Periclase is really problematic because of the small number of peaks in
the usual angular range used for QPA with Cu or Co radiation. Thus, the
particle statistics problem becomes extremely critic, as Pamela already
said. The uncertainties can be compared with the statistical intensity
errors of a single line. See also the textbook of Klug & Alexander
(1954), section "operation of the spectrometer, specimen preparation",
demonstrating that the mean statistical error of the intensity of the
biggest line of quartz of a 15-50 µm fraction is 18.2 %!.
Other systematic errors may result from the description of the peak
profile: If you have only 3-4 peaks, only one peak having significant
intensity bigger than the noise, than the profile shape model of this
peak must be fit perfectly to get a correct intensity fit resp. correct
scale factors. In my opinion, this is hard to do with converted data.
Intensity conversion does not converge the instrumental effect on the
profile shapes. Thus, any physical meaning of profile parameters goes
lost by data conversion, as well as the counting statistics (noise) are
changed by the conversion. In general, Rietveld refinement should not be
done with corrected or modified diffraction data because of the
weighting scheme of the minimization procedure supposes unchanged
counting statistics. You should use a Rietveld program what is able to
refine unchanged ADS data by a physically based peak profile model.
But even if these things are done correctly, an accuracy of < 0.5 wt% of
a minor phase in a complex system like cement can only be acchieved by a
very optimized procedure, from perfect sample prep over high quality
measuring data up to an optimized Rietveld model.
Reinhard
Whitfield, Pamela schrieb:
I’m afraid that 30 microns might be small to the eye but is too big
for quant analysis. The particles need to be in the range of a micron
or two. There’s an old paper by Deane Smith that did a brilliant job
of showing the effects in a statistical manner.
Pam
*From:* José Carlos Cordeiro [mailto:[EMAIL PROTECTED]
*Sent:* May 29, 2008 5:07 PM
*To:* Whitfield, Pamela; rietveld_l@ill.fr
*Cc:* Mitchell, Lyndon
*Subject:* RES: ADS
My samples are cements (fine samples with 30 microns) and don’t need
micronize its. The problem in my refinements is that periclase
quantified with fixed slit is bigger than MgO by Fluorescence, and its
impossible. Ex:
Periclase with fixed slit: 7,14%
MgO by XRF = 6,84%
We know that part of MgO (1,5-2,0%) go to the structure of C3S an C2S
phases, so periclase expected in this sample is 4,8% - 5,3%. If we
consider this, the errors of refinement is not very slim.
Sorry for my bad English!!!
regards
*===========================================*
*José Carlos Cordeiro*
*Diretoria Técnica - Laboratório Central*
Centro Técnico - Curitiba
Rodovia PR-092, 1303, Abranches, 82130-570 Curitiba-PR
Votorantim Cimentos
[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>_
_www.votorantimcimentos.com.br _
Fone: + 55 041 3355-1380
Fax: + 55 041 3355-1358
===========================================
-----Mensagem original-----
*De:* Whitfield, Pamela [mailto:[EMAIL PROTECTED]
*Enviada em:* quinta-feira, 29 de maio de 2008 17:40
*Para:* José Carlos Cordeiro; rietveld_l@ill.fr
*Cc:* Mitchell, Lyndon
*Assunto:* RE: ADS
Hi Jose
Errors in the region of 1wt% aren't unheard of in complex mixtures
(and the liklihood of the second decimal place being meaningful is
very slim). However there are other possible explanations and one in
particular comes to mind.
Did you micronize your sample? If yes then feel free to ignore the
following! If not then it might be worth thinking about.
We did a systematic study on the effect of particle statistics on the
repeatability of quantitative analysis on one of the NIST cements last
year (presented at the ICCC in Montreal - should really go somewhere
more accessible). We compared different divergence slits, micronized
and unmicronized samples, spun and static, and remounted/repeated the
experiments 3 times each for real statistics as opposed to esds.
Funnily enough the phase that showed the biggest anomaly in one of the
unground sample was the periclase - obviously a big (by XRD standards)
lump fell into the sample from the kiln that made up the NIST batch
and made it through the mill at the plant. Without grinding it we got
0.1, 0.1 and 2wt% one series of 3, versus 0.6, 0.6 and 0.6 wt% (this
is from memory so don't don't quote me!) for the micronized sample.
This was the result that we should have got theoretically but it was
nice to see it work in practice, and the periclase did such a nice job
for us by sticking out like a sore thumb even without a Rietveld
analysis! BTW there was no statistically significant difference (95%
limits) in the results of the C3S and C2S phases between ground and
unground samples - mostly because the errors in the unground samples
were so large! The minor phases were another matter. In the micronized
samples the choice of divergence slit size made little difference, but
with the unground samples they made a significant difference.
To cut a long story short it all comes down to particle statistics -
with an ideal powder the choice of fixed versus variable slits (or
even different fixed slits) should make no difference (assuming that
the Rietveld software can handle the difference in profiles). Moral of
the story - sample prep, sample prep and more sample prep!
Pam
-----Original Message-----
*From:* José Carlos Cordeiro [mailto:[EMAIL PROTECTED]
*Sent:* May 29, 2008 3:53 PM
*To:* rietveld_l@ill.fr
*Subject:* ADS
I have a Programmable Divergence Slit (incident beam) in my
difratctometer, that can work in automatic mode (irradiated
length) or fixed mode (0.5 degree or 1.0 degree etc). What is the
best collect mode for Rietveld Refinament??
I did refinement two difractograms collected in booth
configuration (fixed 0.5 degree and automatic mode), the results
in the fase “Periclase” (cement fase) changed, see below:
Periclase with 0.5 degree = 6.08%
Periclase wich automatic mode = 7.14%
Note.: The refinement of the difractograms collected with
automatic mode went converted to fixed mode by software.
What do you think about this?
regards
*===========================================*
*José Carlos Cordeiro*
*Diretoria Técnica - Laboratório Central*
Centro Técnico - Curitiba
Rodovia PR-092, 1303, Abranches, 82130-570 Curitiba-PR
Votorantim Cimentos
[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>_
_www.votorantimcimentos.com.br _
Fone: + 55 041 3355-1380
Fax: + 55 041 3355-1358
===========================================