BigDFT

Full CI calculation

Asked by Jyh-Pin Chou

Hello all

Thanks BigDFT group, it is a nice ab initio calculation software. For some reason, I would like to do wavefunction analyzing. Therefore, I start from a simple example, e.g. H2, and do bond length calculation in Hartree-Fock level. I do full CI calculation by using BigDFT output wavefunction and then add electron-ion potential (1/r) to obtain the eigenvalue. The results are here

http://imgbox.com/TXRFL5cG

As you can see, I used two different hgrid values 0.298765 and 0.3. The black curves in (b) and (d) are the total energy obtained from BigDFT results. At beginning, I used 0.3 hgrid, obtain a smooth E(FCI). But the curve becomes fluctuated after adding 1/r and it seems to have a period of 3. When I slightly change hgrid to 0.298765, this fluctuated condition is more or less solved. This results show a grid-dependent feature for BigDFT wavefunction. I was wondering is that true? Is that a kind of weakness for real space calculation? Or, anyone has suggestion?

Question information

Language:
English Edit question
Status:
Solved
For:
BigDFT Edit question
Assignee:
Thierry Deutsch Edit question
Solved by:
Jyh-Pin Chou
Solved:
Last query:
Last reply:
Revision history for this message
Jyh-Pin Chou (jpchou) said : 		#1

Here are the input parameters

 dft:
   hgrids: 0.45
   rmult: [3.0, 7.0]
   nrepmax: accurate
   nspin: 2
   mpol: 0
   inputpsiid: 2
   output_denspot: 0
   disablesym: No
   norbv: 2
   ixc: 100
 perf:
   exctxpar: BC
 lin_general:
   output_wf: 1
 lin_basis:
   nit_ig: 100
 output:
   orbitals: text
 lin_basis_params:
   H:
     nbasis: 1
     ao_confinement: 5.0E-003
     confinement: [1.0E-003, 0.0]
     rloc_kernel: 7.0
     rloc_kernel_foe: 20.0
 posinp:
   units: bohr
   positions:
   - H: [ 0.0, 0.0, 0.0]
     IGSpin: 1
   - H: [ 1.4, 0.0, 0.0]
     IGSpin: -1
 psppar.H:
   Pseudopotential type : HGH-K
   Atomic number : 1
   No. of Electrons : 1
   Pseudopotential XC : 1
   Local Pseudo Potential (HGH convention):
     Rloc : 0.2
     Coefficients (c1 .. c4): [-4.1802368, 0.72507482, 0.0, 0.0]
   Radii of active regions (AU):
     Coarse : 1.463418464633951
     Fine : 0.2
     Coarse PSP : 0.

Revision history for this message
Thierry Deutsch (thierry-deutsch) said : 		#2

Dear Jyh-Pin,

  It is tricky to answer to your question because I have no information about what you did to calculate full CI energy and eigenvalue especially how you apply the electron-ion potential 1/r.

First of all with the value 1 for output_wf, the files of wavefunctions contain the wavelet coefficients and not the values of the wavefunctions on the grid points in real space. You can convert the files in the data directory using the tool bigdft-tool as:

bigdft-tool -a export-wf FILE:
    Export the compressed wavefunction from FILE to a scalar-field
    representation in Cube format.

Then you will have coefficient in real space and could apply 1/r.

Real space calculations are not generally invariant by translation but in the case of BigDFT, because we use Daubechies wavelets to the order 8, polynomial functions are exact up to the order 7 in our basis set so we preserve invariance by translation for this class of functions i.e. up to the order 7.

Regards, Thierry

Revision history for this message
Jyh-Pin Chou (jpchou) said : 		#3

Dear Thierry

You've made a point precisely and concisely!!
What I'm doing now is to use the wavefunction files directly obtained from the standard output of BigDFT, which are the the wavelet coefficients as you said instead of wavefunction coefficient I needed. This is a big misunderstanding! Thank you very much for solving my problem.

Regards, JP