Читать книгу Encyclopedia of Glass Science, Technology, History, and Culture - Группа авторов - Страница 312

To complement standard MC simulations, the so‐called reverse Monte‐Carlo (RMC) method has been developed to study disordered structures [8]. It enables three‐dimensional structural models to be constructed in a manner consistent with experimental results. The data most commonly used are PDF and their Fourier transforms obtained from diffraction experiments (Chapter 2.2). In RMC calculations the standard procedure is to

1 set up an initial configuration in a periodic boundary cell,

2 calculate the set of quantities relevant to the experimental data considered (e.g. PDF),

3 calculate the mean square deviations χ 2 of the calculated from the observed results(11)

where ρ is an appropriate measure of experimental accuracy,

1 select an atom and give it a random displacement,

2 accept the move if it leads to a χ 2 decrease, but keep the former configuration otherwise,

3 repeat from 2) to 5).

To quote a single example, the structural role of “insufficient” network formers has been successfully studied by RMC in a simulation of the atomic configuration of Mg₂SiO₄ glass in which the measured total structural factors were well fitted [9]. The estimated bond distance of Mg─O differs in the glass and in the crystal (Figure 3), whereas the difference in the peak position of Mg─O between the two phases reflects the structural feature the glass network is built by the corner‐ and edge‐sharing of highly distorted Mg─O‐bearing species.

Figure 3 As defined by Eq. (20), total correlation functions T(r) of Mg₂SiO₄ glass (a) and crystal (b) [9].

The advantage of the RMC method is that knowledge of interatomic potentials is not required, but its drawback is that it is not applicable to novel glass systems for which no experimental data can be compared with model values. Besides, there is a risk of arriving at an incorrect structure if the iterative procedure leads to a local, and not to the true minimum of χ ². A simple way to avoid such a pitfall is to add a set of effective constraints on bond lengths, bond angles, or coordination numbers (CN) that will prevent spurious results from being obtained.