What is a good estimate for "Much less G vectors"?

[obm]

Hi all.
I am a newbie of YAMBO, and trying to get a sense of things.

I wanted to do a convergence study for a largish molecule, by comparing ALDA spectra.

I have 8670457 RL g-vectors from pwscf, I used 0670457 RL for maxgvecs , which I consider to be a huge cut of corners, but still, the code is running for ever to calculate shells.

I wanted to ask experienced users how much is the maximum limit they have used for production calculations.

Best,
Baris

[Daniele Varsano]

Dear Baris,
you are using a huge number of Gvectors,
first of all check the number of Gvectors that pwscf gives you for the wavefunction, and not the density.
Next this number can be reduced. How can be reduced, is something system dependent, that you can verify
"at posteriori", looking at the orthonormalization of the wavefunction. About the calculation of the shells,
it can be quite slow, it is recommendable to run it in parallel.

Cheers,

Daniele

[claudio]

Dear Baris

just a small addition, to check the number of G vectors in the wave-function
just type

yambo -D

and look at:
[RD./SAVE//ns.db1]------------------------------------------
....
WF G-vectors : 17253
....

this it the number you have to reduce

Cla

[obm]

Hi!

Ok, thanks for the clarification.
However, the number of maxGvecs used above is still %21 less than Gvecs used for KS states (please see below).
Since it takes a lot of time to do convergence tests for a very diverse set of values, could you please tell me if this a normal value or is it a big compromise if you have previous experience in hydrocarbons? Am I truncating virtual states or am I truncating components from each KS state?

C.
Baris



Here is the PWSCF output for G-vector grids

Parallelization info
--------------------
sticks: dense smooth PW G-vecs: dense smooth PW
Min 5175 5175 1293 541903 541903 67738
Max 5177 5177 1295 541904 541904 67740
Sum 82821 82821 20707 8670457 8670457 1083821

and here is the Yambo -D


[RD./SAVE//s.db1]-------------------------------------------
Bands : 189
K-points : 1
G-vectors [RL space]: 8670457
Components [wavefunctions]: 1083821
Symmetries [spatial+T-rev]: 2
Spinor components : 1
Spin polarizations : 1
Temperature [ev]: 0.000000
Electrons : 178.0000
WF G-vectors : 1083821
Max atoms/species : 40
No. of atom species : 2
- S/N 005415 ---------------------------- v.03.02.05 r.*** -
[RD./SAVE//s.wf]--------------------------------------------
Bands in each block : 48
Blocks : 4
- S/N 005415 ---------------------------- v.03.02.05 r.*** -

[Daniele Varsano]

Dear Baris,
no experience in hydrocarbons, anyway you have to set the MaxGvecs once at the beginning,
next you can reduce the Gvector you use for the different calculations (self-energies, local field, linear response).
For this is reason is useful to store a lot of G vector at the beginning, and reduce the number later on when
performing the calculations of the observables.

Am I truncating virtual states or am I truncating components from each KS state?

Reducing the Gvectors, you are truncating the components of each KS states, the number of virtual states is
set by the number of unoccupied states you calculate in the nscf run of pwscf.

Cheers,

Daniele

[obm]

Thank you very much!

[claudio]

Dear Baris

try also to reduce the vacuum in your cell, this will reduce automaticaly the number of G-vectors

Cla