|           Line data    Source code 
       1             : // The libMesh Finite Element Library.
       2             : // Copyright (C) 2002-2025 Benjamin S. Kirk, John W. Peterson, Roy H. Stogner
       3             : 
       4             : // This library is free software; you can redistribute it and/or
       5             : // modify it under the terms of the GNU Lesser General Public
       6             : // License as published by the Free Software Foundation; either
       7             : // version 2.1 of the License, or (at your option) any later version.
       8             : 
       9             : // This library is distributed in the hope that it will be useful,
      10             : // but WITHOUT ANY WARRANTY; without even the implied warranty of
      11             : // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
      12             : // Lesser General Public License for more details.
      13             : 
      14             : // You should have received a copy of the GNU Lesser General Public
      15             : // License along with this library; if not, write to the Free Software
      16             : // Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
      17             : 
      18             : #include "libmesh/variational_smoother_system.h"
      19             : 
      20             : #include "libmesh/elem.h"
      21             : #include "libmesh/face_tri3.h"
      22             : #include "libmesh/face_tri6.h"
      23             : #include "libmesh/fe_base.h"
      24             : #include "libmesh/fe_interface.h"
      25             : #include "libmesh/fem_context.h"
      26             : #include "libmesh/mesh.h"
      27             : #include "libmesh/numeric_vector.h"
      28             : #include "libmesh/parallel_ghost_sync.h"
      29             : #include "libmesh/quadrature.h"
      30             : #include "libmesh/string_to_enum.h"
      31             : #include "libmesh/utility.h"
      32             : #include "libmesh/enum_to_string.h"
      33             : #include <libmesh/reference_elem.h>
      34             : 
      35             : // C++ includes
      36             : #include <functional> // std::reference_wrapper
      37             : 
      38             : namespace libMesh
      39             : {
      40             : 
      41             : /*
      42             :  * Gets the dof_id_type value corresponding to the minimum of the Real value.
      43             :  */
      44      133736 : void communicate_pair_min(std::pair<Real, dof_id_type> & pair, const Parallel::Communicator & comm)
      45             : {
      46             :   // Get rank where minimum occurs
      47             :   unsigned int rank;
      48      133736 :   comm.minloc(pair.first, rank);
      49      133736 :   comm.broadcast(pair.second, rank);
      50      133736 : }
      51             : 
      52             : /*
      53             :  * Gets the dof_id_type value corresponding to the maximum of the Real value.
      54             :  */
      55      133736 : void communicate_pair_max(std::pair<Real, dof_id_type> & pair, const Parallel::Communicator & comm)
      56             : {
      57             :   // Get rank where minimum occurs
      58             :   unsigned int rank;
      59      133736 :   comm.maxloc(pair.first, rank);
      60      133736 :   comm.broadcast(pair.second, rank);
      61      133736 : }
      62             : 
      63             : /**
      64             :  * Function to prevent dividing by zero for degenerate elements
      65             :  */
      66     6752976 : Real chi_epsilon(const Real & x, const Real epsilon_squared)
      67             : {
      68     7315532 :   return 0.5 * (x + std::sqrt(epsilon_squared + Utility::pow<2>(x)));
      69             : }
      70             : 
      71             : /**
      72             :  * Given an fe_map, element dimension, and quadrature point index, returns the
      73             :  * Jacobian of the physical-to-reference mapping.
      74             :  */
      75     6779041 : RealTensor get_jacobian_at_qp(const FEMap & fe_map,
      76             :                           const unsigned int & dim,
      77             :                           const unsigned int & qp)
      78             : {
      79     6779041 :   const auto & dxyzdxi = fe_map.get_dxyzdxi()[qp];
      80     1127192 :   const auto & dxyzdeta = fe_map.get_dxyzdeta()[qp];
      81     1127192 :   const auto & dxyzdzeta = fe_map.get_dxyzdzeta()[qp];
      82             : 
      83             :   // RealTensors are always 3x3, so we will fill any dimensions above dim
      84             :   // with 1s on the diagonal. This indicates a 1 to 1 relationship between
      85             :   // the physical and reference elements in these extra dimensions.
