The ABINIT Users Mailing List ( CLOSED )

["Anglade Pierre-Matthieu" <anglade@gmail.com>]

Hi,

first question first:
I think that there is nothing you can do about the first message. I
don't know if you know about minimization algorithms. Basically lots
of them, at each step, try to optimize your functions along a line.
For this a "best guess" is to do a parabolic approximation of your
function along the line. To do this they usualy find three points, two
high and a third in the middle. By those points there is only a single
second degree polynomial which goes throught. And obviously at first
glance your minimum should be the one of the polynomial...  For your
calculation it looks like something is just fooling the algorithm in
abinit. I think that there is few things you can do for that unless
you intend to implement a new minimization scheme. Anyway it's not a
big problem since for other steps, in next direcctions the problem
don't happens.

Second:
- To accelerate your calculation you might wan't to use better scf algorithms.
iscf 5 might be more suitable. If you use abinit 5 you may try "iscf
7" which is sometimes better.
- The preconditioner you use is not suitable for non bulk materials
because it is a fourier space only preconditioner. This means that it
don't take care og inhomogeneous materials. You might prefer using a
RPA preconditioner. See the variable "iprcel" in abinit doc. It is
usualy very suitable and quite efficient.
- If your nanotube is metallic you may be interested in using "tsmear"
and "occopt". This may help and at least improves absolute convergency
if not convergency rate.
- May be you've already carried out a full study about pseudo
potentials. If you haven't, do it. You might be able to find or create
(see opium http://opium.sourceforge.net/ for instance) some better
pseudopotentials with lesser ecut.
- On P4 it is quite interesting to compile abinit yourself and
optimize it for those proc. They can be quite efficient if you take
care of them and perfectly hopelessly slow if you don't.

