Add comments

ferroteam · Oct 10, 2019 · bf80706 · bf80706
1 parent 3a6eea5
commit bf80706
Show file tree

Hide file tree

Showing 3 changed files with 51 additions and 7 deletions.
diff --git a/include/dft.h b/include/dft.h
@@ -141,7 +141,7 @@ namespace dftfe {
  const bool useSingleAtomSolution = false );
 
  /**
- * @brief Selects between only electronic field relaxation or combined electronic and geometry relxation
+ * @brief Selects between only electronic field relaxation or combined electronic and geometry relaxation
  */
  void run();
 

diff --git a/src/dft/dft.cc b/src/dft/dft.cc
@@ -335,7 +335,7 @@ namespace dftfe {
  {
  pcout <<" Warning: User has requested the number of Kohn-Sham wavefunctions to be less than or"
  "equal to half the number of electrons in the system. Setting the Kohn-Sham wavefunctions"
- "to half the number of electrons with a 15 percent buffer to avoid convergence issues in"
+ "to half the number of electrons with a 20 percent buffer to avoid convergence issues in"
  "SCF iterations"<<std::endl;
  }
  d_numEigenValues = (numElectrons/2.0) + std::max(0.2*(numElectrons/2.0),20.0);

diff --git a/src/triangulation/triangulationManager/generateMesh.cc b/src/triangulation/triangulationManager/generateMesh.cc
@@ -413,8 +413,8 @@ namespace dftfe {
 
 
  //
- // refine cells on periodic boundary if their length is greater than two times the
- // mesh size around atom
+ // refine cells on periodic boundary if their length is greater than
+ // mesh size around atom by a factor (set by smootheningFactor)
  //
  if (smoothenCellsOnPeriodicBoundary)
  {
@@ -484,6 +484,9 @@ namespace dftfe {
  cell = parallelTriangulation.begin_active();
  endc = parallelTriangulation.end();
 
+ //
+ // populate maps refinement flag maps to zero values
+ //
  std::map<dealii::CellId,unsigned int> cellIdToLocallyOwnedId;
  unsigned int locallyOwnedCount=0;
  for(;cell != endc; ++cell)
@@ -506,8 +509,13 @@ namespace dftfe {
  cellElectroForce = electrostaticsTriangulationForce.begin_active();
  }
 
- const unsigned int faces_per_cell=dealii::GeometryInfo<3>::faces_per_cell;
 
+ //
+ // go to each locally owned or ghost cell which has a face on the periodic boundary->
+ // query if cell has a periodic neighbour which is coarser -> if yes and the coarse
+ // cell is locally owned set refinement flag on that cell
+ //
+ const unsigned int faces_per_cell=dealii::GeometryInfo<3>::faces_per_cell;
  bool isAnyCellRefined=false;
  for(;cell != endc; ++cell)
  {
@@ -545,7 +553,7 @@ namespace dftfe {
  }
 
  //
- // check that triangulation has consistent refinement across periodic boundary including
+ // check that triangulation has consistent refinement across periodic boundary
  //
  bool triangulationManager::checkPeriodicSurfaceRefinementConsistency(parallel::distributed::Triangulation<3>& parallelTriangulation)
  {
@@ -849,7 +857,16 @@ namespace dftfe {
  d_serialTriaCurrentRefinement.clear();
 
  //
- //Multilayer refinement
+ //Multilayer refinement. Refinement algorithm is progressively modified
+ //if check of parallel consistency of combined periodic
+ //and hanging node constraitns fails (Related to https://github.com/dealii/dealii/issues/7053).
+ //
+
+ //
+ //STAGE1: Call refinementAlgorithmA and consistentPeriodicBoundaryRefinement alternatively.
+ //In the call to refinementAlgorithmA there is no additional reduction of adaptivity performed
+ //on the periodic boundary. Multilevel refinement is performed until both refinementAlgorithmA
+ //and consistentPeriodicBoundaryRefinement do not set refinement flags on any cell.
  //
  unsigned int numLevels=0;
  bool refineFlag = true;
@@ -873,6 +890,7 @@ namespace dftfe {
  //This sets the global refinement sweep flag
  refineFlag= Utilities::MPI::max((unsigned int) refineFlag, mpi_communicator);
 
