[Pw_forum] Warning: negative or imaginary starting charge

Axel Kohlmeyer axel.kohlmeyer at theochem.ruhr-uni-bochum.de
Mon Oct 18 10:15:44 CEST 2004


On Sun, 17 Oct 2004, Aleksandr Tkachenko  wrote:


dear aleksandr,

i may have recently stumbled upon a similar problem. 
in my case it turned out, that the fftw headers and
libraries did not match. i had compiled and installed
a custom fftw library without the type prefix. since
espresso also has an fftw header (which defines a 
double precision ABI), it can easily happen, that you
compile for a double precision fftw, but link to a
single precision library. in my case that produced
NaNs, but it only fully showed after a fresh recompile,
as some parts were previously compiled with a matching
header.

so you could check where you pick up the fftw headers 
and libraries.

in that light, it may also be a good idea to rename 
the internal fftw header file, and add support to
prefer 'type-renamed' header (dfftw.h) and library
(libdfftw.a) to the default names if available.

best regards,
	axel kohlmeyer.


AT> Dear Eyvaz:
AT> 

AT>  What seems strange to me is that it takes 26 iterations for me to
AT> converge al.scf.david input while it takes only 3 iterations in the
AT> "reference" case. Another thing is that cu.scf.cg and few other tests
AT> from example01 do not converge at all with "too many bands are not
AT> converged" error.
AT>  It seems that the initial density is not same in my case. Could it be some 32-64 bit or endian issue in the compilation?

AT> 
AT> Best Regards,
AT>  Alexander
AT> 
AT> 

-- 


=======================================================================
Dr. Axel Kohlmeyer                        e-mail: axel.kohlmeyer at rub.de
Lehrstuhl fuer Theoretische Chemie          Phone: ++49 (0)234/32-26673
Ruhr-Universitaet Bochum - NC 03/53         Fax:   ++49 (0)234/32-14045
D-44780 Bochum  http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/
=======================================================================





More information about the Pw_forum mailing list