For NCPUS>56 get 'internal ERROR RSPHER'

Message

chris_feigl · #1 Post by **chris_feigl** » Thu Apr 22, 2010 3:22 am

Hi,

I'm performing an atom optimisation of crystalline ZnS slab and get the following error only when NCPUs is greater the 56 (the system is sufficiently large to require more than 56 CPUs):

vasp.4.6.31 08Feb07 complex
POSCAR found : 2 types and 160 ions
LDA part: xc-table for Ceperly-Alder, standard interpolation
POSCAR, INCAR and KPOINTS ok, starting setup
WARNING: wrap around errors must be expected
FFT: planning ... 16
reading WAVECAR
internal ERROR RSPHER:running out of buffer 0 0
18 1 0

I've been informed that forcing the projection operators to run in real space causes the crash for NCPU>56, however running them in recip. space with NCPU=120 is highly inefficient, no better than at NCPU=56 in fact.

As can be seen above I'm running on 4.6.31 so shouldn't be experiencing the bug in earlier versions, as described in earlier threads.

Thanks,
Chris

KPOINTS
gamma centered MP
0
Gamma
8 8 1
0 0 0

INCAR
System = WZ ZnS ( 1 0 0) 20 layer slab
LREAL = Auto
IALGO = 48
NELMIN = 4 ! do a minimum of four electronic steps
IBRION = 2 ! use CG algorithm for ions
ISMEAR=-5
ISIF=2
NSW=100
ENCUT=500
ADDGRID= .TRUE.
LWAVE= .FALSE.
LCHARG= .FALSE.
PREC=ACCURATE
NPAR=1

POTCAR
PAW GGA PW91
<span class='smallblacktext'>[ Edited ]</span>

#2 Post by **admin** » Mon Apr 26, 2010 1:41 pm

I am not aware of such a behaviour. Have you tested whether vasp.5.2 behaves the same way?

chris_feigl · #3 Post by **chris_feigl** » Tue Apr 27, 2010 12:46 am

I haven't tested with 5.2, however d_farrell2 has (http://cms.mpi.univie.ac.at/vasp-forum/ ... php?3.7143) and is encountering a similar or same bug.

Other accounts which appear similar or same or here:
http://cms.mpi.univie.ac.at/vasp-forum/ ... php?3.4635
http://cms.mpi.univie.ac.at/vasp-forum/ ... php?3.5423
http://cms.mpi.univie.ac.at/vasp-forum/ ... php?3.5825
http://cms.mpi.univie.ac.at/vasp-forum/ ... php?3.7133

Thanks.

chris_feigl · #4 Post by **chris_feigl** » Tue Apr 27, 2010 1:40 am

The last line of the OUTCAR file when the bug occurs is always
Maximum index for non-local projection operator xxx

I have been running some tests and this is what I have come up with.

Starting with a system of 214 carbon atoms and the following:

INCAR
LREAL= Auto
ISMEAR = -5
IALGO = 48
NSW = 20
ISIF =2
NELMIN = 4 ! do a minimum of four electronic steps
IBRION = 1 ! use CG algorithm for ions
PREC = Accurate
ADDGRID = TRUE
NPAR=1

KPOINTS
Gamma centered Monkhorst Pack
0
Gamma
8 8 8
0 0 0

POTCAR
TITEL = PAW_PBE C 08Apr2002
LULTRA = F use ultrasoft PP ?

The system converges successfully. From OUTCAR:
real space projection operators:
total allocation : 40747.19 KBytes
max/ min on nodes : 1212.38 530.81
number of electron 868.0000000 magnetization
Maximum index for non-local projection operator 404

The system is reduced to 14 atoms and encounters ERROR RSPHER.
From OUTCAR:
real space projection operators:
total allocation : 2629.12 KBytes
max/ min on nodes : 310.31 0.00
number of electron 56.0000000 magnetization
Maximum index for non-local projection operator 148
(^ last line in OUTCAR for all jobs with ERROR RSPHER)

Changed to LREAL=FALSE and did not receive ERROR RSPHER

Changed to LREAL=TRUE, got ERROR RSPHER with:
real space projection operators:
total allocation : 1718.44 KBytes
max/ min on nodes : 202.75 0.00
number of electron 56.0000000
Maximum index for non-local projection operator 82

Kept LREAL=TRUE but removed NPAR=1 from INCAR. Job runs OK for at least 6 iterations.
real space projection operators:
total allocation : 1718.44 KBytes
max/ min on nodes : 1718.44 1718.44
number of electron 56.0000000
Maximum index for non-local projection operator 1980

Also runs OK with NPAR=1 with LREAL=FALSE
<span class='smallblacktext'>[ Edited Tue Apr 27 2010, 04:20PM ]</span>