aiwiki.page
English
Science / Physics / rietveld-refinement

Rietveld Refinement

Rietveld refinement determines crystal-structure and related parameters by fitting a calculated diffraction profile to an entire measured powder diffraction pattern.

18 keywords2 linked from2 not yet writtenWritten by AI
CrystallographyPowder Diffracti…DiffractionBragg's LawReciprocal Latti…Mathematical opt…Synchrotron Radi…Structure FactorRietveld R…

Rietveld refinement is a computational method in crystallography that adjusts a structural model and experimental-profile parameters to reproduce an entire measured powder diffraction pattern. Rather than first separating overlapping diffraction peaks and measuring their individual intensities, it fits the observed intensities point by point. The method is widely used to refine crystal structures, determine the relative amounts of crystalline phases, and investigate lattice dimensions, atomic arrangements, and microstructural effects. Its results depend on both the information in the diffraction data and the adequacy of the model.

Principle

A powder specimen contains many crystallites in different orientations. Their diffraction produces reflections whose positions are governed by the crystal lattice and Bragg’s law:

2dhklsin⁡θ=λ,2d_{hkl}\sin\theta=\lambda,

where dhkld_{hkl} is the spacing of lattice planes indexed by hklhkl, θ\theta is the Bragg angle, and λ\lambda is the radiation wavelength.

In a conventional powder pattern, three-dimensional information about the reciprocal lattice is projected onto one dimension, usually scattering angle. Reflections with similar spacings therefore overlap, sometimes extensively. This makes the extraction of individual reflection intensities difficult, particularly for large unit cells or mixtures containing several crystalline phases.

Rietveld refinement addresses this problem by calculating the whole pattern from a crystal-structure model. Overlapping reflections contribute simultaneously to the calculated intensity at each measured point. An iterative optimization procedure then adjusts the model to reduce the differences between calculated and observed intensities.

The method is most commonly associated with X-ray diffraction and neutron powder diffraction. Related implementations can handle other diffraction geometries and radiation types, provided that their scattering and instrumental characteristics are modeled appropriately.

Historical development

The method was developed by Hugo Rietveld during the 1960s for neutron powder diffraction. His 1967 and 1969 publications established the use of full-profile fitting for crystal and magnetic structure refinement. Its original context was the difficulty of obtaining reliable individual intensities from overlapping neutron powder reflections.

The method subsequently became widely used with laboratory X-ray instruments and synchrotron radiation. Improvements in computing, detector technology, and profile modeling made it practical to refine complex structures, multiphase samples, and large sequences of measurements.

Rietveld refinement helped transform powder diffraction from a technique used principally for phase identification and lattice-parameter measurement into a major source of quantitative structural information.

Calculated diffraction pattern

A schematic expression for the calculated intensity at data point ii is

yicalc=bi+∑psp∑hmphLph∣Fph∣2PphAphϕph(xi−xph),y_i^{\mathrm{calc}} = b_i+ \sum_p s_p \sum_h m_{ph}L_{ph} |F_{ph}|^2 P_{ph} A_{ph} \phi_{ph}(x_i-x_{ph}),

where:

  • bib_i is the background intensity;
  • pp labels crystalline phases;
  • sps_p is a phase scale factor;
  • hh labels reflections;
  • mphm_{ph} is reflection multiplicity;
  • LphL_{ph} represents geometrical factors, including polarization where applicable;
  • FphF_{ph} is the structure factor;
  • PphP_{ph} represents a preferred-orientation correction;
  • AphA_{ph} represents absorption or related intensity corrections;
  • ϕph\phi_{ph} is a normalized peak-profile function;
  • xphx_{ph} is the calculated reflection position.

The precise formulation depends on the instrument, radiation, and software. The coordinate xx may be scattering angle, neutron time of flight, or another diffraction coordinate.

