[CPMD-list] Results on different machine!
Juerg Hutter
hutter at pci.unizh.ch
Tue Nov 30 11:09:56 CET 2004
Hi
convergence can depend (but doesn't have to) critical
from the initial guess. The initial guess in CPMD
depends slightly on the numerics (=different on different
computers) and also on the number of CPU's. In addition,
the DIIS optimizer uses a data compression algorithm
that is number of CPU and also dependent on the special
numerics of the computer.
Therefore it can happen, that the same run on different computers
and or different numbers of CPU shows different convergenc
behavior. As long as you arrive at the same result
in the end there is nothing to be worried about.
reagrds
Juerg Hutter
----------------------------------------------------------
Juerg Hutter Phone : ++41 1 635 4491
Physical Chemistry Institute FAX : ++41 1 635 6838
University of Zurich E-mail: hutter at pci.unizh.ch
Winterthurerstrasse 190
CH-8057 Zurich, Switzerland
----------------------------------------------------------
On Mon, 29 Nov 2004, Adrain Zhou wrote:
> Dear all,
>
> I have made a test on wavfunction optimization(CPMD-3.9.1) on AMD64 opteron machine using pgf compiler and HP-COMPAQ alpha machine. I found the convergence is quite different. For the same input file, It will take 106 steps to get convergence on AMD opteron 64 machine and 42 steps on HP-COMPAQ alpha machine. The total energy is almost the same. Could anybody tell me why the convergence is so different?
>
> Any comments are highly appreciated. Thanks
>
> Regards,
> Adrain
>
> PS. output on AMD opteron 64 machine,
> .....
> .....
> 84 4.274E-05 5.622E-07 -592.031939 -5.275E-06 18.13
> 85 4.071E-05 5.489E-07 -592.031945 -5.508E-06 17.92
> 86 3.877E-05 4.758E-07 -592.031949 -3.949E-06 18.05
> 87 3.571E-05 4.757E-07 -592.031954 -4.749E-06 18.02
> 88 3.249E-05 4.701E-07 -592.031958 -4.670E-06 17.88
> 89 3.041E-05 4.002E-07 -592.031961 -2.988E-06 17.87
> 90 2.683E-05 4.148E-07 -592.031965 -4.210E-06 17.92
> 91 2.359E-05 3.833E-07 -592.031969 -3.171E-06 18.11
> 92 2.156E-05 2.996E-07 -592.031970 -1.792E-06 17.93
> 93 1.786E-05 3.259E-07 -592.031973 -2.608E-06 18.01
> 94 1.593E-05 2.973E-07 -592.031974 -1.403E-06 18.14
> 95 1.483E-05 2.333E-07 -592.031975 -8.897E-07 18.00
> 96 1.244E-05 2.449E-07 -592.031976 -1.176E-06 18.10
> 97 1.119E-05 2.090E-07 -592.031977 -5.425E-07 17.88
> 98 9.586E-06 1.906E-07 -592.031978 -5.930E-07 17.81
> 99 8.653E-06 1.809E-07 -592.031978 -4.799E-07 17.73
> 100 8.406E-06 1.563E-07 -592.031978 -2.905E-07 17.85
> 101 7.966E-06 1.550E-07 -592.031979 -3.474E-07 17.81
> 102 7.577E-06 1.416E-07 -592.031979 -2.152E-07 17.80
> 103 7.083E-06 1.317E-07 -592.031979 -2.226E-07 17.87
> 104 6.585E-06 1.325E-07 -592.031979 -2.616E-07 17.90
> 105 6.140E-06 1.161E-07 -592.031980 -1.586E-07 18.01
> 106 5.583E-06 1.092E-07 -592.031980 -1.865E-07 17.83
>
> output on HP-COMPAQ alpha machine,
> ........
> ........
> 23 2.976E-05 5.975E-07 -592.031962 -4.112E-06 29.99
> 24 2.377E-05 5.372E-07 -592.031965 -2.746E-06 29.96
> 25 2.845E-05 4.674E-07 -592.031967 -2.688E-06 29.87
> 26 2.843E-05 4.508E-07 -592.031970 -2.541E-06 29.92
> 27 1.965E-05 4.250E-07 -592.031971 -1.584E-06 30.07
> 28 1.777E-05 3.669E-07 -592.031973 -1.300E-06 30.14
> 29 1.751E-05 3.276E-07 -592.031974 -1.277E-06 29.98
> 30 1.512E-05 3.064E-07 -592.031975 -9.483E-07 30.18
> 31 1.401E-05 2.799E-07 -592.031976 -7.858E-07 30.19
> 32 1.277E-05 2.581E-07 -592.031976 -8.104E-07 29.91
> 33 1.161E-05 2.403E-07 -592.031977 -6.875E-07 29.88
> 34 1.051E-05 2.221E-07 -592.031978 -5.805E-07 29.98
> 35 9.457E-06 2.047E-07 -592.031978 -4.990E-07 30.01
> 36 9.499E-06 1.879E-07 -592.031979 -3.743E-07 29.83
> 37 9.637E-06 1.775E-07 -592.031979 -3.299E-07 29.98
> 38 8.222E-06 1.691E-07 -592.031979 -2.744E-07 29.83
> 39 8.242E-06 1.550E-07 -592.031979 -2.056E-07 30.18
> 40 8.419E-06 1.457E-07 -592.031980 -2.327E-07 29.78
> 41 7.796E-06 1.410E-07 -592.031980 -1.870E-07 30.02
> 42 5.769E-06 1.302E-07 -592.031980 -1.505E-07 29.98
>
>
>
>
>
> ---------------------------------
> Do You Yahoo!?
> 150ÍòÇúMP3·è¿ñËÑ£¬´øÄú´³ÈëÒôÀÖµîÌÃ
> ÃÀÅ®Ã÷ÐÇÓ¦Óо¡ÓУ¬ËѱéÃÀͼ¡¢ÑÞͼºÍ¿áͼ
> 1G¾ÍÊÇ1000Õ×£¬ÑÅ»¢µçÓÊ×ÔÖúÀ©ÈÝ£¡
More information about the CPMD-list
mailing list