[CP2K-user] [CP2K:20779] Re: SCF Convergence Issues with OT and Diagonalization for CZTS System (Vacuum, Sulfur Layer, and Hydrogen Passivation

[Marcella Iannuzzi]

Hi Layla

It should not be too difficult to verify whether your bulk calculations are 
accurate enough. 
You can compare some properties with the literature, for instance band gap 
and density of states. 
The k-point sampling can also be replaced by adding more replicas of the 
unit cell.
If the 2x2x1 is not sufficient in bulk calculations, it is also not 
sufficient for slab models,  and this for sure would affect the forces.
But if I correctly understood, you did not manage to calculate forces yet. 
A geometry optimisation before running MD is for sure meaningful in this 
case, 
but this would not solve the convergence problem, I fear, since the SCF is 
exactly the same. 
So the problem is still the single point calculation for the set of 
coordinates of your model. 
Maybe you can share the output of the SCF, such that we get an idea of how 
bad the problem is. 

Regards
Marcella



On Thursday, October 17, 2024 at 6:30:15 PM UTC+2 lheid... at gmail.com wrote:

> Hi Marcella,
> Thank you for your feedback and suggestions. I wanted to update you on 
> what I’ve tried so far and ask for some additional guidance.
> Since I didn’t include a &KPOINTS block, CP2K defaulted to Γ-point-only 
> sampling. For my 2x2x1 periodic system, I realize this might not be 
> sufficient, especially for accurate TDDFT calculations. should I add a 2 × 
> 2 × 1 k-point grid to sample the Brillouin zone more effectively?
>
> For MD simulations, I’ve been using Γ-point-only sampling since the focus 
> is on atomic forces. Would you recommend defining a small k-point grid 
> (e.g., 2 × 2 × 1) for MD runs as well, or is Γ-point-only sufficient in 
> this case?
>
> I increased the energy cutoff to 500 Ry but still faced convergence issues 
> with the surface model. I also adjusted the mixing parameter ALPHA down to 
> 0.005, but the issue persisted.
> I also applied DFT+U  However, I’m still encountering issues with the 
> surface convergence.
>
> I tried adding a sulfur layer and hydrogen passivation to address dangling 
> bonds, but the SCF still failed to converge. You mentioned surface 
> reconstruction—would you suggest running a geometry optimization on the 
> surface before attempting SCF calculations? Also, are there any specific 
> techniques or guidelines for identifying and capping dangling bonds 
> effectively to stabilize the surface?
>
> I really appreciate your insights so far. If you have any further 
> recommendations, I would be grateful.
>
> Best regards,
> Layla
>
> On Thursday, October 17, 2024 at 4:17:58 AM UTC-4 Marcella Iannuzzi wrote:
>
>> Hi ...
>>
>> It is good that the bulk system converges. Do you also obtain the correct 
>> electronic structure? The energy cutoff seems very low. 
>> Are you using k-points for the bulk?  Is the PBE functional good enough 
>> for this type of systems?  12 grids are too many, just use 4 or 5
>>
>> I suppose that depending on how you cleave the bulk, there might be 
>> dangling bonds that should be saturated.
>> Maybe the surface needs to go through a reconstruction (just guessing), 
>> in this case it might help to adjust the coordinates to avoid too many 
>> dangling bonds.
>>
>> Diagonalization is recommended if the energy gap is very small, which can 
>> be the case if you have unrelaxed dangling bonds at the surface.
>> In this case, a smaller mixing ALPHA parameter might help, like 0.005.
>>
>> Regards
>> Marcella
>>
>>
>>
>>  
>>
>>
>> On Wednesday, October 16, 2024 at 11:57:41 PM UTC+2 lheid... at gmail.com 
