[DFTB-Plus-User] Problems with supercell phonons calculation

Alessandro LANDI alelandi1 at unisa.it
Fri May 22 16:20:08 CEST 2020


Dear Bálint,
1)
*>When you optimize the molecule in the original supercell, do you also*
* allow the cell to relax?*

I think that you are asking whether I keep the unit cell lattice parameters
fixed or not. In this case, yes, I do keep them fixed (LatticeOpt = No).
However, in the first round I only optimize the unit cell (which contains 4
molecules), not a supercell

2)




*>If the simple unit cell has been relaxed properly, there should be no
relaxations in the repeated cell at all, especially if the k-point
samplings are identical (in the sense, that considering the k-point folding
the enlarged supercell generates the same k-points as you have used in the
original cell.)*

I agree and indeed the optimization of the 2x2x1 supercell (constructed by
replicating the optimized unit cell) is done in 1-2 steps, i.e. the
starting supercell is already almost optimized, as expected . The problem
is that this structure leads to negative frequencies. Thus, I distort the
structure along the lowest negative frequency and reoptimize, getting a new
structure (see previous mail) which indeed has positive frequencies, but
showing some "gaps" (see picture attached to previous mail)

3)


*> I do not understand the reasoning about the differences when using
ShowFoldedCoords with Yes or No*

In short, the 2x2x1 supercell *constructed by replicating the optimized
unit cell obtained specifying "ShowFoldedCoords=Yes"* is already almost
optimized (MaximalForceComponent=10^-7 in first step if I optimize the
*supercell*), which is what I expected (see previous point).
On the other hand  the 2x2x1 supercell *constructed by replicating the
optimized unit cell specifying "ShowFoldedCoords=No"* is very far from
optimized geometry (MaximalForceComponent=10^-2 in first step if I optimize
the *supercell*)

Best regards,
Alessandro

Il giorno ven 22 mag 2020 alle ore 16:04 Bálint Aradi <aradi at uni-bremen.de>
ha scritto:

