Referee 1 **************************************************

This is an good paper that discusses some excellent work on 
parallel multigrid methods for unstructured grids. 

Presentation Changes 

Since the English grammar requires extensive improvement 
I have faxed to the editor suggested grammatical changes.
(EDITOR's note -- these are available)

Referee 2 **************************************************

This paper describes an application of the advanced computational technique
of multigrid-based preconditioning approach to solving a class of important
earth science application problems on parallel computers. The computational
algorithms presented are well known in the computational fluid dynamics
community, and the multigrid approach to solving Poisson pressure equation
is almost standard in a Navier-Stokes solver. The parallel computing work
and application results discussed in the paper are probably still interesting
to modeling scientists in the computational earth science community.

It should be made more clear how the preconditioning step is used in the
multigrid V-cycle solution process. Some discussion would be useful
on the tradeoffs between the levels used in the V-cycle and the parallel computing
efficiency, and how to optimize the time-to-solution, similar for tradeoff
between using shallow and deep adaptations.

How less than 1% communication overhead, as claimed in the paper,
would result in the scaling curve in Fig. 9 on upto 128 PE?


Presentation Changes

Some words should be changed to make the sentences more readable.

Last paragraph in Section 2 (Multigrid Method: Overview)
should be made more specific and readable. I am not very clear
what it says.


Referee 3 **************************************************


The paper presents experimental results for solving Poisson equations using a
multigrid preconditioned conjugate gradient iterative method (MGCG). It
compares this method, proposed by the author, with Incomplete Cholesky
Conjugate Gradient. The subject of the paper is relevant to the journal. There
is a reasonable amount of experiments and it shows good results. However, I
think that description of the algorithm proposed by the author is not very
good.  The description is too high level. In some occasions the paper just
presents a picture (with very little associated explanation) and the reader has
to figure it out by himself. In other instances, the description is too short,
assuming the reader's familiarity with the subject. Due to the multidisciplinar
nature of the subject, I would suggest more detailed descriptions of the
algorithms. I suspect a person reading the paper will want to know more about
how to implement MGCG than just how it performs compared to ICCG.  I would
recommend an extended version of this paper for publication. If space is a
limitation here, I think the figures can be scaled down to at least half of its
current size, without compromising readability.


Presentation Changes

The description of the algorithm proposed by the author need to be worked.  I
think a lot of text should be added to give the reader a more clear view of how
to implement MGCG.

Some formatting would do a great deal for this paper. Having the pictures
closer to where they are referred to in the text simplifies the reading a lot.