Structure Solution with Powder Diffraction

Joel Reid,
Industrial Scientist,
Canadian Light
Source

CLS XRD Summer

School 2022

Wednesday,
August 17th, 2022
Why Solve Crystal Structures with Powder Data?

• To perform Rietveld refinement, and get the most information

possible about your material from your powder data, a
reasonable initial crystal structure is required. What if you
don’t have a reasonable starting structure?

• While single crystal XRD is the most used method (by far) for
determining a crystal structure, many new functional materials
are difficult, if not impossible, to synthesize as large single
crystals.

• Powder diffraction structure solution is addictive. If you try it,

you may have trouble stopping. But it can also be very
challenging…
 Why is Powder Structure Solution Challenging?
2D Images
3D crystal Single Crystal
structure (hundreds of images)

1D powder
Powder pattern
(1 image)

• Powder diffraction suffers an inherent loss of information, due to the

compression of 3D crystalline structure information into 1D.
• Peak overlap means crystal structure solution with PXRD generally requires (1)
detailed initial chemical/molecular information, (2) constraints and/or restraints to
keep solutions chemically reasonable, and (3) validation with other techniques.
 What is Required Before Structure Solution?
• Before trying to solve the crystal structure, synthesize the best
possible sample; either phase-pure, or with minimal impurities.

• Based on your chemical synthesis, figure out the probable chemical

formula, and confirm it with elemental analysis (XRF, ICP-MS, etc.).

• Get high quality data, using an appropriate radiation source (lab X-

rays, synchrotron, neutron) based on the type of sample.

• Perform thorough search/match phase identification on your powder

data with appropriate database(s) like the Powder Diffraction File
(PDF-4+), Cambridge Structural Database (CSD), American
Mineralogist Crystal Structure Database (AMCSD), etc., to make sure
no one has already solved the structure. Look for analogous
structures, where an element (often a metal) is substituted by a
similar element.
Correlating Pattern Features to Crystal Structure

Figure taken from: McCusker, L. B. & Baerlocher, C., Chimia 68 (2014) 19-25
Structure Solution from Powder Diffraction Maze

Figure taken from: David, W.I.F. et al., Structure Determination from Powder
Diffraction Data. (Oxford: New York, 2002).
 Main Steps in PXRD Structure Solution
1. Collect high quality data, with source radiation (lab X-rays,
synchrotron, neutron) suitable for the problem.
2. Perform ab initio indexing to determine the unit cell of the material.
Use systematic reflection absences to determine the space group.
3. Use structure-less Le Bail or Pawley refinements to (i) test unit cells
and space groups to confirm the combination providing the best fit,
and (ii) to extract intensities for structure solution.
4. Choose an appropriate structure solution route for your material
(simulated annealing/parallel tempering, direct methods, charge
flipping, etc.) to solve the basic structure.
5. Use Rietveld refinement to refine the structure, while using Fourier
difference maps and additional chemical information to complete the
structure (often an iterative process).
6. If possible, confirm the final Rietveld-refined crystal structure is correct
with density functional theory (DFT) or other experimental techniques.
 What Type(s) of Data Should You Use?

Table taken from:

Hill, R.J..& Madsen, I.C. in
Structure Determination
from Powder Diffraction
Data. Edited by W. I. F.
David, K. Shankland, L. B.
McCusker & Ch. Baerlocher.
(Oxford: New York, 2002)
Indexing & Space Group Determination

Figure taken from: David, W.I.F. et al., Structure Determination from Powder
Diffraction Data. (Oxford: New York, 2002).
 Indexing & Space Group Determination
• Ab initio indexing is finding the correct unit cell (and space group)
description for your crystalline phase.

• Depending on the data quality and size/symmetry of the unit cell, ab

initio indexing can be the easiest or hardest step.

• There are many programs for indexing powder patterns (DICVOL,

TREOR, GSAS2, McMaille, ITO, X-Cell, etc.), which use a variety of
different algorithms and methodologies.

