Automatic Derivation of Joint Probability Distributions of Normalized Structure Factors and Optimal Symbolic Phase Determination

Thesis from Dr. R. Peschar

 Summary



Although the currently available direct-method computer programs solve most crystal structures in a routine way, in some cases standard application fails and human intervention is necessary. The phase problem involved in solving the crystal structure consists of calculating a large number of non-measurable variables, the phases, from an even much more numerous number of equations, the invariant n-phase relationships, each consisting of a sum of n of the unknown phases. So once the phase sum of a phase relation and all but one of its phases are known, the remaining can be calculated. However, in general none of the invariant phase-sum values are known exactly. Instead, under general statistical assumptions, they can be estimated with a certain probability. The direct-method phase-determination procedure consists then of expressing all unknowns via the most reliable estimated phase sums in some starting values. This thesis deals with two problems involved in this process. Firstly with the problem of the reliability of the phase-sum estimates. The more reliable the phase-sum estimates are, the more likely the phasing process will be successful. The second problem is the order in which the unknown phases must be calculated, via the estimated phase sums, and how many starting values should be used. Because of the nature of the estimates of the phase sums, statistically each calculated phase will be in error. In order to avoid a propagation and cumulation of phase errors, a correct calculation order is of utmost importance. The reliability of phase-sum estimates are based on joint probability distributions (jpd's) of structure factors. Hitherto all jpd's were derived by hand, following certain well-established mathematical rules. The resulting distributions can be expressed in either a series-expansionform, containing usually only a limited number of terms due to the length of the derivations, or in an exponential form, which is a low-order approximation as well.
In chapter two a new derivation procedure is described, valid for the triplet phase sum in space group P1 for equal-atom structures. The resulting distribution, a series expansion containing a selection of the higher-order terms, is not derived by hand. Instead, the derivation algorithm has been programmed, resulting in a computer-aided derivation. Test results show that with the new distribution the absolute value of the triplet phase sum is estimated with less systematical error.

In chapter three this technique is generalized to jpd's of any number of structure factors in P1 for equal-atom structures and it is applied to the distribution of the quartet phase sum. Test results show that with this computer-derived distribution the quartet phase sums are estimated with considerably less systematical error.

The technique to obtain computer-aided derivation of jpd's is further developed in chapter four : a general theoretical and practical procedure is presented to derive jpd's of structure factors in any space group. The series-expansion distributions are correct up to any preset order. Optionally, the program transforms the series expansion into an exponential expression. With a few exceptions, in low-order approximation these exponential expressions turn out to be identical to expressions known from literature.

In chapter five extensively test results are discussed obtained with the computer-derived distributions for, amongst others, the triplet, quartet and quintet phase sums. The phase determination technique is discussed in chapter six. A new symbolic phase determination procedure is introduced, based on the dynamic-programming technique, which calculates under the given basic probabilistic assumptions the statistically most reliable phasing order and starting values. In contrast to existing methods the starting set of phases is not determined a priori and kept fixed during the phasing process but build-up gradually. Test results show that this more flexible and systematic technique is an enormous improvement over the conventionally employed symbolic addition phasing procedure.

E-mail Voor meer informatie of aanvraag van een exemplaar.

Overzicht proefschriften kristallografie Amsterdam.

Home Hoofdpagina van Kristallografie.