Tech isolv bse

[sangallidavide]

• New tddft implementation for eh_space. Collisions and kernel are computed in G-space. Older implementation in R-space is kept for checking purpposes.

• TDDFT_ALDA_R_space.o

• TDDFT_ALDA_eh_space_G_collisions_L.o

• TDDFT_ALDA_eh_space_G_collisions_R.o

• TDDFT_ALDA_eh_space_G_kernel.o

• TDDFT_ALDA_eh_space_R_kernel.o

  
  The new implementation is much faster for 2 reasons:
  a) Collisions are computed before and then assembled together, thus 2N times instead of N^2 times
  b) Collisions and kernel are done in G-space, and ng_closed << fft_size. Moreover a cutoff can be used

• New slepc algorithm takes into account the structure of the BSE matrix with coupling. Developed by the slepc team. See #66

• Configure upgraded to work with development version of slepc, and ndb.BS_PAR file now contains also the BS_energies and the BS_table. This makes it possible to fully reconstruct the BSE externally to yambo. (See #83 )

[sangallidavide]

The robot are green in serial, there is one test failing in parallel. I'm checking this

https://media.yambo-code.eu/robots/tech-isolv-bse/unimi-XPS-8930.php

[sangallidavide]

The robot are green in serial, there is one test failing in parallel. I'm checking this

Fixed bug in parallel. Now the robots are green:

https://media.yambo-code.eu/robots/tech-isolv-bse/unimi-XPS-8930.php

[attacc]

Just to understand these changes require a special branch of PETSc/Slepc? And I guess the flag petsc-branch will disappear when a new version of PETSc will be available?

[sangallidavide]

Just to understand these changes require a special branch of PETSc/Slepc?

All the changes are already available with the released Petsc/Slepc version linked in yambo

[attacc]

Ciao Davide

may you test these changes with INTEL I got a fail in the test:

C6H3Cl3

[serial] 06_BSE_cpl_slepc1: optics T-space static screening 4.5s o-06_BSE_cpl_slepc1.alpha_q1_slepc_bse [ NO in OUTPUT ] o-06_BSE_cpl_slepc1.jdos_q1_slepc_bse [ NO in OUTPUT ]

ciao Claudio

[sangallidavide]

Which version of the slepc library did you use?

[attacc]

You are right... I'm using an old version... I will update it and run it again

[sangallidavide]

I'll see if I can add a check on the slepc version