Calculations apparently is hanging

Message

Rad Balu · #1 Post by **Rad Balu** » Tue Jan 03, 2006 12:10 am

Hi Everybody,

I am running a relaxation (electroinic and ionic) for a system with 56 atoms in the unit cell. The calculation has been running for the last couple of days and there was no output of any king and the cpu usage in all the nodes is around 99%. How do I know what is going on and diagonize the problems if there are any.

Thank you for your suggestions in advance
Rad Balu

Rad Balu · #2 Post by **Rad Balu** » Sat Jan 07, 2006 2:58 pm

Hello All,

Here is a follow up question. One of the reason I think it might be hanging is due to the KPOINTS definitions I used. Is there a way to specify that I want the calculations over only Gamma points even thogh the code id compiled for k-points. If so could somebody tell me how to specify the input?

Thanks everybody
Rad

tjf · #3 Post by **tjf** » Sat Jan 07, 2006 6:16 pm

I doubt that using only the Gamma point would cause a hang.

Have you tried running the job on a single CPU, with a serial binary?

You should get output of some description a few seconds at most after launching VASP (reading POTCAR/POSCAR etc.)

Specifying Gamma-centred automatic k-points and then 1 1 1 with zero shift works fine for just the Gamma point. (But if you're doing a lot of these it's apparently worth compiling a special binary for it.

Rad Balu · #4 Post by **Rad Balu** » Sun Jan 08, 2006 1:17 am

Hi,

Thank you for your reply and that is how ( Gamma and 1 1 1) I had set up the KPOINTS file . Here are my INCAR and POSCAR contents. Please let me know if there are any conflicting parameters and anything that is a glaring mistake.

Thanks again
Rad

INCAR
=================================
System = L-Cysteine

NWRITE = 2
ISTART = 0
ICHARG = 0
PREC = High
ISIF = 3
EDIFF = 1E-6 ! high accuracy for electronic groundstate
NELMIN = 8 ! do a minimum of ten electronic steps
IBRION = 1 ! quasi-Newton (variable metric) algorithm is used to relax the ions into their instantaneous groundstate
ISYM = 2
EDIFFG = -0.01 ! small tolerance for ions
NSW = 20 ! 20 ionic steps should do
NGX=30 ; NGY=30 ; NGZ=30 ; NGXF=48 ; NGYF=48 ; NGZF=48
LREAL = .FALSE. ! real space projections for 56 atoms
! BMIX = 2.5 mixing parameter
LWAVE=.TRUE.
LCHARG=.TRUE.
! MAXMIX = 80 ! keep dielectric function between ionic movements
K = Gamma

Electronic Relaxation
NSW = 0 number of steps for IOM
NBLOCK = 1 ; KBLOCK = 5
!SMASS = 0.5 Nose mass-parameter (am)
POTIM = 5.00 time-step for ion-motion
!TEBEG = 423 temperature

Ionic Relaxation
NSW = 0 number of steps for IOM
NBLOCK = 1 ; KBLOCK = 5
!SMASS = 0.5 Nose mass-parameter (am)
POTIM = 5.00 time-step for ion-motion
========================================
POSCAR
========================================
Monoclinic L-Cysteine
9.4 5.2 11.3
0.105921 -0.000000 0.036472
0.000000 0.191498 -0.000000
0.000000 0.000000 0.093289
12 8 4 4 28
Direct
0.207 0.637 3.677
1.603 0.016 3.516
1.986 -0.141 2.047
5.543 3.248 7.043
4.147 2.627 7.203
3.764 2.470 8.672
0.759 1.159 6.925
-0.571 0.695 7.525
-0.481 0.564 9.047
4.991 3.770 3.795
6.321 3.306 3.194
6.231 3.175 1.672
-0.670 -0.031 4.277
0.037 1.781 3.184
6.420 2.580 6.442
5.713 4.392 7.536
1.621 0.256 6.700
0.924 2.381 6.721
4.129 2.867 4.020
4.826 4.992 3.998
1.658 -1.285 4.234
4.092 1.326 6.485
-0.904 -0.642 6.957
6.654 1.969 3.762
0.724 -0.924 0.997
5.026 1.687 9.722
-0.357 2.151 9.894
6.107 4.762 0.825
-1.114 5.258 1.00
-1.816 3.925 1.00
-1.835 4.020 1.00
0.691 4.010 1.00
-0.748 2.001 1.00
0.869 1.645 1.00
-0.006 1.226 1.00
1.488 5.464 1.00
=========================================

tjf · #5 Post by **tjf** » Sun Jan 08, 2006 3:19 pm

Well, I have to come back to my earlier point. Does that input work OK with a serial binary? This really is the important question.

#6 Post by **admin** » Tue Jan 10, 2006 9:55 am

I fully agree with tjf, please start an interactive run with the serial executable to get at least some information where the calculation hangs up and/or whether the problems are related to parallelisation problems.
concerning your INCAR: please check that the NGiF you specify are NOT smaller than those set automatically, according to PREC.
This certainly does NOT cause the calculation to hangup, but it affects the accuracy of your calculations.

