Energy is NaN with CUFFT

[Axel]

On Mar 14, 9:21 am, Ondrej Marsalek <ondrej.... at gmail.com> wrote:
> Hi,
>
> I have a problem with CUFFT. All seems fine except for a "small
> detail" - some energies in the output are NaN, zero, or infinity. The
> strange thing is that SCF seems to converge and reports sensible
> numbers in the "convergence" column. It looks like this:

hi,

i don't see this. just recompiled on the tesla machine at uiuc.

as usual, please provide the smallest possible input that reproduces
it,
as well as the compiler flags and hardware specs that you were using.
otherwise it is impossible to track it down.

please note, that the current cuda support implementation is to be
considered alpha quality software. it expects users to use it
correctly
and has next to no heuristics installed to prevent bad things to
happen.

i just ran the a minimal Ar.inp, and 32 and 64 water benchmark inputs
and
they all worked fine for me, after making the changes required for
cuda
(%GLOBAL%CUDA%MEMORY).

cheers,
   axel.



>
>   ----------------------------------- OT --------------------------------------
>      1  OT DIIS        0.15E+00      9.652     0.0017576776                 NaN
>      2  OT DIIS        0.15E+00     12.499     0.0007020786                 NaN
>      3  OT DIIS        0.15E+00     12.474     0.0003829465                 NaN
>      4  OT DIIS        0.15E+00     12.426     0.0001190093                 NaN
>      5  OT DIIS        0.15E+00     12.452     0.0000853514                 NaN
>      6  OT DIIS        0.15E+00     12.439     0.0000527586                 NaN
>      7  OT DIIS        0.15E+00     12.459     0.0000396126                 NaN
>      8  OT DIIS        0.15E+00     12.452     0.0000282061                 NaN
>      9  OT DIIS        0.15E+00     12.470     0.0000229232                 NaN
>     10  OT DIIS        0.15E+00     12.483     0.0000185011                 NaN
>     11  OT DIIS        0.15E+00     12.494     0.0000153723                 NaN
>     12  OT DIIS        0.15E+00     12.462     0.0000133772                 NaN
>     13  OT DIIS        0.15E+00     12.462     0.0000115968                 NaN
>     14  OT DIIS        0.15E+00     12.515     0.0000104773                 NaN
>     15  OT DIIS        0.15E+00     12.479     0.0000093015                 NaN
>
>   *** SCF run converged in    15 steps ***
>
>   Total electronic density (r-space):          0.0000000000      256.0000000000
>   Total core charge density (r-space):       255.9999999917       -0.0000000083
>   Total charge density (r-space):                                255.9999999917
>   Total charge density (g-space):                                      Infinity
>
>   Overlap energy of the core charge distribution:              0.00000059138020
>   Self energy of the core charge distribution:             -1419.10895989560481
>   Core Hamiltonian energy:                                    84.83397930334846
>   Hartree energy:                                                           NaN
>   Exchange-correlation energy:                                 0.00000000000000
>   Dispersion energy:                                          -0.08192283398992
>
>   Total energy:                                                             NaN
>
>   outer SCF iter =    1 RMS gradient =   0.93E-05 energy =                  NaN
>   outer SCF loop converged in   1 iterations or   15 steps
>
>  ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):                              NaN
>
> Is it apparent what the problem is or should I just send in a testcase?
>
> Best,
> Ondrej