      86     6779041 :   switch (dim)
      87             :   {
      88        6288 :     case 1:
      89             :       return RealTensor(
      90             :         dxyzdxi(0), 0, 0,
      91             :                      0, 1, 0,
      92             :                      0, 0, 1
      93         524 :       );
      94             :       break;
      95             : 
      96      326297 :     case 2:
      97             :       return RealTensor(
      98             :         dxyzdxi(0), dxyzdeta(0), 0,
      99             :         dxyzdxi(1), dxyzdeta(1), 0,
     100             :         0,              0,               1
     101       26942 :       );
     102             :       break;
     103             : 
     104     6446456 :     case 3:
     105     1072260 :       return RealTensor(
     106             :         dxyzdxi,
     107             :         dxyzdeta,
     108             :         dxyzdzeta
     109      536130 :       ).transpose();  // Note the transposition!
     110             :       break;
     111             : 
     112           0 :     default:
     113           0 :       libmesh_error_msg("Unsupported dimension.");
     114             :   }
     115             : }
     116             : 
     117             : /**
     118             :  * Compute the trace of a dim-dimensional matrix.
     119             :  */
     120     6752976 : Real trace(const RealTensor & A, const unsigned int & dim)
     121             : {
     122      562556 :   Real tr = 0.0;
     123    26674176 :   for (const auto i : make_range(dim))
     124    19921200 :     tr += A(i, i);
     125             : 
     126     6752976 :   return tr;
     127             : }
     128             : 
     129        6834 : VariationalSmootherSystem::~VariationalSmootherSystem () = default;
     130             : 
     131       31020 : void VariationalSmootherSystem::assembly (bool get_residual,
     132             :                                           bool get_jacobian,
     133             :                                           bool apply_heterogeneous_constraints,
     134             :                                           bool apply_no_constraints)
     135             : {
     136             :   // Update the mesh based on the current variable values
     137        1736 :   auto & mesh = this->get_mesh();
     138       31020 :   this->solution->close();
     139             : 
     140      917960 :   for (auto * node : mesh.local_node_ptr_range())
     141             :   {
     142     1813062 :     for (const auto d : make_range(mesh.mesh_dimension()))
     143             :     {
     144     1345764 :       const auto dof_id = node->dof_number(this->number(), d, 0);
     145             :       // Update mesh
     146     1345764 :       (*node)(d) = libmesh_real((*current_local_solution)(dof_id));
     147             :     }
     148       29284 :   }
     149             : 
     150       31020 :   SyncNodalPositions sync_object(mesh);
     151       61172 :   Parallel::sync_dofobject_data_by_id (mesh.comm(), mesh.nodes_begin(), mesh.nodes_end(), sync_object);
     152             : 
     153             :   // Compute and update mesh quality information
     154       31020 :   compute_mesh_quality_info();
     155       31020 :   const bool is_tangled = _mesh_info.mesh_is_tangled;
     156             : 
     157             :   // Update _epsilon_squared_assembly based on whether we are untangling or
     158             :   // smoothing
     159       31020 :   if (_untangling_solve)
     160             :     {
     161         172 :       const Real & min_S = _mesh_info.min_qp_det_S;
     162             :       // This component is based on what Larisa did in the original code.
     163        6312 :       const Real variable_component = 100. * Utility::pow<2>(_ref_vol * min_S);
     164        6312 :       _epsilon_squared_assembly = _epsilon_squared + variable_component;
     165             :     }
     166             : 
     167             :   else
     168       24708 :     _epsilon_squared_assembly = 0.;
     169             : 
     170       31020 :   FEMSystem::assembly(get_residual, get_jacobian, apply_heterogeneous_constraints, apply_no_constraints);
     171             : 
     172       31020 :   if (_untangling_solve && !is_tangled)
     173             :     {
     174             :       // Mesh is untangled, artificially reduce residual by factor 0.9 to tell
     175             :       // the solver we are on the right track. The mesh may become re-tangled in
     176             :       // subsequent nonlinear and line search iterations, so we will not use
     177             :       // this reduction factor in the case of re-tangulation. This approach
     178             :       // should drive the solver to eventually favor untangled solutions.
