NEB input parameters

Asked by rezabma on 2019-09-30

How can I change following NEB parameters:
spring constant list
dt initial current
alpha initial current
Tolerance Maximum change

For example, I need separate k for different images. Where do I enter them? in my fdf? with which keywords? please leave an example.

Question information

English Edit question
Siesta Edit question
No assignee Edit question
Last query:
Last reply:
Nick Papior (nickpapior) said : #1

As alwas, the Lua interface takes very little from the Siesta fdf input file.
The neb_simple.lua example clarifies that it retrieves the maximum displacement and the maximum force tolerance for image relaxation. So those need not be set in the Lua interface.

The NEB class in lua is instantiated via this with individual spring constants.

-- Lua code
run = NEB(images, {k={0.1, 0.2, 0.3, 0.1}})

where images has 4 images and the different images has a spring constant 0.1, 0.2, 0.3 and the last image has 0.1.

Does dt and alpha correspond to variables for the FIRE algorithm? If so then your lua script contains the image relaxation class FIRE(...) somewhere.
dt is named dt_init (since dt is variable along the relaxation)
alpha is named alpha_init (since alpha is variable along the relaxation)

If dt and alpha refer to other things, please let me know since I don't know what you mean. :)

rezabma (rezabma) said : #2

Thanks Nick, I guessed that I should set fire parameters in the flos/flos/optima/fire.lua, but I thought that maybe I set them in neb_simple.lua in the working directory.
I play with fire parameters, and try to set them according to the ASE-Siesta and see the NEB convergence rate. As I told earlier, siesta is very fast in scf and relaxation. It also is very fast in NEB-LUA-Siesta, but the convergence rate of NEB-LUA-Siesta is very slow. I think this maybe concerning to the spring constants, global/local, dt, ... parameters. I compared the NEB calculation of OPENMX, ASE-SIESTA, and NEB-LUA-SIESTA codes. The last one is the fastest one, but with the lowest convergence rate. OPENMX uses hybrid method for NEB, and its convergence rate is very good (and also its SCF is good but not as good as Siesta), but doesn't have climbing method.
ASE-Siesta used python3 modules in ASE to run SCFs by Siesta (similar to lua linking but with python), but its SCFs takes much longer time than LUA-Siesta. I used VASP also in the past year to do NEB job,it is also very slow in convergence rate (similar to LUA-Siesta).

Can you help with this problem?

Provide an answer of your own, or ask rezabma for more information if necessary.

To post a message you must log in.