Rad Balu · #7 Post by **Rad Balu** » Wed Jan 11, 2006 8:27 pm

Hi,

Thanks everybody for all the suggestions. I finally managed to compile the code for serial run using pgf90 and tested the installation using bench.Hg input.

Now, when I tried to run the input posted earlier, the following is the output I got:
===============================
vasp.4.5.8 28Aug03
POSCAR found : 0 types and 0 ions
LDA part: xc-table for , standard interpolation
================================
The only other file that was created is listed below:
IBZKPT

Is there anything else I can do about this run.

I also have a related question. My plan is optimize the structure first and then to set up an MD run. During each step of the MD, I want to output dipoles (using Berry phase) so that after accumulating enough data points I can do an FFT to generate the spectrum. I come from the CPMD camp where I used to do this in routine fashion. I would like to know whether it is possible to do the same thing using VASP and if so, can somebody provide me details on it. That is, how to specify the calculation of a property at every step of the MD run.

PS: I am yet to fix the NGiF value as suggested by the admin.

Thanks again and I appreciate any help that is provided in this regard.

Rad

Rad Balu · #8 Post by **Rad Balu** » Wed Jan 11, 2006 9:15 pm

Hello Again,

I fixed one of the issue in the POSCAR that is related to formatting and it read all the ions and not their positions but failed with the following message:
=====================================
POSCAR found : 5 types and 56 ions

-----------------------------------------------------------------------------
| |
| ADVICE TO THIS USER RUNNING 'VASP/VAMP' (HEAR YOUR MASTER'S VOICE ...): |
| |
| You have a (more or less) 'large supercell' and for larger cells |
| it might be more efficient to use real space projection operators |
| So try LREAL=.TRUE. in the INCAR file. |
| Mind: If you want to do an extremely accurate calculations keep the |
| reciprocal projection scheme (i.e. LREAL=.FALSE.) |
| |
-----------------------------------------------------------------------------

LDA part: xc-table for Ceperly-Alder, standard interpolation
POSCAR, INCAR and KPOINTS ok, starting setup
LAPACK: Routine ZPOTRF failed! 1
FORTRAN STOP
=======================================

I compiled the code with the latest version of pgf90 (version 7.x).
Any suggestions on this issue ?

Thanks
Rad

#9 Post by **admin** » Wed Jan 18, 2006 3:30 pm

ZPOTRF computes the Cholesky factorization of a complex Hermitian
positive definite matrix A.
if the error occurs at such an early stage of the calculation, it is most probable due to one of the follwoing reasons
1) there is an error in your LAPACK installation (if the Hg benchmark gave the same results as the provided output files,
this does not seem to be the case)
2) the generated density is not correct (most often this is due to errors in the input geometry)
--> please have a look at your POSCAR file, (and the interatomic distances written in OUTCAR),
a) the unit cell seems very small to me even for a molecule crystal (the dimensions are in the range of 1 Ã…
b) please mind that if you give Direct coordinates, all positions are scaled with the matrix of the Bravais lattice! and transferred back into the unit cell.

Rad Balu · #10 Post by **Rad Balu** » Thu Jan 26, 2006 1:47 pm

Hi,

Thanks for all the suggestions. I compiled the code using several different math libraries, mkl, amd based, the one supplied with the package and in all the cases I am getting the same error. Since it is working for the sample input, the problem may not be because of the math libraries.
I have set up the POSCAR file based on a structue information from the Cambridge Crystal Structure database (CCDC)entry. I read the entry through Mercury software (available from CCDC) and stored the coordinates in pdb format. In that process I got the vectors for the unit cell, the parameters and the coordinates. I built the POSCAR file based on this. I used one of the VASP utility to read the POSCAR file in jmol format and it gives bonding that are inconsistent. Please let me know if there are better ways of preparing POSCAR file for a molecular crystal.

any sugesstion in this regard will be greatly appreciated.

I haven't received any reply for my earlier query regarding the output of dipoles in each step of the molecular dynamics run. Could somebody point me in the right direction for this issue.

Rad

#11 Post by **admin** » Thu Jan 26, 2006 2:15 pm

if you have the coordinates and the unit cell informations (take care of the units, VASP uses Ã… throughout) I would make it as
simple as possible, following the instructions of the VASP manual
http://cms.mpi.univie.ac.at/vasp/vasp/node63.html

1) setup and scale the Bravais matrix with the overall a_0 (scalar or vector) given in the second line of POSCAR
2) give the coordinates of the atoms either in
Carthesian units (just scaled by a_0) or
Direct units (scaled with the Bravais matrix)
please mind that atoms lying outside the unit cell will be transformed back into the unit cell due to periodicity.

#12 Post by **admin** » Thu Jan 26, 2006 2:31 pm

concerning the change of dipol moments during an MD:
if you read the manual carefully, you will find out that it takes 8 separate steps to calculate the total change of the dipole moment of a crystal for each single displacement mode. Hence you would have to break the MD after each step to do this, if I understand your question correctly.