[CP2K-user] [CP2K:16985] Re: IR spectm for NaCl under PBC
Jürg Hutter
hutter at chem.uzh.ch
Mon May 16 10:56:03 UTC 2022
Hi
just a few thoughts on geometry optimizations and frequency calculations using CP2K.
It might help for the present discussion.
In an ideal setting one expects that the energy of a molecule is constant for translations and rotations.
For periodic systems, the energy should be constant for translations.
In CP2K the XC energy is calculated using equidistant grids (the same for all PW codes) and this
breaks translational invariants and rotational invariants for molecules.
If you calculate the energy for a system while translating it rigidly you will observe a small periodic change
of the energy with a frequency given by the grid spacing. This phenomenon is usually called 'ripples'.
The size of the ripples depends on many parameters (cutoff, pseudopotential, smearing, etc).
The GPW method in CP2K shows rather large ripples, which cause problems in tight geometry optimizations
and frequency calculations. However, forces are consistent with the ripples, so there is no problem
in MD.
For geometry optimizations the optimizer gets "confused" by the ripples and starts to jump back and forth
and might have problems to converge.
In frequency calculations, you will observe non-zero translational and rotational modes. These modes
can have positive or negative frequencies. In CP2K these modes are projected out from the Hessian
assuming they don't couple to the internal molecular modes.
If you want to reduce the ripples in CP2K you can explore some options:
- use GAPW, it will reduce the size considerably
- use smearing of density and gradients (this might have side effects, use with care!)
- some functionals are worse than others (BLYP!)
best regards
Juerg Hutter
________________________________________
From: 'Václav Havel' via cp2k <cp2k at googlegroups.com>
Sent: Saturday, May 14, 2022 3:51 AM
To: cp2k
Subject: Re: [CP2K:16978] Re: IR spectm for NaCl under PBC
Dear Viktor,
I have the same question about imaginary frequencies.
Have you tackled the problem of imaginary frequency ? It seems that just increasing the optimization criteria cannot eliminate all imaginary frequencies .
With best regards
Vaclav
在2022年3月6日星期日 UTC+8 00:29:21<volk... at gmail.com> 写道：
Dear Fabian:
cannot say that evening is any good.
Relatives are in leveled Starobelsk,
and in Hadich, where the front is incoming.
Science ...
Here,
are the results of calculations of the normal modes fro NaCl.at the Gamma point.
Also, I attach the optimization files.
It seems to me that after increasing the optimization criteria
MAX_DR 8.0E-07
RMS_DR 4.0E-07
MAX_FORCE 1.0E-07
RMS_FORCE 1.0E-08
The results are curious.
May be there are some mistakes I did?
Again,
I wish to ask if CP2K has an issue
about the Fourier grid
that may generate imaginary frequencies
if the coords are not properly positioned?
If you do not have energy/time/will,
please,
forgive and disregard the inquiry.
With best regards,
Victor
On Friday, January 28, 2022 at 3:42:08 PM UTC+3 fabia... at gmail.com wrote:
Dear Viktor,
Several things:
- If you choose RUN_TYPE GEO_OPT then you should have MOTION&GEO_OPT and not MOTION&CELL_OPT and vice versa.
- Set EPS_DEFAULT to 1e-12
- For CELL_OPT: you should have a &CELL_REF section with dimensions approx. 1.3-1.5 times larger (test this!) than &CELL
- You need MULTIPLE_UNIT_CELL in &CELL_REF too
Cheers,
Fabian
On Friday, 28 January 2022 at 09:30:16 UTC+1 volk... at gmail.com wrote:
Dear Fabian:
good morning.
According to your notice
I redefined MAX_FORCE to be under
&MOTION
&CELL_OPT
However, the cp2k failed convergence - energy was just oscillating.
After studying your replies to other members,
I did two changes in the input.
1.
While keeping the definitions in
&MOTION
&CELL_OPT
....
in GLOBAL
I changed to
RUN_TYPE GEO_OPT
The logic of the guess is that apparently &CELL_OPT can not lead.
Then, may be,
GEO should lead, while CELL should be set to follow.
It is not clear, if under such setting STRESS TENSOR would have any meaning.
Perhaps it has not role no more and becomes redundant.
2.
Also, I introduced
CELL_REF
I attach the input.
This is the job I am testing.
Would you have a minute to comment, criticize - I be thankful.
In any way,
thank you for your time,
and nice weekend
:-)
Best!
Victor
On Thu, Jan 27, 2022 at 2:30 PM fabia... at gmail.com <fabia... at gmail.com> wrote:
cp2k works just fine and does everything you tell it to do. The issue is that you specified MAX_FORCE 1.0E-07 etc. for &GEO_OPT instead of &CELL_OPT...
On Thursday, 27 January 2022 at 11:39:21 UTC+1 volk... at gmail.com wrote:
Dear Fabian:
Dear Dr. Hutter:
is it possible that CP2K is present not to follow the input instructons?
According to my instructions (pleas, see attached)
in the MOTION section there are instructions
MAX_DR 8.0E-07
RMS_DR 4.0E-07
MAX_FORCE 1.0E-07
RMS_FORCE 1.0E-08
but
as Fabian has noticed:
The output you attached shows for step 28:
Max. gradient = 0.0002958006
Conv. limit for gradients = 0.0004500000
Conv. in gradients = YES
So that output is for MAX_FORCE 4.5E-4 and not 1.0E-7. You should also decrease EPS_SCF (to 1e-7 or 1e-8) and probably EPS_DEFAULT.
Is it possible that there is a bug
that prevents a certain level of precision?
Thank you.
With best regards.
