Empyrean HighScore Rietveld refinement tutorial 2nd

2021.11.17
Ning Ma, NW.E69, [email protected] for complements/complains
Let us assume you have acquired a reasonably good powder XRD pattern and are now
interested in the crystallinity, the phase identification, refining crystallographic parameters
and/or estimating the phase proportions of your sample. All of this can be theoretically
achieved using this HighScore software albeit with varying difficulty and precision.
Usually in Rietveld mode

Main windows

Various display options for XRD pattern Rietveld refinement panel

and Rietveld refinement models,

Detailed
Main window for viewing results refinement
parameters

Right click
X-Y scale

2 Candidate phases
Legends 3 searching results
Not usually performed
4

Can switch to different mode by right click

Quick identification program 1. Background fit, 2 search peak, 3 search and match candidate phases

1. Load data file in high score either by draging or File-open, (double click will only open
the file in Data Viewer). One can of course load multiple files simutaneously. If you
choose to load them by draging or open different files sequentially, you will have
each sample opened in separate windows, but if you wish to directly compare them
on the same figure then load the first by File-open and then load all subsequnet files
by File-Insert. Then you can view them simutaneous under compare view, but for
Rietveld refinement only (the first) one file can be processed under analyze view.
2. The first step towards a crystallographic refinement is background determination, you
want to fine the position and intensity of reflections/peaks, for this you first need to
determine and then remove the effect of background. Right click in the main window
choose Determine background,adjust bending factor (0-20) and granularity, you will
see a smooth curved generated below you XRD pattern, this is the calculated
background, adjust both factors untill this curve matches well you pattern, click
background-accept. Do not click subtract which will directly subtract the intensity on
your pattern by calculated background, then you have changed you raw data.
3. Then you want to trace all the diffraction ‘peaks’, whose position (2θ and d-spacing)
and peak height (intensity) will be what we are mainly interested thereafter. Right
click search peaks-accept, there are various mathematical defination one can set to a
‘peak’ from a fluctuating background. Usually one can just accept the default method.
Once you clicked accept, numerous of small vertical arrows will be displayed above
the main figure, which signifies the list of peaks the programm has discovered (you
can also view/select/delete them in the middle panel under peak list.

Peak labels, dashed line for Kα2 peak

As Cu Kα2 peak cannot be filtered, effectively

every reflection has two peaks Kα1 and Kα2
(higher angle and weaker), these two peaks
are closer/indistinguishable at low 2θ but
distance themselves with increasing 2θ.

Peak labels, dashed line for Kα2 peak

Background curve

Often weak peaks at high angles becomes entangled

with fluctuating background, one can either manually
delete these peaks or just leave it (in the end it will
make the overall fit a bit worse.

4. Now we can start thinking about what minerals/phases are present in this sample.
We do this by compare our pattern with existing patterns from available
crystallographic database (i.e. COD 2021, PDF-2/4 etc. already installed). One can
start a search either by right click and search & match or click analysis (top panel)-
search and match-execute search and match. Then in the pop-out window choice
Edit and put some constrains before click search and OK.

Here you could specify constrains of a search

(chemical/crystallographic)

To speed up the search, one should at least

select some subfiles like minerals/inorganic
 Drag candidate upwards to confirm its presence

