[QE-users] SCF convergence issue after first ionic step
Laurent Pizzagalli
laurent.pizzagalli at univ-poitiers.fr
Tue Dec 19 14:32:58 CET 2023
Dear Paolo,
thank you for your answer. I also do not know why the electronic
structure calculation fails after the update of ionic positions (the
forces being reasonable). As far as I understand there is no simple way
to solve the issue, since there is no keyword to start the scf
calculation from scratch after an ionic relaxation step (Information
from the previous scf calculation being used as input). It is also not
possible to play with 'diag_thr_init', since it is apparently used only
for the scf relaxation of the first ionic step, but not the next ones...
Am I right ?
In that case the only way I see is to perform step by step. I hope that
after a few ionic iterations, the issue will disappear...
Best regards
Laurent
PS You mentioned that the starting wf are anyway taken from the previous
steps. Then what is the meaning of keyword 'wfc_extrapolation' ? Is it
only used during MD calculations ?
On 19/12/2023 13:55, Paolo Giannozzi wrote:
> On 12/19/23 11:08, Laurent Pizzagalli wrote:
>
>> It seems to me that with 'pot_extrapolation=none' and
>> 'wfc_extrpolation =none', the scf calculation at the n+1 ionic
>> relaxation step does not use information from the electronic
>> structure relaxed at ionic step n (i.e. it is equivalent to running a
>> scf calculation from scratch).
>
> it is not exactly the same calculation, because
> - if I remember correctly, the starting wavefunctions and charge
> densities are anyway taken from the previous step
> - thresholds on scf ans iterative diagonalization are computed
> differently
> Hard to say why these rather minor differences lead to a different
> outcome.
>
> Paolo
>
>
>>
>> On 19/12/2023 10:29, Giovanni Cantele wrote:
>>> Hi, I do not have a precise explanation of that, but I can say that
>>> I found such a behavior several times, in the years,
>>> for completely different systems.
>>> And this concerns not only the convergence of the scf step, but also
>>> the relaxation
>>> of the atomic positions.
>>> At some iteration it seems that the algorithm falls in some
>>> "potential well" and cannot exit it at all.
>>>
>>> In this respect, it helps often just to restart the calculation from
>>> scratch using one of the last coordinate sets found by
>>> the pw.x, or in some cases restart the code introducing in the
>>> coordinates even very small perturbations (e.g. delete the
>>> digits from the fourth to the last). In other cases, you might try
>>> to increase the smearing, even just for a few steps, mostly if
>>> the problem is the convergence of the scf cycle.
>>> This is just empirical, but it is the signature for tricky
>>> calculations (where there is no parameter
>>> to identify when a calculation is going to become "tricky"!), where
>>> the presence of local minima or other numerically hard problems,
>>> either with respect to the charge density or with respect to the
>>> atomic positions, can make the convergence non as easy as expected.
>>>
>>> Maybe, someone more expert might add some more specific hints.
>>>
>>> Giovanni
>>>
>>> --
>>>
>>> Giovanni Cantele, PhD
>>> CNR-SPIN
>>> c/o Dipartimento di Fisica
>>> Universita' di Napoli "Federico II"
>>> Complesso Universitario M. S. Angelo - Ed. 6
>>> Via Cintia, I-80126, Napoli, Italy
>>> e-mail: giovanni.cantele at spin.cnr.it
>>> <mailto:giovanni.cantele at spin.cnr.it>
>>> Phone: +39 081 676910
>>> Skype contact: giocan74
>>>
>>> ResearcherID: http://www.researcherid.com/rid/A-1951-2009
>>> Web page: https://sites.google.com/view/giovanni-cantele/home
>>>
>>>
>>> Il giorno mar 19 dic 2023 alle ore 08:49 Laurent Pizzagalli
>>> <laurent.pizzagalli at univ-poitiers.fr> ha scritto:
>>>
>>> Dear all,
>>>
>>> I came across an issue I never encountered before. I performed a
>>> simple
>>> ionic relaxation of a W cluster. The initial scf convergence is
>>> achieved, the ionic forces are calculated (and are reasonable for a
>>> first ionic relaxation step) and the new atomic positions are
>>> obtained.
>>> The issue arises during the next scf calculation, which diverges
>>> in a
>>> few steps. However, if I start a new calculation with the updated
>>> ionic
>>> positions as input structure, the scf calculation runs smoothly and
>>> converges (but the issue occurred again for the next ionic
>>> relaxation
>>> steps).
