[sdpd] Re: I am new here! some comments by Julian Gale

["Dr. Jaap Vente" <jaap... @jade.cieamer.conacyt.mx>]

Dear Jaap,
	Thanks for copying me your email that you sent out.
I would make a couple of comments. Firstly, I think it is
rather optimistic to say that the genetic algorithm would
work without any diffraction data. So far we have found it
works OK for simple binary and ternary compounds with fixed
unit cells (determined from diffraction data!). However, if
you were to let the cell go as well I think it might be a
long process to search all of conformation space. Also the
method in principle may produce a number of candidate
structures with fairly similar energies (as you would expect
given that more than polymorph may exist in some cases). To
trust an interatomic potential to get the relative energies
with high precision is unreasonable and hence the final stage
is to generate the diffraction pattern for the best, say ten,
candidate structures and select according to this. Hence these
methods should, at present, be considered as a compliment to
diffraction not a replacement.
	Secondly, the idea of combined GA/potentials with
diffraction solution methods is something we've thought about
but not done anything about yet. However, it is a good idea
worth pursuing!
	Best wishes,

		Julian