[Smeagol-discuss] failed diagonalisation (leading minor of order 45 of B is not positive definite)

[Ivan Rungger]

Hello Zeila,

  generally that error I think appears when there is something wrong
with the atomic coordinates setup, but the dependence on the k-points
you describe is weird. Maybe you might try to do a calculation with
InitTransport True, in that way you will make a larger supercell, maybe
that will work. Another thing to try might be to make the leads cell
twice as long, just to see if it runs (also with Inittransport T).  The
input file seems OK to me, but anyway here a few notes on the fdf file:
1)Maybe remove the UseSaveData option, just to make sure you don't read
in any old stuff, and put the coordinates in the coor file of in the
fdf, the same for the lattice parameters.
2)You might try to change the basis cutoff radii on the C, which might
help. Let's say r_c=6 or 4.

Cheers,

 Ivan

Zeila Zanolli wrote:
> Dear Smeagolers,
>
> I was trying to run 'leads'-like calculations for a piece of graphene but
> I've got the following warning and error:
>
>   xijorb: WARNING: orbital pair      1   537 is multiply connected
>
>   *Maximum dynamic memory allocated =   602 MB
>   The leading minor of order           45  of B is not positive definite.
>   The factorization of B could not be completed
>   and no eigenvalues or eigenvectors were computed.
>   Terminating due to failed diagonalisation
>   Stopping Program
>
>
> Now, since this diagonalisation is the same as in siesta, i've also tried
> to make a normal siesta2.0 calculation with the same input and i had
> (withour big surprise) a
>
>   Failure to converge standard eigenproblem
>
> error. I have posted this problem on the siesta mailing list as well, but
> i haven't got any answer yet.
> Note that I've tried both DivideAndConquer true and false, but neither
> option works.
>
> Note also that there is no error if i'm using a SZ basis (default
> EnergyShift and SplitNorm), but the error is there for a DZ and DZP basis
> (default EnergyShift and SplitNorm).
>
> As further weird thing, when I did the 'lead' calculation with 1x1x30
> k-mesh there was no error, so - i've thought - the input coordinates
> should be ok.
> Then I've did the same 1x1x30 with siesta2.0, and the code crashed with
> the 'failure to converge'error as before.
>
> Another thing that I didn't get is that for a similar input structure (a
> graphene stripe along the x direction) the lead calculation succeeded
> (from  a k-mesh greather or equal to 48x1x100, but crashed for smaller
> k-meshes).
> In addition, as a test, i did a siesta calculation for the 48x1x100 mesh,
> and that crashed!
>
> So... can anyone help in understanding what is going on?
>
> I'm attaching here .fdf, .out and note that i'm reading the coordinates
> from the .XV file.
>
> nb: I'm using intel LAPACK
>
>
>
>
> ----------------------------------------------------------------------
> Dr. Zeila Zanolli
>
> Université Catholique de Louvain (UCL)
> Unité Physico-Chimie et de Physique des Matériaux (PCPM)
> Place Croix du Sud, 1 (Boltzmann)
> B-1348 Louvain-la-Neuve, Belgium
> Phone: +32 (0)10 47 2617
> Mobile: +32 (0)487 556699
> Fax: +32 (0)10 47 3452
> e-mail: zeila.zanolli at uclouvain.be
> ----------------------------------------------------------------------
> ------------------------------------------------------------------------
>
> _______________________________________________
> Smeagol-discuss mailing list
> Smeagol-discuss at lists.tchpc.tcd.ie
> http://lists.tchpc.tcd.ie/listinfo/smeagol-discuss


-- 
=================================================
Ivan Rungger,

School of Physics and CRANN,  
Trinity College Dublin,  
Dublin 2,  IRELAND  
Phone: +353-1-8968454  
Email: runggeri at tcd.ie
=================================================