     179        4466 :       rhs->close();
     180        4466 :       (*rhs) *= 0.9;
     181             :     }
     182       31020 : }
     183             : 
     184        2414 : void VariationalSmootherSystem::solve()
     185             : {
     186        2414 :   const auto & mesh_info = get_mesh_info();
     187        2414 :   if (_verbosity > 10)
     188           0 :     libMesh::out << "Initial " << mesh_info << std::endl;
     189        2414 :   if (mesh_info.mesh_is_tangled)
     190             :     {
     191             :       // Untangling solve
     192         142 :       _untangling_solve = true;
     193             : 
     194             :       // Untangling seems to work better using only the distortion metric.
     195             :       // For a mixed metric, I've seen it *technically* untangle a mesh to a
     196             :       // still-suboptimal mesh (i.e., a local minima), but not been able to
     197             :       // smooth it well because the smoothest solution is on the other side of
     198             :       // a tangulation barrier.
     199         142 :       const auto dilation_weight = _dilation_weight;
     200         142 :       _dilation_weight = 0.;
     201             : 
     202         142 :       if (_verbosity > 10)
     203           0 :         libMesh::out << "Untangling the mesh" << std::endl;
     204         142 :       FEMSystem::solve();
     205             : 
     206             :       // Reset the dilation weight
     207         142 :       _dilation_weight = dilation_weight;
     208             : 
     209         142 :     if (_verbosity > 10)
     210           0 :       libMesh::out << "Untangled " << mesh_info << std::endl;
     211             :     }
     212             : 
     213             :   // Smoothing solve
     214        2414 :   _untangling_solve = false;
     215        2414 :   if (_verbosity > 10)
     216           0 :     libMesh::out << "Smoothing the mesh" << std::endl;
     217        2414 :   FEMSystem::solve();
     218        2414 :   libmesh_error_msg_if(mesh_info.mesh_is_tangled, "The smoothing solve tangled the mesh!");
     219        2414 :   if (_verbosity > 10)
     220           0 :     libMesh::out << "Smoothed " << mesh_info << std::endl;
     221        2414 : }
     222             : 
     223        2414 : void VariationalSmootherSystem::init_data ()
     224             : {
     225         136 :   auto & mesh = this->get_mesh();
     226          68 :   const auto & elem_orders = mesh.elem_default_orders();
     227        2414 :   libmesh_error_msg_if(elem_orders.size() != 1,
     228             :       "The variational smoother cannot be used for mixed-order meshes!");
     229        2414 :   const auto fe_order = *elem_orders.begin();
     230             :   // Add a variable for each dimension of the mesh
     231             :   // "r0" for x, "r1" for y, "r2" for z
     232        8520 :   for (const auto & d : make_range(mesh.mesh_dimension()))
     233       12040 :     this->add_variable ("r" + std::to_string(d), fe_order);
     234             : 
     235             :   // Do the parent's initialization after variables are defined
     236        2414 :   FEMSystem::init_data();
     237             : 
     238             :   // Set the current_local_solution to the current mesh for the initial guess
     239        2414 :   this->solution->close();
     240             : 
     241       75952 :   for (auto * node : mesh.local_node_ptr_range())
     242             :   {
     243      151812 :     for (const auto d : make_range(mesh.mesh_dimension()))
     244             :     {
     245      112980 :       const auto dof_id = node->dof_number(this->number(), d, 0);
     246             :       // Update solution
     247      112980 :       (*solution).set(dof_id, (*node)(d));
     248             :     }
     249        2278 :   }
     250             : 
     251        2414 :   this->prepare_for_smoothing();
     252        2414 : }
     253             : 
     254        2414 : void VariationalSmootherSystem::prepare_for_smoothing()
     255             : {
     256             :   // If this method has already been called to set _ref_vol, just return.
     257        2482 :   if (std::abs(_ref_vol) > TOLERANCE * TOLERANCE)
     258           0 :     return;
     259             : 
     260        2482 :   std::unique_ptr<DiffContext> con = this->build_context();
     261          68 :   FEMContext & femcontext = cast_ref<FEMContext &>(*con);
     262        2414 :   this->init_context(femcontext);
     263             : 
     264         136 :   const auto & mesh = this->get_mesh();
     265             : 
     266        2414 :   Real elem_averaged_det_S_sum = 0.;
     267             : 
     268             :   // Make pre-requests before reinit() for efficiency in
     269             :   // --enable-deprecated builds, and to avoid errors in
     270             :   // --disable-deprecated builds.
