What is Rietveld Refinement Method? Thierry Roisnel Laboratoire de Chimie du Solide et Inorganique Moléculaire UMR6511 CNRS– Université de Rennes 1 (France) The Rietveld Method, T.R. (Mexico 21-25 June 2004) A powder diffraction pattern from the data treatment point of view A powder diffraction pattern can be recorded in numerical form for a discrete set of scattering angles, times of flight or energies. We will refer to this scattering variable as : T. The experimental powder diffraction pattern is usually given as two arrays : {Ti , yi }i =1,...,n The Rietveld Method, T.R. (Mexico 21-25 June 2004) Powder diffraction profile: scattering variable T: 2θ, TOF, Energy Bragg position Th yi yi-yci zero Position “i”: Ti The Rietveld Method, T.R. (Mexico 21-25 June 2004) 1. The profile of powder diffraction patterns profile The profile can be modelled using the calculated counts: yci at the ith step by summing the contribution from neighbouring Bragg reflections plus the background. The model to calculate a powder diffraction pattern is (for a single phase): yci = bi + h2 ∑ I Ω(T − T ) h = h1 The Rietveld Method, T.R. (Mexico 21-25 June 2004) h i h The profile of powder diffraction patterns yci = bi + ∑ I h Ω(Ti − Th ) h ∫ +∞ −∞ Ω( x)dx = 1 Profile function characterized by its full width at half maximum (FWHM=H) and shape parameters (η, m, ...) Ω( x) = g ( x) ⊗ f ( x) = instrumental ⊗ intrinsic profile The Rietveld Method, T.R. (Mexico 21-25 June 2004) The profile of powder diffraction patterns yc i = ∑ I h Ω(Ti − Th ) + bi h Ih = Ih ( β I ) Ω = Ω( xhi , β P ) bi = bi ( β B ) Contains structural information: atom positions, magnetic moments, etc Contains instrumental (spectral dispersion, resolution parameters …) and sample informations (defects,crystallite size, …) Background: noise, diffuse scattering, ... The Rietveld Method, T.R. (Mexico 21-25 June 2004) 1.1 Integrated intensities I h = { jL p O AC S 2 } h Integrated intensities are proportional to the square of the structure factor S (cryst. and magn. structure factors contributions). The factors are: Multiplicity (j), Lorentz-polarization (Lp), preferred orientation (O), absorption (A), other “corrections” (C) ... The Rietveld Method, T.R. (Mexico 21-25 June 2004) Crystallographic structure factor FH F = C H unit cell ∑ i =1 r r N i . f i .exp − Bi . ( sin θ λ ) .exp(2iπ H .ri ) 2 H: scattering vector Ni: % site occupation factor fi: X-rays: Neutrons: Bi: Debye-Waller displacement factor ri : coordinates of the atom i in the unit cell atomic scattering factor coherent scattering length The Rietveld Method, T.R. (Mexico 21-25 June 2004) Magnetic structure factor FM general formula of Halpern and Johnson: r r 2 r r 2 r r r F = F⊥ ( h ) = Fm ( h ) − e .Fm ( h ) ( 2 r h r r Fm (h ) r e magnetic structure factor unit vector along scattering vector r h scattering vector r k propagation vector r r r h=H +k The Rietveld Method, T.R. (Mexico 21-25 June 2004) ) 2 Magnetic structure Magnetic structure refined by FullProf must have a distribution of magnetic moments than can be expanded as a Fourier series: r r µl , j = ∑ Skr , j .exp( −2iπ .k.Rl ) r k {} In such a case, the magnetic structure factor is given by: { ( ) } nc r r r r r r r r Fm ( H + k ) = p.