CERN Accelerating science

If you experience any problem watching the video, click the download button below
Download Embed
Preprint
Report number EXT-2001-033 ; PNPI-2421
Title LikelihoodLib - Fitting, Function Maximization, and Numerical Analysis
Author(s) Smirnov, I B
Affiliation (PNPI, St.Petersburg, Russia)
Imprint 16 May 2001. - 32 p.
Subject category Computing and Computers
Abstract A new class library is designed for function maximization, minimization, solution of equations and for other problems related to mathematical analysis of multi-parameter functions by numerical iterative methods. When we search the maximum or another special point of a function, we may change and fit all parameters simultaneously, sequentially, recursively, or by any combination of these methods. The discussion is focused on the first the most complicated method, although the others are also supported by the library. For this method we apply: control of precision by interval computations; the calculation of derivatives either by differential arithmetic, or by the method of finite differences with the step lengths which provide suppression of the influence of numerical noise; possible synchronization of the subjective function calls with minimization of the number of iterations; competitive application of various methods for step calculation, and converging to the solution by many trajectories.
Submitted by Igor.Smirnov@cern.ch

 


 Record created 2001-06-01, last modified 2009-07-09


Access to fulltext document:
Download fulltextPDF