     271          68 :   const auto & fe_map = femcontext.get_element_fe(0)->get_fe_map();
     272          68 :   const auto & JxW = fe_map.get_JxW();
     273             : 
     274       40040 :   for (const auto * elem : mesh.active_local_element_ptr_range())
     275             :     {
     276       35280 :       femcontext.pre_fe_reinit(*this, elem);
     277       35280 :       femcontext.elem_fe_reinit();
     278             : 
     279             :       // Add target element info, if applicable
     280       38220 :       if (_target_jacobians.find(elem->type()) == _target_jacobians.end())
     281             :         {
     282        1878 :           const auto [target_elem, target_nodes] = get_target_elem(elem->type());
     283        1878 :           get_target_to_reference_jacobian(target_elem.get(),
     284             :                                            femcontext,
     285        1878 :                                            _target_jacobians[elem->type()],
     286        3756 :                                            _target_jacobian_dets[elem->type()]);
     287             :         }// if find == end()
     288             : 
     289             :       // Reference volume computation
     290        2940 :       Real elem_integrated_det_S = 0.;
     291      334464 :       for (const auto qp : index_range(JxW))
     292      324116 :         elem_integrated_det_S += JxW[qp] / _target_jacobian_dets[elem->type()][qp];
     293       35280 :       const auto ref_elem_vol = elem->reference_elem()->volume();
     294       35280 :       elem_averaged_det_S_sum += elem_integrated_det_S / ref_elem_vol;
     295             : 
     296        2278 :     } // for elem
     297             : 
     298             :   // Get contributions from elements on other processors
     299        2414 :   mesh.comm().sum(elem_averaged_det_S_sum);
     300             : 
     301        2414 :   _ref_vol = elem_averaged_det_S_sum / mesh.n_active_elem();
     302        2414 :   if (_verbosity > 10)
     303           0 :     libMesh::out << "Reference volume: " << _ref_vol << std::endl;
     304        2278 : }
     305             : 
     306       69172 : void VariationalSmootherSystem::init_context(DiffContext & context)
     307             : {
     308        2640 :   FEMContext & c = cast_ref<FEMContext &>(context);
     309             : 
     310        2640 :   FEBase * my_fe = nullptr;
     311             : 
     312             :   // Now make sure we have requested all the data
     313             :   // we need to build the system.
     314             : 
     315             :   // We might have a multi-dimensional mesh
     316             :   const std::set<unsigned char> & elem_dims =
     317        2640 :     c.elem_dimensions();
     318             : 
     319      138344 :   for (const auto & dim : elem_dims)
     320             :     {
     321       69172 :       c.get_element_fe( 0, my_fe, dim );
     322        2640 :       my_fe->get_nothing();
     323             : 
     324        2640 :       auto & fe_map = my_fe->get_fe_map();
     325        2640 :       fe_map.get_dxyzdxi();
     326        2640 :       fe_map.get_dxyzdeta();
     327        2640 :       fe_map.get_dxyzdzeta();
     328        2640 :       fe_map.get_JxW();
     329             : 
     330             :       // Mesh may be tangled, allow negative Jacobians
     331        2640 :       fe_map.set_jacobian_tolerance(std::numeric_limits<Real>::lowest());
     332             : 
     333       69172 :       c.get_side_fe( 0, my_fe, dim );
     334        2640 :       my_fe->get_nothing();
     335             :     }
     336             : 
     337       69172 :   FEMSystem::init_context(context);
     338       69172 : }
     339             : 
     340             : 
     341      377160 : bool VariationalSmootherSystem::element_time_derivative (bool request_jacobian,
     342             :                                              DiffContext & context)
     343             : {
     344       31406 :   FEMContext & c = cast_ref<FEMContext &>(context);
     345             : 
     346       62812 :   const Elem & elem = c.get_elem();
     347             : 
