The typical directory structure should have a main directory which will
include the script "Link" to create the executable.  It will have a 
subdirectory "subs" which will hold the FORTRAN code and the shell
script "Fort" to compile the code.  The main directory will also have a 
subdirectory "obj" to hold the object files created in compilation.
A further subdirectory off of main will be "data" which will hold the
arndt92 phase shifts.  An input directory must be created, and a further 
subdirectory within it for each target nucleus.  Each of these, then, will
hold the input files.  To mirror the input directory, there must be an
output directory with the same target subdirectories.

A typical command to run the code would look like this:
                   arndtp he3/testprot&
The output then goes to out/he3/testprot.  Modification of the input 
switches is done in the above mentioned input file.  In order to speed
the code for parallelization, we have included an inconvenient improvement
in the natrix inversion.  If the number of grid points, ngp in the input
file is changed, the matrix zf must be re-dimensioned to (2*(ngp+1),2*ngp+1))
in lpotp.f, tmatrx.f, and fmatrx.f.  This is the result of using the zgetri
subroutine rather than cmatin to do the matrix inversion.  To do the comp-
ilation, the command "Fort subroutine" would be issued on our system in the 
subs subdirectory.  This will move subroutine.o to obj.  The command Link
is finally issued from the main directory, creating the executable.  These
commands, which use xlf, are included.  Finally, the code has been specialised
for spin 1/2 x 1/2, I doubt that the old 0 x 1/2 switches are all still in 
place.