[DFTB-Plus-User] Dispersion correction in DFTB+NEGF

[Gabriele Penazzi]

On 06/10/2014 12:19 PM, anz118466 at cse.iitd.ac.in wrote:
> Hi Gabriele,
>
> As suggested, I tried to simulate with a smaller system in the 32-bit
> system. It produced no error (with and without dispersion correction). I
> believe it must be due to the larger system that the allocation error
> occurred. Any way to circumvent this?

Dear Anusha,

did you remove stack memory limitation? (see troubleshooting section of
the manual)
If you already did, then your system is just too big and doesn't fit in
memory.

Are you defining a partitioning (correct definition of FirstLayerAtoms
field, see also the online cookbook)? An efficient partitioning may
limit the memory consumption a lot and you can go far beyond 1500 atoms,
but this works good for quasi-1d systems. If the system is very bulky
and no partitioning is possible, you have no other chance than moving to
a machine with more memory or reduce (if you can) the system size.

Best
Gabriele




>
> Thanks & Regards,
> Anusha Iyer
>
>
>
>> On 06/10/2014 07:52 AM, anz118466 at cse.iitd.ac.in wrote:
>>> Hi Gabriele,
>>>
>>> I tried to simulate a testcase I have been running on another machine
>>> (64-bit). The 64-bit machine gives no such error, but the current
>>> machine
>>> exits with the error.
>>>
>>> I believe that the simulation works in 64-bit, but the current machine
>>> is
>>> a 32-bit one (i686). I had compiled the i686 DFTB+ from the source few
>>> days back. When I run the transport simulations, the dftb+ gives no
>>> error
>>> for the source/drain calculations. However, when I try to run the device
>>> input file, it exits with the 'ALLOCATION ERROR'. This 32-bit machine
>>> has
>>> 8 cores and 24GB RAM.
>>>
>>> In addition to above, I had a more general query. In the course of
>>> performing DFTB+ calculations, I have realised that the SCC convergence
>>> highly depends on the input geometry. For small geometries, it would
>>> make
>>> sense to optimise in DFTB+ first. But, I am handling a 1500+ atom
>>> system.
>>> Kindly share how an optimised geometry can be generated in such a case
>>> to
>>> achieve convergence.
>> Dear Anusha,
>>
>> I don't see how this is related to the dispersion correction. You should
>> try a SMALL test case, to see if you can reproduce the bug with the
>> minimum amount of information possible. Moreover, I can not estimate if
>> your memory is enough or not without details on your system, an
>> allocation error may be due as well to insufficient memory. The
>> source/drain calculation is typically much less memory demanding.
>>
>> Unluckily there is no brute force way to relax a large structure,
>> however you could be able to relax a 1500 atoms structure with the
>> parallel version of DFTB+. Otherwise you'll have to rely on cheaper
>> methods for the relaxation, like a classical force field.
>>
>> Best
>> Gabriele
>>
>>
>>> Thanks and Regards,
>>> Anusha
>>>
>>>
>>>
>>>> On 06/08/2014 03:25 PM, anz118466 at cse.iitd.ac.in wrote:
>>>>> Dear all,
>>>>>
>>>>> I had a small query regarding the DFTB+NEGF package. Is dispersion
>>>>> correction supported for the same? The DFTB+NEGF manual mentions about
>>>>> the
>>>>> use of SlaterKirkwood model. Can I include this in my transport
>>>>> calculations for considering van der waals forces?
>>>> Dear Anusha,
>>>>
>>>> we are generally trying to keep all the usual DFTB+ features in the
>>>> transport trunk, therefore we did not specifically remove anything for
>>>> dispersion correction but neither we tested a case like you mention.
>>>>
>>>> The reason is that van der Waals forces are long electrostatic forces
>>>> and they do not enter directly the electronic Hamiltonian, therefore
>>>> they do not affect charge transport. One case I could imagine would be
>>>> if you want to calculate forces out of equilibrium and include also
>>>> dispersion correction, but such a feature is not supported yet.
>>>>
>>>> Going back to the allocation error, which should not be there in any
>>>> case, can you give us some more information?
>>>> How large is the system? Did you try a small test case? Does the normal
>>>> transport calculation without SlaterKirkwood correction work properly?
>>>>
>>>> Best,
>>>> Gabriele
>>>>
>>>>
>>>>> Also, when I incorporated the dispersion block in my dftb+negf
>>>>> simulation,
>>>>> I encountered the following error:
>>>>>
>>>>> ALLOCATION ERROR
>>>>>
>>>>> Is this due to some memory shortage?
>>>>>
>>>>> Thanks & Regards,
>>>>> Anusha Iyer
>>>>>
>>>>> _______________________________________________
>>>>> DFTB-Plus-User mailing list
>>>>> DFTB-Plus-User at dftb-plus.info
>>>>> http://www.dftb-plus.info/mailman/listinfo/dftb-plus-user
>>>> --
>>>> Dr. Gabriele Penazzi
>>>> BCCMS - University of Bremen
>>>>
>>>> http://www.bccms.uni-bremen.de/
>>>> http://sites.google.com/site/gabrielepenazzi/
>>>>
>>>> phone: +49 (0) 421 218 62337
>>>> fax: +49 (0) 421 218 62770
>>>>
>>>> _______________________________________________
>>>> DFTB-Plus-User mailing list
>>>> DFTB-Plus-User at dftb-plus.info
>>>> http://www.dftb-plus.info/mailman/listinfo/dftb-plus-user
>>>>
>>> _______________________________________________
>>> DFTB-Plus-User mailing list
>>> DFTB-Plus-User at dftb-plus.info
>>> http://www.dftb-plus.info/mailman/listinfo/dftb-plus-user
>>
>> --
>> Dr. Gabriele Penazzi
>> BCCMS - University of Bremen
>>
>> http://www.bccms.uni-bremen.de/
>> http://sites.google.com/site/gabrielepenazzi/
>>
>> phone: +49 (0) 421 218 62337
>> fax: +49 (0) 421 218 62770
>>
>> _______________________________________________
>> DFTB-Plus-User mailing list
>> DFTB-Plus-User at dftb-plus.info
>> http://www.dftb-plus.info/mailman/listinfo/dftb-plus-user
>>
>
> _______________________________________________
> DFTB-Plus-User mailing list
> DFTB-Plus-User at dftb-plus.info
> http://www.dftb-plus.info/mailman/listinfo/dftb-plus-user


-- 
Dr. Gabriele Penazzi
BCCMS - University of Bremen

http://www.bccms.uni-bremen.de/
http://sites.google.com/site/gabrielepenazzi/

phone: +49 (0) 421 218 62337
fax: +49 (0) 421 218 62770