Regensburg 2007 – scientific programme

Parts | Days | Selection | Search | Downloads | Help

DY: Fachverband Dynamik und Statistische Physik

DY 28: Nonlinear stochastic systems

DY 28.10: Talk

Thursday, March 29, 2007, 16:15–16:30, H2

Detrended Fluctuation Analysis in Stochastic Global Optimization — •Kay Hamacher — Max-Planck-Institut fuer Physik komplexer Systeme, Dresden

Global optimization (GO) is one of the key numerical tools in computational physics. Among the GO algorithms the ones originating in statistical physics (e.g. Monte Carlo) are particular powerful. We show how an approach to time series analysis (detrended fluctuation analysis) can be leveraged to analyze the dynamics of GO algorithms. The emergence of random walks indicates suboptimal, diffusive behavior. This can be dealt with by adaptive schemes for general GO procedures, such as stochastic tunneling [1,2] and energy landscape paving [3].

[1] K. Hamacher. Adaptation in Stochastic Tunneling Global Optimization of Complex Potential Energy Landscapes, Europhys.Lett. 74 944, 2006

[2] W. Wenzel and K. Hamacher. A Stochastic tunneling approach for global minimization. Phys. Rev. Lett. 82 3003, 1999

[3] K. Hamacher. Energy Landscape Paving As A Perfect Optimization Approach Under Detrended Fluctuation Analysis, Physica A 2007, in press