[molpro-user] some help

[Pedro Jorge Caridade]

Dear Molpro users
After some years using non-parallel molpro, my group bought the parallel 
version to run on a AMD Opteron cluster. Following the instructions on 
the installation guide, I compiled Global Arrays with tcgmsg (without 
mpi). I would like to know the following:

1. I have seen some discussion about local scratch and ethernet-based 
scratch. Obviously, I would prefer doing scratch on each node. This is 
my first question. How molpro and GA works using node scratching?  If I 
have, for example, 300 Gb of scratch on a parallel job and 30 nodes, it 
means that each node will have 10 Gb of local scratch? And what about 
common scratch necessary to do some calculations, for example wave 
function?

2. I'm having some difficulties on running MP2 using parallel nodes, and 
the direct option. Are there some limitations on the number of nodes 
that can used for such calculation? Can we switch direct to non-direct 
on a calculation?

3. A more practical problem. In a do cycle, I would like to keep the 
wave function for each step, but in binary mode. I tried the 
file,2,$name.wfu, with "$name" a string variable, but it doesn't work, 
since the string is not evaluated. Can this be done be done this way? 
Are any solution for this? I have seen also some question about I/O of 
ascii wave function previously on the list, but didn't saw any reply.

Thank for any help
Pedro Caridade