∑ f j ( H + k ).Skr , j .exp 2iπ H + k .rj j =1 p r r fj ( H + k ) r rj Conversion constant Magnetic form factor Vector position of atom j The Rietveld Method, T.R. (Mexico 21-25 June 2004) 1.2 The profile function Ω(x) for cwl data ¾ analytical function to describe the observed diffraction profile. Ω(x) is defined to model: . each single Bragg line profile ¾ Full Width at Half Maximum: H (in deg. for cwl data) ¾ Integral breath β or β* (β* = β2θ.cosθ/λ) ¾ Profil parameter (form factor) φ = H/β . angular dependence of the profile parameters ¾ The choice of Ω(x) should obey to some criteria for structural and/or microstructural analysis: . Fit the observed data . Possible deconvolution (microstructural study) . Refined parameters are physically meaningfull. The Rietveld Method, T.R. (Mexico 21-25 June 2004) Ω(x) modelisation: 1- the Gauss function G( x ) = aG .exp( −bG .x 2 ) aG = 2 ln(2) . H π 4. ln(2) bG = H2 βG = φG = The Rietveld Method, T.R. (Mexico 21-25 June 2004) 1 H π = . aG 2 ln(2) HG βG = 2. ln(2) π = 0.9394 Ω(x) modelisation: 2- the Lorentz function aL L( x ) = 1 + bL .x 2 2 aL = πH 4 bL = 2 H 1 πH = 2 aL HL 2 φL = = = 0.6366 βL π βL = The Rietveld Method, T.R. (Mexico 21-25 June 2004) Ω(x) modelisation: combination of Gauss and Lorentz ¾ pseudo-Voigt function: . Linear combination of Gauss and Lorentz pV ( x ) =η.L( x ) + (1 −η ).G( x ) (0. ≤ η ≤ 1.0) PV ( x ) = PV ( x , η, H ) Normalized pseudo-Voigt function: β pV H 2 = η + (1 − η ). π .ln(2) π. The Rietveld Method, T.R. (Mexico 21-25 June 2004) Ω(x) modelisation: combination of Gauss and Lorentz ¾ Voigt function: . Convolution of Gauss and Lorentz V ( x, H G , H L ) =V ( x, β G , β L ) = G ( x, H G ) ⊗ L( x, H L ) Complex function: approximation of Voigt by a pseudo-Voigt function [Thompson, Cox, Hastings, J. Appl. Cryst. (1987), 20,79-83)] ( H ,η ) F ( H G , H L ) The Rietveld Method, T.R. (Mexico 21-25 June 2004) Ω(x) angular dependence (for cwl data) Caglioti, Paoletti, Ricci formula (1958, neutron case): FWHM 2 = U tan 2 θ + V tan θ + W U,V,W = f(αi, βM) The Rietveld Method, T.R. (Mexico 21-25 June 2004) Ω(x) angular dependence (for cwl data) ¾ Gauss function: ( ) H 2 = U + Dst2 tg 2θ + Vtgθ + W + ¾ Lorentz function ( ) H 2 = U + Dst2 tg 2θ + Vtgθ + W + ¾ pseudo-Voigt function η = η0 + X 2θ ¾ Voigt function IG cos2 θ ( IG cos2 θ ) H G2 = U + U st + (1 − ξ )Dst2 tg 2θ + Vtgθ + W + H L = ( X + X st + ξ .Dst Y + F (Sz )] [ ).tgθ + IG cos2 θ cosθ Instrumental parameters (beam divergency, spectral distribution, …) Sample dependent parameters: strain, size The Rietveld Method, T.R. (Mexico 21-25 June 2004) 1.3 Background bi ¾ TDS (thermal diffuse scattering) ¾ incoherent scattering ¾ sample environment ¾ … The Rietveld Method, T.R. (Mexico 21-25 June 2004) 2. Rietveld method The Rietveld Method consist of refining a crystal (and/or magnetic) structure by minimising the weighted squared difference between the observed and the calculated pattern against the parameter vector: β n χ = ∑ wi { yi − yci ( β )} 2 i =1 σ 2 i: n: 2 wi = σ12 i is the variance of the "observation" yi number of points in the pattern The Rietveld method is a structure refinement method and not a structure determination one. Structural parameters (space group, atomic positions, magnetic configuration …) has to be known as well as possible. The Rietveld Method, T.R. (Mexico 21-25 June 2004) Some points of the L.S. refinement χ 2 ∂χ 2 =0 ∂α α =αopt mini Taylor expansion of shifts δα 0 yic (α ) around initial α 0 obtained by solving linear system of equations (normal equations): Aδα 0 = b A b p x p matrix (p free parameters) Akl = ∑w . i i vector (dimension p) The Rietveld Method, T.R. (Mexico 21-25 June 2004) bk = ∂yic (α0 ) ∂yic (α0 ) . ∂α k ∂αl ∂yic (α 0 ) − w . y y . ( ) ∑i i i ic ∂α k Some points of the L.S. refinement Standard deviations of αk : σ (α k ) = ak χν = 2 χ2 n− p −1 2 Akk χν (reduced χ2) ak coefficient of the codeword of the parameter αk p parameters to refine n points in the pattern having Bragg contributions The Rietveld Method, T.R. (Mexico 21-25 June 2004) Some points of the L.S. refinement Cycle #1: α1 = α 0 + δα 0 New α parameters are considered as starting ones in the next cycle and the process is repeated until a convergence criterion is satisfied. The Rietveld Method, T.R. (Mexico 21-25 June 2004) Two sets of refinable parameters (1) (1) Profile parameters * peaks position: A, B, C, D, E, F, zeroshift 1/d*2 = A.h2 + B.k2 + C.l2 + D.kl + E.hl + F.hk * profile: U, V, W, X, Y … * asymetry P The Rietveld Method, T.R. (Mexico 21-25 June 2004) Two sets of refinable parameters (2) (2) Structural parameters * scale factor * overal isotropic temperature parameter * xi, yi, zi: atomic coordinates * Bi: atomic temperature parameter * Ki: components of the magnetic moment vector The Rietveld Method, T.R. (Mexico 21-25 June 2004) 3. Profile R-factors used in Rietveld Refinements The Rietveld Method, T.R. (Mexico 21-25 June 2004) R-factors and Rietveld Refinements (1) ∑ y −y = 100 ∑y obs ,i Rp calc ,i i R-pattern obs ,i i Rwp w y −y i obs ,i calc ,i ∑ = 100 i 2 wi yobs ,i ∑ i Rexp (N − P + C) = 100 ∑ w y2 i i obs ,i 2 1/ 2 R-weighted pattern 1/ 2 Expected R-weighted pattern N: number of points in the pattern P: number of refined parameters The Rietveld Method, T.R. (Mexico 21-25 June 2004) C: number of constraints R-factors and Rietveld Refinements (2) Rwp χν = Rexp 2 S= Rwp Rexp 2 Reduced Chi-square Goodness of Fit indicator The Rietveld Method, T.R. (Mexico 21-25 June 2004) R-factors and Rietveld Refinements (3) Two important things: • The sums over “i” may be extended only to the regions where Bragg reflections contribute • The denominators in RP and RWP may or not contain the background contribution The Rietveld Method, T.R. (Mexico 21-25 June 2004) Crystallographic R-factors used in Rietveld Refinements ∑ ' I '− I = 100 ∑ 'I ' obs , h RB calc , h h Bragg R-factor obs , h h ∑ ' F '− F = 100 ∑ 'F ' obs , h RF calc , h h obs , h h The Rietveld Method, T.R. (Mexico 21-25 June 2004) Crystallographic RF-factor. Calculation of the “observed” integrated intensities Ω(Ti − Th )( yobs ,i − Bi ) Provides ‘observed’ ' I obs ,h ' = I calc ,h ∑ integrates intensities for ( ycalc ,i − Bi ) i calculating Bragg R-factor ' Fobs ,h ' = ' I obs ,h ' jLp In some programs the crystallographic RFfactor is calculated using just the square root of ‘Iobs,h’ The Rietveld Method, T.R. (Mexico 21-25 June 2004) Important points (1) The absolute value of the profile R-factors has little significance because their values depend on the quality of the data as well as on the goodness of the structural model. The Rfactors obtained by a refinement of the whole pattern without structural model provide the “expected” values for the best structural model. (2) The most important assessments of the quality of the final refinement are: • the absence of important disagreement in the plot of the observed versus calculated