restarting NEB calculations.

[Jörg Saßmannshausen]

Dear all,

I know that question has been asked before but I could not find the answer to
it so I am posting it again.

How do I restart a NEB calculation?

I tried it that way here:

! Possibly restart (if explicitly requested)
@IF 1
&EXT_RESTART ON
RESTART_DEFAULT F
RESTART_FILE_NAME sm-scan3-1_275.restart
RESTART_POS T
RESTART_COUNTERS T
RESTART_VEL T
RESTART_THERMOSTAT T
&END EXT_RESTART
@ENDIF

in the same directory where I done the calculation but alas! it appears it
started again from scratch. The file sm-scan3-1_275.restart is the last restart
file I got so I assumed it contains all the required information.

If I cannot restart it that way, and as I am close to the convergence
criteria, can I simply use the last frames and use that as a starting point? I
don't have that much experience with NEB and I would like to get a sound
result and not a biased one, if that makes sense.

All the best from a mild London!

Jörg


--
*************************************************************
Jörg Saßmannshausen
University College London
Department of Chemistry
Gordon Street
London
WC1H 0AJ

email: j.sassma...@ucl.ac.uk
web: http://sassy.formativ.net

Please avoid sending me Word or PowerPoint attachments.
See http://www.gnu.org/philosophy/no-word-attachments.html

[Teodoro Laino]

On Sep 26, 2013, at 1:31 PM, Jörg Saßmannshausen <j.sassma...@ucl.ac.uk> wrote:

> in the same directory where I done the calculation but alas! it appears it
> started again from scratch. The file sm-scan3-1_275.restart is the last restart
> file I got so I assumed it contains all the required information.

the energies that you will get from your NEB must be identical (up to the SCF convergence) to the one of the frame corresponding to the restart file.
Is this the case?

If not, check the keywords that control aligment..
Also, not all information are stored in the restart file. For instance, the DIIS vectors are not. This means that the first steps may be offline regarding the gradients, but the energy of the first point MUST be identical to the one of the frame when restart was written.

> If I cannot restart it that way, and as I am close to the convergence
> criteria, can I simply use the last frames and use that as a starting point? I

yes

Teo

[Jörg Saßmannshausen]

Hi Teo,

thanks for the quick feedback!

On Donnerstag 26 September 2013 Teodoro Laino wrote:
> On Sep 26, 2013, at 1:31 PM, Jörg Saßmannshausen <j.sassma...@ucl.ac.uk> wrote:
> > in the same directory where I done the calculation but alas! it appears
> > it started again from scratch. The file sm-scan3-1_275.restart is the
> > last restart file I got so I assumed it contains all the required
> > information.
>
> the energies that you will get from your NEB must be identical (up to the
> SCF convergence) to the one of the frame corresponding to the restart
> file. Is this the case?

No, it is not. That is the last frame which was computed in the first run:

*******************************************************************************
BAND TYPE = SM
BAND TYPE OPTIMIZATION = DIIS
STEP NUMBER = 275
NUMBER OF NEB REPLICA = 16
DISTANCES REP = 2.530978 2.530977 2.530977 2.530976
2.530975 2.530975 2.530977 2.530978
2.530978 2.530977 2.530977 2.530977
2.530977 2.530978 2.530978
ENERGIES [au] = -648.478940 -648.467777 -648.450539 -648.436808
-648.424076 -648.411575 -648.401455 -648.395091
-648.394535 -648.397824 -648.402001 -648.407490
-648.415564 -648.423551 -648.425201 -648.422192
BAND TOTAL ENERGY [au] = -10374.75462048212466
*******************************************************************************
*******************************************************************************
RMS DISPLACEMENT = 0.00053 [ 0.00010] ( NO)
MAX DISPLACEMENT = 0.00885 [ 0.00020] ( NO)
RMS FORCE = 0.00274 [ 0.00050] ( NO)
MAX FORCE = 0.02016 [ 0.00100] ( NO)
*******************************************************************************

and that is the first frame after the restart:

*******************************************************************************
BAND TYPE = SM
BAND TYPE OPTIMIZATION = SD
STEP NUMBER = 1
NUMBER OF NEB REPLICA = 16
DISTANCES REP = 4.611085 4.611083 4.611081 4.611071
4.611053 4.611042 4.611057 4.608585
4.363173 4.583945 4.610633 4.610733
4.611017 4.611071 4.611085
ENERGIES [au] = -648.478940 -648.388234 -648.213414 -648.033912
-647.902317 -647.837233 -647.892035 -648.067775
-648.274512 -648.336806 -648.031108 -647.791068
-647.757054 -647.962000 -648.273950 -648.422192
BAND TOTAL ENERGY [au] = -10369.66255104071388
*******************************************************************************
*******************************************************************************
RMS DISPLACEMENT = 0.05464 [ 0.00010] ( NO)
MAX DISPLACEMENT = 0.26217 [ 0.00020] ( NO)
RMS FORCE = 0.02618 [ 0.00050] ( NO)
MAX FORCE = 0.18566 [ 0.00100] ( NO)
*******************************************************************************

So there is a large difference in the engergy and it does no get any better after 6 more frames:

*******************************************************************************
BAND TYPE = SM
BAND TYPE OPTIMIZATION = SD
STEP NUMBER = 6
NUMBER OF NEB REPLICA = 16
DISTANCES REP = 4.167181 4.167174 4.167158 4.167129
4.167112 4.167054 4.166423 4.159775
4.127302 4.118230 4.162667 4.166636
4.167056 4.167160 4.167181
ENERGIES [au] = -648.478940 -648.433865 -648.352353 -648.268776
-648.195851 -648.133911 -648.104171 -648.144445
-648.236878 -648.266347 -648.180945 -648.131997
-648.148447 -648.242456 -648.366991 -648.422192
BAND TOTAL ENERGY [au] = -10372.10856378791868
*******************************************************************************
*******************************************************************************
RMS DISPLACEMENT = 0.02265 [ 0.00010] ( NO)
MAX DISPLACEMENT = 0.12833 [ 0.00020] ( NO)
RMS FORCE = 0.01330 [ 0.00050] ( NO)
MAX FORCE = 0.10403 [ 0.00100] ( NO)
*******************************************************************************

So I think I got a problem here.

> If not, check the keywords that control aligment..

I did not set the ALIGN_FRAMES keyword and the default is .T. here.
I am doing a SM band type here and in that section I have set
ROTATE_FRAMES F
Was that correct or is that where the problem is?

> Also, not all information are stored in the restart file. For instance, the
> DIIS vectors are not. This means that the first steps may be offline
> regarding the gradients, but the energy of the first point MUST be
> identical to the one of the frame when restart was written.

That is what I would have expected. However, unless I am blind and do not see something,
as the energy is different between step 275 and the first step in the restart I got a problem here.

>
> > If I cannot restart it that way, and as I am close to the convergence
> > criteria, can I simply use the last frames and use that as a starting
> > point? I
>
> yes

Are there any problems associated with that method?

[Teodoro Laino]

USE NEB and not SM method. It's only available a linear interpolation and this is by far not enough.

CI-NEB or IT-NEB should be your choice.

The reason why you see different energies is because of the SM method and the reparameterization (see papers referring to this method)

Basically even if you restart, the initial points are moved along the string. This will never allow you to have identical energies after a restart.

Go back to IT-NEB or CI-NEB and everything will work ok.

Teo

[Jörg Saßmannshausen]

Hi Teo,

thanks for the feedback here. I will try the NEB methods you suggested.

All the best from London

Jörg

[Zhendong Guo]