-
3rd LHC Detector Alignment Workshop, June 15-16, 2009, CERN (http://indico.cern.ch/conferenceDisplay.py?confId=50502). new features: ------------ - new preconditioning for GMRES * not limiting the number of parameters to 46340 * always applied, even if 'bandwidth' not chosen * one method with explicit bandwith, another without (and width 1) (recommendation: do NOT specify bandwidth option) - fix problem in hash values for label pairs that was limiting to 46340 parameters - line search with slope = 0, not minimum (-> better stability with outlier rejection) - re-calculates the matrix in case of large number of rejects due to many outliers (might be slower, but more robust) - new options * 'savespace': reduce memory needs of LOOP2 that was larger than needed for final matrix (also speeds up significantly, so recommended) * 'presigma 0.1': assign presigma of 0.1 to all labels where no explicit presigma was set * 'regularisation 1. 0.04': regularisation with tau = 1, i.e. add tau * ||x|| to the minimised function where x is the vector of alignment parameters, devided by their individual presigma (includes 'presigma 0.04') This should reduce the influence of weak modes, but needs detailed studies. Note also: ---------- - remaining matrix solution methods are GMRES, sparseGMRES, diagonalisation and inversion, the others are removed - no global correlations stated in inversion - millepede.log not that readable anymore (but pede.dump is better now) - pede now has an exit code that can differ from 0. git-svn-id: http://svnsrv.desy.de/public/MillepedeII/trunk@13 3547b9b0-65b8-46d3-b95d-921b3f43af62
6ee1abb0