patterns (detailed examination of the “Rietveld plot”: Yobs, Ycalc, Yobs-Ycalc) • the chemical significance of the final structural model The Rietveld Method, T.R. (Mexico 21-25 June 2004) (3) All these reliability factors are not satisfactory from a statistical point of view. Number of more significant parameters are calculated in FullProf: o deviance (A. Antoniadis et al. Acta Cryst A46, 692-1990) y D = 2∑ yi ln i − ( yi − yci ) i yc ,i o from D, one can derive two others measures of discrepancy which are useful as model selection criteria. They take account of both the g.o.f. of a model and the number of parameters used to achieve that fit. Q = D + a. p p number of parameters to refine a represents the coast of fitting an additionnal parameter (Akaike: a=2; Schwartz: a=ln(p)) The Rietveld Method, T.R. (Mexico 21-25 June 2004) Q can be compared to Hamilton criteria used in single crystal crystallography, to use the opportunity to add a refinable parameter. p= p +1 Î χ2 ↓ . If QA ↓ and QS ↓: . If QA ↑ and QS ↑ : The Rietveld Method, T.R. (Mexico 21-25 June 2004) valid parameter not valid parameter o Durbin-Watson statistic parameters: d and QD Measure of the correlation between adjacent residuals (serial correlation) n d= ∑ i =2 { ( w y − y c i i i ) 2 ( ∑{ ( wi yi − yic i ) } 2 − wi −1 yi −1 − yic−1 )} 2 n − 1 3.0901 QD = 2 − n+2 n − p ¾ d < QD : positive serial correlation: sussessive value of the residuals tends to have the same sign (most common situation in profile refinement) ¾ QD < d < 4-QD : no correlation ¾d > 4-QD : negative serial correlation: sussessive value of the residuals tends to have the opposite sign The Rietveld Method, T.R. (Mexico 21-25 June 2004) 4. Some References on R.M. ¾ Rietveld H.M., A Profile Refinement Method for Nuclear and Magnetic Structures, J. Appl. Cryst 2, 65 (1969) ¾ The Rietveld Method, edited by R.A. Young, IUCr, Oxford Science Publications (1996) ¾ A.W. Hewatt, High Resolution Neutron and Synchrotron Powder Diffraction, Chemica Scripta 26A, 119 (1985) ¾ A. Albinatti and B.T.M. Willis, The Rietveld Method in Neutron and X-Ray Powder Diffraction, J. Appl. Cryst. 15, 361 (1982) ¾ A.K. Cheetham and J.C. Taylor, Profile Analysis of Powder Diffraction Data: Its Scope, Limitations and Applications in Solid State Chemistry, J. Solid State Chem. 21, 253 (1977) The Rietveld Method, T.R. (Mexico 21-25 June 2004) Some References on R.M. ¾ L.B. McCusler, R.B. Von Dreele, D.E. Cox, D. Louër and P. Scardi Rietveld refinement guidelines J. Appl. Cryst. (1999) 32, 36-50 ¾E. Jansen, W. Schäfer and G. Will R values in analysis of powder diffraction data using Rietveld refinement J. Appl. Cryst. (1994) 27, 492-496 ¾A. Antoniadis, J. Bernoyer and A. Filhol · Maximum-likelihood methods in powder diffraction refinements Acta Cryst A46, 692 (1990) ¾R.J. Hill and H.D. Flack · The use of the Durbin-Watson d statistic in Rietveld analysis J. Appl. Cryst. 20, 356 (1987) The Rietveld Method, T.R. (Mexico 21-25 June 2004) The R.M. on the Web ¾ http://home.wxs.nl/~rietv025/ * mailing list * R.M. FAQ’S * seminal papers *… ¾ http://www.ccp14.ac.uk Collaborative Computationnal Project in Single Crystal and Powder Diffraction * Tutorials * Software download *… ¾ http://www-llb.cea.fr/fullweb/fp2k/fp2k.htm FullProf web site: * FullProf manual * Tutorials * examples * ... The Rietveld Method, T.R. (Mexico 21-25 June 2004)
© Copyright 2025