The ABINIT Users Mailing List ( CLOSED )

[Yuanliang Xie <xylnew@gmail.com>]

Dear Nuno Galamba,
   Thank you. I  UNDERSTOOD your trick.  it is  one of  MD basic skills as discussed by many
   text books or other mail lists. 
   After  skip the ABINIT documents, I cannot get  a detailed  introduction on controlling  MD calculation in 
ABINIT (e.g.  how to set the ensemble  ).
  I have just tested a system ( 64B + 64N ,gamma point). It seems not very efficient as I want.
Xie Y.L.

On Tue, Mar 10, 2009 at 12:18 AM, Nuno Galamba <ngalamba@cii.fc.ul.pt> wrote:

As I referred to that trick of changing a little bit the coordinates to break
symmetry I assumed that was the case. That is, that Xie Y.L. does not have an
equilibrated configuration for his system and he is simply looking for a fast
way of looking at the computational cost of the simulation.



Nuno Galamba

On Monday 09 March 2009 15:30:42 matthieu verstraete wrote:
> Ok, as a rule there should be no need to increase msym (it corresponds to
> an upper bound for any "normal" space group). If you are using a supercell,
> there should be a reason, and you will be introducing some breaking of the
> full symmetry (liquid state, defect, etc...) usually to much lower
> symmetry. The only reason I can think of to increase msym forcibly is to
> test the size-consistency of the code; you should have enough with a
> defected supercell calculation and a single-unit-cell calculation to get
> all the quantities you need. This means you don't even have to set chkprim
> to 0...
>
> As abinit enforces symmetries quite strongly, if you keep them and try to
> do an MD the positions will stay fully symmetric (or almost) and you will
> not at all be representing a MC ensemble properly. Very important to break
> the symmetry of the initial positions. The ideal is to have a thermalized
> average set of positions, eg from an empirical potential run, but this is
> not always available.
>
>
> Matthieu
>
> On Mon, Mar 9, 2009 at 4:19 PM, Yuanliang Xie <xylnew@gmail.com> wrote:
> > Dear Nuno Galamba,
> >   Thank you for your advice. It give me new idea about the problem.
> > Xie Y.L.
> >
> > On Mon, Mar 9, 2009 at 10:29 PM, Nuno Galamba
<ngalamba@cii.fc.ul.pt>wrote:
> >> Dear Xie Yuanliang,
> >>
> >> I also had this problem as explained by Boris D., when I considered the
> >> solid.
> >> My MD simulations of NaCl and NaI were for the liquid state. However as
> >> I recall one could surmount the problem by reducing the symmetry of the
> >> system
> >> through the very coordinates of the system. That is, rather than having:
> >>
> >>  0.75 0.25 0.75
> >>  0.25 0.25 0.25
> >> ...
> >>
> >> you can do something like:
> >>
> >>  0.750001 0.250002 0.750003
> >>  0.250004 0.250005 0.250001
> >> ...
> >>
> >> Something like this. I had my fcc positions being generated for some
> >> density I
> >> wanted and therefore this was simple to do.
> >>
> >> I made this calculations a long time ago when I was exploring ABINIT and
> >> I don't recall the details anymore, but I do have the idea of
> >> introducing some "noise" on the positions to avoid the max. number of
> >> symmetry operations
> >> dimensioned in ABINIT.
> >>
> >> Hopefully I am not  mystaken.
> >>
> >> This trick is 0k for a MD simulation starting configuration since we
> >> don't care about optimized supercells anyway.
> >>
> >> Take care.
> >>
> >> Nuno Galamba
> >>
> >> --
> >> Nuno Galamba
> >> Grupo de Fisica-Matematica da Universidade de Lisboa
> >> Av. Prof. Gama Pinto, 2, 1649-003, Lisboa, Portugal
> >> URL: http://gfm.cii.fc.ul.pt/people/ngalamba/

> >> e-mail <http://gfm.cii.fc.ul.pt/people/ngalamba/%0Ae-mail>:

