[cig-commits] r13197 - mc/3D/CitcomS/trunk/lib

[tan2 at geodynamics.org]

Author: tan2
Date: 2008-10-29 16:22:10 -0700 (Wed, 29 Oct 2008)
New Revision: 13197

Modified:
   mc/3D/CitcomS/trunk/lib/global_defs.h
Log:
(for r13196) Fixing convergence criterion for Stokes solver.

* New functions global_v_norm2(), global_p_norm2(), global_div_norm2()
  compute the L2 norm of velocity, pressure, and divergence.
* The L2 norm of the vector f is defined as:
  \sum_{all elements} ( \int_{element} (f*f dA) ) / A
  where A is the volume of the domain.
* The convergence criterion is controlled by parameter "accuracy" under
  "[CitcomS.solver.vsolver]". The iteration stops if
  (norm(div(rho*V))/norm(V) < accuracy) or
  ( (norm(dV)/norm(V) < accuracy) and (norm(dP)/norm(P) < accuracy) for
    the last two iterations )
* The default value of "accuracy" becomes 1e-4.
* The input parameters "tole_compressibility" and "relative_err_accuracy"
  under "[CitcomS.solver.vsolver]" are gone.
* The mass matrix E->MASS becomes double precision (was single precision).


Modified: mc/3D/CitcomS/trunk/lib/global_defs.h
===================================================================
--- mc/3D/CitcomS/trunk/lib/global_defs.h	2008-10-29 23:17:11 UTC (rev 13196)
+++ mc/3D/CitcomS/trunk/lib/global_defs.h	2008-10-29 23:22:10 UTC (rev 13197)
@@ -106,6 +106,8 @@
 #define NCS      14   /* max. number of sphere caps */
 
 /* type of elt_del and elt_c arrays */
+/* double precision doesn't help,
+ * probably due to the coordinate transformation c33matrix */
 #if 1
     typedef float higher_precision;
 #else