I need some advice re: pattern fitting (decomposition?).
I am not refining my XRD patterns, I am just fitting the (overlapping)
peaks to Voigt functions. I have code that can include Kalpha1/Kalpha2
splitting for a variety of wavelengths, can refine the amplitude ratios,
optionally includes asymmetry from axial divergence, and can include a four
parameter cubic spline background. The number of peaks included is only
limited by complexity and time. Sometimes small peaks get very broad and
are effectively included in the background. This can be avoided by holding
some peak parameters and/or the background constant.
I set the diffractometer's graphite monochromator myself, so I am refining
the amplitude ratio of the Kalpha1/Kalpha2 splitting. One factor that I do
not account for that will affect this ratio is the atomic form factor.
To find the amplitude ratio, I fit the ratio for the last peak, last 2
peaks, last 3 peaks, ..., whole pattern. I then plot the ratio vs the
number of peaks used. For low numbers of peaks (where the form factor is
~constant), the plot is approx. linear, and the ratio falls when peaks at
lower 2Theta (where the form changes rapidly) are included.
I take the average of the linear region of this plot as my amplitude ratio
and fit the whole pattern using this value.
First, does this sound reasonable? Second, is there any (moderately simple
or at least not extremely complex) way I can include the effect of the form
factor in these calc's? Third, are my peak parameters (shape, width) valid
without this correction? That is, would a refinement (or fitting) that
included the form factor, etc. come up with the same or similar peak
parameters as a fitting procedure that does not include the form factor. I
notice that MacDiff does not include much diffraction physics, it just fits
the data. This said, I suppose I could include corrections for Lorentz and
polarization factors. Is this necessary (advisable)?
I guess this is a question about what is maintained by convolution. If a
voigt function is convoluted with something like a correction for Lorentz
polarization, how is it changed?
thanks!
And the reason I'm fitting, rather than refining, my patterns is that I
have not been able to extract reasonable physical trends from my
refinements, so I am falling back upon Williamson-Hall type plots for
strain analysis... I think my diffractometer(s) is not aligned well enough
for Rietveld. Or something.
Charles Witham
Graduate Research Assistant, Materials Science
California Institute of Technology
Mail Stop 138-78 [EMAIL PROTECTED]
Pasadena, CA 91125 (626) 395-3627
fax 795-6132
JPL lab (818) 393-4209
