[CP2K-user] Error with "Cholesky decompose failed"

[Travis]

Hi,

SIGFPE is a GNU compiler catch-all signal for bad maths (e.g., division by 
zero). What is the input file you tried running? An input deck like this is 
most appropriate for HPC platforms with many nodes. You could also try 
lowering the cost of the calculation by eliminating k-points,




*#  &KPOINTS#   SCHEME  MONKHORST-PACK  4  4  4#   FULL_GRID  .TRUE.#  &END 
KPOINTS*

This would just be a test, k-points are necessary for accurate properties 
of metallic systems.

-T

On Wednesday, October 2, 2019 at 12:00:02 AM UTC-3, Jiapeng Liu wrote:
>
> Hi Travis,
>
> I just tried with your suggested script on my computer, but I got 
> following errors
>
> Program received signal SIGFPE: Floating-point exception - erroneous 
> arithmetic operation.
>
> Backtrace for this error:
>
> Program received signal SIGFPE: Floating-point exception - erroneous 
> arithmetic operation.
>
> For your reference, I just installed cp2k on Ubuntu 18.04 by sudo apt-get 
> install -y cp2k and cp2k version is 5.1 with cp2k.popt. Can you shed some 
> light on this problem, thanks.
>
> Regards,
> Jiapeng
>
>
> On Tuesday, October 1, 2019 at 10:50:17 PM UTC+8, Travis wrote:
>>
>> Hi,
>>
>> Metallic systems need Fermi-Dirac smearing to converge. Something like 
>> this is more fitting. Add in a section for GEO_OPT or CELL_OPT and add 
>> coordinates and a KIND section. The bits in bold are most pertinent to your 
>> case.
>>
>> &FORCE_EVAL
>>  METHOD QUICKSTEP
>>  STRESS_TENSOR ANALYTICAL
>>  &DFT
>>   UKS T
>>   CHARGE 0
>>   MULTIPLICITY 1
>>   BASIS_SET_FILE_NAME data/BASIS_MOLOPT
>>   POTENTIAL_FILE_NAME data/POTENTIAL
>>   &MGRID
>>    CUTOFF 800
>>    NGRIDS 5
>>    RELATIVE_CUTOFF 50
>>   &END MGRID
>>   &QS
>>    EPS_DEFAULT 1.0E-12
>>    METHOD GPW
>> *   EXTRAPOLATION USE_GUESS *
>>   &END QS
>>   &SCF
>>    EPS_SCF 1e-06
>>    MAX_SCF 200
>>    SCF_GUESS RESTART
>>    *ADDED_MOS  *
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> *400   &SMEAR  T    METHOD  FERMI_DIRAC    ELECTRONIC_TEMPERATURE     
>> 3.0000000000000000E+02   &END SMEAR   &MIXING  T    METHOD  BROYDEN_MIXING  
>>   ALPHA     4.0000000000000002E-01    NMIXING  5    NBUFFER  8   &END** 
>> MIXING*
>>   &END SCF
>>   &XC
>>    FUNCTIONAL_ROUTINE NEW
>>    DENSITY_CUTOFF 1.0e-12
>>    GRADIENT_CUTOFF 1.0e-12
>>    TAU_CUTOFF 1.0e-12
>>    &XC_FUNCTIONAL
>>     &PBE
>>      PARAMETRIZATION Orig
>>     &END PBE
>>    &END XC_FUNCTIONAL
>>    &XC_GRID
>>     USE_FINER_GRID  T
>>    &END XC_GRID
>>   &END XC
>>   &POISSON
>>    POISSON_SOLVER  PERIODIC
>>    PERIODIC  XYZ
>>   &END POISSON
>>
>>
>>
>> *  &KPOINTS   SCHEME  MONKHORST-PACK  4  4  4   FULL_GRID  .TRUE.  &END 
>> KPOINTS*
>>  &END DFT
>>  &SUBSYS
>>   &CELL
>>    ABC     6.7638 6.7638 6.7638
>>    ALPHA_BETA_GAMMA   90.000000  90.000000  90.000000
>>    PERIODIC XYZ
>>    MULTIPLE_UNIT_CELL  1  1  1
>>   &END CELL
>>  ...
>>  &END SUBSYS
>> &END FORCE_EVAL
>>
>> Additionally, note that the Cholesky decomposition error is usually 
>> related to using too large a basis set or too diffuse a basis set for the 
>> problem. For a double-zeta flavor of basis set, the EPS_PGF_ORB fix is a 
>> reasonable approach to correct the issue. But for triple-zeta and 
>> quadruple-zeta (or higher) basis sets, it is usually better to select a 
>> smaller or less diffuse basis set. It is very common to use the MOLOPT-SR 
>> variants as they do well for solids and condensed phase simulations. The 
>> short-ranged (SR) variant is purpose built to be less diffuse, so it's more 
>> likely any overlap issues come from poor initial geometry, even with one of 
>> the triple-zeta SR basis sets. Setting NGRIDS 5 as above is useful for 
>> accelerating calculations with the MOLOPT basis sets.
>>
>> -T
>>
>> On Monday, September 30, 2019 at 1:38:35 PM UTC-3, Jiapeng Liu wrote:
>>>
>>> Dear CP2K users,
>>>
>>> I am just learning CP2K from the official website and trying to relax a 
>>> Li bcc structure where 16 atoms are included. I modified from the example 
>>> shown here 
>>> https://www.cp2k.org/exercises:2019_conexs_newcastle:ex3#part_1optimizing_geometry for 
>>> MgO. But I met some problems running my input file. The output says that 
>>> "Cholesky decompose failed: the matrix is not positive definite or 
>>> ill-conditioned". I am quite confused where is the error from. Can you guys 
>>> help me on this. Thanks for your kind help.
>>>
>>> I also attach my input file here, thank you very much.
>>>
>>> [image: WeChat Screenshot_20191001003731.png]
>>>
>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20191002/a137b58d/attachment.htm>