[CP2K-user] My CP2K problem didn't converge with vibration

Matt W mattwa... at gmail.com
Tue May 7 08:23:40 UTC 2019


Hi,

a few ideas:

I'd only run for say 20-30 scf steps in the inner loop. Only run 2-3 outer 
scf loops with FULL_ALL preconditioner - it can become unstable if 
repeatedly applied to bad problems, in my experience.

If nearly converged do a complete restart from the final wavefunction so 
all history is cleared and hopefully you'll be OK.

Try using CG instead of DIIS.

Matt

On Tuesday, May 7, 2019 at 2:16:47 AM UTC+1, qjh... at gmail.com wrote:
>
> Thans for your help, but it didn't work.
> There is something strange I forgot  mentioning before.
>  I run single point calculation. When the helium atom is 5.5A away from 
> MoS2's Mo atom vertically, the calculation converged; but when the helium 
> atom is 5.6 A away from MoS2's Mo atom, the calculation didn't converge!
> I firstly considered cutoff value. But it does't work by changing he 
> cutoff value.
>
> 在 2019年5月6日星期一 UTC+8下午11:56:54,Travis写道:
>>
>> Try lowering the energy gap to 0.001 au. The FULL_ALL preconditioner 
>> requires an energy gap much lower than the material's band gap. 0.1 au is 
>> greater than the band gap of MoS2 from Materials Project, which may be 
>> leading to your observed instability.
>>
>> -T
>>
>>
>> On Monday, May 6, 2019 at 11:00:47 AM UTC-3, qjh... at gmail.com wrote:
>>>
>>> [image: 1.jpg][image: 2.jpg]
>>>
>>>
>>>
>>> Hello everyone!
>>> I encount a problem about convergence, and my input file has been 
>>> uploaded. It's a single point (energy) calculation about a helium atom and 
>>> a 2D MoS2 material.
>>>
>>> Well, it's not wise to run thousands SCF loops, but I wanted to see if 
>>> it would converge finally. Obvisouly, it didn't converge. 
>>> Was I choosing a wrong  algorithm?
>>>
>>> Thanks for your help!
>>>
>>>
>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190507/b73338b9/attachment.htm>


More information about the CP2K-user mailing list