Strange problem with ISIF=3 and VASP 5.2

[CStar]

I compiled VASP versions 4.6 and 5.2 in serial using similar compiler / flag combinations; I used ifort version 11.1, mkl version 10.2, and no optimization flags (-O0).

If I run any calculations with ISIF=3 (for example, using the input files contained in the directory Tutorial/Example1/ ), then I have problems with version 5.2.
The calculation runs to completion, but if I compare the lattice vectors in the CONTCAR files output by versions 4.6 and 5.2 then I see this:
(Version 4.6:)
Silicon !Comment line
5.40000000000000
0.0000000000000000 0.5061065055690370 0.5061065055690370
0.5061065055690370 0.0000000000000000 0.5061065055690370
0.5061065055690370 0.5061065055690370 0.0000000000000000
(Version 5.2:)
Silicon !Comment line
5.40000000000000
0.0000000000000000 0.5077644554442858 0.5000000000000000
0.5077644554442858 0.0000000000000000 0.5000000000000000
0.5077644554442858 0.5077644554442858 0.0000000000000000

I assume that the output from version 4.6 is correct. The cell does not appear to have been updated correctly for the version 5.2 calculation. I have tried similar 'ISIF=3' calculations with alternative systems, and have seen the same behaviour - version 4.6 seems to behave fine, while the third components of the lattice vectors are not updated correctly for version 5.2. Also, for version 5.2 (but not 4.6) I get the error message to standard output

BRMIX: very serious problems the old and the new charge density differ
old charge density: 8.00001 new 8.00023

I assume that this problem is secondary, and results from the fact that the cell did not update correctly. I have tried compiling with ifort version 10 instead, but the problem remains. The problem is also present in parallel runs using version 5.2. Has anyone else experienced problems like these? I have been able to reproduce the problem (i.e. with version 5.2 only) over multiple machines.

[forsdan]

This sounds very strange. Just to clarify: Is the 3rd component updated at all for any of the systems you have investigated? If not, are you sure that you are using the original source code for vasp 5.2? Because to me the results are consistent with what would happen if you e.g. have modified constr_cell_relax.F to only allow for relaxation in the x and y directions.

Cheers,
/Dan

<span class='smallblacktext'>[ Edited Tue Oct 19 2010, 12:16PM ]</span>

[CStar]

Thanks for the reply forsdan,

I guess the source files I had had become somehow corrupted (no idea how that happened). After copying the original source files again and recompiling, everything is behaving normally.

Cheers