[cig-commits] r4728 - short/3D/PyLith/branches/pylith-0.8/doc

Fri Oct 6 16:02:25 PDT 2006

Author: baagaard
Date: 2006-10-06 16:02:25 -0700 (Fri, 06 Oct 2006)
New Revision: 4728

Removed:
   short/3D/PyLith/branches/pylith-0.8/doc/manual.xml
Log:
Removed obsolete xml (docbook) file.

Deleted: short/3D/PyLith/branches/pylith-0.8/doc/manual.xml
===================================================================

--- short/3D/PyLith/branches/pylith-0.8/doc/manual.xml	2006-10-06 23:00:55 UTC (rev 4727)
+++ short/3D/PyLith/branches/pylith-0.8/doc/manual.xml	2006-10-06 23:02:25 UTC (rev 4728)
@@ -1,127 +0,0 @@
-<!DOCTYPE book PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN">
-<book id="PyLithManual" lang="en">
- 
-<bookinfo>
-<title>PyLith Manual (Draft)</title>
-<authorgroup>
-<author>
-<firstname>Charles</firstname>
-<surname>Williams</surname>
-</author>
-<author>
-<firstname>Matthew</firstname>
-<othername>G.</othername>
-<surname>Knepley</surname>
-</author>
-<author>
-<firstname>Cassie</firstname>
-<surname>Ferguson</surname>
-</author>
-</authorgroup>
-<date>March, 2006</date>
-<releaseinfo>Release tag ???</releaseinfo>
-</bookinfo>
- 
-<chapter id="Introduction">
-<title>Parallelization</title>
-
-<para>In parallelizing PyLith, a principal concern was to preserve the existing structure of the serial Fortran
-code. Active development of purely analytic features in PyLith, such as new material models or discretization schemes,
-depends on the familiarity of application scientists with the traditional Fortran programming paradigm. Global,
-topological operation should be strictly segregated from the existing code. In fact, with the exception of integrating
-PETSc for serial linear algebra and solver operations, PyLith can be run purely in serial without activating any of the
-parallel capabilities.</para>
-
-<para>In order to accomplish this separation, we use the PETSc <classname>Sieve</classname> structure to create a model of the
-serial PyLith mesh. This model is then partitioned and distributed to a set of processes. Each process receives a
-self-consistent mesh, meaning the pieces are overlapping. Each process then executes a serial PyLith step on that
-particular mesh piece. The PETSc linear algebra operations are overloaded, using the <classname>Sieve</classname>
-information, to produce a globally consistent field.</para>
-
-<para>First, we must preprocess the PyLith input mesh and return a submesh, which we accomplish by changing the root filename
-on each process.
-<programlisting>
-  lithomop3d.PetscInitialize()
-  self.inventory.scanner.inventory.fileRoot, mesh = lithomop3d.processMesh(self.inventory.scanner.inventory.fileRoot)
-</programlisting>
-We must also provide the mesh information to the routine which creates the linear algebra objects,
-<programlisting>
-  lm3dsetup.sparsesetup(mesh)
-</programlisting>
-which dispatches into the C++ interface
-<programlisting>
-  self.A, self.rhs, self.sol = lithomop3d.createPETScMat(mesh)
-</programlisting>
-</para>
-
-<para>We model the displacement field over the mesh using a fiber bundle. Each vertex is intially associated with three
-degrees of freedom, which are then restricted by the boundary conditions. After which we create local order, and then a
-unique global order which is used to construct PETSc vector scatters.
-<programlisting>
-  field->setPatch(mesh->getTopology()->leaves(), patch);
-  field->setFiberDimensionByDepth(patch, 0, 3);
-  for(ALE::Two::Mesh::sieve_type::traits::depthSequence::iterator v_itor = vertices->begin(); v_itor != vertices->end(); v_itor++) {
-    int numConstraints = 0;
-
-    for(int c = 0; c < numBoundaryComponents; c++) {
-      numConstraints += boundaries->getFiberDimension(patch_type(patch, c+1), *v_itor);
-    }
-
-    if (numConstraints > 0) {
-      field->setFiberDimension(patch, *v_itor, 3 - numConstraints);
-    }
-  }
-  field->orderPatches();
-  field->createGlobalOrder();
-</programlisting>
-We also create an alternative ordering which allows retrieval of field values on each element with the correct
-orientation. This order has previously been constructed based on the ``cell-tuple" method for a simpler bundle over
-vertices, and we are thus able to merely copy it.
-<programlisting>
-  ALE::Obj&lt;ALE::Two::Mesh::sieve_type::traits::heightSequence&gt; elements = mesh->getTopology()->heightStratum(0);
-  ALE::Obj&lt;ALE::Two::Mesh::bundle_type&gt; vertexBundle = mesh->getBundle(0);
-  std::string orderName("element");
-
-  for(ALE::Two::Mesh::sieve_type::traits::heightSequence::iterator e_iter = elements->begin(); e_iter != elements->end(); e_iter++) {
-    // setFiberDimensionByDepth() does not work here since we only want it to apply to the patch cone
-    //   What we really need is the depthStratum relative to the patch
-    ALE::Obj&lt;ALE::Two::Mesh::bundle_type::order_type::coneSequence&gt; cone = vertexBundle->getPatch(orderName, *e_iter);
-
-    field->setPatch(orderName, cone, *e_iter);
-    for(ALE::Two::Mesh::bundle_type::order_type::coneSequence::iterator c_iter = cone->begin(); c_iter != cone->end(); ++c_iter) {
-      field->setFiberDimension(orderName, *e_iter, *c_iter, field->getFiberDimension(patch, *c_iter));
-    }
-  }
-  field->orderPatches(orderName);
-</programlisting>
-</para>
-
-<para>In addition to fields and operators, we must represent boundary conditions. We do this using a fiber bundle over the mesh
-which is foliated over spatial dimension, which in our interface means that patches are labeled by the pair (patch
-number, dimension). A particular set of boundaries values consists of a section of this bundle.</para>
-
-<para>We replace the usual <methodname>MatSetValues()</methodname> call with <methodname>assembleMatrix()</methodname>.
-We do not always require that vectors be managed through the PETSc interface, but can instead be treated as Fortran
-arrays which are subsequently wrapped by serial PETSc vectors, which are then assembled into a global vector for the
-solve using <methodname>assembleVectorComplete()</methodname>. In a similar fashion, the local vector may be extracted
-using the <methodname>restrictVector()</methodname> method.
-<programlisting>
-  call VecCreateSeqWithArray(MPI_COMM_SELF, neq, bresid, locRhs, ierr)
-  call assembleVectorComplete(rhs, locRhs, ADD_VALUES, ierr)
-  call VecDestroy(locRhs, ierr)
-  call VecView(rhs, PETSC_VIEWER_STDOUT_WORLD, ierr)
-  call KSPCreate(MPI_COMM_WORLD, ksp, ierr)
-  call KSPSetfromOptions(ksp, ierr)
-  call KSPSetOperators(ksp,A,A,DIFFERENT_NONZERO_PATTERN,ierr)
-  call KSPSolve(ksp, rhs, sol, ierr)
-  call KSPDestroy(ksp, ierr)
-  call VecView(sol, PETSC_VIEWER_STDOUT_WORLD, ierr)
-  call VecCreateSeqWithArray(MPI_COMM_SELF, neq, dispvec, locSol, ierr)
-  call restrictVector(sol, locSol, INSERT_VALUES, ierr)
-  call VecDestroy(locSol, ierr)
-</programlisting>
-</para>
-
-</chapter>
-
-</book>

