[CP2K-user] [CP2K:15307] Convergence issue for GAPW without ADMM

[ma...@gmail.com]

Hi Marcella and Matt

Thanks! Restarting from the PBE wavefunction really helps converging a lot. 
The smaller EPS_SCHWARZ and EPS_SCHWARZ_FORCES decrease the calculation 
speed significantly. So I normally use EPS_SCHWARZ=1E-6 and 
EPS_SCHWARZ=1E-5. 

Regards,
Hongyang

在2021年5月8日星期六 UTC+10 上午2:28:52<Matt W> 写道：

> Having said that the HSE calc is a bit on the edge. Really needs a bigger 
> cell and EPS_SCHWARZ a bit smaller than I had above.
>
> On Friday, May 7, 2021 at 5:25:49 PM UTC+1 Matt W wrote:
>
>> Yes. Runs fine for me starting from PBE guess. 
>> Make sure you get the electron repulsion integrals into memory - you'll 
>> get a x100+ speed up.
>> Matt
>>
>>
>> On Friday, May 7, 2021 at 3:36:34 PM UTC+1 Marcella Iannuzzi wrote:
>>
>>>
>>> Dear Hongyang,
>>>
>>> Have you tried by starting from the RESTART.wfn of the PBE calculation?
>>> Best
>>> marcella
>>> On Friday, May 7, 2021 at 2:29:21 PM UTC+2 ma... at gmail.com wrote:
>>>
>>>> Dear Augustin,
>>>>
>>>> I just tried adding smearing, but it still doesn't improve too much. It 
>>>> still takes more than 300 cycles to converge. This is really strange 
>>>> because the system is just pure Si and the pob-TZVP as well as HSE06 are 
>>>> very good basis set and functional. I don't understand why it is so hard to 
>>>> converge. When I use pure PBE, the convergence is really good (maybe around 
>>>> 20 cycles). However, HSE06 performs terrible. Do you have some suggestions 
>>>> in this situation?
>>>>
>>>> Regards,
>>>> Hongyang
>>>>
>>>> 在2021年5月7日星期五 UTC+10 下午9:42:06<Augustin Bussy> 写道：
>>>>
>>>>> Dear Hongyang,
>>>>>
>>>>> I tested your input file and it does actually converge, although after 
>>>>> 322 SCF steps. The total energy oscillates along the way, which usually 
>>>>> happens with metals and small gap semiconductors. Adding some smearing may 
>>>>> help.
>>>>>
>>>>> Note that you have an important HFX warning about the use of short 
>>>>> range potentials. You should probably use a larger cell to insure your 
>>>>> results are physically sound.
>>>>>
>>>>> Best,
>>>>> Augustin
>>>>>
>>>>> ----... at googlegroups.com wrote: -----
>>>>> To: "cp2k" <c... at googlegroups.com>
>>>>> From: "ma... at gmail.com" 
>>>>> Sent by: c... at googlegroups.com
>>>>> Date: 05/07/2021 11:37AM
>>>>> Subject: Re: [CP2K:15307] Convergence issue for GAPW without ADMM
>>>>>
>>>>>
>>>>> Dear Augustin,
>>>>>
>>>>> I actually have tried lowering EPS_SCHWARZ to even 1E-16, but the SCF 
>>>>> still didn't converge. 
>>>>> The calculation was run using GAPW without ADMM. I didn't use ADMM in 
>>>>> this calculation. But it is quite weired that pob-TZVP with GAPW cannot 
>>>>> converge. I don't know what is the reason causing instability.
>>>>>
>>>>> Thanks,
>>>>> Hongyang
>>>>>
>>>>> 在2021年5月7日星期五 UTC+10 下午6:29:13<Augustin Bussy> 写道：
>>>>>
>>>>>> Dear Hongyang,
>>>>>>
>>>>>> In condensed systems, there is generally less tolerance to screening. 
>>>>>> I would try lowering EPS_SCHWARZ. Note that the ADMM basis set you are 
>>>>>> currently using is meant to be used with the MOLOPT basis, which is not 
>>>>>> all-electron. For the use of ADMM in an all-electron setting, I would 
>>>>>> recommand the pcseg basis set family (and corresponding admm) by Jensen, 
>>>>>> available on Basis Set Exchange.
>>>>>>
>>>>>> Best,
>>>>>> Augustin
>>>>>>
>>>>>>
>>>>>> ----... at googlegroups.com wrote: -----
>>>>>> To: "cp2k" <c... at googlegroups.com>
>>>>>> From: "ma... at gmail.com" 
>>>>>> Sent by: c... at googlegroups.com
>>>>>> Date: 05/07/2021 03:00AM
>>>>>> Subject: [CP2K:15305] Convergence issue for GAPW without ADMM
>>>>>>
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> I'm a new user of cp2k 8.1. I'm currently using ALL electron basis 
>>>>>> set and GAPW method to do a simple Si supercell calculation. But without 
>>>>>> ADMM, the SCF process does not converge. I do not want to use ADMM in this 
>>>>>> case because, firstly, the speed is slower than calculations without ADMM 
>>>>>> and, secondly, the accuracy is slightly decreased. Could someone please 
>>>>>> provide some suggestions on solving this? Thanks!
>>>>>>
>>>>>> Regards,
>>>>>> Hongyang
>>>>>>
>>>>>> -- 
>>>>>> You received this message because you are subscribed to the Google 
>>>>>> Groups "cp2k" group.
>>>>>> To unsubscribe from this group and stop receiving emails from it, 
>>>>>> send an email to cp... at googlegroups.com.
>>>>>> To view this discussion on the web visit 
>>>>>> https://groups.google.com/d/msgid/cp2k/c6f28c5e-6216-42ab-bab5-3f3c7b24d663n%40googlegroups.com 
>>>>>> <https://groups.google.com/d/msgid/cp2k/c6f28c5e-6216-42ab-bab5-3f3c7b24d663n%40googlegroups.com?utm_medium=email&utm_source=footer>
>>>>>> .
>>>>>>
>>>>>>
>>>>>> [attachment "si-b-01-out.txt" removed by Augustin Bussy/at/UZH]
>>>>>> [attachment "si-b-01-inp.txt" removed by Augustin Bussy/at/UZH]
>>>>>>
>>>>> -- 
>>>>> You received this message because you are subscribed to the Google 
>>>>> Groups "cp2k" group.
>>>>> To unsubscribe from this group and stop receiving emails from it, send 
>>>>> an email to cp... at googlegroups.com.
>>>>> To view this discussion on the web visit 
>>>>> https://groups.google.com/d/msgid/cp2k/d8c52869-2a0f-432b-b87c-489844cf0979n%40googlegroups.com 
>>>>> <https://groups.google.com/d/msgid/cp2k/d8c52869-2a0f-432b-b87c-489844cf0979n%40googlegroups.com?utm_medium=email&utm_source=footer>
>>>>> .
>>>>>
>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210507/2867b208/attachment.htm>