Tables for
Volume H
Powder diffraction
Edited by C. J. Gilmore, J. A. Kaduk and H. Schenk

International Tables for Crystallography (2018). Vol. H, ch. 3.4, pp. 271-272

Section Figures of merit

A. Altomare,a* C. Cuocci,a A. Moliternia and R. Rizzia

aInstitute of Crystallography – CNR, Via Amendola 122/o, Bari, I-70126, Italy
Correspondence e-mail: Figures of merit

| top | pdf |

An important task is the introduction of a figure of merit (FOM) that is able to (a) describe the physical plausibility of a trial cell and its agreement with the observed pattern, and (b) select the best cell among different possible ones. de Wolff (1968[link]) made an important contribution in this direction. He developed the M20 figure of merit defined by[M_{20} = {Q_{20} \over 2\langle\varepsilon\rangle N_{20}},\eqno(3.4.4)]where Q20 is the Q value corresponding to the 20th observed and indexed peak, N20 is the number of different calculated Q values up to Q20, and [\langle\varepsilon \rangle] is the average absolute discrepancy between the observed and the calculated Q values for the 20 indexed peaks; the factor 2 is a result of statistics, explained by the larger chance for an observed line to sit in a large interval as compared with sitting in a small interval. The rationale behind M20 is as follows: the better the agreement between the calculated and the observed peak positions (the smaller the [\langle\varepsilon \rangle] value) and the smaller the volume of the unit cell (the smaller the N20 value), the larger the M20 value and consequently the confidence in the proposed unit cell. A rule of thumb for M20 is that if the number of unindexed peaks whose Q values are less than Q20 is not larger than 2 and if M20 > 10, then the indexing process is physically reasonable (de Wolff, 1968[link]; Werner, 2002[link]). This rule is often valid, but exceptions occur. The use of the first 20 peaks is a compromise (coming from experience) between introducing a quite large number of observed peaks (depending on the number of parameters of the unit cell) and avoiding the use of high-angle peak positions, which are more affected by errors. M20 is statistically expected to be 1 in case of completely arbitrary indexing. It has no upper limit (it can be very large when [\langle\varepsilon \rangle] is very small).

Smith & Snyder (1979[link]) proposed the FN criterion in order to overcome the limits of M20 with respect to its dependence on the 20 lines and on crystal class and space group. The FN figure of merit is given by[F_N = {1 \over\langle |\Delta 2\theta | \rangle} {N \over N_{\rm poss}}, ]where [\langle |\Delta 2\theta | \rangle] is the average absolute discrepancy between the observed and calculated 2θ peak position values and Nposs is the number of possible diffraction lines up to the Nth observed line. The values of [\langle |\Delta 2\theta | \rangle] and Nposs, ([\langle |\Delta 2\theta | \rangle], Nposs), are usually given with FN. With respect to M20, FN is more suitable for ranking the trial solutions and less for indicating their physical plausibility (Werner, 2002[link]).

Both M20 and FN, being based on the discrepancies between observed and calculated lines, are less reliable if there are impurity peaks; if the information about the unindexed lines is not taken into account, the risk of obtaining false solutions increases. Alternative FOMs based on joint probability have also been proposed (Ishida & Watanabe, 1967[link], 1971[link]). Among the recently developed FOMs, we mention:

  • (1) Qpartial (Bergmann, 2007[link]):[Q_{\rm partial} = \sum\limits_{i} \min \left[w_i,\left({x_i - \hat{x_i} \over \delta _i} \right)^2 \right], ]where the summation is over the number of observed lines, wi is the observed weight of line i, x and [ \hat{x_i}] are the observed and simulated line positions, respectively, and δi is the observed random error of line i. Qpartial is multiplied by a factor that depends on the symmetry of the simulated lattice (triclinic, …, cubic), on the unit cell volume and on the number of ignored peaks.

