[QE-users] Problems computing cholesky error in Ni-MoS2 system

Oleksandr Motornyi oleksandr.motornyi at polytechnique.edu
Wed Jun 19 16:25:43 CEST 2019


Dear Jonathan Backman,

I suggest you to try and set diago_david_ndim = 2 instead of the default 
value of diago_david_ndim = 4 if you are using Davidson diagonalization 
- it helped me to resolve a similar issue in the past. The other thing 
that helped me was changing a pseudo potenial to a much hard 
norm-conserving pseudo that contained more semi-core states, however, 
for performance reason I edned up using diago_david_ndim =2 option.

Best regards,

On 19/06/2019 12:47, Jonathan Backman wrote:
> Dear Quantum Espresso community,
>
> I am trying to perform a relax calculation on two different Ni-MoS2 
> contacts. One top contact (Ni over single layer MoS2) and one edge 
> contact (MoS2 connected to the side of Ni), see attached input file. 
> In both cases I get the error "problems computing cholesky". It has 
> previously been suggested that this can happen in the case of "bad" 
> atom positions, but I do not think this is the case for me, especially 
> since it occurs in both my contacts. The problem also remains when 
> changing the pseudo potentials.
>
> It has also been suggested that the error can be caused by a 
> parallelization issue. For the edge contact I am running on 792 
> processor cores (distributed over 33 nodes) divided over 11 pools 
> (since I have 11 kpoints) with 72 processors per pool. With current 
> ecutwfc the FFT dimensions are ( 360, 384,  72). To my understanding 
> should give a effective parallelization, sine number of 
> processors/pool is smaller or equal to FFR_3 and is a divisor of it. I 
> have also tried with 264 processors, 24 processors/pool, but the 
> problem still remains.
>
> The calculation is able to perform 2-3 bfgs steps before showing the 
> error at the beginning of the next scf cycle. If I then take the last 
> atom positions and perform  a new relaxation calculation it can once 
> again go through 2-3 bfgs steps before error. Using this method I can 
> continue restarting it, but it is not very effective. If I instead 
> restart the calculation using the last saved charge density it shows 
> the error immediately in the first scf iteration.
>
> I would be happy for any suggestions on how to solve this problem.
> Please ask if you want more information.
>
>
> Best regards,
> Jonathan Backman, PhD student
> Nano-TCAD Group, ETHZ Switzerland
> jbackman at iis.ee.ethz.ch
>
>
> _______________________________________________
> Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso)
> users mailing list users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users

-- 
Oleksandr Motornyi
PhD

Laboratoire de Solides Irradies
Ecole Polytechnique (Palaiseau, France)

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20190619/0056123d/attachment.html>


More information about the users mailing list