[Octopus-users] Segmentation fault in periodic casida calculation

[O. Baris Malcioglu]

Dear Xavier,

Thanks for your prompt reply,

First of all, please find the compilation environment as reported by
octopus attached.

The reason I wanted to do a periodic cell is that I would like to
compare the results with a code that uses plane wave basis sets, thus
the supercell approach. Of course, using a periodic or non-periodic
cell does not affect the (well-)bound levels that much, however, I am
curious about whether X-C is affected significantly due to periodic
scattering coming from images
i. e. looking at the first few virtual states from ground state DFT of octopus

Non-Periodic
 #st  Spin   Eigenvalue     Occupation       Error
   1   --   -16.760013       2.000000      (9.1E-07)
   2   --    -9.382138       2.000000      (9.1E-07)
   3   --    -9.380361       2.000000      (6.8E-07)
   4   --    -9.380361       2.000000      (8.0E-07)
   5   --    -0.126850       0.000000      (9.3E-07)
   6   --     0.502674       0.000000      (1.0E-05)
Periodic
 #st  Spin   Eigenvalue     Occupation       Error
   1   --   -16.741408       2.000000      (1.9E-03)
   2   --    -9.363976       2.000000      (9.3E-04)
   3   --    -9.363976       2.000000      (2.3E-04)
   4   --    -9.363976       2.000000      (1.0E-03)
   5   --    -0.413110       0.000000      (2.4E-04)
   6   --     0.191874       0.000000      (1.8E-04)

the estimation of the optical gap from DFT is different by 0.3 eV, and
this is so just by playing with periodicity, assuming that the
computer I have done this test is sane enough.

I suspect that if this is the case, it might have some non-trivial
consequences in a Casida calculation.

Best,
Baris









Version                : 3.2.0
Revision               : 6124 [td_write.F90 2009-11-24]
Build time             : Wed Feb 24 15:16:59 CET 2010
Configuration options  : max-dim=3 mpi
Optional libraries     : mpi2 metis libnbc
Architecture           : x86_64
C compiler             : mpicc (icc)
C compiler flags       : -g -O2 -I/u/cm/mbaris/Prog/libs//include
Fortran compiler       : mpif90 (ifort)
Fortran compiler flags : -O3

2010/4/21 Xavier Andrade <xavier at tddft.org>:
> Hi Baris,
>
> Now that I checked your input file in detail, I noticed that you are trying
> do to a calculation for a molecule with a supercell. In Octopus you don't
> need to do that, by default finite boundary conditions are used (the density
> and orbitals are zero at the boundaries of the box). This is a simplified
> version of the input for the calculation (I think) you are trying to do:
>
> =================================================================
>
> Units=ev_angstrom
>
> %CalculationMode
>  gs              | unocc            | casida
>  "ground_state_" |  "unocc_states_" | "lrtddft_"
>  1               | 2                | 3
> %
>
> Radius=6.6
> Spacing=0.15
>
> unocc_states_RestartDir = "ground_state_restart"
> lrtddft_RestartDir = "unocc_states_restart"
>
> %Coordinates
>  "C" |     4.000000000 - 4|  4.000000000 - 4|  4.000000000 - 4
>  "H" |     4.642814093 - 4|  4.642814093 - 4|  4.642814093 - 4
>  "H" |     3.357185907 - 4|  3.357185907 - 4|  4.642814093 - 4
>  "H" |     4.642814093 - 4|  3.357185907 - 4|  3.357185907 - 4
>  "H" |     3.357185907 - 4|  4.642814093 - 4|  3.357185907 - 4
> %
>
> extrastates = 2
>
> XCFunctional = lda_x + lda_c_pz_mod
>
> NumberUnoccStates = 50
>
> LinMinEnergy=0
> LinMaxEnergy=40
> LinEnergyStep=0.001
> LinBroadening=0.1
>
> =================================================================
>
> As you see, I shifted your molecule, since in octopus the center of the box
> is at zero. I hope it helps.
>
> Cheers,
>
> Xavier
>
> On Wed, 21 Apr 2010, O. Baris Malcioglu wrote:
>
>> Dear all,
>>
>> I am getting a segmentation fault with the following input in the
>> casida stage. I am using Octopus 3.2.0. The output file suggests
>> everything goes fine up to unoccupied states. (No NaN, everything
>> converges, etc...)
>>
>> Best,
>> Baris Malcioglu
>>
>>
>>
>> #DebugLevel = 0
>>
>> Units=ev_angstrom
>> %CalculationMode
>> gs              | unocc            | casida
>> "ground_state_" |  "unocc_states_" | "lrtddft_"
>> 1               | 2                | 3
>> %
>>
>> #DoubleGrid=yes
>> BoxShape=parallelepiped
>> Lsize=8 | 8 | 8
>> #Radius=6.6
>> Spacing=0.15
>> PeriodicDimensions=3
>>
>> unocc_states_RestartDir = "ground_state_restart"
>> lrtddft_RestartDir = "unocc_states_restart"
>>
>> FromScratch=no
>>
>> %Coordinates
>> "C" |     4.000000000 |  4.000000000 |  4.000000000| no
>> "H" |     4.642814093 |  4.642814093 |  4.642814093| no
>> "H" |     3.357185907 |  3.357185907 |  4.642814093| no
>> "H" |     4.642814093 |  3.357185907 |  3.357185907| no
>> "H" |     3.357185907 |  4.642814093 |  3.357185907| no
>> %
>>
>> extrastates = 2
>>
>>
>> XCFunctional = lda_x + lda_c_pz_mod
>>
>> #ConvRelDens = 1e-5
>> #ConvAbsDens = 1e-5
>> EigenSolverMaxIter = 300
>>
>> NumberUnoccStates = 50
>>
>> LinMinEnergy=0
>> LinMaxEnergy=40
>> LinEnergyStep=0.001
>> LinBroadening=0.1
>> _______________________________________________
>> Octopus-users mailing list
>> Octopus-users at tddft.org
>> http://www.tddft.org/mailman/listinfo/octopus-users
>>
>