2D kpoint grid being interpreted as 3D

[Daniele Varsano]

Dear Alan,
to be precise, your problem is in the G vectors grid and not in k point grid.
Most probably it is due to the quite distorted cell of your system.
I suggest you to raise a bit the kinetic energy cutoff your ground-state calculation,
50Ry seems to be a bit small for ONCV pseudo and in this way you have a larger grid.
Also, rise a bit the number of Gvectors in the BSENGBlk in the bse input (in conjunction with NGsBlkXs) and
see if it fix the problem. If this does not solve the problem raise the threshold in the G_index function.

Best,
Daniele

[arb83@cam.ac.uk]

Dear Daniele

Many thanks, I will give these a try. I am trying a cutoff energy of 75Ry.

When you say to increase the number of Gvectors in BSENGBlk "in conjunction with NGsBlkXs", do you mean I also need to increase the number of Gvectors in NGsBlkXs, or that those in BSENGBlk should be close to that in NGsBlkXs?

Best wishes,

Alan

[arb83@cam.ac.uk]

Dear Daniele

One more thoughts. When calculating the dielectric screening function for BSE calculations, I decreased the value of "FFTGvecs" to ~15% of that used in DFT calculations (as per http://www.yambo-code.org/wiki/index.ph ... rgy_cutoff) - could this in fact be the source of the problem? Or am I barking up the wrong tree?!

All the best,

Alan

[Daniele Varsano]

Dear Alan,
yes, try to keep FFTGvecs to its original value.
Daniele

[arb83@cam.ac.uk]

Dear Daniele

I have now tried a number of different things to try and reduce the value of G1-G2 in the third direction. I found that increasing Bblk or FFTGvec had no effect. However, increasing the cutoff energy from 50Ry to 70Ry caused the minimum value of G1-G2 (in the three directions) to change from

2.5342684E-05 9.5129013E-05 0.0045070052

to

2.5342684E-05 9.5129013E-05 0.0032340884.

I have two questions:

- it looks like I will need to increase the cutoff energy beyond computationally reasonable bounds to get G1-G2 to be on the order of 1E-5 on the third axis. Should I just increase the tolerance by about 400 times? And if so will this have any unwanted effects on the first or second directions?

Any thoughts would be much appreciated!

With best wishes,

Alan

[Daniele Varsano]

Dear Alan,
70/90 Ry it is a more than reasonable value for ONCV pseudopotential.

And if so will this have any unwanted effects on the first or second directions?

G vectors are compared vector by vector and not direction by direction.
I would increase the kinetic cutoff and your tolerance and verify that the minimum distance between two vectors along Gz is larger than your threshold.

Best,
Daniele

[arb83@cam.ac.uk]

Dear Daniele

Sorry, I'm confused, do you mean smaller than the threshold?

Is increasing the tolerance by a factor of 400 an OK approach to take?

Best wishes,

Alan

[Daniele Varsano]

Dear Alan,
the difference between two Gvectors, in general, should be larger than the threshold otherwise you will have a problem in finding a Gvector such that it is equal to a G-G'. What should be smaller is the condition you are looking at GmG-(G-G') < tolerance that is the condition to assign it.

Is increasing the tolerance by a factor of 400 an OK approach to take?

You can try it, I see it quite dangerous.

Daniele

[arb83@cam.ac.uk]

Dear Daniele

Do you have any other suggestions besides increasing the tolerance? I agree it is dangerous but I can't see anything else which might solve this?

Best wishes,

Alan

[Daniele Varsano]

Dear Alan,
what you can try is to compile the code in double precision adding the flag:

Code: Select all

 --enable-dp

in the configure and see if ti solves the problem.

Best,
Daniele