Bonn 2010 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 52: Supersymmetrie: Parameterbestimmung

T 52.1: Vortrag

Dienstag, 16. März 2010, 16:45–17:00, HG ÜR 2

Optimization of Markov chains for a SUSY Fitter: Fittino — •Xavier Prudent¹, Philip Bechtle², Klaus Desch³, and Peter Wienemann³ — ¹IKTP, Technische Universität Dresden, Germany — ²DESY, Hamburg, Germany — ³Universität Bonn, Germany

A Markov chains is a "random walk" algorithm which allows an efficient scan of a given profile and the search of the absolute minimum, even when this profil suffers from the presence of many secondary minima. This property makes them particularly suited to the study of Supersymmetry (SUSY) models, where minima have to be found in up-to 18-dimensional space for the general MSSM. Hence the SUSY fitter "Fittino" uses a Metropolis*Hastings Markov chain in a frequentist interpretation to study the impact of current low -energy measurements, as well as expected measurements from LHC and ILC, on the SUSY parameter space. The expected properties of an optimal Markov chain should be the independence of final results with respect to the starting point and a fast convergence. These two points can be achieved by optimizing the width of the proposal distribution, that is the "average step length" between two links in the chain. We developped an algorithm for the optimization of the proposal width, by modifying iteratively the width so that the rejection rate be around fifty percent. This optimization leads to a starting point independent chain as well as a faster convergence.