Starting a fit

If the program calcsta has been set up, the fitting is started by the command (in the example you should go to the directory /mcphas/examples/cecu2a/fit) :

simannfit 10 [-t 1000][-s 100000] [-n 200] mcphas.jjj [and possible other input parameter files].

• The ”10” in the command means that the initial statistical ”temperature” of the algorithm is set to 10.

OPTIONS for ...

• STEPWIDTHS

  -w 1.4

  before starting simannfit, multiply all stepwidths by factor 1.4

  -r 0.2

  before starting simannfit, set all stepwidths to parameter range=max-min times 0.2,

  -f 0.2

  before starting simannfit, set all stepwidths to parameter value times 0.2, however never smaller than parameterrange/1000

  -p 20

  probing parameter space: stepwidths are not decreased during fitting, new set of pars $parn$ are generated and for all sets par in results/simannfit.p and results/simannfit.n are scanned and distance $d=\sum_i^N (par_i-parn_i)^2/stepwidth_i^2$ is calculated. If $d>N$(=number of parameters) for all sets par, then $sta$ is calculated. if $sta<=sta$ of initial parameters then $parn$ is appended to the list in results/simannfit.p up to 20 parameters are appended to this list, after that the program stops. the program also stops if step ratio reaches a value $> 11$ (step ratio is a factor applied to all (maximum) stepwidths before generating a step in the random walk It is decreased each successful step in the random walk and increased to find an acceptable step (distance criteria) to try thus, to probe a large region of parameter space do not set statistical temperature too low to enable efficient stepping. On the other hand to probe only the vicinity of a optimum in every direction of parameter space a small statistical inital temperature is necessary - in this way a series of equally good solutions can be explored.

• PARALLEL PROCESSING

  -s0 filename

  instead of results/simannfit.n use filename to store/read

  -s1 filename

  instead of results/simannfit.p use filename to store/read

  -i 23

  for storing in results/simannfit.* use index .23 instead of .0 as suffix

• LOGGING

  -n 50

  allows to store every 50 steps the parametersets in a table in results/simannfit.n (appending existing file).

  -jpglog 3.5e-1 file.jpg

  if sta is less then 3.5e-1, then the image file.jpg (which should be created by calcsta) is copied to results/parsetnr.jpg a html tag is added to results/simannfit.n or results/simannfit.p if stored

  -log 1.3 batchfile.bat

  if sta is less then 3.5e-1, then execute the file batchfile, with sta as argument, which can be adressed in the batch file with $1 (linux) or %1 (windows)

  -h

  Histograms are store in results/par*.hst for review of the variation of parameters during the run of simannfit.

  -d

  store percentage of different contributions to sta in file results/simannfit.dst

• TERMINATION

  -l 0.2

  sets limit - program will end if $sta<0.2$

  -t 100

  sets time limit until program end to 100 seconds

  -s 132

  gives maximal number of iteration steps 132 to be done

  -c

  continue at end of program - do not ask for pressing enter

Note that the user written program calcsta has to be in the directory where simannfit is started from.

The program generates for each parameter a histogram file showing the number of occurrences of a certain parameter value in solutions, where sta decreased.