     348      377160 :   unsigned int dim = c.get_dim();
     349             : 
     350       31406 :   unsigned int x_var = 0, y_var = 1, z_var = 2;
     351             :   // In the case of lower dimensions, we will not access z (1D/2D) or y (1D)
     352      377160 :   if (dim < 3)
     353             :   {
     354        3132 :     z_var = 0;
     355       37872 :     if (dim < 2)
     356         100 :       y_var = 0;
     357             :   }
     358             : 
     359             :   // The subvectors and submatrices we need to fill:
     360             :   // system residual
     361             :   std::reference_wrapper<DenseSubVector<Number>> F[3] =
     362             :   {
     363             :     c.get_elem_residual(x_var),
     364             :     c.get_elem_residual(y_var),
     365             :     c.get_elem_residual(z_var)
     366       31406 :   };
     367             :   // system jacobian
     368             :   std::reference_wrapper<DenseSubMatrix<Number>> K[3][3] =
     369             :     {
     370             :       {c.get_elem_jacobian(x_var, x_var), c.get_elem_jacobian(x_var, y_var), c.get_elem_jacobian(x_var, z_var)},
     371             :       {c.get_elem_jacobian(y_var, x_var), c.get_elem_jacobian(y_var, y_var), c.get_elem_jacobian(y_var, z_var)},
     372             :       {c.get_elem_jacobian(z_var, x_var), c.get_elem_jacobian(z_var, y_var), c.get_elem_jacobian(z_var, z_var)}
     373       31406 :     };
     374             : 
     375             :   // Quadrature info
     376       31406 :   const auto & quad_weights = c.get_element_qrule().get_weights();
     377             : 
     378      377160 :   const auto distortion_weight = 1. - _dilation_weight;
     379             : 
     380             :   // Get some references to cell-specific data that
     381             :   // will be used to assemble the linear system.
     382             : 
     383       31406 :   const auto & fe_map = c.get_element_fe(0)->get_fe_map();
     384             : 
     385       31406 :   const auto & dphidxi_map = fe_map.get_dphidxi_map();
     386       31406 :   const auto & dphideta_map = fe_map.get_dphideta_map();
     387       31406 :   const auto & dphidzeta_map = fe_map.get_dphidzeta_map();
     388             : 
     389      377160 :   const auto & target_jacobian_dets = _target_jacobian_dets[elem.type()];
     390      377160 :   const auto & target_jacobians = _target_jacobians[elem.type()];
     391             : 
     392             :   // Integrate the distortion-dilation metric over the reference element
     393     3604056 :   for (const auto qp : index_range(quad_weights))
     394             :     {
     395             :       // Compute quantities needed to evaluate the distortion-dilation metric
     396             :       // and its gradient and Hessian.
     397             :       // The metric will be minimized when it's gradient with respect to the node
     398             :       // locations (R) is zero. For Newton's method, minimizing the gradient
     399             :       // requires computation of the gradient's Jacobian with respect to R.
     400             :       // The Jacobian of the distortion-dilation metric's gradientis the Hessian
     401             :       // of the metric.
     402             : 
     403             :       // Transform quad weight from reference element to quad weight for target element
     404     3495708 :       const auto quad_weight = quad_weights[qp] / target_jacobian_dets[qp];
     405             : 
     406             :       // Note that the term "Jacobian" has two meanings in this mesh smoothing
     407             :       // application. The first meaning refers to the Jacobian w.r.t R of the
     408             :       // gradient w.r.t. R, (i.e., the Hessian of the metric). This is the K
     409             :       // variable defined above. This is also the Jacobian the 'request_jacobian'
     410             :       // variable refers to. The second meaning refers to the Jacobian of the
     411             :       // physical-to-target element mapping. This is the Jacobian used to
     412             :       // compute the distorion-dilation metric.