Likelywood

Candidate lists

When click on one candidate, whose diffraction

pattern will also be shown here for comparison

5. Now you will see a list (often excessive amount) of candidates displayed in the
middle-lower panel which are ranked based on the quality of match (score) to your
pattern. Compare carefully if one phase is likely present in your sample. There
all over 100’000 XRD patterns with each containing multiple peaks, and your sample
is also often a mixture. Therefore you should first have a good estimation on the
chemistry and what mineral phases you could potentially have, before
dragging random weird minerals upwards into pattern list (candidate list).
Typically due to various experimental errors (temperature error, sample height
displacement, preferred orientation etc, non-perfect match between sample and
database i.e. for solid solutions, an olivine with different Mg# number), the
experimentally measured peak position/relative intensity is somewhat deviated from
the database value (usually should <0.3º otherwise the program will have problem of
recognition).
Usually you could starting viewing the candidates with highest scores, zoom in
on the main window to see how far the measured peak position differs from
database value (and if choose the same phase with a different composition with help
to reduce this difference), check the match/mismatch of respective candidates not
just at one reflection but rather examine the match across the entire scan range
(some mismatch is usually, the intense peaks should match reasonably well). Apart
from the peak position, another important check is peak intensity. Double click on a
candidate file that you find plausible, score down. You will see a table containing
miller index, d-spacing, 2θ and relative intensity (% over the most intensive peak in
this phase). Check if the most intense peaks (60% or above) is reasonably present in
your patter as well. Accepting candidate phase by dragging it upwards to the
upper panel (pattern list). Afterwards, you will see a number of ‘arrows’ above
the figure disappeared, meaning these peaks are tentatively assigned to the
candidate phase and accounted for. Now repeat this process until all major peaks
are accounted for. When this is done, select all accepted patterns from ‘pattern
list’ panel, right click and say convert pattern to phase.
Ideally all peaks should be accounted for so no more arrows and no more high score
candidates present as well. For single phase and pure compound, this should be what we
aim for, but for complex mixture like rocks which are multi-phase mixture with chemical
substitution, some remaining peaks are no big problems.
6. After converting pattern to phase, we need to delete all the peak list to avoid
confusing in the next step, so go to peak list, select all and delete. Now we are ready
to start the Rietveld refinement.
 7. We now go to refinement control (middle upper panel). Rietveld refinement is a
crystallographic model that fits/tries to reproduce experimentally measured XRD
pattern by 1. Input one or multiple existing crystallographic models (these are the
candidates that we have just selected, they provide a starting point of our
refinement/model including parameters like space ground/symmetry, chemistry, site
occupancy, lattice parameter, intrinsic intensity of X-ray diffraction for each reflection.
It is not possible to start a Rietveld refinement without a priori crystallographic model
(at least a space group), 2. Refine these parameters/models so as to better fit our
experimental results (which basically the program iterates these parameters to match
the shape of our XRD pattern).
As you will soon find out, given the excessive number of parameters involved, the
refinement strategy as well as some rational sense is critical to achieve a reasonably
good result.
During sample preparation, there is often a slight error induced by sample height errors, always tick

Systematic/constant ‘error’/’shift’ of all 2θ, since it has been pre-calibrated on instrument, so 0 is fine. For simple samples use polynomial, tick flat, 1/x
and 1-5 coefficients.

For complex mixtures, single click Global

Account for sample porosity, for
variables/background, and on the right panel,
our rocks, usually can just leave it.
click background, choose Chebyshev 1, tick 6-7
variable.

In the end you want to have a reasonable match

between your measured and fitted background.

Each phase has a subfolder, the basic strategy is to refine the more abundant phases and more important
parameters first, others later and do it in a progressive manner
Scale factor = wt.% proportion of this phase in the entire mixture
March-Dollase factor, a model to account for preferred orientation

Uni cell parameter, tick according to crystal class, i.e. for cubic crystals only tick a.
One thing, for trigonal system (e.g. quartz) the database usually treat it as hexagonal so just change a and c.

Caglioti W left and peak shape 1 left, Parameters to adjust peak shapes at high 2θ
 Instead of manually selecting the parameters to refine, it is also possible to do an automatic/default Rietveld refinement by:
Under automatic mode, click F, select Default Rietveld.
Note, this should only be used once you complete all previous steps, importantly converting pattern to phase, delete peak list.

In addition, there are relatively less parameters varied in this default refinement. Therefore one should only use it to get a first
and crude idea, before then manually check and refine additional parameters.