• Indexing is always most straightforward with a single-phase powder

pattern. Some indexing programs allow for small numbers of impurity
reflections (typically 2-4 peaks maximum in the low angle reflections),
but the purer the sample is, the better. If you know peaks are impurity
peaks, you should exclude them from indexing.

DICVOL: Boultif, A. & Louer, D.,

J. Appl. Cryst. 37 (2004) 724-731.
 Indexing & Space Group Determination
• Most (but not all) indexing programs work optimally using
approximately the 20 largest d-spacing (lowest 2θ angle) Bragg
reflections.

• Two figures-of-merit (FOM) are typically used for evaluating ab initio

indexing:

• The de Wolff FOM: 104

𝑄ℎ𝑘𝑙 = 2
𝑄20 𝑑ℎ𝑘𝑙
𝑀20 =
2 ∙ 𝑄 ∙ 𝑁20

• The Smith-Snyder FOM:

1 𝑁
𝐹𝑁 = ∙ Werner, P. E. ‘Autoindexing’,’ in:
∆2𝜃 𝑁 𝜃𝐺 Structure Determination from Powder
Diffraction Data. Ed: David, W.I.F. et al.
(Oxford: New York, 2002).
 Indexing & Space Group Determination
• Some programs work better in different circumstances, but it is
always helpful to compare results with more than one program.

• A few software platforms facilitate access to multiple indexing

programs, like WinPLOTR/FullProf Suite (DICVOL, TREOR, ITO)
and EXPO2014 (DICVOL, TREOR, McMaille).

• Once you have a tentative unit cell, a number of programs can

examine the observed reflections for systematic absences, to
determine the most likely space groups (ChekCell, EXPO2014,
GSAS2, FOX, Topas, etc.).

• You always need to make sure your unit cell volume and the general
multiplicity of your space group are compatible with the formula. The
cell contents (formula, Z) need to make sense with the unit cell.
Estimating Cell Contents (Z) from Formula, Volume
• The number of non-
hydrogen atoms in the
unit cell for an organic
compound can be
estimated by dividing the
volume by 15 to 20
(typically 17 or 18, often
called the ‘Rule of 18’).

• Some software packages

like EXPO have
calculators for estimating
the unit cell volume based
on the formula and Z
value.
Le Bail & Pawley Refinements

Figure taken from: David, W.I.F. et al., Structure Determination from Powder
Diffraction Data. (Oxford: New York, 2002).
 Le Bail & Pawley Refinements

• The Le Bail and Pawley methods were developed to allow full profile
fitting of powder patterns using (1) a unit cell (with space group) to
define the peak positions and (2) the peak shape/width parameters,
but without requiring a crystal structure.

• The peak intensities are optimized to best fit the experimental data
without any structural model constraints (i.e. atomic coordinates).

• These methods are critical for getting estimates of the integrated

reflection intensities (or structure factor amplitudes, |Fhkl| ) from
powder patterns for the solution of unknown crystal structures.

Pawley, G. S. Le Bail, A, Duroy, H. & Fourquet, J. L.

J. Appl. Cryst. 14 (1981) 357-361. Mater. Res. Bull. 23 (1988) 447-452.
 Le Bail & Pawley Refinements

• The Pawley method treats individual reflection intensities as least-

squares parameters to be refined (resulting in a very large least-
squares matrix for large, low symmetry unit cells).

• The Le Bail method uses an iterative profile intensity partitioning

method to estimate the calculated intensities from the observed
pattern.

• The Le Bail method is more similar mathematically to the Rietveld

method, therefore it has been more widely incorporated in Rietveld
programs than the Pawley method. But one or both methods are
available in all major Rietveld programs (GSAS2, FullProf,
JANA2006, Topas etc.).

Pawley, G. S. Le Bail, A, Duroy, H. & Fourquet, J. L.

