* add support for sparse-matrix Jacobians ==> works!

* change default in param.ccl to use this
* change default in src/include/config.h to default to no longer
  link in LAPACK routines
* update documentation


git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinAnalysis/AHFinderDirect/trunk@931 f88db872-0e4f-0410-b76b-b9085cfa78c5

1 files changed, 43 insertions, 38 deletions

diff --git a/param.ccl b/param.ccl
index 3abfcd9..d908172 100644
--- a/param.ccl
+++ b/param.ccl
@@ -108,7 +108,7 @@ boolean print_timing_stats \
 # ***** parameters for the Jacobian matrix *****
 #
 
-keyword Jacobian_method "how do we compute the Jacobian matrix?"
+keyword Jacobian_compute_method "how do we compute the Jacobian matrix?"
 {
 # for debugging only"
 "numerical perturbation"    :: "n.b. this is *very* slow"
@@ -123,17 +123,22 @@ keyword Jacobian_method "how do we compute the Jacobian matrix?"
 } "symbolic differentiation with finite diff d/dr"
 
 #
-# This parameter lists all known storage methods.  See
+# This parameter lists all known methods for storing the Jacobian and
+# solving linear systems with the Jacobian as the LHS matrix.  See
 # "src/include/config.hh" for which of these are actually compiled in at
 # the moment.  N.b. each compiled-in method requires linking with the
 # corresponding linear-solver library; see "src/make.configuration.defn"
 # for details on these libraries.
 #
-keyword Jacobian_storage_method "how do we store the Jacobian matrix?"
+keyword Jacobian_store_solve_method \
+  "how do we store/linear-solve the Jacobian matrix?"
 {
-"dense matrix"	:: "dense matrix (inefficient at high angular resolution)"
-"row-oriented sparse matrix" :: "sparse matrix, row-oriented storage format"
-} "dense matrix"
+"dense matrix/LAPACK"	:: \
+  "store as (Fortran) dense matrix, solve with LAPACK routines"
+"row-oriented sparse matrix/ILUCG" :: \
+  "store as sparse matrix (row-oriented storage format), \
+   solve with ILUCG (incomplete LU decomposition - conjugate gradient) method"
+} "row-oriented sparse matrix/ILUCG"
 
 #
 # This parameter controls two different sorts of one-sided finite
@@ -156,7 +161,7 @@ real Jacobian_perturbation_amplitude \
 # if AHFinderDirect::method = "test Jacobian", should we test all
 # known methods for computing the Jacobian, or just the numerical perturbation
 # method (the latter may be useful of some other methods are broken)
-boolean test_all_Jacobian_methods \
+boolean test_all_Jacobian_compute_methods \
   "should we test all Jacobian computation methods, or just NP?"
 {
 } "true"
@@ -441,35 +446,7 @@ keyword patch_system_type[5] "what type of patch system should we use?"
 } "match Cactus grid symmetry"
 
 #
-# This parameter sets the width of the interpatch ghost zones in the
-# patch system.  Note that this thorn uses the terminology "ghost zone"
-# for any of what Cactus in general now calls a "boundary zone" or a
-# "symmetry zone" or a "patch zone".
-#
-# This parameter must be at least
-# ... 2 if FINITE_DIFF_ORDER is set to 4 in "src/include/config.hh"
-# ... 1 if FINITE_DIFF_ORDER is set to 2 in "src/include/config.hh"
-# The code checks for this being too small, and reports a fatal error if so.
-#
-int ghost_zone_width "number of ghost zones on each side of a patch"
-{
-0:* :: "any integer >= 0"
-} 2
-
-#
-# Our code that computes surface integrals over patches (used for
-# computing BH diagnostics like centroids, areas, masses, etc) silently
-# assumes that this parameter is == 1, so you should probably leave
-# it at that setting.
-#
-int patch_overlap_width \
-  "number of grid points that nominally-just-touching patches should overlap"
-{
-1:*:2 :: "any integer >= 0; current implementation requires that it be odd"
-} 1
-
-#
-# This parameter sets the angular resolution of all the patch systems:
+# This parameter sets the angular resolution of each patch systems:
 # the angular grid spacing in degrees is 90.0/N_zones_per_right_angle.
 #
 # In practice the error in the horizon position is usually dominated
@@ -496,10 +473,38 @@ int patch_overlap_width \
 # the 4th/6th power of this parameter!  For example, doubling the resolution
 # takes 16 times as much memory, and 64 times as long to run!
 #
-int N_zones_per_right_angle "sets angular resolution of patch systems"
+int N_zones_per_right_angle[5] "sets angular resolution of patch systems"
 {
 1:* :: "any integer >= 1; must be even for patch systems other than full-sphere"
-} 12
+} 18
+
+#
+# This parameter sets the width of the interpatch ghost zones in the
+# patch system.  Note that this thorn uses the terminology "ghost zone"
+# for any of what Cactus in general now calls a "boundary zone" or a
+# "symmetry zone" or a "patch zone".
+#
+# This parameter must be at least
+# ... 2 if FINITE_DIFF_ORDER is set to 4 in "src/include/config.hh"
+# ... 1 if FINITE_DIFF_ORDER is set to 2 in "src/include/config.hh"
+# The code checks for this being too small, and reports a fatal error if so.
+#
+int ghost_zone_width "number of ghost zones on each side of a patch"
+{
+0:* :: "any integer >= 0"
+} 2
+
+#
+# Our code that computes surface integrals over patches (used for
+# computing BH diagnostics like centroids, areas, masses, etc) silently
+# assumes that this parameter is == 1, so you should probably leave
+# it at that setting.
+#
+int patch_overlap_width \
+  "number of grid points that nominally-just-touching patches should overlap"
+{
+1:*:2 :: "any integer >= 0; current implementation requires that it be odd"
+} 1
 
 ################################################################################