In general, the structure determination of a disordered crystal should start in the usual way by solving the average structure. The effectiveness of this procedure strongly depends on the distribution of integrated intensities of sharp and diffuse reflections. In cases where the integrated intensities of Bragg peaks are predominant, the maximum information can be drawn from the averaged structure. The observations of fractional occupations of lattice sites, split positions and anomalous large and anharmonic displacement parameters are indications of the disorder involved. Since these aspects of disorder phenomena in the averaged structure may be interpreted very easily, a detailed discussion of this matter is not given here (see any modern textbook of Xray crystallography). Therefore, the anomalies of the average structure can give valuable hints on the underlying disorder and, vice versa, can be used to check the final disorder model derived from the diffuse scattering.
Difficulties may arise from the intensity integration, which should be carried out very carefully to separate the Bragg peaks from the diffuse contributions, e.g. by using a highresolution diffraction method. The importance of this may be understood from the following argument. The averaged structure is determined by the coherent superposition of different structure factors. This interpretation is true if there is a strictly periodic subcell with longrange order that allows for a clear separation of sharp and diffuse scattering. There are important cases, however, where this procedure cannot be applied without loss of information.
(a) The diffuse scattering (other than thermal diffuse scattering) is concentrated near the Bragg peaks for a large number of reflections. Because of the limited resolution power of conventional singlecrystal methods, the separation of sharp and diffuse scattering is impossible. Hence, the conventional study of integrated intensities does not really lead to an averaged structure. In this case, a refinement should be tried using an incoherent superposition of different structure factors (from the average structure and the difference structure). Application of this procedure is subject to conditions which have to be checked very carefully before starting the refinement: first, it is necessary to estimate the amount of diffuse scattering not covered by intensity integration of the `sharp' reflections. Since loss in intensity, hidden in the background scattering, is underestimated very frequently, it should be checked whether nearly coinciding sharp and diffuse maxima are modulated by the same structure factor. It may be difficult to meet this condition in some cases; e.g. this condition is fulfilled for antiphase domains but the same is not true for twin domains.
(b) The concentration of diffuse maxima near Bragg peaks is normally restricted to domain structures with a strictly periodic sublattice. Cases deviating from this rule are possible. Since they are rare, they are omitted here. Even structures with small deviations from the average structure do not necessarily lead to structure factors for diffuse scattering that are proportional to those of the average structure. This has been shown in the case of a twin structure correlated by a mirror plane, where the reflections of a zone only have equal structure factors (Cowley & Au, 1978). This effect causes even more difficulties for orthogonal lattices, where the two twins have reflections in exactly the same positions, although differing in their structure factors. In this particular case, the incoherent or coherent treatment in refinements may be seriously hampered by strains originating from the boundary. Unsatisfactory refinements may be explained in this way but this does not improve their reliability.

The integrated intensity within a Brillouin zone of any structure is independent of atomic positions if the atomic form factors remain unchanged by structural fluctuations. Small deviations of atomic form factors owing to electrondensity changes of valence electrons are neglected. Consequently, the integrated diffuse intensities remain unchanged if the average structure is not altered by the degree of order. The latter condition is obeyed in cases where a geometrical longrange order of the lattice is independent of the degree of order, and no longrange order in the structure exists. This law is extremely useful for the interpretation of diffuse scattering. Unfortunately, intensity integration over coinciding sharp and diffuse maxima does not necessarily lead to a structure determination of the corresponding undistorted structure. This integration may be useful for antiphase domains without major structural changes at the boundaries. In all other cases, the deviations of domains (or clusters) from the averaged structure determine the intensities of maxima, which are no longer correlated with those of the average structure.
If the integrated intensity of diffuse scattering is comparable with, or even larger than, those of the Bragg peaks, it is useful to begin the interpretation with a careful statistical study of the diffuse intensities. Intensity statistics can be applied in a way similar to the intensity statistics in classical structure determination. The following rules are briefly discussed in order to enable a semiquantitative interpretation of the essential features of disorder.
(1) First, it is recommended that the integrated intensities are studied in certain areas of reciprocal space.
(2) Since lowangle scattering is very sensitive to fluctuations of densities, the most important information can be drawn from its intensity behaviour. If there is at least a onedimensional sublattice in reciprocal space without diffuse scattering, it may often be concluded that there is no important lowangle scattering either. This law is subject to the condition of a sufficient number of reflections obeying this extinction rule without any exception.
(3) If the diffuse scattering shows maxima and minima, it should be checked whether the maxima observed may be approximately assigned to a lattice in reciprocal space. Obviously, this condition can hardly be met exactly if these maxima are modulated by a kind of structure factor, which causes displacements of maxima proportional to the gradient of this structure factor. Hence this influence may well be estimated from a careful study of the complete diffuse diffraction pattern.