First switch to manel mode, tick the variables you want to refine (again progressively), click this F function to start refinement

8. Now to start refinement, first switch to manel mode, tick the variables you want to
refine (again progressively), click this F function to start refinement.
Refinement sequence:
First stage: Global variables including specimen displacement, background fitting
curve (polynomial for simple mixture, Chebyshev 1 for complex background shapes),
tick all major phases (maybe >15-20%), scale factor, unit cell parameters,
profile variable-Caglioti W left, peaks shape 1 left.
Now refine the minor phases (parameters listed above) and less critical parameters
of major phases:
Theoretically, a typical XRD reflection should be a mixture of Gaussian+Lorentzian
profile (approximately 50:50), but in practice the measured peaks tend to be more
asymmetric in shape (especially the left side). Another possibility is preferred
orientation of crystals (during sample preparation).
Second stage, only for major phases, refine the asymmetry (click on phase, go
to right panel, find the asymmetry item, choose split width/shape or other
models), PLO (click on phase, go to right panel, find preferred orientation, use
March-Dollase/spherical harmonics, specify the type or PLO, Caglioti U and V
left parameters.
Run refinement and check the fitting quality/residue at every step by right click the
Left bottom panel and chose difference plot, the lower the RWP the better (optimal
value 1).
However it is always important to think about whether refining some parameters
make sense. E.g. if changing a lot parameters does not improve the fit significantly
then maybe it is not worthy. Typically for most mineralogical/petrological we are
either interested in the crystallinity (in which case we just see if there are peaks or
just a broad bump as is typically for glass), lattice parameter/chemical composition or
phase proportion of mixture (model abundance of minerals in a rock). If quantitative
 analysis is not desired (when you just want to know what phase can be present),
then one can simply search and match the phase manually without performing the
Rietveld refinement.

This sample is mixed at a ratio of 3:3:4

9. The ultimate refinement model should be checked by residue (rwp value), the
goodness of match between modelled profile vs. measured profile and absence of
unexplainable feature (unidentified/unattributed peaks, mismatch of intensity etc.)

Finally one word about using XRD to get phase proportion, for a natural rock with
multiple mineral phases which all have solid solutions, it is not always easy to find a
suitable database candidate. As minerals with low symmetry has much more
complicated XRD pattern, feldspar is particularly difficult to fit/refine (actually it is
almost not possible to do it perfectly).
Another point is the sensitivity of XRD on phase abundance, as the phase proportion
is directly proportional to the refined scale factor (in addition to also other factors like
PLO, chemistry), XRD is intrinsically only capably to get a model abundance for
phases that scatters strong enough (only they offer at least one relatively strong
reflection in your XRD pattern, which usually means they should be at least 5 wt. %
or more, unless this phase scatters much stronger than others i.e. a high Z phase like
Fe or Au metal). In the above mentioned example we see for very ideal mixture (3
phases, approximate proportion, all pure components with good quality database
reference model) this estimation can be extremely accurate, but for a natural rock
powder/mixtures 10-20% error should always be born in mind.
Red measured profile
Blue modeled profile
The more matching the better
 Bug shooting

Forgot to define background

If you see something like this when starting a refinement, where modeled pattern looks
completely detached from the measured one, this usually comes from the fact that you did
not specify the background pattern at refinement step (different from first step).
Solution:
In refinement control, select background and fit new background using polynomial or
Chebyshev function.
About XRD database
For now we have two XRD diffraction pattern database:
PDF-2 2021 (suitable for phase identification, more mineralogical data, somewhat weird
name/nomenclature for minerals, alas, not suitable for Rietveld refinement (not enough
crystallographic information).
COD 2021, free XRD database, suitable for both phase identification and Rietveld
refinement (default database).

If necessary one can switch the database under customize-manaage database-and just
unclick the other one before activate the one you need (no compatible simultaneously).
Please when you are done always go back to COD 2021 database.