It has been shown that a satisfying RDM model may constitute a starting model for a successful RMC simulation with constraints.
 

The RDM best models correspond to crystal structures in which the glasses devitrify, in all three cases.
 
 
 
 
 
 

One can think about what would happen if the three partial structure factors for SiO₂ or ZnCl₂ or the ten partial structure factors for NaPbM₂F₉ had been experimentally available.

It is not possible to assert that the actual models proposed by either the RMC or the RDM methods would lead necessarily to low R factors on the lacking structure factors without any adjustment.
 
 
 
 

Those three dimensional structures are simply models that are consistent with the data, constraints and external knowledge.

In other words, the RDM best model is one structure in the group of possible RMC solutions.

RMC tends to produce the most disordered structure if the starting configuration is random, and RDM produces the most ordered.

Combining the two methods produces intermediate order, as expected.
 
 

Both methods have pro and con.
 

Testing a model by RDM is fast, but finding a model having the exact glass composition can be a problem.

Obtaining convergence by RMC may be quite long when drastic constraints are imposed, however, the model size brings more credibility than for the generally small RDM models.

Nevertheless, a strategy is essential for succeeding in building models consistent with external knowledge (no edge sharing if undesired, strict coordinations and so on).

Such a strategy is not always easy to establish with the current existing RMC code, and a strategy avoiding trigonal prisms when octahedra were exclusively required was not found here.
 
 
 
 

It is expected that confidence in RDM modelling will increase as a consequence of the present study, showing that a good RDM model is always an excellent RMC candidate, reconciling both methods.

Trying to go further in combining both methods could be attempted.

The idea would be to decide of atom moves in the RDM method by testing random displacements instead of using the least-squares process, while still using a mean small microstrained model.

Next time, may be.

References