Re: [FreeON-devel] problem of compiling Modules/LinAlg.F90 using mpif90 (backend compiler ifort)

[Gao Bin]

Hi, Matt

Thank you again for your help!

> If you are not using NEB, you shouldn't need
> "--enable-parallel-clones" either. It will simplify your life to leave
> it off.
> clones == NEB images.

I see, I have disabled parallel-clones. Moreover, as mentioned in your last 
email:

> Also, if you are running on SMP type boxes, it would be much easier to just
> throw in a few open mp pragmas to get the same (or better) effect....
> I'd be much more interested in trying to support that as an interim
> approach rather than trying to fix the broken MPI.

But I did not find any openmp directive in the source codes. What should I do 
to use FreeON, let us say, in a node with 8 CPUs using openmp?

> (1) You could get these matrices with a little effort.  Since they may
> be big, getting them well will depend
> on the format you want them in.  This is something I'm willing to help you 
> with.

My code could read HDF5 file, so it would be better that I could have the 
atomic orbital density matrix and Fock matrix in a HDF5 file after SCF 
calculations. Should I modify the code?

> (2) It really depends on what kind of molecule.  If you are computing
> something with a very small band gap,
> it will be hard.  So, like graphene or a zero gap nano tube will be
> hard.  If its more protein like and well behaved,
> 200-300 atoms with 6-31G** is certainly possible.  It also depends on
> how many digits in the total energy and
> forces you need, how fast the calculations will be.  You can probably
> do more than
> 3-4K atoms with 6-31G (depending on memory, # of digits you need, your
> patience etc).

I only need SCF calculations, and will probably require 5-6 digits. The 
molecule is like protein, so the band gap should be OK. Our computer clusters 
here either have 8 CPUs or 12 CPUs per node (around 2.7 GHz), with 16GB or 24 
GB memory. May I ask, from your experience, how long a calculation of 3-4K 
atoms with 6-31G will take on this machine? Thank you.

> (3) 6-D, 10-F Cartesians.  The indexing is localized to the file
> Modlues/Indexing.F90.
> Look at the function, LMNDex(L,M,N).

I have looked the code, and tried the following simple routine to print the 
indices of f-orbitals.

  LMN = 3
  do N = 0, LMN
    do M = 0, LMN-N
      L = LMN-(M+N)
      print *,LMNDex(L,M,N),L,M,N
    end do
  end do

what I got are

          11           3           0           0
          12           2           1           0
          13           1           2           0
          14           0           3           0
          15           2           0           1
          16           1           1           1
          17           0           2           1
          18           1           0           2
          19           0           1           2
          20           0           0           3
which look like FreeON starts from xxx to the yyy corner, and then xxz down to 
yyz, and finally the zzz corner. This is great, since my code uses the same 
indexing.

> I would suggest just trying out a light weight trial run of what you
> want to do (ie STO-2G),
> so that you don't make a big investment, only to find something big
> missing at the end.  I'm happy to
> help to an extent, and  likely to help more if you give back by
> commenting in  the code etc as
> you discover things  :)

Yes, I should first try small basis set to get some idea about the CPU time and 
memory usage. Thank you again for your help.

Cheers

Gao