Whole Pattern Fitting
Using Jade to further interrogate your X-ray
diffraction data
Phase Quantification
Although semi-quantitative methods such as RIR have proven very
effective in the past, our computational abilities have progressed to
the point where more accurate methods are available, i.e. whole-
pattern fitting (WPF) methods. WPF methods, used correctly, are
significantly more accurate than RIR.
WPF methods can also allow the user to move beyond mere phase
quantification and actually account for other properties of the sample,
such as preferred orientation, changes in lattice parameters, and
elemental substitution.
The program used by PROTO for these methods is JADE.
WPF Methods
There are many WPF methods that have been developed over
the course of XRD’s history.
Where RIR methods focus on the intensities of a select few peaks
to determine the relative abundances of phases in multi-
component samples, WPF methods utilize all data points within
a diffraction pattern to compute relative abundances.
The three main WPF methods are the Pawley, Le Bail, and
Rietveld techniques.
Pawley and Le Bail Methods
Pawley Le Bail
• Least-squares method • Least-squares method
• Requires no structure model (i.e. no specific unit • Requires no structure model (i.e. no specific unit
cell data) cell data)
– Somewhat more general data are required, such as – Somewhat more general data are required, such as
space group symmetry, and estimates of unit cell space group symmetry, and estimates of unit cell
parameters parameters
• Calculates intensity • Uses observed intensity
– Makes the process slower – Makes the process faster
• Relatively difficult to use, but mathematically • Relatively easy to use, but not mathematically
perfect perfect due to use of observed intensity
• Has trouble with overlapping peaks • Has trouble with overlapping peaks
The benefit of this method is that since no structural models are included in the fit,
the fit is not structure-model biased.
However, for situations where the context of a sample is known, search/match can
drastically narrow down the phase options, and structural data can be incorporated
into your WPF model… This is the Rietveld method!
Rietveld Method
Often, PXRD scientists have some knowledge about the
composition of their sample, and the composition is generally
not novel. As discussed earlier, a major benefit of PXRD is the
ability to qualify the phases present within a sample. The crystal
structures of many phases, man-made and naturally occurring,
have been identified and defined, allowing a PXRD scientist to
match the peaks in their diffraction pattern to known crystal
structures. This knowledge is necessary for Rietveld refinement.
Rietveld Method (cont’d)
• Least squares method
• Uses structure data that you input to model every data point of a
powder diffraction pattern
• Each modeled point includes background and instrument
contributions to peak width and shape, in addition to the structural
information the user provides
• Goal is similar to other WPF methods: find the best possible fit
between a structure-based model and the observed powder pattern
There is a formula to calculate the intensity at any given step in the
scan…
Calculated Intensity at Any Step
𝑦𝑖𝑐 = 𝑆𝑝 𝑝𝑘 𝐿𝑘 𝐹𝑘 2 𝐺 Δθ𝑖𝑘 𝑃𝑘 + 𝑦𝑖𝑏 𝑐
𝑃 𝑘
In which:
• 𝑆𝑝 is the scale factor for phase p
• 𝐿𝑘 is the Lorentz and polarization factors for the Kth Bragg reflection
• 𝐹𝑘 is the structure factor
• 𝑝𝑘 is the multiplicity factor
• 𝑃𝑘 is the preferred orientation function
• θ𝑖𝑘 is the Bragg angle for the Kth reflection
• 𝐺 Δθ𝑖𝑘 is the reflection profile function
• 𝑦𝑖𝑏 𝑐 is the refined background
Remember…
• Because the Rietveld method uses structure factors in
calculating a model, it assumes that each phase present is 3-
dimensionally ordered. This means that any amorphous
phases in your sample will not be considered during model
refinement, and any weight percentages obtained for the
phases considered will be inaccurate.
• Also, the weight percentages produced by this method are
relative and not absolute. To obtain absolute masses, a known
quantity of an internal standard (such as corundum or quartz)
must be added.
JADE
MDI JADE is a powder XRD program capable of XRD pattern
viewing/processing/data reduction, with a focus on phase ID and
quantification.
JADE may be operated manually (i.e. the user selects the peaks,
the phases, and performs the model refinement themselves) or
it may automated.
First, import your data file
Next, identify your peaks
Peak and Phase List
The peak list is viewable, showing the angle and corresponding d-
spacing, as well as other data such as peak height, peak area, and
FWHM.
You can search the crystal structure database used by JADE, which includes the American Mineralogist
Crystal Structure Database (AMCSD) for geological and metallurgical phases, or you can download phase
files (cif cards) directly from the Crystallography Open Database (COD) and create a custom database.
This is particularly useful for pharmaceutical/non-mineralogical samples. You may also purchase the
commercial database provided by the International Center for Diffraction Data (ICDD), which is
compatible with JADE.
Now that you’ve selected your phases, refine your model!
Difference plot
The pink overlay represents the
calculated simulation JADE generates
to model your observed powder
diffraction pattern
Site Occupancy Variation
In addition to modeling the
diffraction pattern based on
phases, the user is able to
allow JADE to refine site
occupancy. Olivine, a Mg-Fe
bearing mineral, can have Mg
or Fe in its octahedral sites.
Because they substitute for
each other so well, even end-
member compositions such
as the Mg-olivine, forsterite,
contain small quantities of Fe.
This can be accounted for
with JADE by allowing the
program to refine octahedral
site occupancy. This can
provide the user with a more
accurate model of their
observed data.
Refinement with Site Occupancy Variation
JADE Output
Remember that when you are refining a model, the
R2 value indicates how well the model fits your data.
The lower the value, the better the fit.
R2 Example: Gypsum and Calcite Mixture,
Preferred Orientation
This first model was refined using The second model was refined using
gypsum and calcite without the same mineral phases, but with the
preferred orientation. Note that the preferred orientation function turned
R2 value is relatively high at 48.26%. on. The R2 value dropped to 14.08%.
R2 Example: Gypsum and Calcite Mixture
This sample contains two minerals: gypsum Next, the model was refined using gypsum,
and calcite. There are no other minerals calcite, and carobbite, a mineral phase
present in the sample. It was modeled in known to not be present in this sample.
JADE first with only those two phases.
Note! The R2 value is 14.08% But now the R2 value is 12.91%.
That must mean the model is
with just gypsum and calcite... better!
Beware!
In the previous examples, two different parameters were added to the model:
preferred orientation, and an extraneous third phase. Both additions resulted in a
significant decrease in the R2 value.
The preferred orientation function is reasonable because both phases exhibit
strong preferred orientation due to cleavage. The addition of the third phase,
although it also lowers the R2 value, is not valid because that phase is not present.
Be careful when adding phases and altering parameters during refinement. NOTE:
– The more parameters allowed to vary in the model, the better the fit regardless
of whether those parameters are actually necessary. At a certain level of
complexity, this becomes a problem you will have to grapple with… Each sample
will be different, and will require different handling!
Final Review
• Whole pattern fitting is an umbrella term that encompasses
the Pawley, Le Bail, and Rietveld techniques
– WPF is significantly more accurate than semi-quant methods like RIR
• Only in Rietveld method can you vary atomic occupancy
• Remember, you are modeling your diffraction pattern, so you
will need to vary multiple parameters to achieve the bets fit
– not every diffraction pattern will require the same parameters to
vary
– Too many model parameters (more than necessary) can improve
your fit, but that doesn’t mean your model is better
