[Pw_forum] Looking for some guidance with failing tests

[Paolo Giannozzi]

On Mon, May 15, 2017 at 6:52 PM, Barry Moore <moore0557 at gmail.com> wrote:

> pw_langevin - langevin.in: **FAILED**.
> pw_langevin - langevin_smc.in: **FAILED**.
> [...]
> Is this expected?

yes. The "langevin" test is stochastic and you get different results
if you run in serial and in parallel. It should actually be removed.

> There are about 30 failing tests total. They are all "very close" to the correct answer, but not the correct answer.

Then they are correct. There are two reasons why the results may differ:
- randomized initial wavefunctions depend upon the number of
processors and may result in slightly different paths to convergence
and to a slightly different end point. Such difference may be
minimized by tightening convergence criteria
 (conv_thr)
- Those oh so fast mathematical libraries are not always oh so accurate

> I have never run QE in my academic career

time to start

Paolo

-- 
Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222