The ABINIT Users Mailing List ( CLOSED )

[PGanesh <pganesh@ciw.edu>]

Dear Petr,

This is very helpful and interesting.  I am trying to understand it. Are 
r1, r2 and r3 the three lattice vectors (i.e. the first three rows of 
rprim in Abinit)? What you call the compliance tensor, you use to get 
the strain that needs to be applied to the structure (r_(n-1))) from the 
stress-residue.  Size of its elements will determine the rate of 
convergence and by selecting certain components to zero one can allow 
certain lattice vectors to remain constant, while others relax (I don't 
see how though. For example if I chose 'du' such that it makes e[1]=1, 
e[6]=0, e[5]=0, then I can fix the 'x' component of @r1, but I am not 
changing 'x' component of @r3 either!. One can of course fix @r1 and @r2 
while @r3 changes by hand as you showed in the direct method).  

In short, for my case of interest, I think what you are saying is that 
the matrix (''e'") that transforms the vector r3(old)->r3(new) has 
elements proportional to the stress tensor (assuming the target stress 
is all zero).  I need to study your code to see how you choose the 
proportionality constants that control the convergence.

Please correct me if I am wrong.  I will then try to test it myself.

Thanks,
Ganesh

Petr Sestak wrote:
> Hi Matthieu and P. Ganesh,
>
> There are two ways how to affect the relaxation procedure. The first is to set
> some component of array @du (tensor bit) to zero value to avoid the relaxation
> of corresponding stress tensor component. However, this tensor is affecting
> only the final stress tensor, but we cannot control directly the translation
> vectors. I think, that the relaxation procedure in the Abinit code is very
> similar or same except the tensor bit as you wrote. May I ask, if the Abinit 
> is
> able to avoid relaxation of some stress tensor component during the relaxation
> to the given values of stress tensor (parameter STRTARGET)? Because I guess,
> that this is the only difference between the mentioned relaxation procedures. 
>
> The second way is direct modification of the relaxation procedure. I am using
> this in some very special cases when the first way can not be used. This
> modification I used at the PGanesh problem as I demonstrated in my previous
> answer. I changed the following code lines
>
>   @r1=(0, $r1[1]*$e[1]+$r1[2]*$e[6]+$r1[3]*$e[5],
> $r1[1]*$e[6]+$r1[2]*$e[2]+$r1[3]*$e[4],
> $r1[1]*$e[5]+$r1[2]*$e[4]+$r1[3]*$e[3]);
>   @r2=(0, $r2[1]*$e[1]+$r2[2]*$e[6]+$r2[3]*$e[5],
> $r2[1]*$e[6]+$r2[2]*$e[2]+$r2[3]*$e[4],
> $r2[1]*$e[5]+$r2[2]*$e[4]+$r2[3]*$e[3]);
>   @r3=(0, $r3[1]*$e[1]+$r3[2]*$e[6]+$r3[3]*$e[5],
> $r3[1]*$e[6]+$r3[2]*$e[2]+$r3[3]*$e[4],
> $r3[1]*$e[5]+$r3[2]*$e[4]+$r3[3]*$e[3]);
>
> to
>   @r1=(0, $r1[1], $r1[2], $r1[3]);
>   @r2=(0, $r2[1], $r2[2], $r2[3]);
>   @r3=(0, $r3[1]*$e[1]+$r3[2]*$e[6]+$r3[3]*$e[5],
> $r3[1]*$e[6]+$r3[2]*$e[2]+$r3[3]*$e[4],
> $r3[1]*$e[5]+$r3[2]*$e[4]+$r3[3]*$e[3]);
>
> I am afraid, that the direct modification of the relaxation procedure can not
> be implemented to the source code because it is different for each structure
> and relaxation conditions. This is the reason why I am using the external
> script.
>
> I have some idea how to solve this problem. I will try to implement it to my
> scipt and if I get some good results I will let you know.
>
> Yes, my indices i,j are in the cartesian coordinates.
>
> Best regards
>
> Ing. Peter Sestak
> Institute of Physical Engineering
> Faculty of Mechanical Engineering
> Brno University of Technology
>