Page 1 of 1
lda+paw phonon mode anot consistent from v-7.10.5 to v-8.0.8 [SOLVED]
Posted: Wed Jul 20, 2016 7:36 pm
by bmatt
Hello abinitors,
I am calculating response at Gamma to get an IR spectrum using LDA+PAW in abinit-8.0.8
* Ti_1.0N_1.0 is fine during all calcs, the problem occurs when doing TiN_0.75X_0.25
* the calculated frequencies (with an E-field perturbation rfelfd = 2 or 3) are garbage, where 50% of the optical modes are at ~-2000.0 cm-1 (obviously wrong.
* when using abinit-7.10.5 (just to calculate the modes without an E-field perturbation rfelfd = 0) the frequencies are good. When performing the same calculation with abinit-8.0.8, they are back to being very bad.
* everything is fine when running LDA+nC or PBE+NC for each calculation type
looking to get an infrared spectrum using anaddb, so just getting the modes with v-7.10.5 is not helpful, so I need v-8.0.8 to apply the E-field (rfelfd=2 or 3) using paw
Re: lda+paw phonon mode anot consistent from v-7.10.5 to v-8
Posted: Thu Jul 21, 2016 6:55 pm
by torrent
Dear bmatt,
It's difficult to help you without an input file, showing that 7.10.5 and 8.0.8 give different results.
Re: lda+paw phonon mode anot consistent from v-7.10.5 to v-8
Posted: Fri Jul 22, 2016 12:48 am
by bmatt
Here are the 4 output files that shows my problem...
ir-eos_LDApaw8-TiN-X.out is the pure TiN (a metal) with no problems
v7 = v-7.10.5 and v8 = v-8.0.8
ir-eos_LDApaw8-TiNB.out is the issue where the formula is Ti_4N_3B_1
v7 = v-7.10.5 and v8 = v-8.0.8
* did not apply the rfelfd = 3 for v8 calc just to show my point
Re: lda+paw phonon mode anot consistent from v-7.10.5 to v-8
Posted: Sat Jul 23, 2016 12:15 am
by torrent
Ok, found!
it seems that you have identified a (recently introduced) bug. Thanks!
First of all, I guess you could explain some differences between version 7.10.5 and version 8.0.8 version. The main one is that the default value for tsmear (smearing temperature) has changed: from 0.0.4 Ha to 0.0.1 Ha.
This explains the differences for the datasets 1 and 2 in your case. Note also that, for the v7.10.5, your value of nband is too low (the last band is still occupied). With the new value of tsmear (v8.0.8), this is OK.
Then, thanks to your example files, I have found the origin of the inconsistency beween 7.10.5 and 8.0.8.
This will be corrected in the next version.
In the meanwhile, if you want to apply the correction now, just apply the following patch to the src/66_nonlocal/nonlop.F90 file (one line to change):
Code: Select all
-- src/66_nonlocal/nonlop.F90 2016-07-22 23:57:35.000000000 +0200
***************
*** 545,551 ****
ABI_ALLOCATE(enl_,(dimenl1,1,hamk%nspinor**2))
do ispden=1,hamk%nspinor**2
if (dimenl2==hamk%natom) then
! enl_(:,1,ispden)=enl_ptr(:,iatm,ispden)
else if (dimenl2==hamk%ntypat) then
enl_(:,1,ispden)=enl_ptr(:,itypat,ispden)
else
--- 545,551 ----
ABI_ALLOCATE(enl_,(dimenl1,1,hamk%nspinor**2))
do ispden=1,hamk%nspinor**2
if (dimenl2==hamk%natom) then
! enl_(:,1,ispden)=enl_ptr(:,iatom_only_,ispden)
else if (dimenl2==hamk%ntypat) then
enl_(:,1,ispden)=enl_ptr(:,itypat,ispden)
else
Re: lda+paw phonon mode anot consistent from v-7.10.5 to v-8
Posted: Mon Aug 08, 2016 8:27 am
by gonze
Dear all,
The bug fix posted by Marc is OK for PAW, but introduces another bug for norm-conserving pseudos.
Actually, Marc has introduced the correct bug fix in ABINITv8.0.8b, available (soon) on the
abinit Web site. There are two lines to modify :
if (dimenl2==hamk%natom .and.hamk%usepaw==1) then
enl_(:,1,ispden)=enl_ptr(:,iatom_only_,ispden)
Best wishes,
Xavier