     413             :       //
     414             :       // Grab the physical-to-reference mapping Jacobian matrix (i.e., "S") at this qp
     415     3495708 :       RealTensor S = get_jacobian_at_qp(fe_map, dim, qp);
     416             : 
     417             :       // Apply target element transformation to get the physical-to-target jacobian
     418     3226896 :       S *= _target_jacobians[elem.type()][qp];
     419             : 
     420             :       // Compute quantities needed for the smoothing algorithm
     421             : 
     422             :       // determinant
     423     3226896 :       const Real det = S.det();
     424     3226896 :       const Real det_sq = det * det;
     425     3226896 :       const Real det_cube = det_sq * det;
     426             : 
     427     3226896 :       const Real ref_vol_sq = _ref_vol * _ref_vol;
     428             : 
     429             :       // trace of S^T * S
     430             :       // DO NOT USE RealTensor.tr for the trace, it will NOT be correct for
     431             :       // 1D and 2D meshes because of our hack of putting 1s in the diagonal of
     432             :       // S for the extra dimensions (see get_jacobian_at_qp)
     433     3226896 :       const auto tr = trace(S.transpose() * S, dim);
     434     3226896 :       const Real tr_div_dim = tr / dim;
     435             : 
     436             :       // Precompute pow(tr_div_dim, 0.5 * dim - x) for x = 0, 1, 2
     437     3226896 :       const Real half_dim = 0.5 * dim;
     438             :       const std::vector<Real> trace_powers{
     439     3226896 :         std::pow(tr_div_dim, half_dim),
     440     3226896 :         std::pow(tr_div_dim, half_dim - 1.),
     441     3226896 :         std::pow(tr_div_dim, half_dim - 2.),
     442     3495708 :       };
     443             : 
     444             :       // inverse of S
     445     3495708 :       const RealTensor S_inv = S.inverse();
     446             :       // inverse transpose of S
     447      537624 :       const RealTensor S_inv_T = S_inv.transpose();
     448             : 
     449             :       // Identity matrix
     450     5378544 :       const RealTensor I(1, 0, 0,
     451     5378544 :                          0, 1, 0,
     452     3226896 :                          0, 0, 1);
     453             : 
     454             :       // The chi function allows us to handle degenerate elements
     455     3226896 :       const auto chi = chi_epsilon(det, _epsilon_squared_assembly);
     456     3226896 :       const Real chi_sq = chi * chi;
     457     3226896 :       const Real sqrt_term = std::sqrt(_epsilon_squared_assembly + det_sq);
     458             :       // dchi(x) / dx
     459     3226896 :       const Real chi_prime = 0.5 * (1. + det / sqrt_term);
     460     3226896 :       const Real chi_prime_sq = chi_prime * chi_prime;
     461             :       // d2chi(x) / dx2
     462     3226896 :       const Real chi_2prime = 0.5 * (1. / sqrt_term - det_sq / Utility::pow<3>(sqrt_term));
     463             : 
     464             :       // Distortion metric (beta)
     465             :       //const Real beta = trace_powers[0] / chi;
     466     4302144 :       const RealTensor dbeta_dS = (trace_powers[1] / chi) * S - (trace_powers[0] / chi_sq * chi_prime * det) * S_inv_T;
     467             : 
     468             :       // Dilation metric (mu)
     469             :       //const Real mu = 0.5 * (_ref_vol + det_sq / _ref_vol) / chi;
     470             :       // We represent d mu / dS as alpha(S) * S^-T, where alpha is a scalar function
     471     3226896 :       const Real alpha = (-chi_prime * det_cube + 2. * det_sq * chi - ref_vol_sq * det * chi_prime) / (2. * _ref_vol * chi_sq);
     472     3226896 :       const RealTensor dmu_dS = alpha * S_inv_T;
     473             : 
     474             :       // Combined metric (E)
     475             :       //const Real E = distortion_weight * beta + _dilation_weight * mu;
     476     4033332 :       const RealTensor dE_dS = distortion_weight * dbeta_dS + _dilation_weight * dmu_dS;
     477             : 
     478             :       // This vector is useful in computing dS/dR below
     479    13982832 :       std::vector<std::vector<std::vector<Real>>> dphi_maps = {dphidxi_map, dphideta_map, dphidzeta_map};
     480             : 
     481             :       // Compute residual (i.e., the gradient of the combined metric w.r.t node locations)
     482             :       // Recall that when the gradient (residual) is zero, the combined metric is minimized
     483    32600448 :       for (const auto l : elem.node_index_range())
     484             :       {
