We have tested our method successfully on a number of small molecules,
and computed their NMR chemical shift[1]. We are now
applying it to larger, more interesting systems. We have performed
calculations on ice (12 atoms, matrix size N=8500, number of
eigenvalues m=16), a diamond surface (22 atoms, N=12600, m=41),
amorphous carbon (76 atoms, N=39000, m=134), and liquid water (96
atoms, N=64000, m=128), all with excellent outcome. Just for the
fun of it, we show in Figure 5 a plot of the trace of
the susceptibility tensor 
as a function of position in the
plane of a water molecule.
Figure 5: Trace of the susceptibility tensor 
as a function of position in the plane of a water molecule. We were
surprised about the fast decay of trace(
) away from the oxygen
atom, and the anisotropy of the distribution caused by the hydrogen
atoms.
Unlike many non-commercial scientific codes, our program is very user friendly, since it has dynamic memory allocation throughout. Also, the data layout is done on the fly and entirely by the program itself, such that parallel runs do not require additional user input. Currently there are about eight active users. The largest calculations (liquid water) typically run within 4-6 hours on 32 processors of the Cray T3E. On the SGI PowerChallenges, we normally run on 16 processors in dedicated mode. Our total group uses about 50000 node hours of parallel CPU time a year. We have accomplished the transition from vector to parallel architectures successfully. The available parallel platforms (T3E, PowerChallenge, SP2) are all stable production machines. In fact, we have not bothered yet to port the code to the vector J90/C90 platforms, although they will certainly perform well on the FFT.