[QE-users] Convergence issue in relaxation calculation

Mon Sep 9 15:58:15 CEST 2024

On 2024-09-08 01:32, Bhargab Kakati wrote:
> Dear Omar A. Ashour,
> 
> Thank you so much for your detailed guidance to tackle the issue. I
> will take a step back and re-implement the input parameters as per
> your suggestions and try to run the calculation again.
> 
> Regards,
> Bhargab
> 
> On 8 Sept 2024 1:20 am, "Omar A. Ashour via users"
> <users at lists.quantum-espresso.org> wrote:
> 
>> Be careful translating VASP input files directly into QE because
>> the implementation details can differ, so just converting the units
>> isn’t enough.
>> 
>> electron_maxtep = 3000 —> if you can’t converge in a couple
>> hundred steps, you should really take a step back and assess what
>> your system is doing. Your calculation did 628 steps, and the SCF
>> accuracy was all over the place--the last step is ~2 Ry. Something
>> is clearly wrong.
>> 
>> Increasing your cutoff to at least 60 Ry will possibly fix your
>> convergence problems. VASP’s PAW sets are incredibly soft; you
>> can’t just convert their cutoff to Ry and expect it to work with a
>> different pseudopotential. Read the UPF files for your pseudos: Ga
>> has a suggested cutoff of 60 Ry, and that’s the minimum you should
>> use.
>> 
>> Other notes:
>> 
>> * Your pseudos are wrong. QE implements SOC differently than VASP.
>> You need to use fully relativistic pseudos.
>> * If this is metallic, use cold smearing with a larger degauss;
>> your k-grid is somewhat coarse. It will probably help convergence,
>> too.
>> * conv_thr is very large; there is no good reason to go above the
>> default 1e-6 for a cell of this size, and you may need to go lower
>> depending on what you’re trying to do. Note that QE’s
>> implementation is different from VASP’s, see equation A7 in this
>> paper, and I believe it’s missing a factor of V^2/2 with V the
>> cell volume.
>> * for_conv_thr is far too large; again, don’t go above the
>> default unless you have a good reason. If I recall correctly, QE
>> applies the force convergence criteria to each component, whereas
>> VASP applies it to the magnitude of the force, which is something to
>> be aware of.
>> * It won’t make a difference here for this system, but if you use
>> ibrav=0 you should have more decimal places in your CELL_PARAMETERS
>> card (10 or so) or the symmetry analysis may not work properly.
>> 
>> You’re relaxing a noncolinear magnet, so due to the complicated
>> energy landscape, you should expect more difficult convergence, even
>> with a proper input file.
>> 
>> Best,
>> 
>> Omar A. Ashour
>> UC Berkeley
>> 
>> On Sep 7, 2024 at 1:53:16 AM, Bhargab Kakati
>> <bhrgabkakati at iasst.res.in> wrote:
>> 
>> On 2024-09-05 23:58, Bhargab Kakati wrote:
>> Thank you so for your suggestion, I'll try and get back yo you if I
>> encounter some issues.
>> 
>> Regards,
>> Bhargab
>> 
>> On 5 Sept 2024 5:13 pm, sogenyi at student.eaifr.org wrote:
>> 
>>> Dear Kakati,
>>> 
>>> I think you need to apply same condition for ecut and ecutrho when
>>> 
>>> using ultrasoft potential to PAW pseudopotential i.e ecutrho = 8
>> or
>>> 10
>>> times ecut.
>>> 
>>> Sunday.
>>> 
>>> Quoting Bhargab Kakati <bhrgabkakati at iasst.res.in>:
>>> 
>>>> Dear experts,
>>>> 
>>>> I am trying to relax my alloy system but it is not converging
>> even
>>> 
>>>> after running for a very long time. I hava attached here the
>> input
>>> 
>>>> and outputs of the run. Any help in resloving the issue would be
>>> 
>>>> highly appreciated.
>>>> 
>>>> Regards,
>>>> Bhargab Kakati
>>> 
>>> _______________________________________________
>>> 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 [1])
>>> users mailing list users at lists.quantum-espresso.org
>>> https://lists.quantum-espresso.org/mailman/listinfo/users
