Book reference "The Theory and Operation of Spectral Analysis
Using ROBFIT". AIP 1991 ISBN 0-88318-941-0
This has since been bought by Springer
Robert L.
Coldwell and Gary J. Bamford,
The codes in srobfit.zip[i] fit the entire background of a spectrum as a cubic spline in which the location of the knots are also parameters. The zip file Srobdocs.zip contains a number of sample runs. Overlapping peaks are fitted together, but non-overlapping peaks are fitted separately. For most fits, the background and the peaks can easily be fitted independently Stanfit3.html. The codes iterate the fits to achieve maximum accuracy. There was some development of the code after the book went to press, but the algorithms and usage are the same. The code has recently been extended to work using g77 as found in linux combined with the tektronics feature of xterm for graphics srobunix.zip. I have recently gone through a couple of examples of the code in practice. The first concerns the ability of the code to be forced into breaking peaks into their components Adventures with srobfit.htm. The second involves a simple fit using the stgen and xcaliber codes aug2002.htm. Sample fits of possible interes to G. Dunnam are in gdtrob.htm, Gdunnt2.htm, and Gdunnt3.htm.
The background is compressed by using a weighted average over each set of 32 points to represent the background in the region. I want to know if anyone finds strange results coming from this. I have experimented with “better” alogithms in the intervening years and with challenging real data would like to see if they really are.
The robust minimization routine used in this work is that described in the book. I have possibly improved this routine as described in nlfit\Welcome.htm. If anyone has data for which the current code seems stressed in its ability to fit data, I would be interested in knowing how the new routine compares to the old. Most of the recent work has been done in collaboration with George P. Lasche and involves Unkfit which forms a complete response function for each suspected nuclide in the data. Many parts of this are in the zip package srobunix.zip.
The MS 5.1 Fortran and Watfor in which the routines were originally written no longer exists in the commercial market. Intel 10 works well. The exe’s in the runable directory were generated by me using the mkgen and mkcopro bat files in the dos directory after properly setting the fortran paths. They worked on my machines. The dos graphics goes into a full screen window – no longer supported I copied more modern graphics into the runable file to make this a full screen window. The command <prt scr> puts this in the clipboard, while <esc> gives a limited menu of options. I routinely dump these from there into Microsoft Word.
The \ is a special character in unix which caused no end of problems. The graphics produces a separate screen with relatively good resolution, but no color. Instead there is a set of dot and dot dashes to separate the fits. The makefile is in the srobfit/fortran/unix. The directions are in unixread.me.
[i] March 9, 2008 – Windows Vista does not support the dos full screen graphics used by the original srobfit. The three routines rawdd, stdis, and fspdis have been copied from srobunix into this zip.