  • (2) McM20 (Le Bail, 2008[link]):[McM_{20}=[100/(R_pN_{20})]B_rS_y,]where N20 is the number of possible lines that might exist up to the 20th observed line (for a primitive P lattice). Rp is the profile R factor (Young, 1993[link]). Br is a factor arbitrarily set to 6 for F and R Bravais lattices, 4 for I, 2 for A, B and C, and 1 for P. Sy is a factor equal to 6 for a cubic or a rhombohedral cell, 4 for a trigonal, hexagonal or tetragonal cell, 2 for an orthorhombic cell, and 1 for a monoclinic or triclinic cell.

  • (3) WRIP20 (Altomare et al., 2009[link]):[{\rm WRIP}20 = {\rm RAT}_{Rp}^2\times {\rm RAT}_{\rm Ind}\times {\rm RAT}_{\rm Pres}\times w_u\times {\rm RAT}_{M20}^{1/2}.\eqno(3.4.5)]Based on M20 (M20 and FN remain the most widely used FOMs), WRIP20 has been developed for exploiting the full information contained in the diffraction profile. The factors that appear in (3.4.5)[link] are[\displaylines{{\rm RAT}_{Rp} = {1 - R_p \over 1 - (R_p)_{\rm min}}, \quad {\rm RAT}_{\rm Ind} = {{\rm PERC}_{\rm Ind} \over ({\rm PERC}_{\rm Ind})_{\rm max}}, \cr {\rm RAT}_{\rm Pres} = {({\rm PERC}_{\rm Pres})_{\rm min} \over {\rm PERC}_{\rm Pres}},\quad w_u= (N_{\rm obs } - N_u)/N_{\rm obs},\cr {\rm RAT}_{M20} = {M_{20} \over (M_{20})_{\rm max}},\quad {\rm PERC}_{\rm Pres} = \textstyle\sum\limits_{\rm Pres} {\rm mult} /\textstyle\sum\limits_{\rm all} {\rm mult}.}]Rp is the profile-fitting agreement calculated after the Le Bail (Chapter 3.5[link] ) decomposition of the full pattern using the space group with the highest Laue symmetry compatible with the geometry of the current unit cell and no extinction conditions. PERCInd, the percentage of independent observations in the experimental profile, is estimated according to Altomare et al. (1995[link]). For each extinction symbol compatible with the lattice geometry of the current unit cell, normalized intensities are calculated and subjected to statistical analysis in order to obtain a probability value associated with each extinction symbol in accordance with Altomare et al. (2004[link], 2005[link]). For the extinction symbol with the highest probability value, the value of PERCPres is calculated: [\textstyle\sum_{\rm all}{\rm mult}] is the total number of reflections (symmetry-equivalent included) for the space group having the highest Laue symmetry and no extinction conditions. (It varies with the volume of the unit cell and the data resolution.) [\textstyle\sum_{\rm Pres} {\rm mult}], which varies according to the extinction rules of the current extinction symbol, coincides with the number of non-systematically absent reflections (with the symmetry equivalents included). The subscripts min and max mark the minimum and the maximum values of each factor respectively, calculated for the possible unit cells that are to be ranked. Nobs and Nu are the number of observed and unindexed lines, respectively. All the terms in (3.4.5)[link] are between 0 and 1, so ensuring that WRIP20 also lies between 0 and 1. In addition, WRIP20 has the following properties: (a) it is continuous, that is, definable in any interval of the experimental pattern; (b) it takes into account the peak intensities, the number of generated peaks and their overlap, and the systematically absent reflections (through the extinction-symbol test); and (c) it is not very sensitive to the presence of impurity lines (these usually have low intensities). WRIP20 is effective in finding the correct cell among a number of possible ones and selecting the corresponding most probable extinction symbol (see Example 3[link] in Section[link]).

