.
This is a visualization of some of the first data
to come off the new 128-node
Cray
T3E
recently installed at NERSC.
The still frames below show
part of an eigenmode of the Dirac operator. Since we
work in four dimensions the full structure cannot be
shown in one picture; these images show single three
dimensional slices. If your browser is Java-capable,
click on one of the images below to animate the frames
and see the full four-dimensional object as we loop
over the supressed dimension.
Big (1.5MB) version
Small (0.5MB) version
Technical details:
We use the simple Wilson plaquette action on a 32*32*32*64 lattice,
with beta=6.4, and staggered quarks. The boundaries are periodic
in each of the four dimensions.
The configuration was cooled slightly (20 steps)
to erase short-distance fluctuations. We used inverse iteration
to converge to the eigenvector with the smallest eigenvalue.
More precisely, we have isolated an eigenvector of Dslash-squared.
It is a linear combination of the pair of related lowest
eigenvectors of Dslash (with eigenvalues lambda and -lambda).
The fact that the eigenmode is localized is interesting.
(Higher modes would look like uniform uninteresting planewaves.)
Though we have not yet checked explicitly, we expect that the
eigenmode is "pinned" on top of instantons, topologically
nontrivial knots in the underlying gauge field.
These localized modes are a special feature of non-Abelian
gauge theories, and are part of what makes QCD so hard to
solve. Ultimately, the size and shape of such eigenmodes
is believed to be responsible for basic properties of matter,
such as the the size and mass of the proton.