[Pw_forum] NEB initial path change
Guido Fratesi
fratesi at mater.unimib.it
Mon Feb 4 09:17:04 CET 2008
Dear Dong-Hee,
DHL> I was wondering if I can manually specify the initial path of NEB calculation.
Ideally, you can prepare manually a prefix.path file, and use the
"restart" option in the input of pw.x, but it's simpler to use
intermediate images.
DHL> Can I make the initial images unequispaced? Since I approximately know
DHL> the final path of a reaction, I think the NEB calculation would be
DHL> finised earlier if I make the initial path similar to the final path.
DHL> Manual also says we can specify "intermediate_image". Can I do what I
DHL> want using "intermediate_image" option, or does "intermediate_image"
DHL> mean another geometry optimized configuration between the initial and
DHL> final images?
If you know a few configurations which are likely to occur during the
reaction, in my opinion the best option is generally that you specify them
as intermediate images. Those are just taken once, in order to
construct the starting reaction path by some interpolation. Images in
this path will still be equispaced, but will "pass close" to the images
you have specified. This will generally be sufficient, unless you really
know that the path will pass exactly by a given configuration.
If you do exactly know this (how???), say for two configurations C1 C2,
then you could split the path React-C1, C1-C2, C2-Products, and:
* either run these separately
* either merge these paths into a single one and run it: run each of them
with pw.x with 0 neb steps (this will generate linear interpolation
files), and then paste these files, manually or with path_merge.bash in
pwtools/.
Notice: if you know that the reaction proceeds through some (meta)stable
intermediate, definitely the best option is to optimize this state,
and split the NEB calculation before/after it.
Hope this helps.
Best wishes,
Guido
--
Guido Fratesi
Dipartimento di Scienza dei Materiali
Universita` degli Studi di Milano-Bicocca
via Cozzi 53, 20125 Milano, Italy
More information about the Pw_forum
mailing list