>>> I obtained the same behavior with two different pw.x versions (6.7
>>> and
>>> 7.1). I also tried to set 'pot_extrapolation = none', but with no
>>> difference...
>>>
>>> Do any of you have an idea about the origin (and solution) of this
>>> issue ?
>>>
>>> Thanks
>>>
>>> Laurent
>>>
>>> -- ,,, __,
>>> /'^'\ |__|
>>> ( o o ) |
>>> --------------------------------------------------oOOO--(_)--OO|o------
>>> <Laurent.Pizzagalli at univ-poitiers.fr>
>>> http://laurent.pizzagalli.free.fr/ Tel +33 549 49 74 99
>>> ------------------------------------------ Fax +33 549 49 66 92
>>> Institut P'
>>> Departement de Physique et de Mécanique des Matériaux
>>> CNRS UPR 3346
>>> Université de Poitiers
>>> SP2MI
>>> TSA 41123 .oooO
>>> 86073 Poitiers Cedex 9, FRANCE ( ) Oooo.
>>> ----------------------------------------------------\ (----(
>>> )-------
>>> \_) ) /
>>> (_/
>>>
>>> _______________________________________________
>>> The Quantum ESPRESSO community stands by the Ukrainian
>>> people and expresses its concerns about the devastating
>>> effects that the Russian military offensive has on their
>>> country and on the free and peaceful scientific, cultural,
>>> and economic cooperation amongst peoples
>>> _______________________________________________
>>> Quantum ESPRESSO is supported by MaX (www.max-centre.eu
>>> <http://www.max-centre.eu>)
>>> users mailing list users at lists.quantum-espresso.org
>>> https://lists.quantum-espresso.org/mailman/listinfo/users
>>>
>>>
>>> _______________________________________________
>>> The Quantum ESPRESSO community stands by the Ukrainian
>>> people and expresses its concerns about the devastating
>>> effects that the Russian military offensive has on their
>>> country and on the free and peaceful scientific, cultural,
>>> and economic cooperation amongst peoples
>>> _______________________________________________
>>> Quantum ESPRESSO is supported by MaX (www.max-centre.eu)
>>> users mailing listusers at lists.quantum-espresso.org
>>> https://lists.quantum-espresso.org/mailman/listinfo/users
>>
>> --
>> ,,, __,
>> /'^'\ |__|
>> ( o o ) |
>> --------------------------------------------------oOOO--(_)--OO|o------
>> <Laurent.Pizzagalli at univ-poitiers.fr>
>> http://laurent.pizzagalli.free.fr/ Tel +33 549 49 74 99
>> ------------------------------------------ Fax +33 549 49 66 92
>> Institut P'
>> Departement de Physique et de Mécanique des Matériaux
>> CNRS UPR 3346
>> Université de Poitiers
>> SP2MI
>> TSA 41123 .oooO
>> 86073 Poitiers Cedex 9, FRANCE ( ) Oooo.
>> ----------------------------------------------------\ (----( )-------
>> \_) ) /
>> (_/
>>
>>
>> _______________________________________________
>> The Quantum ESPRESSO community stands by the Ukrainian
>> people and expresses its concerns about the devastating
>> effects that the Russian military offensive has on their
>> country and on the free and peaceful scientific, cultural,
>> and economic cooperation amongst peoples
>> _______________________________________________
>> Quantum ESPRESSO is supported by MaX (www.max-centre.eu)
>> users mailing list users at lists.quantum-espresso.org
>> https://lists.quantum-espresso.org/mailman/listinfo/users
>
--
,,, __,
/'^'\ |__|
( o o ) |
--------------------------------------------------oOOO--(_)--OO|o------
<Laurent.Pizzagalli at univ-poitiers.fr>
http://laurent.pizzagalli.free.fr/ Tel +33 549 49 74 99
------------------------------------------ Fax +33 549 49 66 92
Institut P'
Departement de Physique et de Mécanique des Matériaux
CNRS UPR 3346
Université de Poitiers
SP2MI
TSA 41123 .oooO
86073 Poitiers Cedex 9, FRANCE ( ) Oooo.
----------------------------------------------------\ (----( )-------
\_) ) /
(_/
More information about the users
mailing list