> Dear Alessandro,
>
> A few thoughts:
>
> - When you optimize the molecule in the original supercell, do you also
> allow the cell to relax? Otherwise you may still have stress in the
> system, which has not be relaxed out. However, since you have only 1
> molecule per supercell, there is no way to reduce the stress with fixed
> lattice vectors. (If one molecule moves within the cell, all its
> periodic images will move as well, keeping the distance between the
> constant...)
>
> - If the simple unit cell has been relaxed properly, there should be no
> relaxations in the repeated cell at all, especially if the k-point
> samplings are identical (in the sense, that considering the k-point
> folding the enlarged supercell generates the same k-points as you have
> used in the original cell.)
>
> - I do not understand the reasoning about the differences when using
> ShowFoldedCoords with Yes or No. If you have periodicity, folding the
> coordinates should have no effect on any results. If you repeat the
> supercell endlessly (which the periodic boundary conditions do), you
> would get exactly the same system, whether you used the folded or
> unfolded-coordinates.
>
> Best regards,
>
> Bálint
>
>
>
> On 22.05.20 09:55, Alessandro LANDI wrote:
> > Dear DFTB+ users, I have some problems concerning the evaluation of the
> > phonons of a supercell.
> > In particular I am working on a 2x2x1 supercell of Rubrene (1120 atoms).
> >
> > 1)  In order to get the phonons I first perform an optimization of the
> > unit cell with a 2x2x1 k-point sampling (see the bottom of this email
> > for the input code); then, starting from the optimized structure, I
> > construct a 2x2x1 supercell and perform an optimization of the whole
> > supercell with a 1x1x1 k-point sampling.
> > In other words, the input file is the same I used for the unit cell
> > optimization but for the k-point section.
> >
> >  As expected, the starting structure is very near to the minimum of
> > energy and optimization (Maximal force component 10^-8) ends in less
> > than 5 cycles. Let us call this optimized supercell  "supercell A"
> > (picture not atttached to avoid exceeding max dimensions of the
> > message). Computation of the Hessian and the modes for "supercell A"
> > leads to a quite high number of negative frequencies (up to 40),
> > suggesting that the structure is not in the real minimum of energy.
> > Thus, I slightly distort the structure along the lowest-frequency normal
> > mode and restart the optimization, obtaining a structure that we can
> > refer to as "supercell B". Problem is, this time I get a completely
> > different structure from "supercell A": the molecules are not equally
> > spaced; on the contrary, they come closer to each other leaving a quite
> > big gap (see the image attached). Nevertheless, this time I get positive
> > frequencies, thus indicating that "supercell B" should be the real
> minimum.
> >
> > Is it normal? Since (I hope) I am working in Periodic Boundary
> > Conditions, I expect that replicating the optimized supercell in order
> > to construct a (virtually endless) crystalline structure I do not have
> > gaps between the repeating unit (which is the 2x2x1 supercell). Indeed,
> > I would have no gaps if I replicate the "supercell A" (which however
> > leads to negative frequencies), while I do have gaps in the "supercell
> > B" which however should be the correct one, having positive frequencies.
> > What am I doing wrong?
> >
> > 2) You can notice that I have "fragmented" molecules in the pictures
> > attached. This happens because I have used "ShowFoldedCoords=Yes" in the
> > input file and the starting geometry has some negative coordinates. Of
> > course, by setting"ShowFoldedCoords=No", I get molecules which are not
> > fragmented. However, in this case the 2x2x1 supercell obtained by
> > replicating the optimized unit cell (again with a 2x2x1 k-point
> > sampling) is quite far from the energy minimum (Maximal force component
> > about 10^-3 at cycle 1). This is very different from what I obtained
> > when using "ShowFoldedCoords=Yes" (see my description at point 1)
> > above), where the 2x2x1 supercell was very near to the minimum of energy
> > right from the start.
> >  Is this  difference between two computations, differing only for
> > "ShowFoldedCoords", normal? I remark that the 2x2x1 supercell is simply
> > obtained by replicating the unit cell properly shifted by a displacement
> > equal to the lattice constants expressed in cartesian coordinates.
> > Taking advantage of the fact that the unit cell is orthorhombic in
> > pseudo-code it can be written as:
> > for i in range(2):
> >    for j in range(2):
> >     for k in range(1):
> >         x=x+i*a_axis
> >         y=y+j*b_axis
> >         z=z+k*c_axis
> >
> >
> > Thank you for any help you can give me
> > Best regards,
> > Alessandro
> >
> > INPUT FILE:
> >
> > Geometry = GenFormat {
> > <<< "geom_start.gen"
> > }
> > Driver = ConjugateGradient {
> >   MovedAtoms = 1:-1
> >   MaxForceComponent = 1E-8
> >   MaxSteps = 10000
> >   LatticeOpt = No
> >   OutputPrefix = "geom_opt"
> > }
> > Hamiltonian = DFTB {
> >   SCC = Yes
> >   SCCTolerance = 1e-10
> >   MaxSCCIterations = 20000
> >   Mixer = Broyden {
> >     MixingParameter = 0.99
> > }
> >   SlaterKosterFiles = Type2FileNames{
> >     Prefix = "Slako_3ob/"
> >     Separator = "-"
> >     Suffix = ".skf"
> > }
> >   MaxAngularMomentum = {
> >     H = "s"
> >     C = "p"
> > }
> >   Filling = Fermi {
> >      Temperature [Kelvin] = 300.0
> >   }
> >   KpointsAndWeights = SuperCellFolding {
> >     2 0 0
> >     0 2 0
> >     0 0 1
> >     0.0 0.0 0.0
> > }
> >   Eigensolver = DivideAndConquer {}
> >   Differentiation = Richardson {}
> >   ThirdOrderFull = Yes
> >   DampXH = Yes
> >   DampXHExponent = 4.0
> >   HubbardDerivs {
> >    H = -0.1857
> >    C = -0.1492
> > }
> >   Dispersion = DftD3{}
> > }
> > Options {
> >   ShowFoldedCoords       =       Yes
> > }
> > Parallel {
> >   UseOmpThreads = Yes
> > }
> > ParserOptions = {
> >   ParserVersion = 5
> > }
> >
> >
> > --
> > Alessandro Landi, Assegnista di Ricerca
> > Dipartimento di Chimica e Biologia "Adolfo Zambelli"
> > Università degli Studi di Salerno
> > Via Giovanni Paolo II, 132 - 84084 - Fisciano (SA)
> > Phone number 089969390
> >
> > _______________________________________________
> > DFTB-Plus-User mailing list
> > DFTB-Plus-User at mailman.zfn.uni-bremen.de
> >
> https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user
> >
>
>
> --
> Dr. Bálint Aradi
> Bremen Center for Computational Materials Science, University of Bremen
> http://www.bccms.uni-bremen.de/cms/people/b-aradi/
>
>
> _______________________________________________
> DFTB-Plus-User mailing list
> DFTB-Plus-User at mailman.zfn.uni-bremen.de
> https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user
>


-- 
Alessandro Landi, Assegnista di Ricerca
Dipartimento di Chimica e Biologia "Adolfo Zambelli"
Università degli Studi di Salerno
Via Giovanni Paolo II, 132 - 84084 - Fisciano (SA)
Phone number 089969390
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20200522/0a9a5478/attachment-0001.htm>


More information about the DFTB-Plus-User mailing list