neb : question about the number of scf optimizations for each band step for each replica

[afloris]

Dear cp2k users,

 

I am running for the first time some NEB calculations...and I am using cp2k since few days only.

From the output of a particular replica, e.g.  xxx-BAND2.out,  

(xxx being the name of the project, 2 is the replica), I see that for each band step, several scf optimizations are performed for each replica.

I have some questions: 

1. why  more than one scf optimization is performed? (every optimization being achieved with ~40 scf steps) 

2. Are these purely electronic optimizations, or there is a geometrical minimization involved?

(this question is motivated because different forces are found after each scf optimization...although no "new" atomic positions are printed...) 

Can someone clarify this to me?

Thank you very much 

Kind Regards

Andrea

-- 

 Dr Andrea Floris

 Research Associate

 King's College London

 Strand, London WC2R 2LS

 United Kingdom

 Phone: +44 (0) 207 848 2064

 Fax    : +44 (0) 207 848 2420

 Location:  Strand Building, 4th floor, Room 4.02 

 Emails: andrea...@kcl.ac.uk, an.f...@gmail.com

[Teodoro Laino]

On May 17, 2013, at 7:22 PM, afloris <an.f...@gmail.com> wrote:

Dear cp2k users,

 

I am running for the first time some NEB calculations...and I am using cp2k since few days only.

From the output of a particular replica, e.g.  xxx-BAND2.out,  

(xxx being the name of the project, 2 is the replica), I see that for each band step, several scf optimizations are performed for each replica.

I have some questions: 

1. why  more than one scf optimization is performed? (every optimization being achieved with ~40 scf steps) 

it depends on how many processors you are using and how many processors per replica you requested. On the top of that, the number of scf optimisations depends also on the type of optimiser you are using. 

Again without an input is almost useless to cover all possible things you may see.

2. Are these purely electronic optimizations, or there is a geometrical minimization involved?

the geometry optimisation steps are not shown in the *BAND* output

Regards

(this question is motivated because different forces are found after each scf optimization...although no "new" atomic positions are printed...) 

Can someone clarify this to me?

Thank you very much 

Kind Regards

Andrea

-- 

 Dr Andrea Floris

 Research Associate

 King's College London

 Strand, London WC2R 2LS

 United Kingdom

 Phone: +44 (0) 207 848 2064

 Fax    : +44 (0) 207 848 2420

 Location:  Strand Building, 4th floor, Room 4.02 

 Emails: andrea...@kcl.ac.uk, an.f...@gmail.com

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To post to this group, send email to cp...@googlegroups.com.
Visit this group at http://groups.google.com/group/cp2k?hl=en.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

[Teodoro Laino]

>>
>> 1. why more than one scf optimization is performed? (every optimization being achieved with ~40 scf steps)
> it depends on how many processors you are using and how many processors per replica you requested. On the top of that, the number of scf optimisations depends also on the type of optimiser you are using.
> Again without an input is almost useless to cover all possible things you may see.

Correction: forget about the parallel setup - you should always have a number of BAND* files equal to the number of frames of your NEB.
So not sure at all what you are seeing… (probably it is only the output of the OUTER_SCF cycle)

In any case, without input..

[Andrea Floris]

Dear Teo, 

I attach the input and output and all relevant files. 

My problem is in the BAND* files , whose number is indeed equal to the number of replica (three in my case). 

If you in one of these files (prefix is neba) do  

grep -i  'Band  Step  Nr.'  neba-BAND1.out

you have as output something like

... 

 ** Band  Step  Nr. :   13                                                    **

 ** Band  Step  Nr. :   13                                                    **

 ** Band  Step  Nr. :   13                                                    **

 ** Band  Step  Nr. :   13                                                    **

 ** Band  Step  Nr. :   13                                                    **

 ** Band  Step  Nr. :   14                                                    **

 ** Band  Step  Nr. :   14                                                    **

 ** Band  Step  Nr. :   14                                                    **

...

so, e.g. 5 optimizations (but geometrical or purely electronic scf???) in the band  step 13, and 3 in the band step 14...

That's the point I don't understand...

Moreover, I also do not understand why this is done *also* for the end points, when you choose to not optimize 

the initial and final replica...For example in my case here replica 1 was fixed.  In fact in the 

neba-1.ener

file, the energy of replica 1 is fixed, i.e. it has always the same energy through the band optimization steps. But in the file  neba-BAND1.out many total energies are found...

one for each scf optimization.

Thanks in advance for you answer,

Best

Andrea

--
You received this message because you are subscribed to a topic in the Google Groups "cp2k" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/cp2k/_icaeFBVnZ4/unsubscribe?hl=en.
To unsubscribe from this group and all its topics, send an email to cp2k+uns...@googlegroups.com.

To post to this group, send email to cp...@googlegroups.com.
Visit this group at http://groups.google.com/group/cp2k?hl=en.
For more options, visit https://groups.google.com/groups/opt_out.

[Teodoro Laino]

DIIS performs a LS - what you see is the result of the line search.

Use NO_LS if you don't like that.

Moreover, I have no idea if this is a production run or simply a test, but you want to keep the distance between your replica much smaller than 3.0+ Angstrom.

Hope this helps.

Teo

You received this message because you are subscribed to the Google Groups "cp2k" group.

To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.

To post to this group, send email to cp...@googlegroups.com.
Visit this group at http://groups.google.com/group/cp2k?hl=en.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

<neb_test_3replicas.tgz>

[Andrea Floris]

On Mon, May 20, 2013 at 4:08 PM, Teodoro Laino <teodor...@gmail.com> wrote:

DIIS performs a LS - what you see is the result of the line search.

Use NO_LS if you don't like that.

What are the consequences of using NO_LS in a NEB calculation? Using NO_LS would result in a single optimization of the image

(per band step) and  (maybe separate issue) no electronic optimization at all for "fixed" images?

I am just running a test, with 3-5 replica only.  I see quite a strong dependence on the K_SPRING constant if I vary it from 0.05 to 0.02 

(the barrier decreases of about 0.05eV and the path is different).  

Is is correct to expect a much reduced variation of the barrier with K_SPRING, if the number of

images is increased, e.g. to 10-15? Which distance is typically OK between adiacent images, 0.5 A?

Thank you for your help

Best

Andrea

[wolfshow]

Dear All,

I am facing with the similar questions.

In my calculations, if I used NO_LS, the results seem more reasonable (the total energy of each image converged). I want to know if there is some disadvantageous consequence of using NO_LS in a NEB calculation

Thank you in advance.

Best Regards,

Fengchao

在 2013年5月21日星期二UTC+8上午1时52分23秒，afloris写道：