missmatch in npw and mpw

Total energy, geometry optimization, DFT+U, spin....

Moderator: bguster

Locked
ianabinit
Posts: 2
Joined: Mon Jul 23, 2018 8:09 pm

missmatch in npw and mpw

Post by ianabinit » Mon Jul 23, 2018 8:15 pm

Hi everyone,

I am currently running a single k-point calculation using abinit to compare the result with my own code. Since I want to compare the number of bases used in the abinit and my approach, and my calculation is only a single k-point calculation, I take mpw number in the ouput file as the number of basis functions being used in abinit. However, then I found in the ouput file, there are two npw, one is consistent with mpw, the other is npw (full set), which is two times larger than npw.

wfconv: 160 bands initialized randomly with npw=623757, for ikpt= 1
_setup2: Arith. and geom. avg. npw (full set) are 1247513.000 1247513.000

I think this is in the same case as in this page
http://estest.ucdavis.edu/exist/estest/ ... enzene.out

Could anyone kindly help me with the questions
a. why would the latter npw (full set) is 2 times greater than mpw?
b. which one should I refer to as the reference number of basis functions used in abinit?

Thanks a lot in advance for you guys kind help.

Annelinde
Posts: 6
Joined: Tue Jul 03, 2018 2:32 pm

Re: missmatch in npw and mpw

Post by Annelinde » Thu Jul 26, 2018 11:02 am

Hi,

Did you parallelise this calculation? If I multiply the smallest npw with the number of cores I use, I got roughly the value of npw (full set). Did you parallelise this calculation over 2 cores?

In that case I think you'd need to take into account the full set value.

with kind regards,

Annelinde

ianabinit
Posts: 2
Joined: Mon Jul 23, 2018 8:09 pm

Re: missmatch in npw and mpw

Post by ianabinit » Sat Jul 28, 2018 8:26 pm

Thanks for your reply, I did parallelize the code, but I am definitely sure I was parallelizing the code over much more than 2 processors. And I checked the code which computing mpw and npw locating at

ABINIT_DIR/src/56_recipspace/getmpw.F90
ABINIT_DIR/src/52_fft_mpi_noabirule/m_fftcore.F90

The formulae used to compute the npw/mpw seems to be (2*ecut)**(3/2) * ucvol / (6*pi**2), which is identical to the 1247513 result. But In the .out file, mpw and npw are shown to be 623757, which is just half. I'm not sure if I misunderstood anything or there is a bug in calculating npw/mpw when doing gamma point calculation.

I appreciate a lot if anyone could help me with this.

Locked