It should then be checked whether the corresponding lattice represents a sub or a superlattice of the structure. An increase of the width of reflections as a function of increasing indicates strained clusters of this sub or superlattice.
(4) The next step is to search for extinction rules for the diffuse scattering. The simplest is the lack of lowangle scattering, which has already been mentioned above. Since diffuse scattering is generally given by equation (4.2.3.23b),it may be concluded that this condition is fulfilled in cases where all structural elements participating in disorder differ by translations only (stacking faults, antiphase domains etc.). They add phase factors to the various structure factors, which may become (n = integer) for specific values of the reciprocal vector H. If all are equivalent by symmetry
Other possibilities for vanishing diffuse scattering may be derived in a similar manner for special reflections if glide operations are responsible for disorder. Since we are concerned with disordered structures, these glide operations need not necessarily be a symmetry operation of the lattice. It should be pointed out, however, that all these extinction rules of diffuse scattering are a kind of `anti'extinction rule, because they are valid for reflections having maximum intensity for the sharp reflections unless the structure factor itself vanishes.
(5) Furthermore, it is important to plot the integrated intensities of sharp and diffuse scattering as a function of the reciprocal coordinates, at least in a semiquantitative way. If the ratio of integrated intensities remains constant in the statistical sense, we are predominantly concerned with a density phenomenon. It should be pointed out, however, that a particlesize effect of domains behaves like a density phenomenon (the density changes at the boundary!).
If the ratio of `diffuse' to `sharp' intensities increases with diffraction angle, we have to take into account atomic displacements. A careful study of this ratio yields very important information on the number of displaced atoms. The result has to be discussed separately for domain structures if the displacements are equal in the subcells of a single domain but different for the various domains. In the case of two domains with displacements of all atoms, the integrated intensities of sharp and diffuse reflections become statistically equal for large . Other rules may be derived from statistical considerations.
(6) The next step of a semiquantitative interpretation is to check the intensity distribution of diffuse reflections in reciprocal space. In general this modulation is simpler than that of the sharp reflections. Hence it is frequently possible to start a structure determination with diffuse scattering. This method is extremely helpful for one and twodimensional disorder where partial structure determinations yield valuable information, even for the evaluation of the average structure.
(7) In cases where no sub or superlattice belonging to the diffuse scattering can be determined, a careful check of integrated intensities in the neighbourhood of Bragg peaks should again be performed. If systematic absences are found, the disorder is most probably restricted to specific lattice sites which may also be found in the average structure. The accuracy, however, is much lower here. If no such effects correlated with the average structure are observed, the disorder problem is related to a distribution of molecules or clusters with a structure differing from the average structure. As pointed out in Section 4.2.3.1, the problem of the representative structure(s) of the molecule(s) or the cluster(s) should be solved. Furthermore, their distribution function(s) is (are) needed. In this particular case, it is very useful to start with a study of diffuse intensity at low diffraction angles in order to acquire information about density effects. Despite the contribution to sharp reflections, one should remember that the level of information derived from the average structure may be very low (e.g. small displacements, low concentrations etc.).
(8) A Patterson picture – or strictly speaking a difference Patterson ( Fourier synthesis) – may be very useful in this case. This method is promising in the case of disorder in molecular structures where the molecules concerned are at least partly known. Hence the interpretation of the difference Patterson may start with some internal molecular distances. Nonmolecular structures show some distances of the average structure. Consequently, a study of the important distances will yield information on displacements or replacements in the average structure. For a detailed study of this matter the reader is referred to the literature (Schwartz & Cohen, 1977).

Although it is highly improbable that exactly the same diffraction picture will be found, the use of an atlas of optical transforms (Wooster, 1962; Harburn et al., 1975; Welberry & Withers, 1987) may be helpful at the beginning of any study of diffuse scattering. Alternatively, computer simulations may be helpful, as discussed in Section 4.2.7. The most important step is the separation of the distribution function from the molecular scattering. Since this information may be derived from a careful comparison of lowangle diffraction with the remaining sharp reflections, this task is not too difficult. If the influence of the distribution function is unknown, the reader is strongly advised to disregard the immediate neighbourhood of Bragg peaks in the first step of the interpretation. Obviously information may be lost in this way but, as has been shown in the past, much confusion caused by an attempt to interpret the scattering near the Bragg peaks with specific structural properties of a cluster or molecular model is avoided. The inclusion of this part of diffuse scattering can be made after a complete interpretation of the change of the influence of the distribution function on diffraction in the wideangle region.