>> Thank you so for your suggestion, I'll try and get back yo you if I
>> 
>> encounter some issues.
>> 
>> Regards,
>> Bhargab
>> 
>> On 5 Sept 2024 5:13 pm, sogenyi at student.eaifr.org wrote:
>>> 
>>> Dear Kakati,
>>> 
>>> I think you need to apply same condition for ecut and ecutrho when
>> 
>>> using ultrasoft potential to PAW pseudopotential i.e ecutrho = 8
>> or
>>> 10
>>> times ecut.
>>> 
>>> Sunday.
>>> 
>>> 
>>> Quoting Bhargab Kakati <bhrgabkakati at iasst.res.in>:
>>> 
>>>> Dear experts,
>>>> 
>>>> I am trying to relax my alloy system but it is not converging
>> even
>>>> after running for a very long time. I hava attached here the
>> input
>>>> and outputs of the run. Any help in resloving the issue would be
>> 
>>>> highly appreciated.
>>>> 
>>>> Regards,
>>>> Bhargab Kakati
>>> 
>>> 
>>> _______________________________________________
>>> 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 [1])
>>> users mailing list users at lists.quantum-espresso.org
>>> https://lists.quantum-espresso.org/mailman/listinfo/users
>> 
>> Hello,
>> 
>> I tried to relax my system bu increasing ecutrho = 8 times of ecut
>> and
>> also changed the pseudopotentials too. It is still not converging. I
>> 
>> have attached the new input and output files here.
>> 
>> Regards,
>> Bhargab
>> 
>> _______________________________________________
>> 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 [1])
>> users mailing list users at lists.quantum-espresso.org
>> https://lists.quantum-espresso.org/mailman/listinfo/users
> 
> 
> 
> Links:
> ------
> [1] http://www.max-centre.eu
> Dear Omar A. Ashour,
> 
> Thank you so much for your detailed guidance to tackle the issue. I 
> will take a step back and re-implement the input parameters as per your 
> suggestions and try to run the calculation again.
> 
> Regards,
> Bhargab
> 
> 
> 
> 
> 
> On 8 Sept 2024 1:20 am, "Omar A. Ashour via users" 
> <users at lists.quantum-espresso.org> wrote:
>> 
>> Be careful translating VASP input files directly into QE because the 
>> implementation details can differ, so just converting the units isn’t 
>> enough. 
>> 
>> electron_maxtep = 3000 —> if you can’t converge in a couple hundred 
>> steps, you should really take a step back and assess what your system 
>> is doing. Your calculation did 628 steps, and the SCF accuracy was all 
>> over the place--the last step is ~2 Ry. Something is clearly wrong.
>> 
>> Increasing your cutoff to at least 60 Ry will possibly fix your 
>> convergence problems. VASP’s PAW sets are incredibly soft; you can’t 
>> just convert their cutoff to Ry and expect it to work with a different 
>> pseudopotential. Read the UPF files for your pseudos: Ga has a 
>> suggested cutoff of 60 Ry, and that’s the minimum you should use.
>> 
>> Other notes:
>> Your pseudos are wrong. QE implements SOC differently than VASP. You 
>> need to use fully relativistic pseudos.
>> If this is metallic, use cold smearing with a larger degauss; your 
>> k-grid is somewhat coarse. It will probably help convergence, too.
>> conv_thr is very large; there is no good reason to go above the 
>> default 1e-6 for a cell of this size, and you may need to go lower 
>> depending on what you’re trying to do. Note that QE’s implementation 
>> is different from VASP’s, see equation A7 in this paper, and I believe 
>> it’s missing a factor of V^2/2 with V the cell volume. 
>> for_conv_thr is far too large; again, don’t go above the default 
>> unless you have a good reason. If I recall correctly, QE applies the 
>> force convergence criteria to each component, whereas VASP applies it 
>> to the magnitude of the force, which is something to be aware of.
>> It won’t make a difference here for this system, but if you use 
>> ibrav=0 you should have more decimal places in your CELL_PARAMETERS 
