mirror of https://github.com/abinit/abinit.git
119 lines
3.2 KiB
Plaintext
119 lines
3.2 KiB
Plaintext
Tutorial on Many-Body calculations in parallel
|
|
|
|
alpha-quartz SiO2
|
|
-----------------------------------------
|
|
|
|
First test: Generation of the WFK file
|
|
|
|
Standard GS + NSCF calculation to produce the WFK file
|
|
with a Gamma-centered 4x4x3 mesh (converged calculations
|
|
would require a much denser sampling e.g. 6x6x5 should be ok).
|
|
|
|
This test does not scale very well with the number of processors since we have
|
|
only 9 k-points in the IBZ and we don't use the paralkgb option.
|
|
|
|
Also the number of bands in the WFK file is not enough to converge the GW
|
|
calculations (1200 bands for the screening, 800 for sigma would be needed).
|
|
|
|
How to run the test:
|
|
|
|
(mpirun ...) abinit < tmbt_1.files > log_1
|
|
|
|
then rename the WFK file
|
|
|
|
mv tmbt_1o_DS2_WFK 443_WFK
|
|
|
|
In the next steps, we will be using this file to start the GW calculations.
|
|
|
|
-------------------------------------------
|
|
|
|
Second step: Screening calculation with the Adler-Wiser formulation and gwpara=2.
|
|
|
|
First create a symbolic link for the WFK file
|
|
|
|
ln -s 443_WFK tmbt_2i_WFK
|
|
|
|
change the files file, then run the code as usual
|
|
|
|
(mpirun ...) abinit < tmbt_2.files > log_2
|
|
|
|
The RPA polarizabilty includes 26 empty bands therefore
|
|
this test should scale very well up to 26 processors.
|
|
Mind that, for optimal speedup, the number of processors should divide 26.
|
|
|
|
At the end of the calculation we have to rename the output SCR file
|
|
|
|
mv tmbt_2o_SCR 443_ppm_SCR
|
|
|
|
The prefix _ppm means that this SCR file can be used for GW calculations
|
|
within the plasmon-pole approximation (only two frequencies have been computed).
|
|
|
|
-------------------------------------------
|
|
|
|
|
|
Third step: Screening calculation with the Hilbert transform method and gwpara=2
|
|
|
|
First create a symbolic link for the WFK file
|
|
|
|
ln -s 443_WFK tmbt_3i_WFK
|
|
|
|
change the files file specifying the correct index associated to the test,
|
|
then run the code as usual
|
|
|
|
(mpirun ...) abinit < tmbt_3.files > log_3
|
|
|
|
This test uses the same number of empty states as the previous step
|
|
hence it will scale up to 26 processors.
|
|
|
|
At the end of the calculation we have to rename the output SCR file
|
|
|
|
mv tmbt_3o_SCR 443_cd_SCR
|
|
|
|
The prefix _ppm means that this SCR file can be used for GW calculations
|
|
withing the contour-defomation technique.
|
|
|
|
-------------------------------------------
|
|
|
|
|
|
Fourth step: GW calculation with the plasmon-pole approximation and gwpara=2.
|
|
|
|
First create two symbolic links for the WFK and the SCR file
|
|
|
|
ln -s 443_WFK tmbt_4i_WFK
|
|
ln -s 443_ppm_SCR tmbt_4i_SCR
|
|
|
|
then change the files as usual and issue
|
|
|
|
(mpirun ...) abinit < tmbt_4.files > log_4
|
|
|
|
The number of bands is 50, hence the correlation part will scale up to 50 nodes.
|
|
|
|
The same input file can be used for performing CD calculations provided that we link the correct SCR file using
|
|
|
|
ln -s 443_cd_SCR tmbt_4i_SCR
|
|
|
|
and we replace the line
|
|
|
|
gwcalctyp 0 ppmodel
|
|
|
|
with
|
|
|
|
gwcalctyp 2
|
|
|
|
in the input file
|
|
|
|
-------------------------------------------
|
|
|
|
Fifth step: GW calculation with the plasmon-pole approximation and gwpara=2.
|
|
|
|
First create a symbolic link for the WFK and the SCR file
|
|
|
|
ln -s 443_WFK tmbt_5i_WFK
|
|
ln -s 443_ppm_SCR tmbt_5i_SCR
|
|
|
|
change the files file than run the code as usual
|
|
|
|
(mpirun ...) abinit < tmbt_5.files > log_5
|
|
|
|
This input file should scale up ... processors.
|