Question about the cutoff energy in QE

[Wan Qian]

Dear QMCPACK developers,

 I am calculating the diffusion of a copper atom in beryllium crystals. When I perform the cutoff energy test, I find that the dft total enery is not convergent. ( pseudopotential comes from the library 'https://pseudopotentiallibrary.org/':  

' Be.upf '  and   'Cu.ccECP.upf' ). Can you hlep me?

ecutwfc = 120Ry:              totol energy = -431.85953962 Ry

ecutwfc = 200Ry:              totol energy = -438.22107416 Ry 

ecutwfc = 250Ry:              totol energy = -440.65886417 Ry

ecutwfc = 300Ry:              totol energy = -442.36506504 Ry

ecutwfc = 350Ry:              totol energy = -443.53786755 Ry

ecutwfc = 400Ry:              totol energy = -444.33544518 Ry

ecutwfc = 450Ry:              totol energy = -444.87599579 Ry

ecutwfc = 500Ry:              totol energy = -445.24334703 Ry

ecutwfc = 550Ry:              totol energy = -445.49435969 Ry

ecutwfc = 600Ry:              totol energy = -445.66708433 Ry

ecutwfc = 650Ry:              totol energy = -445.78678787 Ry 

ecutwfc = 700Ry:              totol energy = -445.87029896 Ry

ecutwfc = 750Ry:              totol energy = -445.92893806 Ry

ecutwfc = 800Ry:              totol energy = -445.97036617 Ry

ecutwfc = 850Ry:              totol energy = -445.99978357 Ry

ecutwfc = 900Ry:              totol energy = -446.02079608 Ry

...

...

 

I have attached the scf.in and scf.out files (cutoff=300Ry).

Best, 

Qian Wan 

[Wan Qian]

Wan Qian <xhua...@gmail.com> 於 2021年1月10日週日 下午1:01寫道：

[Guangming Wang]

Hi Qian,

Thank you for posting on the energy convergence issue with your calculation. My anticipation is that the total energy convergence problem may be attributed to Cu. The plane-wave potential is probably too hard and will need very high KE cutoff to converge the energy.

We have also got some reports on the high KE cutoff convergence problem in plane-wave calculations using several transition metal UPF. We are working to solve the problem by generating a new version of ccECPs for plan-wave calculations. Unfortunately, we don't have the new version for Cu now yet. (So far I have Mn, Fe, Co, Ni new version. Cu will come next) We will soon release a soft-version of ccECP UPF to deprecate these problematic ones.

BTW, if you are going to use ccECPs in semi-local gaussian format, they should work well.

Best,

Guangming Wang

[Wan Qian]

Dear Guangming Wang

Thanks for the help. 

If I use a high cutoff energy in QE, will it have any impact on subsequent QMC results in addition to expensive resource requirements? For some reasons, I can only use QE to generate wave functions, so is there any other UPF forms I can choose?  such as QE's standard pseudopotential library. 

Looking forward to your reply！

Best,

Qian Wan

[Guangming Wang]

I think the high KE cutoff will just make your QE calculation more expensive, limiting you from going to a larger cell size. In QMC, we heard about the large variance using high KE cutoff UPFs in one or two cases. I think you should be safe to use QE's standard pseudopotential library. Or use Be ccECP UPF and Cu UPF from QE's standard pseudopotential library.

Best,

Guangming

[Paul R. C. Kent]

Obviously there is a lot of history on making pseudopotentials that are highly accurate and useful for the solid state with QMC. Clearly the problem is not solved - the ccECPs are more accurate in molecular tests than  previous efforts, but many of them are impractically hard (high cutoff). This almost rules out their usage in the solid state. More work is needed to make soft versions that are still accurate but also usable with plane waves. For 3d elements the Troullier-Martins potentials from J. T. Krogel et al. Physical Review B 93 075143 (2016) have been used successfully. These are also on pseudopotentiallibrary.org . Generally the basis set and hence trial wavefunction DFT total energies should be significantly more converged than the energy scales of interest in the QMC. If you pick a potential from a DFT library (say, one of the QE ones), bear in mind that it will inherit any inaccuracies from the DFT approximation used to generate it. It may also have non-local components that are an issue for the locality approximation. For some elements, materials and properties these don't matter, while for others they do. A good minimal test is to check dimer properties.

Good luck, Paul

[Wan Qian]

Thanks a lot for responses.

I'm trying to use  pseudopotentials from the QE's standard pseudopotential library ('Be.pbe-n-rrkjus_psl.1.0.0.UPF' and 'Cu.pbe-dn-rrkjus_psl.1.0.0.UPF'). Unfortunately, when using pw2qmcpack.x to convert the wave functions, it failed.

The error:

     pw2qmcpack found

     MPI rank pool id    npwx     npw  npwx_tot     ngtot       ngm     ngm_g nks

            0       0    1890    1867     22672     24077     77727    932685  14

esh5 destory the existing ./Tr.pwscf.h5

esh5 create ./Tr.pwscf.h5

The wrong norm of k-point   1 band   1 , before collection, is    1.029726371634101

The orbitals read from QE files are incorrect. Perhaps your QE orbital files are corrupted.

I have atteched the  relevant files. Thank you.

Best wishes

Qian Wan

Paul R. C. Kent <prckent...@gmail.com> 於 2021年1月13日週三 上午2:07寫道：

--
You received this message because you are subscribed to a topic in the Google Groups "qmcpack" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/qmcpack/pc5sBa06tJA/unsubscribe.
To unsubscribe from this group and all its topics, send an email to qmcpack+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/qmcpack/d1fcf20e-41fa-4b98-9497-b7f01880229fn%40googlegroups.com.

[Ye Luo]

pw2qmcpack only supports norm conserving PP which is required by QMCPACK.

Neither PAW nor ultrasoft PP is supported.

Best,

Ye

===================
Ye Luo, Ph.D.
Computational Science Division & Leadership Computing Facility
Argonne National Laboratory

You received this message because you are subscribed to the Google Groups "qmcpack" group.
To unsubscribe from this group and stop receiving emails from it, send an email to qmcpack+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/qmcpack/CAOqCNO5B_uUQMVFd7HuCUsVRWt%3Df8CXNZb7PUJnUFo2tqy%3D7sA%40mail.gmail.com.