J. Appl. Cryst. 14 (1981) 357-361. Mater. Res. Bull. 23 (1988) 447-452.
 Comparison of Rietveld & Le Bail Refinements

Rietveld Le Bail
χ2=2.47 χ2=1.36

• The Le Bail refinement sets a lower limit on the achievable reduced

chi-squared (χ2), and provides a reference for how good your
Rietveld refinement is.
• This can be helpful for deciding if your final refinement and structural
model are reasonable.
Comparison of Rietveld & Le Bail Refinements

Rietveld Le Bail
χ2=2.47 χ2=1.36

MoO2(O2)H2O

Reid, J. W., Kaduk, J. A. & Matei, L.

Powder Diffraction 33 (2018) 49-54.
Crystal Structure Solution

Figure taken from: David, W.I.F. et al., Structure Determination from Powder
Diffraction Data. (Oxford: New York, 2002).
 Crystal Structure Solution

Direct-Space Reciprocal-
(DS) or ‘Real- Space (RS)
Space’ Methods Methods

Figure taken from: David, W.I.F. et al., Structure Determination from Powder
Diffraction Data. (Oxford: New York, 2002).
 Crystal Structure Solution

• Crystal structure solution strategies for powder data can

generally be divided into two types of methods:

1. Reciprocal-space methods:
• Direct methods/Patterson methods (EXPO2014, Topas,
XLENS)
• Charge flipping (GSAS2, SUPERFLIP, JANA2006)
• Maximum entropy methods (MICE)

2. Direct-space (real-space) methods:

• Simulated annealing (GSAS2, DASH, Topas, PSSP, FOX)
• Parallel tempering (FOX)
• Monte Carlo and hybrid methods (ESPOIR, POWDERSOLVE)

Structure solution review article: Meden, A. & Radosavljevic Evans, I.,

Cryst. Res. Technol. 50 (2015) 747-758.
 Reciprocal-Space Methods
• Reciprocal-space (RS) methods are based on single-crystal structure
solution methods, which over time have been optimized for powder
data, and use individual Bragg reflection integrated intensities ( |Fhkl| )
extracted from the full PXRD pattern.

• The key to success with these methods is getting a sufficient number

of accurate (non-overlapped) intensities with data content out to
atomic resolution (d = ~1 Å). These methods tend to need minimal to
moderate reflection overlap, which makes them more successful for
smaller, higher symmetry unit cells (and higher resolution data).

• RS methods generally require minimal a priori chemical information

to find a basic solution, but you still want to know the chemical
information, if possible, to confirm (or adjust) the fine details of the
solution.
 Indexing and Solution Demonstration: EXPO2014

• Indexing synchrotron powder data (E = 18 keV, λ = 0.68908 Å)

for cytosine (C4H5N3O) using DICVOL, via EXPO2014, and
solution with direct methods (EXPO2014).

EXPO: Altomare, A. et al.,

J. Appl. Cryst. 46 (2013) 1231-1235.
 Direct-Space Structure Solution Methods
• Direct-space (DS) methods tend to use the whole pattern, rather than
extracted integrated intensities. They follow a global optimization
strategy, where molecules or multiple structural fragments are moved
around the unit cell in ‘random walk’ processes.

• The entire powder pattern is calculated at every step, to compare

with the observed pattern, and the best fit(s) are preserved.

• DS methods tend to be less dependent on resolution (peak breadth)

and data content (minimum d-spacing) than reciprocal-space
methods, because they use the entire pattern.

• But to use DS methods, you need to have reasonable chemical

information like the molecule(s) present in the structure (organics) or
the approximate cell contents and polyhedral fragments (inorganics).

DS review: Cerny, R. & Farve-Nicolin, V.,

Z. Kristallogr. 222 (2007) 747-758.
 Structure Solution Example: FOX
• Laboratory powder data (Cu Kα1, λ = 1.54059 Å) structure solution
for trandolapril (C24H34N2O56) using FOX with parallel tempering.