     485   116832288 :         for (const auto var_id : make_range(dim))
     486             :         {
     487             :           // Build dS/dR, the derivative of the physical-to-target mapping Jacobin w.r.t. the mesh
     488             :           // node locations
     489    80171276 :           RealTensor dS_dR = RealTensor(0);
     490   348526944 :           for (const auto jj : make_range(dim))
     491   348084800 :             dS_dR(var_id, jj) = dphi_maps[jj][l][qp];
     492    94746196 :           dS_dR *= target_jacobians[qp];
     493             : 
     494             :           // Residual contribution. The contraction of dE/dS and dS/dR gives us
     495             :           // the gradient we are looking for, dE/dR
     496    94746196 :           F[var_id](l) += quad_weight * dE_dS.contract(dS_dR);
     497             :         }// for var_id
     498             :       }// for l
     499             : 
     500             : 
     501     3226896 :       if (request_jacobian)
     502             :       {
     503             :         // Compute jacobian of the smoothing system (i.e., the Hessian of the
     504             :         // combined metric w.r.t. the mesh node locations)
     505             : 
     506             :         // Precompute coefficients to be applied to each component of tensor
     507             :         // products in the loops below. At first glance, these coefficients look
     508             :         // like gibberish, but everything in the Hessian has been verified by
     509             :         // taking finite differences of the gradient. We should probably write
     510             :         // down the derivations of the gradient and Hessian somewhere...
     511             : 
     512             :         // Recall that above, dbeta_dS takes the form:
     513             :         // d(beta)/dS = c1(S) * S - c2(S) * S_inv_T,
     514             :         // where c1 and c2 are scalar-valued functions.
     515             :         const std::vector<Real> d2beta_dS2_coefs_times_distortion_weight = {
     516             :           //Part 1: scaler coefficients of d(c1 * S) / dS
     517             :           //
     518             :           // multiplies I[i,a] x I[j,b]
     519     1587928 :           (trace_powers[1] / chi) * distortion_weight,
     520             :           // multiplies S[a,b] x S[i,j]
     521     1720250 :           (((dim - 2.) / dim) * trace_powers[2] / chi) * distortion_weight,
     522             :           // multiplies S_inv[b,a] * S[i,j]
     523     1720250 :           (-(trace_powers[1] / chi_sq) * chi_prime * det) * distortion_weight,
     524             :           //
     525             :           //Part 2: scaler coefficients of d(-c2 * S_inv_T) / dS
     526             :           //
     527             :           // multiplies S[a,b] x S_inv[j,i]
     528     1720250 :           (-(trace_powers[1] / chi_sq) * chi_prime * det) * distortion_weight,
     529             :           // multiplies S_inv[b,a] x S_inv[j,i]
     530     1720250 :           (trace_powers[0] * (det / chi_sq)
     531     1587928 :             * ((2. * chi_prime_sq / chi - chi_2prime) * det - chi_prime)) * distortion_weight,
     532             :           // multiplies S_inv[b,i] x S_inv[j,a]
     533     1720250 :           ((trace_powers[0] / chi_sq) * chi_prime * det) * distortion_weight,
     534     1720250 :         };
     535             : 
     536             :         // d alpha / dS has the form c(S) * S^-T, where c is the scalar coefficient defined below
     537     1852572 :         const Real dalpha_dS_coef_times_dilation_weight = ((det / (2. * _ref_vol * chi_sq))
     538     1852572 :           * (-4. * _ref_vol * alpha * chi * chi_prime - chi_2prime * det_cube
     539     1852572 :              - chi_prime * det_sq + 4 * chi * det - ref_vol_sq * (chi_prime + det * chi_2prime))) * _dilation_weight;
     540             : 
     541             :         // This is also useful to precompute
     542     1587928 :         const Real alpha_times_dilation_weight = alpha * _dilation_weight;
     543             : 
     544             :         /*
     545             : 
     546             :         To increase the efficiency of the Jacobian computation, we take
     547             :         advantage of l-p symmetry, ij-ab symmetry, and the sparsity pattern of
     548             :         dS_dR. We also factor all possible multipliers out of inner loops for
     549             :         efficiency. The result is code that is more difficult to read. For
     550             :         clarity, consult the pseudo-code below in this comment.
