Several difficulties arise when designing Multigrid (MG) solvers
for the systems of linear equations encountered in lattice gauge
theories
(LGT). First, these systems become increasingly ill conditioned as
one probes
in greater detail the microscopic structures of these theories --
much more so than standard elliptic systems of partial
differential equations (PDEs). In addition, the near nullspace
of the system matrix is locally oscillatory and varies with fluctuations
in the background gauge field, which is itself randomly prescribed.
Classical MG methods assume explicit knowledge of the near nullspace
in their design (typically that it is geometrically smooth in algebraic
neighborhoods) and; hence, are not immediately applicable to these
systems.
This talk presents a complex-arithmetic adaptive MG solver for
2D LGT systems. The presented method uses a geometric blocking scheme
to define the sparsity pattern of a local and sparse coarse space basis
and an adaptive smoothed aggregation MG setup procedure to define
its coefficients.
This is a joint work in collaboration with M. Brezina, J. Ruge, T. Manteuffel, and S. McCormick of CU Boulder, R. Falgout of CASC-LLNL, R. Brower, M. Clark, J. Osborne, and C. Rebbi of BU, S. MacLachlan and K. Oosterlee of T.U. Delft, and L. Zikatanov of PSU