  • (4) Two new figures of merit based on de Wolff's method, the reversed figure of merit ([M_n^{\rm Rev}]) and the symmetric figure of merit ([M_n^{\rm Sym}]), have recently been proposed (Oishi-Tomiyasu, 2013[link]). As observed by Oishi-Tomiyasu, the de Wolff figure of merit Mn does not use the observed and calculated lines in a symmetrical way, consequently it is (a) insensitive to computed but unobserved lines (i.e., extinct peaks) and (b) sensitive to unindexed observed lines (e.g., impurity peaks). [M_n^{\rm Rev}] and [M_n^{\rm Sym}] aim to compensate for the disadvantages of Mn. In particular, [M_n^{\rm Rev}] has characteristics opposite to those of Mn with regard to sensitivity to extinct reflections and impurity peaks, and [M_n^{\rm Sym}] has intermediate properties between Mn and [M_n^{\rm Rev}]. They prove useful in selecting the correct solution, particularly in case of presence of impurity peaks. (See also Section[link].)


Altomare, A., Caliandro, R., Camalli, M., Cuocci, C., da Silva, I., Giacovazzo, C., Moliterni, A. G. G. & Spagna, R. (2004). Space-group determination from powder diffraction data: a probabilistic approach. J. Appl. Cryst. 37, 957–966.Google Scholar
Altomare, A., Camalli, M., Cuocci, C., da Silva, I., Giacovazzo, C., Moliterni, A. G. G. & Rizzi, R. (2005). Space group determination: improvements in EXPO2004. J. Appl. Cryst. 38, 760–767.Google Scholar
Altomare, A., Campi, G., Cuocci, C., Eriksson, L., Giacovazzo, C., Moliterni, A., Rizzi, R. & Werner, P.-E. (2009). Advances in powder diffraction pattern indexing: N-TREOR09. J. Appl. Cryst. 42, 768–775.Google Scholar
Altomare, A., Cascarano, G., Giacovazzo, C., Guagliardi, A., Moliterni, A. G. G., Burla, M. C. & Polidori, G. (1995). On the number of statistically independent observations in a powder diffraction pattern. J. Appl. Cryst. 28, 738–744.Google Scholar
Bergmann, J. (2007). EFLECH/INDEX – another try of whole pattern indexing. Z. Kristallogr. Suppl. 26, 197–202.Google Scholar
Ishida, T. & Watanabe, Y. (1967). Probability computer method of determining the lattice parameters from powder diffraction data. J. Phys. Soc. Jpn, 23, 556–565.Google Scholar
Ishida, T. & Watanabe, Y. (1971). Analysis of powder diffraction patterns of monoclinic and triclinic crystals. J. Appl. Cryst. 4, 311–316.Google Scholar
Le Bail, A. (2008). Structure solution. In Principles and Applications of Powder Diffraction, edited by A. Clearfield, J. H. Reibenspies & N. Bhuvanesh, pp. 261–309. Oxford: Wiley-Blackwell.Google Scholar
Oishi-Tomiyasu, R. (2013). Reversed de Wolff figure of merit and its application to powder indexing solutions. J. Appl. Cryst. 46, 1277–1282.Google Scholar
Smith, G. S. & Snyder, R. L. (1979). FN: a criterion for rating powder diffraction patterns and evaluating the reliability of powder-pattern indexing. J. Appl. Cryst. 12, 60–65.Google Scholar
Werner, P.-E. (2002). Autoindexing. In Structure Determination from Powder Diffraction Data, edited by W. I. F. David, K. Shankland, L. B. McCusker & Ch. Baerlocher, pp. 118–135. Oxford University Press.Google Scholar
Wolff, P. M. de (1968). A simplified criterion for the reliability of a powder pattern indexing. J. Appl. Cryst. 1, 108–113.Google Scholar
Young, R. A. (1993). Introduction to the Rietveld method. In The Rietveld Method, edited by R. A. Young, pp. 1–38. Oxford: IUCr/Oxford University Press.Google Scholar

to end of page
to top of page