TARGET: su3_eigen_asqtad

INPUT PARAMETERS:

 Max_Rayleigh_iters : Maximum iterations allowed for Ritz functional
                      minimization. This limit should be rarely reached.
	              Use a large number 200-300
                      
 Restart_Rayleigh : The Ritz functional minimization looses orthogonality
                     and has to be restarted. This should be a number
                     in the range 10 to 30. In my case the optimum was 10
                     But you may find otherwise in you case.

 Kalkreuter_iters : This the maximum number of Kalkruter iterations.
                     One iteration is a Ritz minimization for each
                     eigenvector followed by a diagonalization of the
                     resulting matrix.
            
 eigenval_tolerance : The eigenvalue tolerance need. 
                     See also comments in eigen_stuff.c about 
                     the convergenve criterion  (STRICT_CONVERGENCE).

 error_decrseace : After the error has decrease by a factor of
                   error_decrease in the Ritz minization the Rayleigh
                   iteration stops. Kalkreuter suggests .3 as the optimum
                   value. The reason is that you do not want to waist a
                   lot of time minimizing the Ritz functional because
                   converging a little for each eigenvector and then
                   moving the next one you achieve uniform convergence
                   for all the eigenvectors. 


The rest of the input parameters are the same as the ones in ks_imp_dyn1.

TEST INPUT: in.eigen
TEST OUTPUT: out.eigen
Action used in tests: asqtad_action.h

--------------

Date: Thu, 23 Mar 2000 18:17:48 -0700
From: Kostas Orginos <[email protected]>
To: [email protected]
Subject: One last comment

I just fixed some problems I had in the past with dslash_special. The
code I sent you uses either dslash_special or the regular dslash. 
The dslash_special is chosen by defining USE_DSLASH_SPECIAL in
eigen_stuff.c. All my running was done with out the dslash_special.
I have tested the dslash_special both in vanilla and mpi mode on a cold lattice
and on an instanton. Both versions give the same results. Thus I assume the 
version with the dslash_special is correct. I kept both options available
just in case... 

Kostas.