ABINIT Discussion Forums

I've been trying to calculate the electronic structure of a surface, and my input files work great using norm conserving psuedopotentials. But recently I've wanted to look at some things with spin orbit coupling, so I've switched to PAW calculations.
Because there is a large vacuum, I've used iprcel =45 in the past. But when I try using that with paw calculations I get the following error
--- !ERROR
message: |
The distribfft passed was already allocated for fine grid
src_file: m_distribfft.F90
src_line: 219


Does anyone have any suggestions of how to use the pre-conditioner with paw sets effectively? I've tried other iprcels /= 0, and I've tried pawsushat= 0 and =1, and still get the same problem
Thanks,
James

I wanted to post the simple input below. If I comment out iprcel, everything works in this case. But when I use a larger vacuum and multiple layers, the SCF cycle won't converge. This seems really odd to me, and again this error (error being used loosely, could be MY error) only happens when I use paw instead of NCPPs.
Thanks for any suggestions
James


#Set 1 : ground state self-consistency

##K-Point Generation#######
kptopt 1 # Automatic generation of k points, taking
ngkpt 16 16 1
shiftk 0 0 0.5


##SCF Parameters########
ecut 12
pawecutdg 24
nstep 50
toldfe 1d-8
ixc 11
nband 50
iprcel 45 ## Commenting out the iprcel eliminates the error,


#####Geometry Parameters#####
acell 6.0376 6.0376 26 Bohr

angdeg 90 90 60
natom 3
ntypat 2
typat 1 2 1
znucl 16 42
xred 0.0000000000E+00 0.0000000000E+00 1.8495767284E-01
3.3333333333E-01 3.3333333333E-01 3.0000000000E-01
0 0 4.1504232716E-01

Hi Everyone,
I wanted to make sure that I wasn't just mooching here on the board. I thought that since the output referred to m_distribfft.F90, I thought it could be the way I compiled abinit. So I downloaded the windows binaries, and saw the same thing.
Looking into line 219 of m_distribfft.F90, it is

case('f')
if (distribfft_arg%n2_fine> 0) then
if(n2==distribfft_arg%n2_fine) then
MSG_WARNING("The distribfft passed was already allocated for fine grid");
return
else
MSG_ERROR("The distribfft passed was already allocated for fine grid");
endif
endif

Reading the comments, case('f') refers to whether we're looking at the "c"oarse or "f"ine grid
n2 and n3 are the size of the second and third fft grid

Those are left undefined, and assigned by abinit.
I thought maybe if I did my mixing over the coarse grid,this could at least get a change. Changing the variable pawmixdg doesn't affect any change in this error.
In a normal calculation (ie no defined iprcel), the SCF cycle seems to start normally
"-P-0000 wfconv: 80 bands initialized randomly with npw= 1636, for ikpt= 1
-P-0000 wfconv: 80 bands initialized randomly with npw= 1646, for ikpt= 2
-P-0000 wfconv: 80 bands initialized randomly with npw= 1642, for ikpt= 3
-P-0000 wfconv: 80 bands initialized randomly with npw= 1634, for ikpt= 4
-P-0000 wfconv: 80 bands initialized randomly with npw= 1626, for ikpt= 5
-P-0000 wfconv: 80 bands initialized randomly with npw= 1628, for ikpt= 6
-P-0000 wfconv: 80 bands initialized randomly with npw= 1636, for ikpt= 7
-P-0000 wfconv: 80 bands initialized randomly with npw= 1638, for ikpt= 8
-P-0000 wfconv: 80 bands initialized randomly with npw= 1628, for ikpt= 9
-P-0000 wfconv: 80 bands initialized randomly with npw= 1638, for ikpt= 10
-P-0000 wfconv: 80 bands initialized randomly with npw= 1636, for ikpt= 11
-P-0000 wfconv: 80 bands initialized randomly with npw= 1628, for ikpt= 12
-P-0000 wfconv: 80 bands initialized randomly with npw= 1626, for ikpt= 13

_setup2: Arith. and geom. avg. npw (full set) are 1630.422 1630.405

================================================================================
FFT (fine) grid used in SCF cycle:

getcut: wavevector= 0.0000 0.0000 0.0000 ngfft= 30 30 120
ecut(hartree)= 24.000 => boxcut(ratio)= 2.09285

ewald : nr and ng are 5 and 31"


but when iprcel =45, it goes from band initialization into the m_distribfft error:

"wfconv: 80 bands initialized randomly with npw= 1628, for ikpt= 10
wfconv: 80 bands initialized randomly with npw= 1618, for ikpt= 11
wfconv: 80 bands initialized randomly with npw= 1618, for ikpt= 12
wfconv: 80 bands initialized randomly with npw= 1618, for ikpt= 13

_setup2: Arith. and geom. avg. npw (full set) are 1630.422 1630.405

================================================================================
For input ecut= 4.422023E+00 best grid ngfft= 12 12 50
max ecut= 4.562502E+00

==== FFT mesh ====
FFT mesh divisions ........................ 12 12 50
Augmented FFT divisions ................... 13 13 50
FFT algorithm ............................. 100
FFT cache size ............................ 16
DATA TYPE INFORMATION:
...

m_distribfft.F90:219:ERROR
The distribfft passed was already allocated for fine grid

leave_new : decision taken to exit ...
"

I'd be happy to provide any additional information if anyone has any ideas.
Thanks,
James

I have the exact same problem. I'm using Abinit 7.6.2.

Has anyone got any ideas how to solve this?

Another one here with the same issue.

Any suggestions?

Thx!

Same problem, and setting diemix and diemac also causes the geometric optimizations to crash (so nothing reasonable can be used to help convergence), with surface and vacuum, even using diemac is not all that reasonable.

Was this ever solved? I am having the same problems with 7.4.3 (latest version on this machine).

I am also getting a bug -- albeit a different bug -- using a PAW pseudopotential with pawsushat = 1 and iprcel=149

I have tried this with Abinit versions 8.4.4 and 8.6.1

--- !BUG
src_file: m_mpinfo.F90
src_line: 726
mpi_rank: 0
message: |
Unable to find an allocated distrib for this fft grid
...