The short and medium range order govern diffraction patterns of glasses. From this study, expectations for a unique structure representation for the NaPbM2F9 (M = Fe, V) fluoride glasses from the RDM or RMC methods are disappointed. Several models which may seem very different to crystallographer eyes have led by the RDM method to similar diffraction data fits because they are in fact characterized by similar mean short and medium range orders. The RMC model may be considered as combining the various special arrangements (plus others) characterizing the crystalline-derived models tested by the RDM method. A tridimensional network built up from corner sharing [MF6] octahedra was expected from crystal chemistry considerations before to start the study. Such a network characterizes the NaPbFe2F9 crystal structure which is quite different from that of KPbCr2F9 and of the other related enneafluorides having a monodimensional character drawn by infinite isolated staircase double chains of corner sharing octahedra. In fact, RDM as well as RMC results tend to show that locally the KPbCr2F9 configuration cannot be ruled out. The contrast between the 1950x3 free atomic coordinates implicit in the RMC model and the 14 coordinates refined by the RDM method for the NaPbFe2F9 crystalline structure-derived model is not reflected by a proportionally large difference in fit quality. This suggests that the RDM method could be improved maybe by the introduction of constraints on interatomic distances and of a Monte Carlo process which would allow to describe small cells in the P1 space group. With such a modification, the 14 parameters of the model designed here as the best would extend to 52x3 by a method combining RMC and RDM, leading very probably to some R factor improvement. On another hand, the presence of [MF6] trigonal prisms in the RMC result is theoretically a nonsense regarding fluoride crystal chemistry (but who knows really glass structures ?). A new RMC modelling with a more drastic constraint in order to build exclusively octahedra should be undertaken. Application of RMC and RDM methods to TMFG fluoride glasses evenmore concentrated in [MF6] octahedra is in preparation for the selected typical compositions BaMnMF7 and PbMnMF7 (M = Fe,V) (66% of the cations in sixfold coordination against 50% in the title glasses). More edge sharing is to be expected in these glasses as suggested by the corresponding crystal structures in which up to 50-100% octahedra share at least one edge with another octahedron.

Acknowledgments. Many thanks are due to A-M. Mercier who prepared the glass samples, to the Institut Laue-Langevin for providing neutrons during experiment 6-08-189 (invaluable help in measuring data was ensured by P. Chieux) and to Prof. C. Jacoboni for his constructive criticism of the ILL proposal. Drawings of RDM and RMC models are respectively by the STRUVIR [38] and GLASSVIR softwares [39].

Armel Le Bail - June 1997