• Indexing performed with

DICVOL determined an
orthorhombic cell (a = 19.7145,
b = 15.0499, c = 7.6534 Å).

• The space group was

determined to be P212121 using
ChekCell.

• A trandolapril molecule was built

from two entries in the
Cambridge Structural
Database (CSD) using
Avogadro.
FOX: Farve-Nicolin, V. & Cerny, R.,
J. Appl. Cryst. 35 (2002) 734-743.
 Structure Solution Example: FOX
• Laboratory powder data (Cu Kα1, λ = 1.54059 Å) structure solution
for trandolapril (C24H34N2O56) using FOX with parallel tempering.

• Indexing performed with

DICVOL determined an
orthorhombic cell (a = 19.7145, SIWCAC
b = 15.0499, c = 7.6534 Å).

• The space group was

determined to be P212121 using
ChekCell.

• A trandolapril molecule was built

from two entries in the
Cambridge Structural FEFKEI
Database (CSD) using
Avogadro.
CSD: Groom, C. R. et al.,
Acta Crystallogr. B 72 (2016) 171-179.
 Structure Solution Example: FOX
• Laboratory powder data (Cu Kα1, λ = 1.54059 Å) structure solution
for trandolapril (C24H34N2O56) using FOX with parallel tempering.

• Indexing performed with

DICVOL determined an
orthorhombic cell (a = 19.7145, SIWCAC
b = 15.0499, c = 7.6534 Å).

• The space group was

determined to be P212121 using
ChekCell.

• A trandolapril molecule was built

from two entries in the
Cambridge Structural FEFKEI
Database (CSD) using
Avogadro.
Avogadro: Hanwell, M. D. et al.,
J. Cheminform. 4 (2012) 17.
 Structure Solution Example: FOX
• Laboratory powder data (Cu Kα1, λ = 1.54059 Å) structure solution
for trandolapril (C24H34N2O56) using FOX with parallel tempering.
• A trandolapril molecule was built
from two entries in the
Cambridge Structural
Database (CSD) using
Avogadro.

• A crystallographic information
file (CIF) of the molecule was
converted to a Fenske-Hall Z-
matrix (FHZ file) using Open
Babel.

• The FHZ file was used to load

the molecule into FOX for
structure solution.
Reid, J. W., Kaduk, J. A. & Vickers, M.
Powder Diffraction 31 (2016) 205-210.
Open Babel: O’Boyle, N. et al.,
J. Chem. Inform. 3 (2011) 1-14.
 Structure Solution Example: FOX
• Laboratory powder data (Cu Kα1, λ = 1.54059 Å) structure solution
for trandolapril (C24H34N2O56) using FOX with parallel tempering.

• A trandolapril molecule was built

from two entries in the
Cambridge Structural
Database (CSD) using
Avogadro.

• After solving the crystal

structure with FOX, the
structure was Rietveld refined
with GSAS, using restraints on
the bond distances and angles
generated by Mogul (CSD).
Reid, J. W., Kaduk, J. A. & Vickers, M.
Powder Diffraction 31 (2016) 205-210.
GSAS-II: Toby, B. H. & von Dreele, R. B.
J. Appl. Cryst. 46 (2013) 544-549.
Examining Bond Distances and Angles with Mogul
• Mogul allows you to upload a molecule and compare bond distances
and angles to geometries from structures in the CSD.

Mogul: Bruno, I. J. et al.

J. Chem. Inf. Comput. Sci. 44 (2014) 2133-2144.
Examining Bond Distances and Angles with Mogul
• Mogul allows you to upload a molecule and compare bond distances
and angles to geometries from structures in the CSD.

Mogul: Bruno, I. J. et al.

