Non-integer total number of Alpha/Beta Electrons

[M. Brehm]

Dear Community,

I am performing DFT calculations of a solid-state lattice with CP2k, using 
the classical mixing scheme and Fermi-Dirac-Smearing, like usual for 
lattices. As the total number of electrons in my system is odd, I perform 
unrestricted DFT calculations, setting the spin multiplicity to 2. In the 
beginning, CP2k displays 288 alpha electrons and 287 beta electrons, like 
it should be. The SCF cycle converges, and in the end, I print out the MO 
energies and occupation numbers. Suddenly, there are 287.5 alpha electrons 
and 287.5 beta electrons! For my understanding, the total number of 
alpha/beta electrons is not a degree of freedom of a unrestricted DFT 
calculation, and should remain fixed during the calculations. How can it be 
that in the end I get these fractional electron numbers? With slightly 
different systems or geometries, this is not the case, and the total 
electron count for alpha/beta remains constant.

I assumed that it could have something to do with the initial SCF guess 
(poor guess might lead to convergence into a local minimum which is not the 
ground state), but I re-ran the calculation with SCF guess "random", and 
still get exactly the same results.

My input file and output file are attached to this posting.

Any hints/comments are appreciated.

Best regards,
Martin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20150217/8eacf00d/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: tio2.inp
Type: chemical/x-gamess-input
Size: 6661 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20150217/8eacf00d/attachment.inp>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: tio2.out
Type: application/octet-stream
Size: 69265 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20150217/8eacf00d/attachment.obj>