András Sütö

In this paper, an extremely simple structure solution method termed charge flipping is presented. It works ab initio on high-resolution X-ray diffraction data in the manner of Fourier recycling. The real-space modification simply changes the sign of charge density below a threshold, while in reciprocal space the moduli F(obs) are retained resulting in an F(obs) map without weighting. The algorithm is tested using synthetic data for a wide range of structures, the solution statistics are analysed and the quality of reconstruction is checked. Finally, mathematical aspects of the algorithm are considered in detail, and these show that in this chaotic iteration process the solution is a limit cycle and not a fixed point.

The original charge flipping algorithm [Oszlanyi & Suto (2004). Acta Cryst. A60, 34-141] is an amazingly simple structure solution method which works ab initio on high-resolution X-ray diffraction data. In this paper, a new version of the algorithm is presented that complements the phase exploration in reciprocal space. Instead of prescribing observed moduli of all structure factors, weak reflections are treated separately. For these reflections, calculated moduli are accepted unchanged and calculated phases are shifted by a constant Deltaphi=pi/2. This means that the observed data of weak reflections are not used in the iteration, except for the knowledge that they are indeed weak. The improvement is drastic, in some cases the success rate is increased by a factor of ten, in other cases a previously unsolvable structure becomes solvable by the modified algorithm.

This paper summarizes the current state of charge flipping, a recently developed algorithm of ab initio structure determination. Its operation is based on the perturbation of large plateaus of low electron density but not directly on atomicity. Such a working principle radically differs from that of classical direct methods and offers complementary applications. The list of successful structure-solution cases includes periodic and aperiodic crystals using single-crystal and powder diffraction data measured with X-ray and neutron radiation. Apart from counting applications, the paper mainly deals with algorithmic issues: it describes and compares new variants of the iteration scheme, helps to identify and improve solutions, discusses the required data and the use of known information. Finally, it tries to foretell the future of such an alternative among well established direct methods.