J. Chem. Inf. Comput. Sci. 44 (2014) 2133-2144.
 Crystal Structure Validation

• Once you have a tentative structure, it’s important to take

multiple steps to confirm it is (likely) correct:
• Make sure you can obtain a quality final Rietveld refinement
(reasonable statistical metrics, and more importantly, a decent
visual fit to the pattern).
• Make sure the structure is consistent with the available elemental
and chemical analysis (XRF, ICP-MS, NMR, IR, etc.).
• Check the structure to make sure it is chemically reasonable.
• See how closely the Rietveld-refined structure compares with a
density functional theory (DFT) optimized structure, if possible.

Evaluating statistical metrics: Chemical reasonableness - organics:

Toby, B. Powder Diffr. 21 (2006) 67-70. Kaduk, J. A. Powder Diffr. 22 (2007) 74-82.

Chemical reasonableness - inorganics:

Kaduk, J. A. Powder Diffr. 22 (2007) 268-278.
Structure Validation - Chemical Reasonableness

• For organic structures, run the structure through the

Mogul module of the Cambridge Structural Database
(CSD) to check the bond distances and angles.

• For inorganic structures, perform bond valence sum

calculations to see if the bond lengths and valences are
reasonable (BondStr in the FullProf Suite).

• For either type of structure, run your final CIF through

CheckCIF (https://checkcif.iucr.org/).
Bond Valence Model: Chemical reasonableness - organics:
Brown, I. D. The Chemical Bond in Inorganic Kaduk, J. A. Powder Diffr. 22 (2007) 74-82.
Chemistry: The Bond Valence Model.
(Oxford: New York, 2002). Chemical reasonableness - inorganics:
Kaduk, J. A. Powder Diffr. 22 (2007) 268-278.
 What is Density Functional Theory (DFT)?
• DFT is a quantum mechanical computation method that models
electronic (atomic) structure and related properties in terms of the
electron density in the system.

• Some excellent introductory resources and primers exist regarding

the theory and application of DFT:
• Burke, K., & Wagner, L.O. ‘DFT in a Nutshell,’ Int. J. Quantum Chem.
(2012) 24259.
• Fiolhais, C., Nogueira, F., & Marques, M. (Eds.). A Primer on Density
Functional Theory. Springer: Berlin (2003).
• Thakkar, A. J. Quantum Chemistry: A concise introduction for students of
physics, chemistry, biochemistry and materials science (2nd Ed).
Morgan and Claypool: San Rafael (2017).

• There are many software packages for DFT (CRYSTAL, VASP,

ORCA, CASTEP, Quantum Expresso, Gaussian, Hyperchem, etc.).
 Why Use DFT with Powder Diffraction?

• Of the numerous types of quantum mechanical calculations,

DFT is one of the most used, successful and versatile methods
available.

• DFT can help you:

1. Validate your results and determine if your interpretation
of your data is correct.
2. Provide more accurate structural details (i.e. hydrogen
positions and bonding) where the experimental data is
ambiguous or completely lacking.
3. Gain additional insights into properties of your material
that might not be clear with the experimental data alone.
 Why use DFT with PXRD Structure Solution?

• For small molecule crystal structures, PXRD provides the basic

packing arrangement of the molecules, but DFT can improve
the details:

Figure taken from:

van de Streek, J. ‘Reliable and Highly Accurate Molecular Crystal Structures from a Combination
of XRPD and DFT-D.’ Presented at Accuracy in Powder Diffraction IV (2013).
Benchmarking PXRD Structure Validation with DFT
• The average root-mean-square Cartesian displacement (RMSCD)
between the non-hydrogen atoms of the PXRD and DFT structures
can be used to validate the correctness of the PXRD structure.

Correct ? Incorrect
# Structures

RMSCD between PXRD and DFT (Å)

van de Streek, J. & Neumann, M. A. Acta Cryst B70 (2014) 1020-1032.
Benchmarking PXRD Structure Validation with DFT
• The average root-mean-square Cartesian displacement (RMSCD)
between the non-hydrogen atoms of the PXRD and DFT structures
can be used to validate the correctness of the structure.