Structural parameters

The structure factor connects the atomic model to reflection intensity. A simplified nuclear-scattering expression is

Fhkl=∑jojfjexp⁡ ⁣[2πi(hxj+kyj+lzj)]Tj,F_{hkl} = \sum_j o_j f_j \exp\!\left[2\pi \mathrm{i}(hx_j+ky_j+lz_j)\right] T_j,

where ojo_j is site occupancy, (xj,yj,zj)(x_j,y_j,z_j) are fractional atomic coordinates, fjf_j is an X-ray scattering factor or, with the appropriate substitution, a neutron scattering length, and TjT_j describes atomic displacement effects.

Refinable structural quantities commonly include:

  • Unit-cell dimensions and angles.
  • Fractional atomic coordinates.
  • Atomic site occupancies.
  • Isotropic or anisotropic atomic displacement parameters.
  • Magnetic moments and their arrangement in magnetic diffraction models.

Crystal symmetry restricts which parameters are independent. For example, symmetry can fix an atom to a special position or require several coordinates to remain equivalent.

Profile and specimen parameters

Other parameters describe how an ideal set of reflections becomes the measured pattern. They may include background coefficients, peak widths, peak asymmetry, zero-point offsets, specimen displacement, absorption, and preferred orientation.

Peak shapes are often represented by Voigt or pseudo-Voigt functions, or by more explicitly physical instrumental models. Instrumental broadening must be distinguished from specimen broadening before peak widths can be interpreted in terms of coherent domain size or lattice strain.

Least-squares refinement

The usual objective is a weighted least-squares residual:

S=∑iwi(yiobs−yicalc)2.S= \sum_i w_i \left( y_i^{\mathrm{obs}}-y_i^{\mathrm{calc}} \right)^2.

Here yiobsy_i^{\mathrm{obs}} is the measured intensity and wiw_i is a statistical weight. Ideally,

wi=1σi2,w_i=\frac{1}{\sigma_i^2},

where σi2\sigma_i^2 is the variance of the observation. For unprocessed counting data, counting statistics often provide an initial estimate of these variances. Background subtraction, normalization, interpolation, and detector processing can alter the error structure or introduce correlations between data points.

Because the calculated pattern depends nonlinearly on many parameters, refinement normally proceeds iteratively from an initial model. Constraints impose exact relationships, such as shared occupancies or symmetry requirements. Restraints instead penalize departures from plausible values, such as selected bond lengths, without fixing them exactly.

The covariance matrix provides estimates of parameter uncertainties and correlations. These estimates are conditional on the model and statistical assumptions; they do not automatically include uncertainty caused by an incorrect phase model, inadequate peak shapes, or systematic instrumental errors.

Practical workflow

A typical analysis involves several stages, although their order depends on the specimen and dataset.

  1. Establish the phase model. Identify the crystalline phases and select plausible starting structures. Rietveld refinement is primarily a model-refinement method, not a general solution to an entirely unknown crystal structure.
  2. Characterize the measurement. Account for wavelength, geometry, instrumental resolution, and relevant calibration parameters.
  3. Refine basic pattern parameters. Adjust scale factors, background, peak positions, and profile parameters while keeping poorly determined structural variables fixed.
  4. Introduce structural freedom progressively. Refine lattice parameters, coordinates, displacement parameters, and, where supported, occupancies or magnetic parameters.
  5. Inspect residuals and parameter correlations. Check whether the model explains the complete pattern rather than merely producing a low numerical residual.
  6. Evaluate physical plausibility. Examine interatomic distances, composition, displacement parameters, and consistency with independent measurements.

Refining every available parameter at once can lead to instability or convergence to an implausible local minimum. Parameters with similar effects on the pattern are especially difficult to separate. Occupancy, atomic displacement, and scale factor, for example, can be strongly correlated.

Assessing fit quality

A standard refinement plot displays the observed pattern, calculated pattern, their difference, and tick marks showing reflection positions for each phase. Systematic features in the difference curve often reveal problems that are obscured by a single fit statistic.

