[CPMD-list] discrepant total energies

[Kozin, I (Igor)]

Hi Axel,
thanks for your comment.

> IK> I was recently pointed out to the difference in 
> IK> the initial total energies computed using different
> IK> compilers. Particularly for cpmd-test/std/wat32/inp-1
> IK> IBM           -539.79205456 A.U.  (as provided)
> IK> Intel         -539.79205456 A.U.
> IK> PathScale     -539.79205494 A.U.
> IK> PGI           -534.04296070 A.U.
> IK> 
> IK> There is not much difference using Intel or PGI
> IK> on 32- or 64-bit platforms or using various compiler
> IK> options. Fortunately everything converges nicely to 
> IK> -551.05084246 A.U.
> IK> But still the question is why the initial PGI value
> IK> is so different?
> 
> the pgi compiler is known to occasionally miscompile stuff.
> i have seen it with other plane-wave pseudopotential
> packages as well. in this case it seems to be code to compute
> the slater orbitals for the inital guess.
> there was another post concerning this issue recently in
> combination with vanderbilt pseudopotentials. please check 
> the mailing list archive. i did a cross-check of that
> specific example and found, that PGI compiles correctly
> in case of turning off the optimization (-O0). it would
> be useful to know, _which_ parts of cpmd are actually 
> miscompiled, so we can take care of it in the PGI configuration
> files.

Yes, it would :)
Fortunately PGI 6.0 is out and I am happy to report that even 
with -fastsse 
 (K+E1+L+N+X)           TOTAL ENERGY =         -539.79205494 A.U.

Best,
Igor