>> card (10 or so) or the symmetry analysis may not work properly. 
>> 
>> You’re relaxing a noncolinear magnet, so due to the complicated energy 
>> landscape, you should expect more difficult convergence, even with a 
>> proper input file.
>> 
>> Best,
>> 
>> Omar A. Ashour
>> UC Berkeley 
>> 
>> On Sep 7, 2024 at 1:53:16 AM, Bhargab Kakati 
>> <bhrgabkakati at iasst.res.in> wrote:
>>> 
>>> On 2024-09-05 23:58, Bhargab Kakati wrote:
>>>> 
>>>> Thank you so for your suggestion, I'll try and get back yo you if I
>>>> 
>>>> encounter some issues.
>>>> 
>>>> 
>>>> Regards,
>>>> 
>>>> Bhargab
>>>> 
>>>> 
>>>> On 5 Sept 2024 5:13 pm, sogenyi at student.eaifr.org wrote:
>>>> 
>>>> 
>>>> > Dear Kakati,
>>>> 
>>>> >
>>>> 
>>>> > I think you need to apply same condition for ecut and ecutrho when
>>>> 
>>>> >
>>>> 
>>>> > using ultrasoft potential to PAW pseudopotential i.e ecutrho = 8 or
>>>> 
>>>> > 10
>>>> 
>>>> > times ecut.
>>>> 
>>>> >
>>>> 
>>>> > Sunday.
>>>> 
>>>> >
>>>> 
>>>> > Quoting Bhargab Kakati <bhrgabkakati at iasst.res.in>:
>>>> 
>>>> >
>>>> 
>>>> >> Dear experts,
>>>> 
>>>> >>
>>>> 
>>>> >> I am trying to relax my alloy system but it is not converging even
>>>> 
>>>> >
>>>> 
>>>> >> after running for a very long time. I hava attached here the input
>>>> 
>>>> >
>>>> 
>>>> >> and outputs of the run. Any help in resloving the issue would be
>>>> 
>>>> >
>>>> 
>>>> >> highly appreciated.
>>>> 
>>>> >>
>>>> 
>>>> >> Regards,
>>>> 
>>>> >> Bhargab Kakati
>>>> 
>>>> >
>>>> 
>>>> > _______________________________________________
>>>> 
>>>> > 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
>>>> 
>>>> Thank you so for your suggestion, I'll try and get back yo you if I
>>>> 
>>>> encounter some issues.
>>>> 
>>>> 
>>>> Regards,
>>>> 
>>>> Bhargab
>>>> 
>>>> 
>>>> 
>>>> On 5 Sept 2024 5:13 pm, sogenyi at student.eaifr.org wrote:
>>>> 
>>>> >
>>>> 
>>>> > Dear Kakati,
>>>> 
>>>> >
>>>> 
>>>> > I think you need to apply same condition for ecut and ecutrho when 
>>>> 
>>>> > using ultrasoft potential to PAW pseudopotential i.e ecutrho = 8 or
>>>> 
>>>> > 10 
>>>> 
>>>> > times ecut.
>>>> 
>>>> >
>>>> 
>>>> > Sunday.
>>>> 
>>>> >
>>>> 
>>>> >
>>>> 
>>>> > Quoting Bhargab Kakati <bhrgabkakati at iasst.res.in>:
>>>> 
>>>> >
>>>> 
>>>> > > Dear experts,
>>>> 
>>>> > >
>>>> 
>>>> > > I am trying to relax my alloy system but it is not converging even 
>>>> 
>>>> > > after running for a very long time. I hava attached here the input 
>>>> 
>>>> > > and outputs of the run. Any help in resloving the issue would be 
>>>> 
>>>> > > highly appreciated.
>>>> 
>>>> > >
>>>> 
>>>> > > Regards,
>>>> 
>>>> > > Bhargab Kakati
>>>> 
>>>> >
>>>> 
>>>> >
>>>> 
>>>> > _______________________________________________
>>>> 
>>>> > 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
>>> 
>>> 
>>> Hello,
>>> 
>>> I tried to relax my system bu increasing ecutrho = 8 times of ecut 
>>> and
>>> also changed the pseudopotentials too. It is still not converging. I
>>> have attached the new input and output files here.
>>> 
>>> Regards,
>>> Bhargab
>>> _______________________________________________
>>> 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

Dear Omar A. Ashour,

As per your suggestions, I have made few changes in the input file and 
tried to relax my structure again. this time I got improved "estimated 
scf accuracy    <       0.00002101 Ry" but still it hasn't converged in 
even 895 iterations. I have attached the input and output files here for 
  your convenience.

Regards,
Bhargab
-------------- next part --------------
A non-text attachment was scrubbed...
Name: new-inputs-2.zip
Type: application/zip
Size: 38154 bytes
Desc: not available
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20240909/c08b0c6f/attachment.zip>