1.  Created a new "core specific" directory

mkdir dyn_exp

2.  Create a test directory for experimental core

mkdir test/exp_real

3.  Edited top level Makefile to add targets for same

3.a.  rules to build the framework and then the experimental core

exp_wrf :
        $(MAKE) MODULE_DIRS="$(SLT_MODULES)" ext
        $(MAKE) MODULE_DIRS="$(SLT_MODULES)" toolsdir
        $(MAKE) MODULE_DIRS="$(SLT_MODULES)" framework
        $(MAKE) MODULE_DIRS="$(SLT_MODULES)" shared
        $(MAKE) MODULE_DIRS="$(SLT_MODULES)" exp_core

3.b.  sub-rule to build the expimental core

exp_core :
        @ echo '--------------------------------------'
        ( cd dyn_exp ; $(MAKE) )

3.c.  experimental core initialization

exp_real : exp_wrf
        @ echo '--------------------------------------'
        ( cd main ; $(MAKE) MODULE_DIRS="$(EXP_MODULES)" SOLVER=exp IDEAL_CASE=exp exp_ideal )
        ( cd test/exp ; /bin/rm -f ideal.exe ; ln -sf ../../main/ideal.exe . )
        ( cd test/exp ; ln -sf ../../run/README.namelist . )


3.d.   add macros to specify the modules for this core

EXP_MODULE_DIR = -I../dyn_exp
EXP_MODULES =  $(EXP_MODULE_DIR) $(INCLUDE_MODULES)



4. Edit share/solve_interface.F to add call to experimental core

   ELSE IF ( config_flags%dyn_opt == DYN_EXP  ) THEN

     CALL solve_exp  ( grid ,            &
!
#include <exp_actual_args.inc>
!
               )

4a. share/start_domain.F

5. Create dyn_exp/solve_exp.F

It's all there and very short -- just a trivial relaxation; some of the
stuff in there is just "magic," for right now.  Note that the code in
here is the code to do one step. The time loop is part of the driver
and can be found in frame/module_integrate.F, which call
solve_interface, which calls this routine. Note, too, that solve_exp
doesn't do any computation itself. It is mediation layer. It calls
model layer subroutines (dyn_exp/module_exp.F) to do the actual computation
in 'tile-callable' fashion.

6. Create dyn_exp/module_exp.F

This is the model layer code.  Note that boundary tests are always
encoded explicitly as conditionals using the domain indices passed
in through the arg list; never implicitly as part of the loop range.

7. Edit the Registry file and create the state data assocaited with this
solver.  Single entry:

 state real x ikj dyn_exp 2 - ih "TOYVAR"

This specfies a two timelevel variable 'x' that will be known at the
mediation layer and below as x_1 and x_2 (since it is core associated it
will be known as exp_x_1 and exp_x_2 in the driver layer; the name
of the core is prepended to prevent colllisions with variables of the
same name that are associated with other cores).  The 'ih' means it will
participate in initial data and in history data. The veriable is known
externally as TOYVAR, its data name and the name the variable will have
in data sets.

Note that since the variable is not listed as staggered in any dimension
it's logical (domain) size is ids:ide-1, kds:kde-1, jds:jde-1. This
is important in the the module_exp.F code that tests for northern
and eastern boundaries, and in the init code in module_initialize_exp.F

8. Edit the Registry file and create a halo-exchange for x_1.

 halo     HALO_EXP_A    4:x_1

Note that since halo operations are called from the mediation layer, it
is not necessary to pre-pend the dyncore name to the variable name x_1
when adding it to a comm operation like a halo exchange.

9. Edit the Registry file to set up '4' as the value of the 
namelist variable dyn_opt that means to select our exp dyncore.

package   dyn_exp       dyn_opt==4                   -             -

10. Create a dyn_exp/Makefile  (see that file)

11. Create an file exp/init_modules.F (also includes a couple of stubs
for other dyncores already in WRF; this is not the normal or correct
way of doing that; but I'm hurrying...)

12. Create a file exp/module_initialize_exp.F.  This is not part of the
WRF model itself; rather it is a pre-processor that produces initial
data for the WRF model.

13. Edit frame/module_domain.F to add case for DYN_EXP to
alloc_space_field.  (This is a bug;
one should never have to edit the framework code; will fix this in
coming versions).  Same goes for share/start_domain.F, although this
is not a framework routine.

14. clean commands

15. namelist file
     copy from another dir and change dyn_opt
     boundary conditions