[QE-users] Convergence of hybrids: set dexx >> conv_thr ?

[Stefano de Gironcoli]

in the scf cycle the estimated scf accuracy is an estimate of the 
overestimation of the current energy wrt the scf solution.

in the hybrid case, due to the double loop procedure, it is made of two 
terms:
1) how much the inner scf cycle (when V_x is computed with  frozen HF 
orbitals)  is accurate (this is maesured by dr2 in mix_rho and  targets tr2)
2) how much the scf energy is overestimated because the in/out HF 
orbitals do not agree in the outer loop (this is measured by dexx and 
targets tr2_final)

as you say    tr2_final=tr2

the (ir)rational of this is that what one is asking  is that the overall 
error in energy is  not larger than   tr2 ( actually 2*tr2)
and it would be pointless to ask a tight tr2 in one term when one is 
happy to accept a larger error elsewhere...

but  different ratios of  tr2/tr2_final were not explored so maybe some 
other combinations might be more effective to reach the final result

what one should keep in mind is that the estimated error in the final 
energy (the overall scf accuracy) at the end of the
calculation is     dr2+dexx targeting    tr2+tr2_final

stefano

On 24/04/20 03:18, Christoph Wolf wrote:
> Dear all,
>
> I was wondering if there is a particular reason that a hybrid 
> calculation converges at the same threshold as the SCF as
>
> tr2_final = tr2
>
> Since my dexx converges painfully slowly after the first few 
> iterations I was wondering if it could be "safe" to terminate the 
> calculation earlier?
>
>      est. exchange err (dexx)  =       0.08363460 Ry
>      est. exchange err (dexx)  =       0.00463121 Ry
>      est. exchange err (dexx)  =       0.00159110 Ry
>      est. exchange err (dexx)  =       0.00100574 Ry
>      est. exchange err (dexx)  =       0.00063909 Ry
>      est. exchange err (dexx)  =       0.00037948 Ry
>      est. exchange err (dexx)  =       0.00021565 Ry
>      est. exchange err (dexx)  =       0.00012503 Ry
>      est. exchange err (dexx)  =       0.00007862 Ry
>      est. exchange err (dexx)  =       0.00005529 Ry
>
> ...
> (30 ACE cycles later)
>      est. exchange err (dexx)  =       0.00000155 Ry
>      est. exchange err (dexx)  =       0.00000154 Ry
>      est. exchange err (dexx)  =       0.00000153 Ry
>      est. exchange err (dexx)  =       0.00000152 Ry
>      est. exchange err (dexx)  =       0.00000152 Ry
>
> when looking at eigenvalues at gamma (this is from a molecule) they 
> are basically static at this point.
>
> I am mainly wondering because there is no variable that controls 
> tr2_final independently from tr2 so I guess there must be a good 
> reason for that...
>
> thanks in advance!
>
> Best,
> Chris
> -- 
> Postdoctoral Researcher
> Center for Quantum Nanoscience, Institute for Basic Science
> Ewha Womans University, Seoul, South Korea
>
> _______________________________________________
> 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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20200424/c6d0a320/attachment.html>