> >> ngalamba@cii.fc.ul.pt
> >>
> >> On Monday 09 March 2009 12:24:24 Yuanliang Xie wrote:
> >> > Dear Nuno Galamba and others,
> >> >  Thank Nuno Galamba for reply.
> >> >  Would you please tell me the details how  make a supercell  in ABINIT
> >> > ( as a 64-atoms NaCl supercell  mentioned  the Nuno Galamba's reply
> >> > )? I made a 64-atoms cubic boron nitride supercell in ABINIT V5.6.4,
> >> > but it fails.  The inputfile is attached .
> >> > The output is
> >> >
> >> > symfind : ERROR -
> >> >   The number of symmetries (including non-symmorphic
> >> >   translations) is larger than msym= 384
> >> >   Action : take a cell that is primitive, or at least
> >> >   smaller than the present one.
> >> >
> >> >  leave_new : decision taken to exit ...
> >> >
> >> > Thanks
> >> > Xie Y.L.
> >> >
> >> > On Mon, Mar 9, 2009 at 2:19 AM, Nuno Jorge Lopes Galamba <
> >> >
> >> > ngalamba@cii.fc.ul.pt> wrote:
> >> > >  Dear Xie Yuanliang,
> >> > >
> >> > > ABINIT can make molecular dynamics much as any other ab initio
> >>
> >> molecular
> >>
> >> > > dynamics code although there are differences relative to other
> >> > > codes.
> >>
> >> Of
> >>
> >> > > course there are also many differences between different (other than
> >> > > ABINIT) codes.  ABINIT allows carrying out what can be called
> >> > > Hellman-Feynman MD simulations. This means that forces on the nuclei
> >>
> >> are
> >>
> >> > > obtained via the Hellman-Feynman theorem every time-step. This is a
> >> > > striking difference from Car-Parrinello MD (CPMD) with obvious
> >> > > computational costs since in the later the energy functional is not
> >> > > minimized every time-step. Other than that both ABINIT and CPMD both
> >>
> >> use
> >>
> >> > > plane-waves and pseudopotentials. On the other hand since
> >> > > plane-waves
> >>
> >> are
> >>
> >> > > used rather than localized basis functions, allowing to apply the
> >> > > Hellman-Feynman theorem, this is also different from what is
> >>
> >> frequently
> >>
> >> > > called Born-Oppenheimer MD, which uses localized basis functions and
> >> > > therefore the Hellman-Feynman theorem cannot be applied. These
> >>
> >> different
> >>
> >> > > denominations are obviously not the fundamental issue, rather, the
> >> > > approximations involved are. Thus one can find first principles or
> >> > > ab initio MD simulations using very different schemes most of which
> >> > > motivated by the need for less demanding force calculation
> >> > > algorithms under a DFT description. I think ABINIT is not specially
> >> > > devoted to
> >>
> >> this
> >>
> >> > > issue and therefore large systems are in my view prohibitive. I have
> >> > > carried out MD simulations of ABINIT of NaCl and NaI (64 atoms) and
> >>
> >> they
> >>
> >> > > were obviously very demanding, however other systems can be much
> >>
> >> worst.
> >>
> >> > > This shall depend, broadly, on the one hand on the number of valence
> >> > > electrons in your system and on the other on the plane-wave cut-off
> >>
> >> you
> >>
> >> > > need to have reasonably converged forces.
> >> > >
> >> > > As for the second part of your question this is not difficult. All
> >> > > you need is to have a set of positions and velocities for your atoms
> >> > > and
> >>
> >> to
> >>
> >> > > set the acell variable as to obtain the desired density. Thus you
> >>
> >> simply
> >>
> >> > > treat acell much as if it were the size of the MD box (supercell),
> >>
> >> rather
> >>
> >> > > than the length of the primitive cell. The computations can, for a
> >> > > supercell, be carried out on a single k point (0,0,0) and you are
> >> > > all set. If you can parallelize ABINIT in terms of nbands things can
> >>
> >> improve,
> >>
> >> > > I would guess, although I have never used ABINIP.
> >> > >
> >> > > Please note I have limited experience with ABINIT, and you may want
> >> > > to check with other ABINIT users.
> >> > >
> >> > > Take care.
> >> > > *
> >> > >
> >> > > *Nuno Galamba
> >> > > Grupo de Física-Matemática da Universidade de Lisboa.
> >> > > Portugal
> >> > >
> >> > > *
> >> > >
> >> > > On Sun, 8 Mar 2009 20:51:24 +0800, Yuanliang Xie wrote*
> >> > >
> >> > > > Dear all,
> >> > > >    I want to know if ABINIT able to do molecular dynamics
> >>
> >>  calculation
> >>
> >> > > for  a very  big super cell  as other ab initio molecular dynamics
> >>
> >> codes.
> >>
> >> > > Moreover, how can I make a supercell in a single point energy
> >>
> >> calculation
> >>
> >> > > with ABINIT?
> >> > >
> >> > > >  Thanks.
> >> > > >
> >> > > > Xie Yuanliang
> >> > > > SYIAE, Shenyang, China
> >>
> >> --
> >> Nuno Galamba
> >> Grupo de Fisica-Matematica da Universidade de Lisboa
> >> Av. Prof. Gama Pinto, 2, 1649-003, Lisboa, Portugal
> >> URL: http://gfm.cii.fc.ul.pt/people/ngalamba/

> >> e-mail <http://gfm.cii.fc.ul.pt/people/ngalamba/%0Ae-mail>:

> >> ngalamba@cii.fc.ul.pt

--

Nuno Galamba
Grupo de Fisica-Matematica da Universidade de Lisboa
Av. Prof. Gama Pinto, 2, 1649-003, Lisboa, Portugal
URL: http://gfm.cii.fc.ul.pt/people/ngalamba/
e-mail: ngalamba@cii.fc.ul.pt