[Pw_forum] phonons

[Stefano Baroni]

On Aug 20, 2006, at 3:15 AM, lan haiping wrote:

> Dear  Prof Nicola,

just a note about the "western" usage of names. (I would be glad to  
hear an explanation on how names are used in Chinese!). "Nicola" is  
the "given" name, i.e. it is the name that parents give to the  
newborn baby. It is also called "first" name because in written it is  
the first name that must be used. The other name "Marzari" is the  
family name which is transmitted from one generation to the next in a  
patrilinear way (a baby gets the fathers' family name by default). It  
is a form of confidentiality to address somebody by his/her first  
name. It is perfect here because: i) it is common usage on the web  
and; ii) it is common practice among scientists (particularly in the  
US). However, addressing somebody by his/her title is _very_ formal.  
Not very appropriate in this forum, but above all kind of funny if  
accompanied by the much more confidential usage of the first name.  
Resuming: "Dear Nicola" is confidential and "Dear Prof. Marzari" is  
formal - both ar OK, the first more appropriate to this forum. "Dear  
Prof Nicola" is just funny! ... I was not joking saying that I am  
curious about _your_ usage of "given", "family", or whatever names.  
If you have a  few spare minutes, just let me know ...

Science, now!

>
> Thanks for your long suggestion.
>
> I may not follow some of your words, would you please clarify  ?
>  1,  How to relax the ions more ?

The code must have some default threshold for the atomic forces (and,  
maybe, stress too if you also relax the unit cell) such that  when  
forces are smaller, it stops. The threshold is usually more than  
adequate for most practical purposes, _but to learn_ What I mean is  
that it is a good idea to practice and to see _with one's own eyes_  
how a given property (phonon frequencies, in the present case) depend  
on external parameter and/or convergence criteria. As it is possible  
to vary by hands the values of the convergence thresholds, it is a  
good idea to try a same calculation with different values of the  
relevant thresholds and to see how the frequencies depend on them.  I  
have a comment, though. As Nicola said in a previous post, it is  
unlikely (though not impossible) that phonons go soft because the  
structure is poorely optimized. It is worth stopping a while and  
understand carefully why it is both unlikely _and_ possible. It is  
possible because squared phonon frequencies are related to the  
Hessian (matrix of second derivatives) of the energy with respect to  
the nuclear coordinates. It is perfectly possible that, out of  
equilibrium, one or more of the eigenvalues of this matrix are  
negative. I even believe (but would like not to comment on this  
statement) that the ratio of the configuration space corresponding to  
all positive eigenvalues to the total configuration space decreases  
exponentially with the number of atoms. HOWEVER, phonon frequencies  
only depend on second derivatives of the energy, and not directly on  
first derivatives. So, if you are not too far from the minimum, not  
too good a convergence (i.e. the existence of nonvanishing forces)  
may certainly affect the accuracy of phonon calculations (it would  
not if the crystal were perfectly harmonic), but not the stability of  
the system (i.e. the sign of any of the squared frequencies).

>     2, I knew from most refs that ecutwfs  30 Ry for ultrasoft-PP  
> is enough in pw calculations , Is it right ?

This statement is meaningless. 30 Ry may be too large or too small a  
cutoff, depending on the PP that you are using. It is true that many  
US PP calculations are done with cutoffs in this range, but whether  
or not this value is appropriate in your case ONLY YOU CAN TELL. i.e.  
you have to perform careful convergence tests in order to be sure  
that your PW basis set is adequate to describe the properties that  
you want with the accuracy that you need (the same holds, of course,  
for many other devil's details of the calculation: k-points, gaussian  
smearing, etc.)

> But i donot have much idear about ecutwfs in ph calculations .

By and large, (squared) phonon frequencies are second derivatives of  
the crystal energy for a given kinetic-energy cutoff. If a cutoff is  
OK for geometry optimization, it  is also so for phonon calculation  
(see below)

> Would you please give me some refs of more detail settings about PH  
> calculations ?

