[Pw_forum] pwscf 3.* good for production?

[Paolo Giannozzi]

On Oct 3, 2006, at 24:51 , Antonio Suriano wrote:

> I just wanted to ask if the improvement  in the algorithms could
> "translate" by some mRy results for one atom calculation.

of course it can. Algorithmic changes in the calculation of matrix
elements of pseudopotentials - integration, interpolation etc - can
lead to minor differences in absolute energies (but not in energy
differences or in physically significant quantities). The same applies
to exchange-correlation functionals, where various parameters
are used to decide when to cut off diverging terms. Even purely
numerical differences between different machines can lead to
slightly different total energies. Even rearranging do loops may
lead to small differences.

As a rule it is safer to use the same version of a code to perform
a set of calculations. If it is needed or useful to move to another
version, one has to be careful when calculating quantities (such
as cohesive energies) that depend on small differences of large
numbers.

Paolo