[CPMD-list] CPMD BUG?

[Boris Kiefer]

Dear all:

I am new to business using CPMD - MD.  To familiarize myself with
the code and its variety of options I tried to compile the cpmdv35_3
under pgf90 on linux. The compiler worked fine and generated the
executable cpmd.x.

Running the program for a somewhat artificial test example for
geometry optimization of silicon in the diamond phase (Si-I) I
found that the program stopped with a segmentation fault after
the initial setup and probably when it the SCF loop.
In any case the error seems to be associated with the
BLAS function DDOT. In the routine CSIZE the function DDOT is
called with three arguments. However, DDOT is called from
other routines with 5 arguments. According to the description of
DDOT it needs to be called with 5 arguments. Can anybody
confirm this finding?

After fixing this problem cpmd.x continued just fine.  DDOT
calculated the dot-product of two vectors and in this case the
stride should be one. I think the correct calling sequence for
DDOT should be:

                     DDOT(a,d,c) ==> DDOT(a,b,1,c,1)

Cheers -- Boris
--
--------------------------------------------------------------
Boris Kiefer                 Phone:  USA-609-258-3261
Department of Geosciences        FAX  :  USA-609-258-1274
Princeton University             EMAIL:  bkiefer at princeton.edu
Princeton, NJ 08544-1003
--------------------------------------------------------------