I hope this helps

Best regards

PMA


On 6/18/06, yshtogun@cas.usf.edu <yshtogun@cas.usf.edu> wrote:
>  Dear ABINIT users!
>   I have two questions, one about optimization of the unit cell and the 
> second about speed up optimization for super cell.
>   First, I running optimization of my unit cell, there is my input file:
>
> # bond legth a = 1.428 angstrom
> # the potential files - 6c.pspnc LDA (Troullier - Martins)
>
> #Optimization of the lattice parameters
> optcell 0
> ionmov  3
> ntime  100
> #dilatmx 1.05 (using default value)
> ecutsm  0.5
>
> #Definition of the unit cell
> acell 33.0665018318 33.0665018318 4.2840000000 angstrom   # The size of unit 
> cell in angstrom
>
> #Definition of the atom types
> ntypat 1          # There is only one type of atom
> znucl 6           # The keyword "znucl" refers to the atomic number of the
>                   # possible type(s) of atom. The pseudopotential(s)
>                   # mentioned in the "files" file must correspond
>                   # to the type(s) of atom. Here, the only type is Silicon.
> #Definition of the atoms
> natom 28          # There are 28 atoms
> typat 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
>                   # They all are of type 1, that is, Carbon.
>
> xangst            # This keyword indicate that the location of the atoms
>                   # will follow, one triplet of number for each atom
>                   # Triplet giving the cartesian coordinates of all atoms, in 
> Angstrom
> 19.2887927159  16.5332509159  0.0000000000
> 19.0159083159  17.7288357159  0.7140000000
> ……………………………………………….
> 18.2513031159  14.3788816159  2.8560000000
> 19.0159083159  15.3376661159  2.1420000000
> 19.2887927159  16.5332509159  2.8560000000
>
> #Definition of the planewave basis set
> ecut  32.0         # Maximal kinetic energy cut-off, in Hartree
>
> #Exchange-correlation functional
> ixc 1             # LDA Teter Pade parametrization
> iscf 2
>
> #Definition of the k-point grid
> kptopt 1          # Option for the automatic generation of k points, taking
>                   # into account the symmetry
> ngkpt 1 1 4       # This is a 1x1x4 grid based on the primitive vectors
> nshiftk 1
> shiftk  0 0 0.5
>
>
> #Definition of the SCF procedure
> nstep 100         # Maximal number of SCF cycles
> toldfe 1.0d-6     # Will stop when, twice in a row, the difference
>                     # between two consecutive evaluations of total energy
>                     # differ by less than toldfe (in Hartree)
> diemac 12.0       # Although this is not mandatory, it is worth to
>    ***********************************************************
>
>     after several steps I got next warning, but my calculation still running:
>
> BROYDEN STEP NUMBER   9  -------------------------------------------------
>
>  findmin : WARNING -
>   The 2nd degree equation has no positive root (choice=4).
>
>  findmin : COMMENT -
>   There is a problem, since the new total energy is larger
>   than the old (choice=4).
>   I take a point between the old and new, close to the old .
>    line minimization, algorithm   4
>                         lambda      etotal            dedv        d2edv2
>    old point         :  0.0000E+00 -1.6877357422E+02 -5.9156E-08  4.9407-324
>    new point         :  1.0000E+00 -1.6877357389E+02  1.3913E-07  8.6856-317
>    predicted point   :  2.5000E-01 -1.6877357421E+02  0.0000E+00  0.0000E+00
>
>      What does it mean, maybe I need specified some additional parameter?
>
>     The second question: How can I speed up my calculation for optimization 
> of super cell? I want to optimize three unit cells of (7,0) Carbon nanotubes 
> – 84 atoms per 3 unit cells. There is my input files without coordinate of 
> atoms:
>
> #Optimization of the lattice parameters
> optcell 0
> ionmov  3
> ntime  200
> ecutsm  0.5
>
> #Definition of the unit cell
> acell 33.0665018318 33.0665018318 12.8520000000 angstrom    # The size of 
> unit cell in angstrom
>
> #Definition of the atom types
> ntypat 1          # There is only one type of atom
> znucl 6           # The keyword "znucl" refers to the atomic number of the
>
> #Definition of the atoms
> natom 84          # There are 84 atoms
> typat 84*1
>                   # They all are of type 1, that is, Carbon.
>
> xangst            # This keyword indicate that the location of the atoms
>                   # will follow, one triplet of number for each atom
>                   # Triplet giving the cartesian coordinates of all atoms, in 
> Angstrom
> 19.2887927159  16.5332509159  0.0000000000
> 19.0159083159  17.7288357159  0.7140000000
> ……………………………………………….
> 17.1464166159  19.2197055159  10.7100000000
> 15.9200852159  19.2197055159  11.4240000000
>
> #Definition of the planewave basis set
> ecut  32.0         # Maximal kinetic energy cut-off, in Hartree
>
> #Exchange-correlation functional
> ixc 1             # LDA Teter Pade parametrization
> iscf 2
>
> #Check whether the cell is primitive, if the symetry finder is used, a 
> non-zero velue of chkprim will make the code stop
>                   #if a non-primitive cell is used. If chkprim = 0, a warning 
> is issued, but the run doesnt stop.
> chkprim 0
>
> #Definition of the k-point grid
> kptopt 1          # Option for the automatic generation of k points, taking
>                   # into account the symmetry
> ngkpt 1 1 4       # This is a 1x1x4 grid based on the primitive vectors
> nshiftk 1
> shiftk  0 0 0.5
>
> #Definition of the SCF procedure
> nstep 200         # Maximal number of SCF cycles
> toldfe 1.0d-6     # Will stop when, twice in a row, the difference
>                   # between two consecutive evaluations of total energy
>                   # differ by less than toldfe (in Hartree)
> diemac 12.0       # Although this is not mandatory, it is worth to
>
>      In this calculation I have just 2 – k points, so I used two processor 
> but it is not enough memory for such calculation, that's why I used 4 or 6 
> processors it is running (also give me message that I waist time because  2 – 
> k points),  I using university computer core with next specification:
> I.R.C.E. is a grid system designed to facilitate many forms of computational 
> research.
> I.R.C.E. is currently made up of two main parts, the High-Throughput Grid 
> which is supported by Condor and over 130 Intel Pentium 4-based computer lab 
> machines, the High-Performance Cluster Grid which currently contains 36 
> compute nodes and 5 file server nodes in the following configurations:
> •       12 compute nodes, each with 8.0GB of DDR-ECC RAM, two AMD Opteron 248 
> Processors at 2.2 Ghz, and inter-connected with a high-bandwidth, low-latency 
> Myrinet switching fabric.
>
>    Does somebody have suggestion how to speed up such calculation?
>
>    I appreciate all comments and suggestion!
>
>    Thank you in advance!
>
>    Yaroslav!


--
Pierre-Matthieu Anglade