Initialization stuck at Shells finder

[bob]

Dear all,

while trying yambo on a small molecule toy system, when I try to initialize yambo (yambo -i) the process is stuck at "Shells finder"

Code: Select all

>yambo

 <---> [M 0.015 Gb] Alloc g_vec k_pt dl_sop wf_nc_k wf_igk (0.015)
 __   __  _______  __   __  _______  _______
|  | |  ||   _   ||  |_|  ||  _    ||       |
|  |_|  ||  |_|  ||       || |_|   ||   _   |
|       ||       ||       ||       ||  | |  |
|_     _||       ||       ||  _   | |  |_|  |
  |   |  |   _   || ||_|| || |_|   ||       |
  |___|  |__| |__||_|   |_||_______||_______|

 <---> [01] Job Setup
 <---> [02] Input variables setup
 <---> [02.01] Unit cells
 <---> [02.02] Symmetries
 <---> [02.03] RL shells
 <---> Shells finder |                    | [000%] --(E) --(X)

What can be the reason for that?

Cheers,
Bjoern

[andrea marini]

How many G vectors do you have ? If the number is huge the shell finder may take quite a while to calculate the shells. The easy solution would be then to create an intialization input file with higher verbosity

Code: Select all

> yambo -i -V rl 

and reduce the number of G-vectors defined by the variable MaxGvecs

[bob]

I guess it's just a matter of the number of Gvectors, if I read pw.x output correctly:

Code: Select all

     G cutoff =10945.1917  ( 624469 G-vectors)     FFT grid: (216,135, 90)
     G cutoff = 7004.9227  ( 319643 G-vectors)  smooth grid: (180,108, 72)

When I use a very small number of Gvectors in yambo, it works fine. Is there any estimate of how long the Shell finder can take?

Bjoern

[andrea marini]

Well. It is not a linear calculation as the number of operations to do increases with the number of shells already found. So I coded the way Yambo estimates the running time only for linear series of steps Note however, that, once you calculate the shells those are written in a database that is read in all following calculations. So you may even consider calculating the shells once for all.

To estimate the number of G-vectors needed check the normalization of the wavefunctions. In general you need much less G-vectors than the one in the charge. In your case maybe 10000-50000 could be enough.