ESTIMATES OF
THE UNCERTAINTY OF LATTICE PARAMETERS REFINED FROM NEUTRON POWDER DIFFRACTION
DATA
Burkhard Peplinskia,
Daniel M. Toebbensb, Winfried Kockelmannc, Richard
Ibbersonc
a Federal Institute for Materials
Research and Testing (BAM),
Richard-Willstätter-Str. 11, D-12489 Berlin, Federal Republic of Germany
b Hahn-Meitner-Institute (HMI), BENSC, SF2
Glienicker Str. 100, D-14109 Berlin, Federal Republic of Germany
c ISIS facility, Rutherford Appleton Laboratory (RAL), Chilton,
Didcot, OX11 0QX, UK
Lattice
parameter refinements from neutron diffraction (ND) data may be preferred to
those from diffraction data collected with conventional X-rays or synchrotron
radiation, for example if investigations on coarse-grained powders or bulk
samples, with volumes of up to several cubic centimetres, are meant to be representative
for the whole specimen. For these kind of analyses ND takes advantage of the
high penetration of neutrons for most elements (isotopes), allowing them to
simultaneously probe the whole volume of a thick specimen with nearly no
attenuation. Furthermore, with ND the observed intensities of Bragg reflections
are averaged over a larger number of crystallite orientations than the high-resolution modes of the conventional
X-ray or synchrotron radiation techniques usually allow. This is a consequence
of the large equatorial and axial divergence normally employed in constant
wavelength ND beam optics and of the wide wavelength spectrum and the large
acceptance angles of the detector banks used in time-of-flight (ToF) ND. From the metrological point of view the following three
aspects of the uncertainty of measurement are especially relevant:
1. If the
lattice parameter refinement is carried out by the Rietveld method or by any other whole pattern technique, then the only
available measures of the uncertainty of the refined lattice parameters are the
estimated standard deviations (e.s.d.s) calculated by the full pattern fitting
program. However, e.s.d.s are measures of precision rather than of accuracy and
these two terms must not be confused.
2. e.s.d.s
calculated by many Rietveld programs are not reliable measures of the ‘probable
errors’, because, in many cases, they are systematically too small, due to
‘serial correlation’. Therefore, uncorrected e.s.d.s are not only no measure of
accuracy, but even no reliable measure of
the precision of refined lattice parameters. The e.s.d.s are calculated
under the assumption that the values in the difference curve are independent
observations. However, adjacent individual points in the difference curve are
not independent but correlated by the used profile function. This correlation
depends on the size of the 2θ steps used for data collection and
evaluation. A formula for estimating the corrections that should be applied to
the e.s.d.s has been given by Berar and Lennan [1] who established that e.s.d.s
calculated by Rietveld programs without
consideration of serial correlations are often too small by a factor of
approximately two. Therefore, at the 65% confidence level, the probable errors are calculated from the
uncorrected e.s.d.s by multiplying the latter by a factor of about two.
Consequently, the U95% uncertainties of measurement at the 95%
confidence level are calculated by multiplying the uncorrected e.s.d.-values by
a factor of about four.
3. Bragg’s law involves a 100% correlation
between the d-values (and therefore the lattice parameters) of a
crystallographic phase and the wavelength of the diffracted radiation.
Thus for any assumed value of the wavelength another set of lattice parameters
results, whereas the agreement indices and the e.s.d.s remain the same. Since for all neutron diffractometers the
wavelength is not known per se, as
would be the case with characteristic X-ray radiation, even under the
idealising assumption of the complete absence of any systematic errors from the
model and from the observed data the lattice parameters are completely indeterminate. Therefore,
a determination of accurate lattice parameters by constant wavelength ND or ToF
ND necessarily includes calibration
procedures with reference materials that have lattice parameters certified for
a given temperature by independent
analytical techniques. The propagation of the error resulting from the
calibration procedure should be
considered in the estimation of the total uncertainty of the lattice parameters
of the actual sample under investigation! Consequently, in the case that the
calibration procedure as well as the analysis of the actual sample are carried
out by the Rietveld method or any other whole
pattern technique the U95% uncertainties of the final analysis results
can be as large as the eight-fold of
the uncorrected e.s.d.s from a single refinement!
Further aspects
of the uncertainty of measurement (e.g. line shifts due to absorption or the
question about preferences for the methods of internal or external standard) have
to be analysed separately for each of the following three types of ND
instruments:
a.
High-resolution
powder ToF-ND instruments, see e.g. [2]
b.
High-resolution
powder diffractometers for
monochromatic neutrons, equipped with
parallel collimators placed in front of the detectors and restricting the
equatorial
divergence of the diffracted beam (prototype: D1A and D2B at the ILL, see e.g.
[3])
c.
Powder diffractometers for monochromatic neutrons
in Debye-Scherrer-geometry with
one (or several) position-sensitive
detector(s) providing resolution in the equatorial plane
The
determination of accurate lattice parameters through the refinement of
diffraction data collected with any type of these instruments requires highly
accurate mathematical modelling of the observed diffraction line profiles. For
a constant wavelength ND instrument of type b this can be achieved by the ‘fundamental
parameter approach’ which combines the individual contributions to the
reflection profile by convolution integrals. This leads to the following
writing of Bragg’s law:
(1)
with λeff.
= effective value of the wavelength; θobs. = observed halved diffraction angle;
EPS1,eff = effective value of the zero correction for the detector
bank; Δ = axial divergence; R = diffractometer radius; hd = effective
height of the detector; hs = effective height of the specimen; FWHM =
contribution to the reflection profile caused by the real structure of the
specimen; is the ‘axial
divergence correction function’ which can be determined only approximately due
to strong correlations between some of the parameters.
By
differentiating equation (1) a formula for the uncertainty of refined lattice
parameters can be derived that contains a number of individual contributions,
not all of which are additive. An analogous formula can be derived for ToF-ND.
By the analysis of the individual contributions
and by taking into account the results of the proficiency testing carried out
at several ND instruments (see e.g. [4-5]) it can be shown that the lattice
parameters of cubic, tetragonal, hexagonal and orthorhombic materials can be
refined in the metric system and at the 95% confidence level with a relative
accuracy of Δai/ai ≈ 2-3•10-5 using a high-resolution multi-collimator/multi-detector
powder diffractometer for monochromatic neutrons or a dedicated high-resolution
ToF-ND instrument.
The present study
is based on the analysis of data published by a number of laboratories as well
as on experimental results of the authors. It facilitates the statistically
sound inter-pretation of differences between lattice parameters refined on the
same sample using different ND instruments.
1.
Berar
J.F., Lennan P. (1991) J. Appl.
Cryst. 24, 1-5
2. David W.I.F. et al. (1986) Mater. Sci. Forum 9, 89-101
3.
Hewart A.W. (1986) Mater. Sci. Forum 9, 69-79
4.
Toebbens D.M. et al. (2001) Mater. Sci. Forum 378-381, 288-293
5.
Kockelmann W. et al. (2000) Mater. Sci. Forum 321-324,
332-337