parallel calculation

Posted on : June 29, 2016 (Wed) 21:38:01

by Brix

Hello,
I tried recently the code calculation to perform a DOS calculation on beta FeSi2 because I'm a user of VASP and I wanted to check out this code on a known density of states before performing further calculation. My DOS is absolutely wrong so I would like someone to help me improve my input file.
Here it is c----------------------FeSi2----------------------------------
go data/FeSi2
c------------------------------------------------------------
c brvtyp a c/a b/a alpha beta gamma
so 18.6684 , 0.78995 , 0.79369, , , ,
c------------------------------------------------------------
c edelt ewidth reltyp sdftyp magtyp record
0.001 1.5 sra mjw mag init
c------------------------------------------------------------
c outtyp bzqlty maxitr pmix
update 3 200 0.03
c------------------------------------------------------------
c ntyp
2
c------------------------------------------------------------
c type ncmp rmt field mxl anclr conc
Si 1 1 0.0 2
14 100
Fe 1 1 0.0 2
26 100
c------------------------------------------------------------
c natm
48
c------------------------------------------------------------
c atmicx atmtyp
0.1265171517094217 0.0446015906003794 0.2729726787751403 Si
0.8734828182905758 0.9553983943996229 0.7270272912248572 Si
0.8734828182905758 0.4553983943996229 0.7729727087751428 Si
0.1265171517094217 0.5446016056003771 0.2270273212248597 Si
0.8734828182905758 0.5446016056003771 0.2270273212248597 Si
0.1265171517094217 0.4553983943996229 0.7729727087751428 Si
0.1265171517094217 0.9553983943996229 0.7270272912248572 Si
0.8734828182905758 0.0446015906003794 0.2729726787751403 Si
0.6265171817094242 0.5446016056003771 0.2729726787751403 Si
0.3734828482905783 0.4553983943996229 0.7270272912248572 Si
0.3734828482905783 0.9553983943996229 0.7729727087751428 Si
0.6265171817094242 0.0446015906003794 0.2270273212248597 Si
0.3734828482905783 0.0446015906003794 0.2270273212248597 Si
0.6265171817094242 0.9553983943996229 0.7729727087751428 Si
0.6265171817094242 0.4553983943996229 0.7270272912248572 Si
0.3734828482905783 0.5446016056003771 0.2729726787751403 Si
0.1282169582562460 0.2736468728684613 0.0496612924990245 Si
0.8717830417437540 0.7263530971315433 0.9503387335009705 Si
0.8717830417437540 0.2263531271315387 0.5496612664990295 Si
0.1282169582562460 0.7736469028684567 0.4503387045009717 Si
0.8717830417437540 0.7736469028684567 0.4503387045009717 Si
0.1282169582562460 0.2263531271315387 0.5496612664990295 Si
0.1282169582562460 0.7263530971315433 0.9503387335009705 Si
0.8717830417437540 0.2736468728684613 0.0496612924990245 Si
0.6282169582562460 0.7736469028684567 0.0496612924990245 Si
0.3717830417437540 0.2263531271315387 0.9503387335009705 Si
0.3717830417437540 0.7263530971315433 0.5496612664990295 Si
0.6282169582562460 0.2736468728684613 0.4503387045009717 Si
0.3717830417437540 0.2736468728684613 0.4503387045009717 Si
0.6282169582562460 0.7263530971315433 0.5496612664990295 Si
0.6282169582562460 0.2263531271315387 0.9503387335009705 Si
0.3717830417437540 0.7736469028684567 0.0496612924990245 Si
0.0000000000000000 0.3072269447421974 0.3121266557251516 Fe
0.0000000000000000 0.6927730552578026 0.6878733142748459 Fe
0.0000000000000000 0.1927730552578026 0.8121266857251541 Fe
0.0000000000000000 0.8072269447421974 0.1878733442748484 Fe
0.5000000000000000 0.8072269447421974 0.3121266557251516 Fe
0.5000000000000000 0.1927730552578026 0.6878733142748459 Fe
0.5000000000000000 0.6927730552578026 0.8121266857251541 Fe
0.5000000000000000 0.3072269447421974 0.1878733442748484 Fe
0.2168152914588646 0.0000000000000000 0.0000000000000000 Fe
0.7831847385411379 0.0000000000000000 0.0000000000000000 Fe
0.7831847385411379 0.5000000000000000 0.5000000000000000 Fe
0.2168152914588646 0.5000000000000000 0.5000000000000000 Fe
0.7168152614588621 0.5000000000000000 0.0000000000000000 Fe
0.2831847085411354 0.5000000000000000 0.0000000000000000 Fe
0.2831847085411354 0.0000000000000000 0.5000000000000000 Fe
0.7168152614588621 0.0000000000000000 0.5000000000000000 Fe
c-----------------------------------------------------------



Furthermore I executed the calculation on 8 processors and 32 and I got the same calculation time.
Do you know where it come from?
Thank you.

[Re:03] parallel calculation

Posted on : June 29, 2016 (Wed) 23:20:00

by Administrator

Dear Brix,

also a important tips when you got a wrong DOS for large unit cell:
modify the subroutine source/etaopt.f

after the line
preta=1d0/vc**(2d0/3d0)/pi

add the following line
preta=1.7d0*preta

This increases the number of reciprocal k-points and R vectors used in the calculation of the structural Green's function, but eliminates a strange behavior in DOS calculations. Concerning the number of processor, as I mentioned the code is using only OpenMP. Therefore, mpirun -np does not work unfortunately. Although there exists a version of AkaiKKR which supports mpirun, we have not yet opened it.

KKR admin

[Re:02] parallel calculation

Posted on : June 29, 2016 (Wed) 23:04:21

by Brix

Thank you very much for your answer it will help me a lot.
I checked on my output and I see 8 threads detected but I ran it on 4 processors of 8 threads each by using mpirun -np do you have an idea of what could have been wrong?

[Re:01] parallel calculation

Posted on : June 29, 2016 (Wed) 22:28:22

by Administrator

Dear Brix,

(1) if the system is beta-FeSi2, it contains 2 types of Si and 2 types Fe. Therefore, you have to introduce Si1, Si2, Fe1, and Fe2 for correct calculation. Check carefully if your input is all right. If you atomic coordinates refer to the a,b,c or x,y,z axes, a,b,c or x,y,z should be attached to the values, otherwise the numbers are considered as just the cartesian coordinate. (2) Second, the system is rather large, start with bzqlty=0 (3 is specified in your output) and pmix=0.005~0.01. (3) After having attained considerable improvement of convergence (may be 20 iteration if pmix=0.01 is used), you may increase pmix. (4) It might be safer to always use "2nd" instead of "init". (5) The code is using OpenMP (thread parallel) and automatically detects the number of cores available. Check the number of threads that is output in the end of the output.