>> wrote:
>>
>>> Hi,
>>>   Thank you for the suggestion! I’ve already tested the *bulk CZTS 
>>> system* with the same *basis sets, pseudopotentials, and 
>>> exchange-correlation functional*, and it *converged successfully*. 
>>> However, the *convergence problem arises only after introducing the 20 
>>> Å vacuum layer* and *surface modifications*. I've Adjusted *MAX_SCF* 
>>> values (up to 500 iterations) and tried various *mixing parameters*, 
>>> including Broyden mixing, but no improvement. I have attached my input file 
>>> for more clarification (before switching to OT).
>>> Thank you again for your help and suggestions!
>>> Best regards,
>>> Layla
>>>
>>> On Wednesday, October 16, 2024 at 2:15:52 PM UTC-4 Marcella Iannuzzi 
>>> wrote:
>>>
>>>>
>>>> Hi ..
>>>>
>>>> Maybe it simply needs to run for more iterations to converge.
>>>> With the information you provide it is hard to guess.
>>>> Is the electronic structure calculation of the bulk working fine?
>>>> Can you reproduce with your settings (BS, PP, XC etc) the known bulk 
>>>> properties?
>>>>
>>>> Regards
>>>> Marcella
>>>>
>>>>
>>>>
>>>> On Wednesday, October 16, 2024 at 7:40:32 PM UTC+2 lheid... at gmail.com 
>>>> wrote:
>>>>
>>>>> Hello CP2K community,
>>>>>
>>>>> I am running molecular dynamics (MD) simulations on a *Cu2ZnSnS4 
>>>>> (CZTS)* system using *DFT* in CP2K. Below is a detailed description 
>>>>> of my system and the modifications I applied, followed by the SCF 
>>>>> convergence issue I am facing.
>>>>>
>>>>> Cu2ZnSnS4 (CZTS) system modeled in a periodic box.
>>>>> Unit cell dimensions: 10.8 × 10.8 × 10.8 Å.
>>>>> The goal is to study surface interactions and electronic properties 
>>>>> with a vacuum layer.
>>>>>
>>>>> A 20 Å vacuum layer was added in the Z direction to simulate surface 
>>>>> effects: 10.8 × 10.8 × 30.8 Å
>>>>>
>>>>> A layer of sulfur (S) atoms was added to the surface to stabilize the 
>>>>> system and account for surface states.
>>>>>
>>>>> I attempted hydrogen passivation by capping the dangling bonds with H 
>>>>> atoms to further stabilize the surface.
>>>>> I tried running the SCF loop with and without hydrogen passivation, 
>>>>> but both cases failed to converge.
>>>>>
>>>>> SCF Settings and Methods Tried:
>>>>> *Orbital Transformation (OT):*
>>>>> MINIMIZER: DIIS
>>>>> PRECONDITIONER: FULL_SINGLE_INVERSE
>>>>> ENERGY_GAP: 0.001
>>>>> N_HISTORY_VEC: 7
>>>>> *Diagonalization:*
>>>>> I disabled the OT section and enabled diagonalization as a fallback 
>>>>> method, but the SCF still did not converge. ( I tried different parameters 
>>>>> setting)
>>>>> SCF Parameters:
>>>>> SCF_GUESS: ATOMIC
>>>>> EPS_SCF: 1.0E-6
>>>>> MAX_SCF: 100
>>>>>
>>>>> *The SCF loop exits after a few minutes, failing to converge under 
>>>>> both OT and diagonalization methods.*
>>>>> Are there specific SCF settings or preconditioners that can improve 
>>>>> convergence for systems with large vacuum gaps?
>>>>> Are there alternative strategies for handling surfaces and vacuum 
>>>>> layers that could make the system more stable for electronic structure 
>>>>> calculations?
>>>>> Has anyone successfully applied hydrogen passivation to stabilize 
>>>>> surfaces and improve SCF convergence in CP2K?
>>>>>
>>>>> Any suggestions or advice would be greatly appreciated!
>>>>>
>>>>> Thank you for your help and support.
>>>>>
>>>>>
>>>>>

-- 
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 cp2k+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/97af0a63-a7c3-4474-9891-42d99fadc529n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20241017/01b99e8e/attachment-0001.htm>