Runge-Kutta (Fehlberg) 45 with error estimation.

Fourth order accurate evolution with an additional fifth order step for error estimation. How much sense the error makes is unclear to me, but hey.

For the moment the error is stored in an internal MoL array ErrorEstimate; there is one per evolved variable. At a later point this may be moved out to user thorns who can register their own etc.

As the implementation uses 6 evaluations of the RHS (necessary) and 6 levels of scratch space (one more than necessary - laziness kicked in) then this is very expensive.

This is a partial fix for PR/1840.


git-svn-id: http://svn.cactuscode.org/arrangements/CactusNumerical/MoL/trunk@74 578cdeb0-5ea1-4b81-8215-5a3b8777ee0b

1 files changed, 1 insertions, 0 deletions

diff --git a/param.ccl b/param.ccl
index e982acb..a65216a 100644
--- a/param.ccl
+++ b/param.ccl
@@ -87,6 +87,7 @@ KEYWORD ODE_Method "The ODE method use by MoL to do time integration"
   "ICN-avg"	:: "Iterative Crank Nicholson with averaging"
   "RK2"		:: "Efficient RK2"
   "RK3"		:: "Efficient RK3"
+  "RK45"        :: "RK45 with error estimation"
 } "ICN"
 
 KEYWORD Generic_Type "If using the generic method, which sort"