[Pw_forum] pwscf tests/examples question

[Paolo Giannozzi]

On May 27, 2007, at 13:33 , mbaris at metu.edu.tr wrote:

> when I edit the environment_variables file to use mpich2
> (PARA_PREFIX="mpiexec -n 4") and run run_all_examples,
> the batch halts at example05 in running pp.x

there are a few things that you cannot do in parallel using pp.x

> On a single cpu (i.e. with PARA_PREFIX empty) upto example34
> can be run, which fails stating vdw.x can not be found.

it is not there, it has to be separately downloaded. If you do not
need it, just ignore it

> <      one-electron contribution =     7.88010983 Ry
> <      hartree contribution      =     1.91301246 Ry
> <      xc contribution           =    -7.07049550 Ry
> ---
>>      one-electron contribution =     7.88010934 Ry
>>      hartree contribution      =     1.91301318 Ry
>>      xc contribution           =    -7.07049573 Ry
> As you can see, the contribution weights are different, however the  
> resulting energy is the same!!!

these are minor numerical differences that it is hard to avoid. Since  
the
total energy is variational (i.e. the first-order error wrt the error  
in the
charge density vanishes), it is less sensitive to minor differences  
in the
calculated rho than the various contributions

> check_example seems to ignore this.

check_example was written exactly to ignore such small numerical  
differences

> Lastly, with the encouragement of the manual I have installed intel  
> MKL, but instead of seeing a performance leap, the code runs  
> slightly slower now (order
> of 0.01 sec). I guess more fine tuning is required on my side.

it is unlikely to see much gain in small calculations, but in serious  
calculations
the advantage of using MKL will be visible

Paolo
---
Paolo Giannozzi, Democritos and University of Udine, Italy