A NEW EXTENSION OF THE `LEARNT-PROFILE' METHOD. By Ewa Galdecki, Institute of Low Temperature and Structure Research, Polish Academy of Sciences, ul. Okólna 2, 50-950 Wroclaw, Poland

As proved by the author, the use of `learnt profiles' as models of measured X-ray single-crystal diffraction peaks rather than selected `shape functions' leeds to a very effective data processing. The `learnt profile' can be determined and used in such a way as in appropiate original papers [Clegg, W. (1981), Acta Cryst. A37, 437, and references quoted therein], but the idea may also be extended. In a system of data-processing computer programs written by the author one may learn the shape(s) of either

(i) the double peaks (for various ranges of J values), or

(ii) the single peak (one of the components of the K[[alpha]] doublet), or

(iii) the crystal profile, that is one of factors of the physical convolution model (the other ones being the spectral distribution for given source of X-rays, and some aberration functions).

In the case (i) of the `simple learnt-profile' method, one should collect models of the double-peak shapes within a number (say 25-50) of J ranges (the strongest peaks are selected and averaged), the models are stored in their real angular scale. Thus the widths of the peaks are given. These may be adjusted in the next stage. Using the `basic learnt-profile' method, described in the original papers, case (ii), it is necessary - apart from learning the single-peak shape (stored by the present author in a standardized form, as a spline function) - to derive some information on the width of such single peaks, that is treatet as a function of J and the direction cosines. In the case (iii) of the `convolution' or `learnt crystal-profile' method, the spectral distribution of the given K[[alpha]]₁ - K[[alpha]]₂ line must be known. It should be expressed as a function of angle for respective J values. The function is used for both the crystal-profile determination - via a deconvolution of the observed profile in relation to the spectral distribution - and, in the next stage, for syntheses of the profiles. As results from the author's tests, the efficiency of the methods is very good. Their properties and possibilities will be discussed in detail.