About non periodic systems

[hplan]

Dear Developers:

I noticed that there are several tags concerning with non peirodic systems, such as "NonPDirs", and those for coulomb interaction cutoff. I wonder about these settings and their combinations. My quesions are following:

1, should we trigger on 'NonPDirs' whenever coulomb interaction cutoff applied ?
2, Is there any restriction on cellshape when coulomb interaction cutoff applied ? For instance, whether Coulomb interaction cutoff can be applied for a non-orthorhombic supercell ?
3, Should we apply Coulomb interaction cutoff for nonperiodic systems no matter what calculation to perform , such as RPA, tddft, GW, and BSE calculations ?

Thanks for your time and patience .

Best withes

Hai-Ping

[Daniele Varsano]

Dear Hai-Ping,

1, should we trigger on 'NonPDirs' whenever coulomb interaction cutoff applied ?

The variable "NonPDirs" is intended to activate other variables that you need when treating non-periodic systems, and perform some check about the
"real" non-periopdicity of your system, you can switch it, but
I don't think it is mandatory if you will use cutoff potentials in your calculations because the same checks wil be performed.

2, Is there any restriction on cellshape when coulomb interaction cutoff applied ? For instance, whether Coulomb interaction cutoff can be applied for a non-orthorhombic supercell ?

In line of principle there is not, in practice there is. The code will check the symmetry of your cell, if compatible with cutoff shape of your
coulomb interaction. If you don't have any particular reason, I suggest you to use orthorhombic supercell.

3, Should we apply Coulomb interaction cutoff for nonperiodic systems no matter what calculation to perform , such as RPA, tddft, GW, and BSE calculations ?

Well, while for RPA (and TDLDA) calculations volume effects are not such relevant, they are very important for GW and BSE calculations.
You can see it looking this paper Phys. Rev. B 73, 205119 (2006).
In Fig.5 you can see that also RPA polarizability has a strong dependence on volume, but you can easily obtain the asymptotic results for smaller
volume without using any cutoff coulomb potential just looking the epsilon output (scaled by a volume factor). The Im alfa --> Im eps for infinte
volume (it can be seen analitically i.e. in this paper Sottile et al. INT. J. OF QUANTUM CHEMISTRY, 102, 684, 2005.

Best,

Daniele

[hplan]

Dear Daniele:
Thanks!

The variable "NonPDirs" is intended to activate other variables that you need when treating non-periodic systems, and perform some check about the
"real" non-periopdicity of your system, you can switch it, but
I don't think it is mandatory if you will use cutoff potentials in your calculations because the same checks wil be performed.

I just found yambo would perform polarizability calculation when i trigger on 'NonPDirs' for some direction.

In line of principle there is not, in practice there is. The code will check the symmetry of your cell, if compatible with cutoff shape of your
coulomb interaction. If you don't have any particular reason, I suggest you to use orthorhombic supercell.

But how can i define the cutoff shape ? I found the document for coulomb interaction cutoff is quite incomplete.
And it seems that the format of CUTBox can only be employed for orthorhombic situation since there are only 3 parameters (I just thought the 3 parameters are used to define cutoff of x,y,z directions respectively) . Furthermore, how can we define the coulomb cutoff for a slab calculation ? I just thought the setting like this

Code: Select all

% CUTBox
 0.000    | 0.000    | 16.000    |      # [CUT] [au] Box sides
%

probably define a coulomb cutoff for slab along z direction. But it seems not true since we only give one border side along z direction. I am quite confused about this issue.



Best wishes

[Daniele Varsano]

Dear Hai Ping,

First of all let me say you that the box-shape for this kind of calculations,
less than 3 non-periodic dimension has been not extensively tested.

And it seems that the format of CUTBox can only be employed for orthorhombic situation since there are only 3 parameters (I just thought the 3 parameters are used to define cutoff of x,y,z directions respectively).

Yes, I'm afraid you should use in orthorombic situations, but may someone else can give you more details about it.

But how can i define the cutoff shape ?

The right way to do it is :
CUTGeo= "box xyz"
for a box finite in 3D
CUTGeo= "box xy" or "box yz" etc.
for a box that is infinite in the "z" direction or "x" direction etc.
CUTGeo= "box x" etc.
for a box that is infinite in the "yz" direction and cut in the x direction etc.

than you gives the corresponding values for the cut of your potential in the direction that you want you cut.
Let me add, that in order to construct potential that are not periodic with such shapes, you should also
perform Random Integral Integrations. IN order to do that, in your input you have to include these variables:
RandQpts=1000000 # [RIM] Number of random q-points in the BZ
RandGvec=1 RL # [RIM] Coulomb interaction RS components

Usually, just one Gvec (so only Bz) is enough.

As this part of the coulomb interaction with these kind of shape has not been well-tested, I suggest you to
use with precaution. A way to look at it is to activate this variable:
#CUTCol_test # [CUT] Perform a cutoff test in R-space
Just uncomment it. In this way an inverse Fourier Transform of your coulomb components is performed and you have outputs of the potential in real-space projected onto planes, and you can look if they have the desired shape.
Of course the quality of the real-space representation strongly depends on how much G-vectors you include in
the calculation.


I hope this can help,

Finally, let me ask you to include your affiliation in your posts, you can do ti easily just filling in it
the signature of our profile. This is a rule of the forum.

Best,

Daniele

[hplan]

Dear Daniele:

Thank you for detailed explaination.

First of all let me say you that the box-shape for this kind of calculations,
less than 3 non-periodic dimension has been not extensively tested.

Yes, what i want to deal with is box-shape.

Yes, I'm afraid you should use in orthorombic situations, but may someone else can give you more details about it.

But for other situations, is there any other method to do except increasing supercell ?

The right way to do it is :
CUTGeo= "box xyz"
for a box finite in 3D
CUTGeo= "box xy" or "box yz" etc.
for a box that is infinite in the "z" direction or "x" direction etc.
CUTGeo= "box x" etc.
for a box that is infinite in the "yz" direction and cut in the x direction etc.

than you gives the corresponding values for the cut of your potential in the direction that you want you cut.
Let me add, that in order to construct potential that are not periodic with such shapes, you should also
perform Random Integral Integrations. IN order to do that, in your input you have to include these variables:
RandQpts=1000000 # [RIM] Number of random q-points in the BZ
RandGvec=1 RL # [RIM] Coulomb interaction RS components

Usually, just one Gvec (so only Bz) is enough.

I think it is wise to add these contents to Document or userguide .

Finally, let me ask you to include your affiliation in your posts, you can do ti easily just filling in it
the signature of our profile. This is a rule of the forum.

Yes, i know it, and have already done !

Thanks again for your help and patience.

Regards,
Hai-Ping