[cig-commits] r19171 - short/3D/PyLith/branches/v1.6-stable
brad at geodynamics.org
brad at geodynamics.org
Thu Nov 10 09:52:31 PST 2011
Author: brad
Date: 2011-11-10 09:52:31 -0800 (Thu, 10 Nov 2011)
New Revision: 19171
Modified:
short/3D/PyLith/branches/v1.6-stable/README
Log:
Updated README.
Modified: short/3D/PyLith/branches/v1.6-stable/README
===================================================================
--- short/3D/PyLith/branches/v1.6-stable/README 2011-11-09 22:59:03 UTC (rev 19170)
+++ short/3D/PyLith/branches/v1.6-stable/README 2011-11-10 17:52:31 UTC (rev 19171)
@@ -78,7 +78,56 @@
- Fixed bug in writing tensor data for Xdmf files. Switched Tensor
to Tensor6 to account for symmetry.
+ - Fixed bug in writing HDF5 files in parallel when one processor
+ does not write any information (e.g., faults and boundary
+ conditions).
+ - Added dimensioning of time dataset in HDF5 files. The units are
+ now seconds rather than nondimensional time.
+
+ - Fixed memory allocation error (std::bad_alloc) when a processor
+ did not contain cells for a boundary condition or output. This bug
+ did not show up on all architectures.
+
+ - Increased robustness of spontaneous rupture (fault friction)
+ implementation to broaden the range of conditions it can
+ handle. The implementation now properly handles cases with fault
+ opening and cases with zero shear or normal tractions.
+
+
+* Internal changes
+
+ - Fault implementation
+
+ Several changes have been made to the fault implementation, but
+ none of these affect the user interface. The runtime performance
+ is nearly identical with improved accuracy for spontaneous rupture
+ (fault friction) simulations. These changes involved switching to
+ using tractions (non-integrated quantities) for the Lagrange
+ multipliers in the global coordinate system rather than integrated
+ quantities in the fault coordinate system. Additionally, initial
+ fault tractions are associated with the fault vertices and their
+ interpolation uses the finite-element basis functions.
+
+ - Distribution of mesh among processors
+
+ The data structures used to distribute the mesh among processors
+ have been improved. This reduces memory use and runtime for this
+ stage of the simulations.
+
+
+KNOWN ISSUES
+
+ The custom line search used with the PETSc nonlinear solver (SNES)
+ has difficulty handling cases with fault opening when the fault
+ normal tractions are much larger than the fault shear tractions. The
+ direction of the line search tends to be nearly orthogonal to the
+ residual, resulting in extremely poor convergence. We plan to
+ improve the line search algorithm in a future release in order to
+ resolve this issue and improve the rate of convergence in
+ spontaneous rupture simulations.
+
+
----------------------------------------------------------------------
Version 1.6.1
----------------------------------------------------------------------
More information about the CIG-COMMITS
mailing list