[ase-users] Parallelizing over NEB images
Esben Leonhard Kolsbjerg
esb at inano.au.dk
Tue Nov 21 17:43:22 CET 2017
Hi Divya,
I was in the middle of answering your mail as well when Gaël replied. Here is my take!
1. So, fmax is the convergence criteria for ALL images individually, but rather think you are converging the complete path as a whole. The path is first “fully” optimized when all images have a maximum NUGDED force below fmax. That mean the TRUE force will not be below fmax for other than the climbing image.
I personally think of the NEB module written as a sophisticated constraint – like FixAtoms constraint!
2. You are right that it is the heights energy images which energy and force which are logged. There is no direct why of setting the NEB module to log an individual image. A solution could probably be to write and attach an observer to your optimizer that logs the images individually. I have tried to do write something you can’t start from if you want to proceed with this (a have attached a NEB script below). I don’t promise it is without bugs!
Good luck,
Esben
____________________________________________________________________
#!/usr/bin/env python
# -*- coding: utf-8 -*-
from __future__ import division
from gpaw import *
from ase.io import write,read, Trajectory
from ase.neb import NEB
from ase.optimize import BFGS
import numpy as np
from ase import Atoms
import ase.parallel as mpi
from gpaw import extra_parameters
extra_parameters['blacs'] = True
import time
size = mpi.world.size
rank = mpi.world.rank
from math import sqrt
class logger:
def __init__(self, name, traj, master):
self.name = name
self.traj = traj
self.master = master
if self.master:
self.logfile = open(name, 'a')
self.logfile.write(
'%4s %8s %15s %12s\n' %
('Step', 'Time', 'Energy', 'fmax'))
self.logfile.flush()
self.nsteps = 0
def write(self):
if self.master:
forces = self.traj.get_forces()
fmax = sqrt((forces**2).sum(axis=1).max())
e = self.traj.get_potential_energy()
T = time.localtime()
self.logfile.write('%3d %02d:%02d:%02d %15.6f %12.4f\n' %
(self.nsteps, T[3], T[4], T[5], e, fmax))
self.logfile.flush()
self.nsteps += 1
images = [read('neb000.traj')]
images.append(read('neb001.traj'))
images.append(read('neb002.traj'))
images.append(read('neb003.traj'))
images.append(read('neb004.traj'))
nim = len(images) - 2
n = size // nim # number of cpu's per image
j = 1 + rank // n # my image number
assert nim * n == size
for i in range(nim):
ranks = np.arange(i * n, (i + 1) * n)
if rank in ranks:
calc = GPAW(xc='LDA',
mode='lcao',
basis='dzp',
communicator=ranks)
images[i + 1].set_calculator(calc)
images[i + 1].calc.set(txt='calc_neb%d.out'%j)
neb=NEB(images, climb=True,
parallel=True)
qn = BFGS(neb, logfile='maxE.log')
traj = Trajectory('sneb00%d.traj' % (j), 'w', images[j],
master=(rank % n == 0))
qn.attach(traj)
logger = logger('log00%d.log' % (j), images[j],
master=(rank % n == 0))
qn.attach(logger.write)
qn.run(fmax=0.025)
_____________________________________________________________________________________
D. 21/11/2017 17.08 skrev "ase-users-bounces at listserv.fysik.dtu.dk på vegne af Gaël Donval via ase-users" <ase-users-bounces at listserv.fysik.dtu.dk på vegne af ase-users at listserv.fysik.dtu.dk>:
Hi Divya,
> Dear ASE users,
>
> I am trying to perform my first CI-NEB calculation where I have 4
> images between the initial and final images and the optimization of
> the images is parallelized (script below). I would greatly appreciate
> your help on the below 2 questions:
>
>
> 1. What does the fmax for the optimizer specify exactly? Is
> this taken as the curvature at the initial and final states of the
> NEB interpolation? How does NEB determine the convergence criterion
> for the each of the images on the minimum energy path and where can I
> see this?
The `fmax` keyword is always a maximum force of some sort. To go into
more details, your LBFGS optimizer is going to call the `get_forces`
method to whatever object is provided (in your case a NEB object),
calculate their norm and check if none of them is greater than `fmax`.
For an NEB object, `get_forces` is going to gather all the forces from
all the inner images (that is, all images except the end-points --that
are untouched).
So the `fmax` you provide *is* the convergence criterion for all
images: the images are interdependent; they are coupled; "converging"
one by itself makes little sense.
>
>
> 2. When I read the log file, it seems like instead of logging
> the energy and max force for each image, it is logging the potential
> energy and max force for the image with the highest energy. I want to
> confirm that despite this, each of the images are separately
> optimized towards the convergence criteria and that this logging does
> not affect the optimization itself. Also, how can I make the log the
> forces and energies of each image separately for it to make more
> sense?
The energy in your logfile is indeed the maximum energy. What NEB is
usually used for is determining an energy barrier providing known end-
points. In that context, providing the maximum energy point on the NEB
path as a summary of the advance of the algorithm is the only thing
that really makes sense. Though this is just a log file, a summary: all
the other inner images are constantly optimised too.
Actually, you should really think of all the images as a single system.
Regarding local information, you can extract it either from the
trajectory file generated by LBGFS (provided you gave a value to its
`trajectory` argument); you could extract that from the images
themselves or you could use (and study if you need something more)
`ase.neb.NEBTools`.
Cheers,
Gaël
>
> Thanks in advance.
