Installation query using MPICH2 and gfortran compiler

[sudhakar]

I was trying to compile parallel version of vasp on a P4 cluster using MPICH2. The linker is mpif90 which used "gfortran" for compiling.

I had to change the syntax of some files (diolib.F, dlexlib.F, drdatab.F) in the vasp library to get libdmy.a . This was necessary as gfortran does not recognize "&" sign for continuation of a line.

After this compilation was smooth. But while running a test calculations using mpivasp, OUTCAR gives a warning that the distance between some atoms is very small. But this is not possible as this is a converged calculation from an earlier vasp run. The calculation actually converges to the right answer.

Can some one tell me why I am getting the warning message. Are there any compatibility issues using gfortran or changing the syntax might have caused this.

Thanks in advance..

[tjf]

[quote="sudhakar"]After this compilation was smooth. But while running a test calculations using mpivasp, OUTCAR gives a warning that the distance between some atoms is very small. But this is not possible as this is a converged calculation from an earlier vasp run. The calculation actually converges to the right answer.[/quote]

Check the actual positions of the ions in the output, and compare them to POSCAR an the previous run. (Have you copied CONTCAR from a previous run or something else?) Is this a single geometry calculation?

[sudhakar]

Hi tjf

The -ffree-form option works great. Using that I am able to compile without any problems. But still while running I get the warning that "distance between atoms is very small".

I am using a test case to calculate the bond length of H2 molecule. I have copied the converged CONTCAR from an earlier vasp run to POSCAR for testing. Also while running it gives an error that "BRMIX: very serious problems the old and the new charge density differ" which was not there in the earlier runs.

So I am wondering if compiling with gfortran causes this problem.

Thanks

[tjf]

What are the ionic positions in OUTCAR? How do they compare to the positions in POSCAR?

[priya]

1. Are you reading the WAVECAR converged from a previous run? If you are and have changed the number of KPOINTS from the earlier run, then I have seen the BRMIX warning. I think it is because the weights at each k-point have changed and so the crude integration that the program does initially to find number of electrons, doesn't work.

2. Is the nearest neighbour distance, that the program is writing out correct? The program probably has a preset criterion to print out the warning about the distance. I guess the optimization would have moved the atoms closer and that is why you didn't see it during the earlier run.

[sudhakar]

The ionic positions in the OUTCAR are the same as those in POSCAR. I am starting afresh and not reading any WAVECAR files. The nearest neighbor distance in the OUTCAR is 0.75 which is the same as the H2 bond length.

An interesting thing I did for comparison was to compare the serial versions of vasp compiled with gfortran and ifort. Vasp compiled with ifort did not give any warning messages but still had that BRMIX issue. Vasp compiled with gfortran gave me the warning and also the BRMIX error.

What this leads me to beleive is that there are some issues with vasp when compiling with gfortran. Unfortunately I can only use gfortran to compile mpivasp at the moment as we don't have mpich support for ifort.

[tjf]

I've done a little digging through the code and discovered that the warning about small distances is based on the read RWIGS from the potentials (plus 20%). Check that the RWIGS you are getting in OUTCAR is what you expect. Check also that the nearest neighbour tables printed by the two versions of the code are the same. There's some logic in that part of the code that I haven't fully absorbed.


By the way Vaspmaster, where is the IDIOT flag documented in the manual??? :-D