IMAGES RELAXATION
Posted: Tue Apr 22, 2008 4:06 pm
Dear All,
I am presently runing ab-initio calculations with VASP to study how organic molecules adsorb on metalic surface ( ex. Cu surface). The unit cell which I am using for some of my systems is a 2x2x1 unit cell (which is the experimental found unit cell ). This unit cell contains 50 atoms, from which 32 are Cu atoms and 18 are C+O+H atoms. Such a cell is large enought such that one can run the job on, up to, 256 CPUs without any problems.
Recently, I have got access to some large supercomputers. This is a good thing as I have also larger systems which I am running and which otherwise would need a lot of time to converge.
The problem is, however, that I still have to converged some small systems (as the one described above) and I have difficulties to do this on the large supercomputers. The smallest number of CPUs which I can ask for is 1024 CPUs and I do not have access to smaller machines. 1024 CPUs is however too much for a system of only 50 atoms.
I have tried to solve this problem in several ways:
1) by increasing the E_cut_off, the number of bands and the size of the supercell in the z-direction. Didn't work. I need unrealistic (huge) values to get something working.
2) by going from a 2x2x1 supercell to a 4x4x1 supercell. This means going from 50 atoms to 200 atoms (and of course doubling the supercell in the x and y direction). This works. My question is now: when relaxing the 4x4 unit cell will VASP relax INDEPENDENTLY all the 4 images or will relax only 1 image and then change the other 3 images accordingly
Looking into the OUTCAR I see that VASP notice that the given 4x4 supercell can be decomposed into 4 smaller cells:
########################################################################################
Lattice vectors:
A1 = ( 0.0000000000, 0.0000000000, -10.2853630000)
A2 = ( 0.0000000000, -14.5457000000, 0.0000000000)
A3 = ( -26.8570110000, 0.0000000000, 0.0000000000)
Subroutine PRICEL returns following result:
LATTYP: Found a simple orthorhombic cell.
ALAT = 5.1426810000
B/A-ratio = 1.4142137146
C/A-ratio = 5.2223754497
Lattice vectors:
A1 = ( 0.0000000000, 0.0000000000, -5.1426810000)
A2 = ( 0.0000000000, -7.2728500000, 0.0000000000)
A3 = ( -26.8570110000, 0.0000000000, 0.0000000000)
4 primitive cells build up your supercell.
Analysis of symmetry for initial positions (statically):
Routine SETGRP: Setting up the symmetry group for a
simple orthorhombic supercell.
Subroutine GETGRP returns: Found 1 space group operations
(whereof 1 operations were pure point group operations)
out of a pool of 8 trial point group operations.
The static configuration has the point symmetry C_1 .
Analysis of symmetry for dynamics (positions and initial velocities):
Subroutine DYNSYM returns: Found 1 space group operations
(whereof 1 operations were pure point group operations)
out of a pool of 1 trial space group operations
(whereof 1 operations were pure point group operations)
and found also 4 'primitive' translations
The dynamic configuration has the point symmetry C_1 .
Analysis of constrained symmetry for selective dynamics:
Subroutine DYNSYM returns: Found 1 space group operations
(whereof 1 operations were pure point group operations)
out of a pool of 1 trial space group operations
(whereof 1 operations were pure point group operations)
and found also 4 'primitive' translations
The constrained configuration has the point symmetry C_1 .
####################################################################################
Is not clear for me, however, how the relaxation is done in this case. Will VASP relaxe only the atoms from a primitive cell and then change the coodinates (of the atoms) from the other 3 primitive cells by the same amount? For me this would be, of course, ideally. (The other alternative, i.e. VASP relaxes independently the atoms from the 4 primitive cells, should be as well O.K. IN THE ABSENCE of any numerical inaccuracies).
Eduard
I am presently runing ab-initio calculations with VASP to study how organic molecules adsorb on metalic surface ( ex. Cu surface). The unit cell which I am using for some of my systems is a 2x2x1 unit cell (which is the experimental found unit cell ). This unit cell contains 50 atoms, from which 32 are Cu atoms and 18 are C+O+H atoms. Such a cell is large enought such that one can run the job on, up to, 256 CPUs without any problems.
Recently, I have got access to some large supercomputers. This is a good thing as I have also larger systems which I am running and which otherwise would need a lot of time to converge.
The problem is, however, that I still have to converged some small systems (as the one described above) and I have difficulties to do this on the large supercomputers. The smallest number of CPUs which I can ask for is 1024 CPUs and I do not have access to smaller machines. 1024 CPUs is however too much for a system of only 50 atoms.
I have tried to solve this problem in several ways:
1) by increasing the E_cut_off, the number of bands and the size of the supercell in the z-direction. Didn't work. I need unrealistic (huge) values to get something working.
2) by going from a 2x2x1 supercell to a 4x4x1 supercell. This means going from 50 atoms to 200 atoms (and of course doubling the supercell in the x and y direction). This works. My question is now: when relaxing the 4x4 unit cell will VASP relax INDEPENDENTLY all the 4 images or will relax only 1 image and then change the other 3 images accordingly
Looking into the OUTCAR I see that VASP notice that the given 4x4 supercell can be decomposed into 4 smaller cells:########################################################################################
Lattice vectors:
A1 = ( 0.0000000000, 0.0000000000, -10.2853630000)
A2 = ( 0.0000000000, -14.5457000000, 0.0000000000)
A3 = ( -26.8570110000, 0.0000000000, 0.0000000000)
Subroutine PRICEL returns following result:
LATTYP: Found a simple orthorhombic cell.
ALAT = 5.1426810000
B/A-ratio = 1.4142137146
C/A-ratio = 5.2223754497
Lattice vectors:
A1 = ( 0.0000000000, 0.0000000000, -5.1426810000)
A2 = ( 0.0000000000, -7.2728500000, 0.0000000000)
A3 = ( -26.8570110000, 0.0000000000, 0.0000000000)
4 primitive cells build up your supercell.
Analysis of symmetry for initial positions (statically):
Routine SETGRP: Setting up the symmetry group for a
simple orthorhombic supercell.
Subroutine GETGRP returns: Found 1 space group operations
(whereof 1 operations were pure point group operations)
out of a pool of 8 trial point group operations.
The static configuration has the point symmetry C_1 .
Analysis of symmetry for dynamics (positions and initial velocities):
Subroutine DYNSYM returns: Found 1 space group operations
(whereof 1 operations were pure point group operations)
out of a pool of 1 trial space group operations
(whereof 1 operations were pure point group operations)
and found also 4 'primitive' translations
The dynamic configuration has the point symmetry C_1 .
Analysis of constrained symmetry for selective dynamics:
Subroutine DYNSYM returns: Found 1 space group operations
(whereof 1 operations were pure point group operations)
out of a pool of 1 trial space group operations
(whereof 1 operations were pure point group operations)
and found also 4 'primitive' translations
The constrained configuration has the point symmetry C_1 .
####################################################################################
Is not clear for me, however, how the relaxation is done in this case. Will VASP relaxe only the atoms from a primitive cell and then change the coodinates (of the atoms) from the other 3 primitive cells by the same amount? For me this would be, of course, ideally. (The other alternative, i.e. VASP relaxes independently the atoms from the 4 primitive cells, should be as well O.K. IN THE ABSENCE of any numerical inaccuracies).
Eduard