>
> Best regards,
> Divya
>
> from __future__ import print_function
> from ase.constraints import FixAtoms
> from ase.optimize import BFGS, QuasiNewton, LBFGS, MDMin
> from ase.lattice.cubic import FaceCenteredCubic
> from ase.build import fcc110, add_adsorbate
> from gpaw import GPAW, PW, FermiDirac, Mixer
> from ase.io import write, read,Trajectory
> from ase import Atoms, Atom
> from ase.neb import NEB
> from ase.optimize.fire import FIRE
> from ase.parallel import rank, size
> import numpy as np
>
> calc = GPAW('Ag110_2x3x6_BEEF.gpw')
> CO2_slab=read('CINEB_Ag110_2x3x6_CO2-H_reference.traj')
> COOH_slab=read('Ag110_2x3x6_COOHupLbridge_rot_BEEF.traj')
>
> initial = CO2_slab
> final = COOH_slab
>
> constraint = FixAtoms(mask=[atom.tag > 2 for atom in initial])
>
> n = size // 4 # number of cpu's per image
> j = 1 + rank // n # my image number
> assert 4 * n == size
>
> images = [initial]
>
> for i in range(4):
> ranks = range(i * n, (i + 1) * n)
> image = initial.copy()
>
> if rank in ranks:
> calc=GPAW(mode=PW(450), xc='BEEF-vdW', basis='dzp',
> gpts=(48,48,160), kpts=(3,3,1), txt='CINEB_Ag110_2x3x6_CO2-
> COOH_Tafel_neb%d.txt' % j, occupations=FermiDirac(0.1),
> mixer=Mixer(beta=0.1, nmaxold=5, weight=50.0), communicator=ranks)
> image.set_calculator(calc)
>
> image.set_constraint(constraint)
> images += [image]
>
> images += [final]
>
> neb = NEB(images, parallel=True, climb=True)
> neb.interpolate('idpp')
> #Run NEB calculation
> opt = LBFGS(neb, logfile='CINEB_Ag110_2x3x6_CO2-COOH_Tafel.log')
> opt.run(fmax=0.05)
>
> for i in range(1, 5):
> opt.attach(io.write('CINEB_Ag110_2x3x6_CO2-COOH_Tafel-%d.cif' %
> i, 'w', images[i]))
> opt.attach(io.Trajectory('CINEB_Ag110_2x3x6_CO2-COOH_Tafel-
> %d.traj' % i, 'w', images[i]))
>
>
> #Logfile
>
> LBFGS: 0 18:17:31 -5379.937600 6.8668
> LBFGS: 1 18:44:44 -5380.096707 6.1825
> LBFGS: 2 19:11:00 -5380.155402 4.4230
> LBFGS: 3 19:39:07 -5380.249108 3.3944
> LBFGS: 4 20:07:05 -5380.368769 3.2199
> LBFGS: 5 20:35:12 -5380.521739 3.4305
> LBFGS: 6 21:03:11 -5380.727266 3.9711
> LBFGS: 7 21:31:13 -5380.953959 4.5234
> LBFGS: 8 21:57:42 -5381.208836 5.0993
> LBFGS: 9 22:24:52 -5381.490995 5.6717
> LBFGS: 10 22:51:58 -5381.801246 6.2453
> LBFGS: 11 23:19:50 -5382.139225 6.7877
> LBFGS: 12 23:47:46 -5382.503905 7.2781
> LBFGS: 13 00:14:56 -5382.896339 7.7335
> LBFGS: 14 00:43:06 -5383.314196 8.1039
> LBFGS: 15 01:12:07 -5383.755902 8.3441
> LBFGS: 16 01:42:02 -5384.209731 8.3273
> LBFGS: 17 02:12:50 -5384.660808 7.9158
> LBFGS: 18 02:44:25 -5385.012915 7.1416
> LBFGS: 19 03:14:20 -5385.247200 6.9705
> LBFGS: 20 03:45:01 -5385.395184 7.2958
> LBFGS: 21 04:15:33 -5385.502550 7.5201
> LBFGS: 22 04:46:22 -5385.588210 7.6800
> LBFGS: 23 05:17:00 -5385.661367 7.7952
> LBFGS: 24 05:47:43 -5385.726172 7.8750
> LBFGS: 25 06:18:26 -5385.784934 7.9238
> LBFGS: 26 06:49:07 -5385.839004 7.9446
> LBFGS: 27 07:20:34 -5385.889114 7.9381
> LBFGS: 28 07:52:51 -5385.935424 7.9049
> LBFGS: 29 08:25:06 -5385.974499 7.8480
> LBFGS: 30 08:57:26 -5386.006282 7.7645
> LBFGS: 31 09:29:42 -5386.029139 7.6479
> LBFGS: 32 10:02:02 -5386.039673 7.4886
> LBFGS: 33 10:35:16 -5386.032930 7.2809
> LBFGS: 34 11:07:37 -5386.006100 7.0377
> LBFGS: 35 11:39:01 -5385.960810 6.7818
> LBFGS: 36 12:11:23 -5385.900440 6.5308
> LBFGS: 37 12:44:37 -5385.829873 6.2943
> LBFGS: 38 13:17:49 -5385.753279 6.0716
> LBFGS: 39 13:50:59 -5385.675096 6.1741
> LBFGS: 40 14:23:24 -5385.599475 6.4399
> LBFGS: 41 14:54:54 -5385.530148 6.6489
> LBFGS: 42 15:28:03 -5385.469514 6.8101
>
>
> _______________________________________________
> ase-users mailing list
> ase-users at listserv.fysik.dtu.dk
> https://listserv.fysik.dtu.dk/mailman/listinfo/ase-users
_______________________________________________
ase-users mailing list
ase-users at listserv.fysik.dtu.dk
https://listserv.fysik.dtu.dk/mailman/listinfo/ase-users
More information about the ase-users
mailing list