05.04.12

# "Gray-Box" Optimization

 Typ: Diplom-/ Masterarbeit Bereich: Simulation/Modellierung Termin: sofort/immediately Beschreibung:Overview of Gray-Box Optimization Scheme The current standard in optimization expects simulations to be completed before they pass results back to the main optimizer program. However, in many cases, it may be desirable to use interim results from the simulation, as provided by checkpointing or other output, to guide optimization calculations. This is especially true when error bounds for the accuracy of a simulation are available through analysis of the results already obtained during the course of a simulation. In such cases, it may be possible to short-circuit the decision-making process as soon as an available move can be identified, rather than waiting until the simulations have completed. Consequently, the optimizer works on a gray-box'' code, whose results can be periodically interrogated. We will develop the methodology for gray-box'' optimization by application to molecular dynamics force field parameterization. Molecular dynamics simulations study the thermodynamics and physical evolution of a system through repeated integration of Newton's equations of motion, as applied to a specific force field. However, in many cases, the exact parameters for a few of the terms needed in a force field may not be known, and some means for providing the remaining terms must be found. Typically, data obtained via experiment or other simulations can be used as targets'' for the optimization procedure. In this project, we will provide a proof-of-concept implementation using {\em pattern searching as a template for structuring the optimization calculations. For this project, the student will be responsible for: * Create a pattern search algorithm which can periodically check the status of a running job and collect needed results * Create tool which can pass the results back to the optimizer * Assist in preparation of scripting tools to communicate between the different programs involved Prerequisites for this project are: * Experience in C/C++ programming * Proficiency in English communication Preferred candidates will have some experience with one or more of the following: * Parallel programming (MPI) * Molecular simulations, and in particular molecular dynamics using LAMMPS * Optimization, including the ACRO/COLIN package * Scripting tools, such as Python Lehrstuhl: Prozesstechnik Ansprechpartner: Ahmed E. Ismail, Fady Assassa