A question on dimer method

[hlzya]

Hi,

I am running a dimer calculation. Firstly, the configurations of the initial state and the final state were established by the vasp code. Then I used the CI-NEB method to find the saddle point with the forces on the atoms in all images less than 0.1 eV/A. Using the script neb2dim.pl, I set up a dimer run. I find that the maxmum forces on atoms increase significantly in every four ionic steps. I want to know why?

Enclosed is the output of the vef.pl script, and the INCAR file is also attached.

Thank you very much for your kind attention.

Yi-An

Part of the output of the vef.pl script:

step force energy
138 0.15153900 -154.506968 -0.137655
139 0.05560200 -154.507472 -0.138159
140 0.11063700 -154.507663 -0.13835
141 0.10388200 -154.507779 -0.138466
142 0.25960100 -154.507061 -0.137748
143 0.06703800 -154.507893 -0.13858
144 0.04921900 -154.508027 -0.138714
145 0.04650400 -154.508097 -0.138784
146 0.17947800 -154.507682 -0.138369
147 0.03528000 -154.508248 -0.138935
148 0.07588800 -154.508369 -0.139056
149 0.08171300 -154.508409 -0.139096
150 0.06766100 -154.508356 -0.139043
151 0.02829400 -154.508558 -0.139245
152 0.04291700 -154.508588 -0.139275
153 0.04533000 -154.508605 -0.139292
154 0.09314100 -154.508519 -0.139206
155 0.02928700 -154.508728 -0.139415
156 0.03087800 -154.508713 -0.1394
157 0.03459500 -154.508736 -0.139423
158 0.16828900 -154.508384 -0.139071
159 0.02998100 -154.508860 -0.139547
160 0.07084400 -154.508982 -0.139669
161 0.07640300 -154.509007 -0.139694
162 0.13605500 -154.508792 -0.139479
163 0.02426900 -154.509096 -0.139783
164 0.03879100 -154.509123 -0.13981
165 0.04072500 -154.509169 -0.139856
166 0.18280700 -154.508794 -0.139481
167 0.02352000 -154.509243 -0.13993
168 0.05862800 -154.509400 -0.140087
169 0.06525700 -154.509465 -0.140152
170 0.16501500 -154.509010 -0.139697
171 0.03132200 -154.509510 -0.140197
172 0.06137300 -154.509591 -0.140278
173 0.06740700 -154.509684 -0.140371
174 0.19393500 -154.509246 -0.139933
175 0.03120300 -154.509696 -0.140383
176 0.04542100 -154.509744 -0.140431
177 0.04558300 -154.509850 -0.140537
178 0.16645700 -154.509490 -0.140177
179 0.03194600 -154.509866 -0.140553
180 0.04735000 -154.510065 -0.140752
181 0.05470900 -154.510184 -0.140871
182 0.10704000 -154.509946 -0.140633
183 0.03360000 -154.510147 -0.140834
184 0.07701800 -154.510227 -0.140914

The INCAR file:

SYSTEM = default

Start parameter for this Run:
ISTART = 0
ICHARG = 2
INIWAV = 1

Electronic Relaxation:
ENCUT = 370
ENAUG = 650
PREC = Accurate
IALGO = 48
NELM = 60
NELMIN = 8 # surface = 8, bulk = 4
NELMDL = -12 # When IALGO = 38, 0
EDIFF = 1E-06 # If EDIFFG = -0.001, set EDIFF = 1E-07.
# NBANDS = 7
GGA = 91
VOSKOWN = 1
LREAL = Auto
WEIMIN = 0

Ionic Relaxation:
EDIFFG = -0.01
NSW = 1000
IBRION = 3
ISIF = 2
POTIM = 0.0
ISYM = 0
IOPT = 2

DOS related values:
SIGMA = 0.2
ISMEAR = 1

Spin polarized:
ISPIN = 2
# MAGMOM = 1(amount)*2(initial moment) 3*2

File writing
LWAVE = .FALSE.
LCHARG = .FALSE.

Calculation of DOS
# LORBIT = 1
# RWIGS = 1.286 1.000 0.370
# NPAR = 1

NEB
# ICHAIN=0
# IMAGES = 4
# SPRING = -5
# LCLIMB=.TRUE.

dimer
ICHAIN = 2
DdR = 0.005
DRotMax = 1
DFNMin = 0.01
DFNMax = 1.0

[graeme]

The dimer is moving significantly every four ionic steps:

1. force at new position
2. displacement along the current lowest mode direction
3. small finite difference rotation to find a better lowest mode direction
4. finite difference step for conjugate gradient line optimization step

Then the dimer does a Newton's step along the conjugate gradient (CG) direction, and the forces change significantly. The middle steps (2) and (3) are specific to the dimer, as compared to a regular CG minimization, and are the reason the dimer takes longer to converge than a minimization. These are the steps that find the lowest mode, along which the dimer is maximized to find a saddle.