A survey is presented on existing approaches to determining unit-cell parameters from powder x-ray diffraction data. The relationships between them are examinated along with the advantages and shortcomings.

INTRODUCTION

   The are entire classes of compounds available only in polycristalline form, such as clays; also, some substances are of interest in the polycrysalline state, such as high-temperature superconductors. The polycrystalline material gives a depleted one-dimesional diffraction pattern, and therefore the structure interpretation is difficult. On the other hand, determining the lattice parameters is the first step in interpreting the structure, and the reliability with wich this is done, and even the possibility of doing it, will determine the establishment of the structure. Attention is  concentrated on the development methods, algorithms and programs for the purpose. As additional to my review we will be concentrated on the parallel programming and neural networks for the determination of unit-cells, because I have gained good experience in the software industry due to last 20 years. In order to demonstrate new possibilities I will use the Java programming language with the two packages: Jogamp (OpenCL interface to Java) and Neuroph (neural networks package), which I used for much application, for example Stock Market Prediction or Fraud Detection.

Unit - cell parameter determination in the powder method has specific features and differs substantially from the single-crystal approach, so it represents an independent research line. Methods for using powder diffraction data have been developed almost from the start of diffraction research. The papers are unevenly distributed over the years. The maximum occured in the 1970s., which was associated with computer developments. At the start of the 1980s, there was some fall in the number of papers, although on the whole the interestst in the powder methods increased ( powder methods being used to refine structures). Internet has done more publications, but the new approches were only in the branches artifical intellegence. That fall of approches was due to ambiguity in the solutios for the unit-cell parameters given by the powder method. Several generations of researchers have made effors to overcome this basic difficulty.

The statistics of the publication is impossible to become. What is the difference from old time to today: in the first row - the computer performance. The selection criteria are not changed. Another aspect it is possible today, to automate, for example,  the human assessment of symmetry.

In the next I will concentrate on the two aspects:

1.  Is it the performance of computers so to enable direct try-and-error?

2. Is it possible to have more information from full powder diffraction pattern and how can do it?