+ //try the other type of refinement to prevent while loop from ending prematurely
  if (!refineFlag)
  {
  refineFlag=consistentPeriodicBoundaryRefinement(parallelTriangulation,
@@ -900,6 +918,7 @@ namespace dftfe {
  //This sets the global refinement sweep flag
  refineFlag= Utilities::MPI::max((unsigned int) refineFlag, mpi_communicator);
 
+ //try the other type of refinement to prevent while loop from ending prematurely
  if (!refineFlag)
  {
  refineFlag=refinementAlgorithmA(parallelTriangulation,
@@ -981,6 +1000,16 @@ namespace dftfe {
 
  if (!dftParameters::reproducible_output)
  {
+ //
+ //STAGE2: This stage is only activated combined periodic and hanging node constraints are
+ //not consistent in parallel.
+ //Call refinementAlgorithmA and consistentPeriodicBoundaryRefinement alternatively.
+ //In the call to refinementAlgorithmA there an additional reduction of adaptivity performed
+ //on the periodic boundary such that the maximum cell length on the periodic boundary is less
+ //than two times the MESH SIZE AROUND ATOM. Multilevel refinement is performed until both
+ //refinementAlgorithmAand consistentPeriodicBoundaryRefinement do not set refinement flags
+ //on any cell.
+ //
  if (!checkConstraintsConsistency(parallelTriangulation))
  {
  refineFlag=true;
@@ -1004,6 +1033,7 @@ namespace dftfe {
  //This sets the global refinement sweep flag
  refineFlag= Utilities::MPI::max((unsigned int) refineFlag, mpi_communicator);
 
+ //try the other type of refinement to prevent while loop from ending prematurely
  if (!refineFlag)
  {
  refineFlag=consistentPeriodicBoundaryRefinement(parallelTriangulation,
@@ -1031,6 +1061,7 @@ namespace dftfe {
  //This sets the global refinement sweep flag
  refineFlag= Utilities::MPI::max((unsigned int) refineFlag, mpi_communicator);
 
+ //try the other type of refinement to prevent while loop from ending prematurely
  if (!refineFlag)
  {
  refineFlag=refinementAlgorithmA(parallelTriangulation,
@@ -1099,6 +1130,17 @@ namespace dftfe {
  }
  }
 
+ //
+ //STAGE3: This stage is only activated combined periodic and hanging node constraints are
+ //still not consistent in parallel.
+ //Call refinementAlgorithmA and consistentPeriodicBoundaryRefinement alternatively.
+ //In the call to refinementAlgorithmA there an additional reduction of adaptivity performed
+ //on the periodic boundary such that the maximum cell length on the periodic boundary is less
+ //than MESH SIZE AROUND ATOM essentially ensuring uniform refinement on the periodic boundary
+ //in the case of MESH SIZE AROUND ATOM being same as MESH SIZE AT ATOM.
+ //Multilevel refinement is performed until both refinementAlgorithmA and
+ //consistentPeriodicBoundaryRefinement do not set refinement flags on any cell.
+ //
  if (!checkConstraintsConsistency(parallelTriangulation))
  {
  refineFlag=true;
@@ -1122,6 +1164,7 @@ namespace dftfe {
  //This sets the global refinement sweep flag
  refineFlag= Utilities::MPI::max((unsigned int) refineFlag, mpi_communicator);
 
+ //try the other type of refinement to prevent while loop from ending prematurely
  if (!refineFlag)
  {
  refineFlag=consistentPeriodicBoundaryRefinement(parallelTriangulation,
@@ -1149,6 +1192,7 @@ namespace dftfe {
  //This sets the global refinement sweep flag
  refineFlag= Utilities::MPI::max((unsigned int) refineFlag, mpi_communicator);
 
+ //try the other type of refinement to prevent while loop from ending prematurely
  if (!refineFlag)
  {
  refineFlag=refinementAlgorithmA(parallelTriangulation,