Serious convergence problem (Au and CeO2 mixed system)

[Yong Li]

Dear Matt,

you are right, your solution is very helpful, now it works. But it take 
more more time.

On Wednesday, March 30, 2016 at 10:16:01 PM UTC+8, Matt W wrote:
>
> Hi,
>
> When Marcella says "not the easiest to treat with DFT", I think she means 
> 'an absolute nightmare to treat with DFT':)
>
> At beginning the EPS_SCF is 1.0E-6, I just loose it when SCF does't 
>> converge. For MD simulation, the geometries diverge too much, this is main 
>> reason that SCF doesn't converge. I hope there is something can improve it.
>>
>> You'll need a very small timestep for MD too. The electronic structure is 
> unstable, with multiple solutions (polarons jumping onto the CeO2 from the 
> metal) so if you aren't very careful you'll lose your solution. This occurs 
> with any DFT code, should be papers from Michaelides for instance where I 
> think they used 0.1 fs with a popular plane-wave code.
>
> Matt
>  
>
>> All the best,
>> Yong
>>
>>
>>
>> On Tuesday, March 29, 2016 at 9:41:54 PM UTC+8, Marcella Iannuzzi wrote:
>>>
>>> Dear Yong Li,
>>>
>>> your system is not the easiest to treat with DFT. 
>>> I am not too much surprised that you encounter convergence problems. 
>>> The combination of functional, DFT+U, basis-sets and potentials might 
>>> not be optimal.
>>> According to our last experience with CeO2, to get a good description of 
>>> the electronic structure we needed a small core Ce element, with 30 valence 
>>> electrons. 
>>>
>>> Have you already tried the following things?
>>> - reduce the mixing coefficient alpha
>>> - increase the smearing temperature
>>> - use a different extrapolation method, e.g., prev_wf
>>> - tighten the eps_default
>>>
>>> Kind regards
>>> Marcella
>>>
>>>
>>> On Saturday, March 26, 2016 at 6:40:18 PM UTC+1, Yong Li wrote:
>>>>
>>>> Dear Sir/Madam,
>>>>
>>>> I was running a large AIMD simulation (about 5200 basis functions) for 
>>>> Au surface supported CeO2 cluster ( (Au(111) surface has180 Au atoms, 
>>>> and Ce10O20 cluster). 
>>>>
>>>> At the beginning, simulation goes well, then it has a very serious 
>>>> convergence problems at around step number of 350. it can't get converged 
>>>> no matter what I do to the SCF control. 
>>>>
>>>> Even if I make it converged coincidentally, but it will not converge 
>>>> after considering Kinetic energy and temperature.
>>>>
>>>> I tried every method I can find， but I can‘t figure it out.
>>>>
>>>> I am depressed and frustrated now, hope someone could help me out with 
>>>> it.  Any suggestions are appreciated!!
>>>>
>>>> The input output file were enclosed in attachment, part of input file 
>>>> is as follows
>>>>
>>>>      
>>>>
>>>>      MULTIPLICITY  1
>>>>      PLUS_U_METHOD  MULLIKEN_CHARGES
>>>>      &SCF
>>>>        MAX_SCF  100
>>>>        EPS_SCF     7.0E-05
>>>>        CHOLESKY  INVERSE
>>>>        SCF_GUESS  ATOMIC
>>>>        ADDED_MOS  400
>>>>        &DIAGONALIZATION  T
>>>>          ALGORITHM  STANDARD
>>>>        &END DIAGONALIZATION
>>>>        &OUTER_SCF  T
>>>>          EPS_SCF     8.0000000000000001E-05
>>>>          MAX_SCF  5
>>>>        &END OUTER_SCF
>>>>        &SMEAR  T
>>>>     #     METHOD  ENERGY_WINDOW
>>>>          METHOD  FERMI_DIRAC
>>>>          ELECTRONIC_TEMPERATURE     3.0000000000000000E+02
>>>>        &END SMEAR
>>>>        &MIXING  T
>>>>          METHOD  BROYDEN_MIXING
>>>>          ALPHA     0.1
>>>>          BETA      1.5
>>>>          NBUFFER  8
>>>>        &END MIXING
>>>>        &PRINT
>>>>          &RESTART  SILENT
>>>>            FILENAME =RESTART.wfn
>>>>            &EACH
>>>>              MD  5
>>>>            &END EACH
>>>>          &END RESTART
>>>>        &END PRINT
>>>>      &END SCF
>>>>      &QS
>>>>        EPS_DEFAULT     1.0000000000000000E-11
>>>>        EPS_PGF_ORB     1.0000000000000001E-05
>>>>        EXTRAPOLATION  ASPC
>>>>        EXTRAPOLATION_ORDER  3
>>>>      &END QS
>>>>      &MGRID
>>>>        CUTOFF     5.0000000000000000E+02
>>>>      &END MGRID
>>>>      &XC
>>>>        DENSITY_CUTOFF     1.0000000000000000E-10
>>>>        GRADIENT_CUTOFF     1.0000000000000000E-10
>>>>        TAU_CUTOFF     1.0000000000000000E-10
>>>>        &XC_GRID
>>>>          XC_SMOOTH_RHO  NN10
>>>>          XC_DERIV  SPLINE2_SMOOTH
>>>>        &END XC_GRID
>>>>        &XC_FUNCTIONAL  NO_SHORTCUT
>>>>          &PBE  T
>>>>          &END PBE
>>>>        &END XC_FUNCTIONAL
>>>>      &END XC
>>>>      &PRINT
>>>>        &MULLIKEN  SILENT
>>>>          FILENAME =md.mulliken
>>>>          &EACH
>>>>            MD  5
>>>>          &END EACH
>>>>        &END MULLIKEN
>>>>      &END PRINT
>>>>    &END DFT
>>>>    &SUBSYS
>>>>
>>>>
>>>>
>>>>
>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20160404/fb4cc8d1/attachment.htm>