S. Baroni, A. Dal Corso, S. de Gironcoli, and P. Giannozzi, Rev. Mod.  
Phys. 73, 515 (2001).

Enjoy!

Stefano B.

PS: for us, it is considered polite to sign with our first name, or  
with the first name followed by the family name. It would be unpolite  
(except for formal business correspondence) to prepend the title to  
the signature - "Stefano" or "Stefano B.": confidential, OK; "Stefano  
Baroni" a bit less confidential, also OK in most cases; "Prof.  
Stefano Baroni": very formal and kind of assuming and not-so-polite  
in many cases - OK in a few

>
>
> I am trying to find more problems related to this wrong result  
> following your suggestion.
> Thank you again, it really helps me much.
>
> p.s.  It is nice for me listening  your lectures in Beijing , i  
> really learn much from your 3 lectures .
>
>
> Regards,
>
> Hai-Ping
>
>
> On 8/20/06, Nicola Marzari <marzari at mit.edu> wrote:
>
>
>
> Dear Haiping,
>
>
> there are two possibilities that lead to negative phonon  
> frequencies at
> gamma.
>
> The first one is less likely:
>
> 1) you have not carefully relaxed the *internal* degrees of freedom -
> i.e. the ionic coodinates. Then, the phonon code discovers that moving
> the atoms along a certain pattern (an "eigenvector" of the dynamical
> matrix at gamma) the energy goes down. Perfect - just relax the ions
> more. Relaxation of the unit cell has nothing to do with this
>
> The second one is more likely:
>
> 2) the calculation is not converged with respect
> to the plane wave cutoff needed, and the number of k-points needed,
> for the small or zero frequencies at gamma. Obtaining an accurate,
> faithful description of low energy modes requires cutoffs and  
> especially
> duals much higher than anything you are used to (to give you an  
> example,
> for an ultrasoft C for diamond, you need 100 Ry in the wavefunction  
> and
> 2800 Ry in the charge density to get a convergence, at gamma,
> of the zero modes down to the last cm-1). Does it mean you need
> to use 100/2800 Ry for diamond ? No ! By enforcing the translation sum
> rules that the interatomic force constants must satisfy, you can  
> monitor
> (after enforcing said rules) the convergence of the low frequency  
> modes,
> and discover you can use much more reasonable cutoffs.
>
> Still, even more than a regular calculations, phonons require a
> through control on the parameters that define your calculation.
>
> Note that, as a warm-up, it is easier to converge LDA calculations
> than GGA calculations (the exchange-correlation terms, calculated in
> real space, introduce very high frequencies in reciprocal space).
>
>                        nicola
>
>
> >> Dear All,
> >> I also have calculated gamma phonon frequencies of anatase TiO2 .
> >> I run ph.x caculation after relaxing the structure, but  9 of 18  
> frequencies
> >> are minus.
> >>
> >> Would you please give me some hints and help?
> >>
> >> input settings are below, thank you in advance !
> >>
> >> *phonons of TiO2 at Gamma
> >>  &inputph
> >>   tr2_ph=1.0d-14,
> >>   recover=.true.,
> >>   epsil=.true.,
> >>   prefix='tio2',
> >>   fildyn='TiO2-gamma.dyn ',
> >>   outdir='/home/haiping/tmp/'
> >>  /
> >> 0.0 0.0 0.0*
>
>
> ---------------------------------------------------------------------
> Prof Nicola Marzari   Department of Materials Science and Engineering
> 13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
> tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>

---
Stefano Baroni - SISSA  &  DEMOCRITOS National Simulation Center -  
Trieste
[+39] 040 3787 406 (tel) -528 (fax) / stefanobaroni (skype)

Please, if possible, don't  send me MS Word or PowerPoint attachments
Why? See:  http://www.gnu.org/philosophy/no-word-attachments.html



-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/attachments/20060820/054cdf5f/attachment.htm