Solifenacin hydrogen succinate Mupirocin form I,

RMSCD = 0.078 Å RMSCD = 1.02 Å

Rietveld
DFT
Kaduk, J.A. et al. Powder Kaduk, J.A. et al. Powder
Diffraction 30 (2015) 211-217 Diffraction 31 (2016) 118-125
 DFT Example: Donepezil Hydrochloride
Donepezil HCl, form III
C24H29NO3·HCl

Lab PXRD, Cu Kα1 (λ = 1.54059 Å)

Upper Limit: 2θ = 35° (dmin = ~ 2.5 Å)

Can we actually solve the correct

structure with this data range?
 DFT Example: Donepezil Hydrochloride
Donepezil HCl, form III
C24H29NO3·HCl

RMSCD = 0.14 Å

Lab PXRD, Cu Kα1 (λ = 1.54059 Å)

Upper Limit: 2θ = 35° (dmin = ~ 2.5 Å)

Can we actually solve the correct

structure with this data range? Yes

Rietveld
DFT
 DFT Example: Donepezil Hydrochloride
Donepezil HCl, form III
C24H29NO3·HCl

RMSCD = 0.14 Å

Lab PXRD, Cu Kα1 (λ = 1.54059 Å)

Upper Limit: 2θ = 35° (dmin = ~ 2.5 Å)

Can we actually solve the correct

structure with this data range? Yes
Reid, J.W. & Kaduk, J.A.,
Rietveld Powder Diffraction 36 (2021) 233-240
DFT
 Software for Structure Solution
Step Software
Search/Match and Data Mining Powder Diffraction File (PDF-4+),
Databases Cambridge Structural Database (CSD)
Ab initio indexing DICVOL, TREOR (via FullProf Suite),
McMaille (via EXPO2014), GSAS2
Space Group Determination ChekCell, EXPO2014, GSAS2

Le Bail & Rietveld Refinement GSAS2, FullProf, JANA2006, Topas

Reciprocal-Space Structure Solution EXPO2014, SuperFlip (via JANA2006),

GSAS2, XLENS

Direct-Space Structure Solution FOX, GSAS2, GALLOP

Structure Building, Editing, Plotting & Avogadro, VESTA, JMOL, Mercury

File Conversion (CSD), Open Babel
Bond Distances, Angles, Valences and Mogul (CSD), BondStr, CheckCIF
Structure Validation
Density Functional Theory (DFT) CRYSTAL17, VASP, Quantum Expresso,
CASTEP, Gaussian, ORCA
 Summary

• Structure solution from powder diffraction data is a powerful

tool that comes with challenges. But for some materials, it may
be the only viable route to a crystal structure.

• Much more so than single-crystal XRD, structure solution from

powder data generally requires:
• Significant knowledge of the molecular structure and chemistry of
the material.
• Constraints and restraints during both structure solution and
refinement to keep solutions chemically reasonable.
• Validation and verification of the structure using additional
chemical and physical data and/or DFT optimization, if possible.
 Acknowledgements

• Many thanks are due to Jim Kaduk for multiple

collaborations, many helpful discussions regarding
aspects of structure solution and density functional
theory, and for providing a constant source of inspiration.
 Further Reading

• David, W.I.F., Shankland, K., McCusker, L. & Baerlocher, C. (ed.) Structure

Determination from Powder Diffraction Data.
Oxford: New York, 2002.

• Gilmore C. J., Kaduk, J. A. & Schenk, H. (ed.) International Tables for

Crystallography Volume H: Powder Diffraction.
Wiley: New York, 2019.

• Pecharsky, V.K. & Zavalij, P.Y. Fundamentals of Powder Diffraction and

Structural Characterization of Materials, 2nd Ed.
Springer: New York, 2009.

• Dinnebier, R.E. & Billinge, S.J.L. (ed.) Powder Diffraction, Theory and
Practice.
RSC: Cambridge, 2009.