We consider preconditioners for weighted Toeplitz regularized least squares (WTRLS) problems
Iterative methods using circulant preconditioners has been proposed since 1980's. The efficiency of solving systems are greatly enhanced by these indirect methods. Since then, preconditioners based on circulant matrices were suggested for solving the WTRLS problems. However, most of them do not perform satisfactorily and the results of convergence rates can be disappointing.
Benzi and Ng
considered the above system
where is replaced by
, and proposed a new approach to
solve it which is based on an augmented system formulation. We now
adopt a similar approach that transforms the problem into a
block linear system.
The WTRLS problems turn out to be a quadratic minimization problem, which is equivalent to solving the linear system
The above linear system can overall be rewritten as the following
block system:
There are not any particular results discussing how
to solve a general
system effectively.
The four upper left blocks are considered as one single
block and the corresponding two upper right blocks
together
still attain full rank, so we treat this system as
a
block case
and consider methods developed for solving indefinite systems.
Here we shall adopt constraint
preconditioner
and Hermitian skew-Hermitian splitting (HSS)
preconditioner
and investigate them in details.
The formulation and actual preconditioning settings
for WTRLS problems will be discussed.
A test problem is performed to demonstrate the convergence behaviour
using these iterative methods. We shall see that the number of iterations
mainly depend on the condition numbers of
and
as
well as some factors related to each preconditioners.
In particular, this effect is more significant in
the case of using HSS preconditioner. The iteration results and
the weaknesses for both preconditioners can be explained
with some analyses on the spectra for them.