Victor
On Thu, Jan 27, 2022 at 12:00 PM fabia... at gmail.com <fabia... at gmail.com> wrote:
Dear Viktor,
The output you attached shows for step 28:
Max. gradient = 0.0002958006
Conv. limit for gradients = 0.0004500000
Conv. in gradients = YES
So that output is for MAX_FORCE 4.5E-4 and not 1.0E-7. You should also decrease EPS_SCF (to 1e-7 or 1e-8) and probably EPS_DEFAULT.
Fabian
On Wednesday, 26 January 2022 at 21:12:03 UTC+1 volk... at gmail.com wrote:
Dear Fabian:
please forgive me
If I set
OPTIMIZER BFGS
MAX_DR 8.0E-06
RMS_DR 4.0E-06
MAX_FORCE 1.0E-06
RMS_FORCE 1.0E-07
then the forces are
ATOMIC FORCES in [a.u.]
# Atom Kind Element X Y Z
1 1 Na 0.00000796 0.00000796 0.00000796
2 1 Na 0.00000796 0.00001127 0.00001127
3 1 Na 0.00001127 0.00000796 0.00001127
4 1 Na 0.00001127 0.00001127 0.00000796
5 2 Cl 0.00000052 0.00000052 0.00000052
6 2 Cl 0.00000052 0.00000052 0.00000052
7 2 Cl 0.00000052 0.00000052 0.00000052
8 2 Cl 0.00000052 0.00000052 0.00000052
SUM OF ATOMIC FORCES 0.00004056 0.00004056 0.00004056 0.00007026
and the soectrum is terrible (please, see the attached xxx_F.out)
Then I try to instruct more strict convregence by
OPTIMIZER BFGS
MAX_DR 8.0E-07
RMS_DR 4.0E-07
MAX_FORCE 1.0E-07
RMS_FORCE 1.0E-08
but the forces deteriorate
ATOMIC FORCES in [a.u.]
# Atom Kind Element X Y Z
1 1 Na 0.00009945 0.00009945 0.00009945
2 1 Na 0.00025879 -0.00001513 -0.00001513
3 1 Na -0.00001519 0.00025872 -0.00001518
4 1 Na -0.00001513 -0.00001513 0.00025879
5 2 Cl 0.00031492 0.00004422 0.00004422
6 2 Cl 0.00024177 0.00024177 0.00024177
7 2 Cl 0.00004422 0.00004422 0.00031492
8 2 Cl 0.00004422 0.00031492 0.00004422
SUM OF ATOMIC FORCES 0.00097305 0.00097303 0.00097305 0.00168536
I attach the optimization inp and out pair.
I donot see what is wrong?
May be GPW and BFGS are not for such optimizations?
Thank you.
Victor
On Mon, Jan 24, 2022 at 4:22 PM fabia... at gmail.com <fabia... at gmail.com> wrote:
Dear Victor,
You need a much tighter convergence criterion for the optimization prior to the vibrational analysis. You increased MAX_FORCE to 5e-3 (guessing from the output), instead you have to decrease it to at least 1e-5, maybe even lower.
CP2K prints the forces for the replica in the output. The forces under "Minimum Structure - Energy and Forces:" should be more or less 0 and considerably differ between the other images. Currently you are computing the vibrational analysis from almost pure noise.
Cheers,
Fabian
On Sunday, 23 January 2022 at 19:58:05 UTC+1 volk... at gmail.com wrote:
Dear colleagues:
here is the normal mode analysis I could manage by now for NaCl under PBC.
There is only one unit cell: previously, I reported that I could not configure CP2K to reproduce NaCl structure using multiple cells 2 2 2.
There are strange distortions, which I did confirm to sustain under different CUTOFF values (1000 and 600), and regardless if dispersion is switched off or not.
When a single cell under PBC, optimization leads to the “right” structure, when ESP 10^-12.
However, the negative(imaginary) frequencies are many and they are very large.
I may anticipate that the “miss performance” is enhanced due to the PBC.
Eventhough the results are "devastating",
:-)
nonetheless, I find this might be very good, very helpful – I got an idea.
Dear Dr. Hutter,
is it possible that CP2K may have an “issue” about positioning a structure
in respect to both, in respect to the FT grind and in respect to the PBC.
Perhaps, there could be a relative mismatch phenomenon, which could stimulate artificial transition states with negative frequencies?
May be, what I observe is that PBC on the unit cell helps to sense this better?
Thank you for your time and for any comments.
With best regards,
Sincerely
Victor
--
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/bAo0CtNixo8/unsubscribe.
To unsubscribe from this group and all its topics, send an email to cp2k+uns... at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/dd5077c3-4bac-4fa0-a81d-ef779f43e157n%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/dd5077c3-4bac-4fa0-a81d-ef779f43e157n%40googlegroups.com?utm_medium=email&utm_source=footer>.
--
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/bAo0CtNixo8/unsubscribe.
To unsubscribe from this group and all its topics, send an email to cp2k+uns... at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/d036e57c-90aa-4656-85d8-ce7e1d917d8cn%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/d036e57c-90aa-4656-85d8-ce7e1d917d8cn%40googlegroups.com?utm_medium=email&utm_source=footer>.
--
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... at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/a6d3efcd-5a7e-4243-b67f-fcb79386feben%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/a6d3efcd-5a7e-4243-b67f-fcb79386feben%40googlegroups.com?utm_medium=email&utm_source=footer>.
--
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+unsubscribe at googlegroups.com<mailto:cp2k+unsubscribe at googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/b435519a-f6e2-4059-b6b4-f30b10b19a54n%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/b435519a-f6e2-4059-b6b4-f30b10b19a54n%40googlegroups.com?utm_medium=email&utm_source=footer>.
--
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+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/ZR0P278MB07599F729E36EA6A9DDAC9929FCF9%40ZR0P278MB0759.CHEP278.PROD.OUTLOOK.COM.
More information about the CP2K-user
mailing list