This shows you the differences between two versions of the page.
bfgs_minimization [2014/07/08 11:40] |
bfgs_minimization [2014/07/08 11:40] prokop created |
||
---|---|---|---|
Line 1: | Line 1: | ||
+ | ====== BFGS ====== | ||
+ | it call [[http://users.eecs.northwestern.edu/~nocedal/lbfgsb.html|l-bfgs-b]] routine till: | ||
+ | - the convergence criteria defined in cg.optional are reached (f(i,j)<fmax for all i,j) | ||
+ | - the algorithm is not able to find the smaller energy in spite of f>fmax | ||
+ | - the maximum number of steps exceeds | ||
+ | |||
+ | ===setting of input files:=== | ||
+ | ==fireball.in:== | ||
+ | iquench = -5 | ||
+ | nstepf = 1000 ! ! Max number of opt steps, default value = 100 | ||
+ | dt = 0.5 !/only if FIRE minimization is used - last parameter in cgopt.optional >0/ | ||
+ | |||
+ | ==cgopt.optional:== | ||
+ | This routine uses the same input file as Conjiged Gradients (iquench = -4). It does not use all parameters, but they must be there. | ||
+ | 0.1 ! NOT USED in BFGS | ||
+ | 0.1 ! NOT USED in BFGS | ||
+ | 0.00000001 ! NOT USED in BFGS (total energy tolerance ) | ||
+ | 0.001 ! fmax - force tolerance - end of minimization if fij<fmax | ||
+ | 10000 ! NOT MORE USED IN BFGS (Max number of opt steps) - now the max. number of steps is given by parameter nstepf in firaball.in | ||
+ | 0 ! number of restarts in case ftot>fmax (it seems it nave no effect) | ||
+ | 0 ! if >0 then dt must be defined in fireball.in. Number of FIRE downhill steps after BFGS minimization "fails" | ||