasi-random samples

NATW Seminar by Koen Poppe: Gaussian distributed quasi-random samples

When Feb 22, 2012
from 04:00 PM to 06:00 PM
Add event to calendar vCal
iCal

location: Celestijnenlaan 200A, 200A 05.001, 3001 Leuven-Heverlee

by Koen Poppe (KULeuven)

 

The Box--Muller transformation is a well known method to generate bivariate normally distributed variables. When used in combination with LCGs there is a clearly visible spiral configuration. These spirals also appear when the input points are lattices or other structured point sets.

We developed quality criteria for the resulting two-dimensional point sets in a number of ways: based on geometrical reasonings, by defining discrepancy measures, using a suitable set of test-functions and by means of a worst-case error in a reproducing kernel Hilbert space setting.

The main focus here is the extension of both the Box--Muller transformation and the quality criteria to moderate dimension. This leads to point sets that are suitable in the context of quasi-Monte Carlo methods for integration with Gaussian weight function over the entire space.
 

Comments (0)