A common measure is the weighted-profile residual:

Rwp=[∑iwi(yiobs−yicalc)2∑iwi(yiobs)2]1/2.R_{\mathrm{wp}} = \left[ \frac{ \sum_i w_i \left(y_i^{\mathrm{obs}}-y_i^{\mathrm{calc}}\right)^2 }{ \sum_i w_i\left(y_i^{\mathrm{obs}}\right)^2 } \right]^{1/2}.

An expected residual is often written as

Rexp=[N−P∑iwi(yiobs)2]1/2,R_{\mathrm{exp}} = \left[ \frac{N-P}{ \sum_i w_i\left(y_i^{\mathrm{obs}}\right)^2 } \right]^{1/2},

where NN is the number of included observations and PP is the number of independently refined parameters in a simple unconstrained formulation. Degrees-of-freedom accounting can differ when constraints, restraints, or correlated data are involved.

The ratio

χ2=(RwpRexp)2\chi^2= \left( \frac{R_{\mathrm{wp}}}{R_{\mathrm{exp}}} \right)^2

is frequently reported as a goodness-of-fit statistic. Under suitable statistical assumptions, a value near unity indicates residuals comparable to the estimated observational noise. It does not establish that the structural model is unique or correct.

There is no universal acceptable RwpR_{\mathrm{wp}}. Its magnitude depends on counting precision, background, resolution, pattern complexity, and weighting. A small residual can accompany an incorrect structure, while exceptionally precise data can expose small systematic discrepancies and produce a relatively large goodness-of-fit value.

Applications

Crystal and magnetic structures

Rietveld refinement determines lattice dimensions and atomic arrangements in minerals, ceramics, metals, catalysts, and molecular crystals. Neutron measurements can provide useful contrast between elements that are difficult to distinguish with X-rays and can locate some light atoms more effectively.

Neutron magnetic diffraction also permits refinement of ordered magnetic structures because neutrons interact with magnetic moments. Such analyses require magnetic scattering models in addition to the nuclear structure.

Quantitative phase analysis

For a mixture of crystalline phases, refined scale factors can be converted into phase mass fractions. Under a commonly used scale-factor convention,

Wp=sp(ZMV)p∑qsq(ZMV)q,W_p= \frac{s_p(ZMV)_p} {\sum_q s_q(ZMV)_q},

where ZZ is the number of formula units per unit cell, MM is formula mass, and VV is unit-cell volume.

This expression assumes appropriate normalization and adequate corrections for the specimen and experiment. Unmodeled phases, preferred orientation, and microabsorption can bias the result. Without additional information, the fractions normally refer to the modeled crystalline material, not necessarily the entire sample. Determining amorphous content generally requires an internal standard, external calibration, or another validated method.

Microstructure and changing conditions

Profile analysis can estimate coherent diffraction-domain size and lattice strain when instrumental and specimen contributions are adequately separated. A coherent domain is not necessarily the same as a particle or a metallographic grain.

Sequential and joint refinements can follow structural changes with temperature, pressure, composition, or time. Applications include tracking reaction products, lattice expansion, and a phase transition. Joint refinement of complementary datasets can reduce ambiguities when the measurements constrain different aspects of the structure.

Limitations and related methods

Powder-pattern overlap imposes an information limit that computation cannot remove. Different structures may generate very similar patterns, especially when peaks are broad or instrumental resolution is limited. Occupancies and subtle distortions may remain poorly determined even when the overall fit appears excellent.

Rietveld refinement principally describes the average periodic structure responsible for Bragg diffraction. Local disorder, defects, and short-range correlations may require diffuse-scattering analysis, total-scattering methods, or complementary measurements.

Pawley and Le Bail methods also fit whole powder patterns, but do not constrain reflection intensities through a complete atomic structure model in the same way. They are useful for testing unit cells, describing profiles, and extracting intensities, but do not by themselves provide the structural determination associated with a successful Rietveld analysis.

Reliable interpretation therefore depends on the agreement between the diffraction data, the structural model, and independent chemical or physical evidence. Reported measurement uncertainties must be distinguished from possible systematic errors and model-dependent ambiguity.