     551             : 
     552             :         for (const auto l: elem.node_index_range()) // Contribution to Hessian
     553             :         from node l
     554             :         {
     555             :           for (const auto var_id1 : make_range(dim)) // Contribution from each
     556             :         x/y/z component of node l
     557             :           {
     558             :             // Build dS/dR_l, the derivative of the physical-to-target
     559             :         mapping Jacobin w.r.t.
     560             :             // the l-th node
     561             :             RealTensor dS_dR_l = RealTensor(0);
     562             :             for (const auto ii : make_range(dim))
     563             :               dS_dR_l(var_id1, ii) = dphi_maps[ii][l][qp];
     564             :             dS_dR_l *= target_jacobians[qp];
     565             : 
     566             :             for (const auto p: elem.node_index_range()) // Contribution to
     567             :         Hessian from node p
     568             :             {
     569             :               for (const auto var_id2 : make_range(dim)) // Contribution from
     570             :         each x/y/z component of node p
     571             :               {
     572             :                 // Build dS/dR_l, the derivative of the physical-to-target
     573             :         mapping Jacobin w.r.t.
     574             :                 // the p-th node
     575             :                 RealTensor dS_dR_p = RealTensor(0);
     576             :                 for (const auto jj : make_range(dim))
     577             :                   dS_dR_p(var_id2, jj) = dphi_maps[jj][p][qp];
     578             :                 dS_dR_p *= target_jacobians[qp];
     579             : 
     580             :                 Real d2beta_dR2 = 0.;
     581             :                 Real d2mu_dR2 = 0.;
     582             :                 // Perform tensor contraction
     583             :                 for (const auto i : make_range(dim))
     584             :                 {
     585             :                   for (const auto j : make_range(dim))
     586             :                   {
     587             :                     for (const auto a : make_range(dim))
     588             :                     {
     589             :                       for (const auto b : make_range(dim))
     590             :                       {
     591             :                         // Nasty tensor products to be multiplied by
     592             :         d2beta_dS2_coefs to get d2(beta) / dS2 const std::vector<Real>
     593             :         d2beta_dS2_tensor_contributions =
     594             :                         {
     595             :                           I(i,a)     * I(j,b),
     596             :                           S(a,b)     * S(i,j),
     597             :                           S_inv(b,a) * S(i,j),
     598             :                           S(a,b)     * S_inv(j,i),
     599             :                           S_inv(b,a) * S_inv(j,i),
     600             :                           S_inv(j,a) * S_inv(b,i),
     601             :                         };
     602             : 
     603             :                         // Combine precomputed coefficients with tensor products
     604             :         to get d2(beta) / dS2 Real d2beta_dS2 = 0.; for (const auto comp_id :
     605             :         index_range(d2beta_dS2_coefs))
     606             :                         {
     607             :                           const Real contribution = d2beta_dS2_coefs[comp_id] *
     608             :         d2beta_dS2_tensor_contributions[comp_id]; d2beta_dS2 += contribution;
     609             :                         }
     610             : 
     611             :                         // Incorporate tensor product portion to get d2(mu) /
     612             :         dS2 const Real d2mu_dS2 = dalpha_dS_coef * S_inv(b,a) * S_inv(j,i) -
     613             :         alpha * S_inv(b,i) * S_inv(j,a);
     614             : 
     615             :                         // Chain rule to change d/dS to d/dR
     616             :                         d2beta_dR2 += d2beta_dS2 * dS_dR_l(a, b) * dS_dR_p(i,
     617             :         j); d2mu_dR2 += d2mu_dS2 * dS_dR_l(a, b) * dS_dR_p(i, j);
     618             : 
     619             :                       }// for b
     620             :                     }// for a
     621             :                   }// for j
     622             :                 }// for i, end tensor contraction
     623             : 
     624             :                 // Jacobian contribution
     625             :                 K[var_id1][var_id2](l, p) += quad_weight * ((1. -
     626             :         _dilation_weight) * d2beta_dR2 + _dilation_weight * d2mu_dR2);
     627             : 
     628             :               }// for var_id2
     629             :             }// for p
     630             :           }// for var_id1
     631             :         }// for l
     632             : 
     633             :         End pseudo-code, begin efficient code
     634             : 
     635             :         */
     636             : 
     637    15941360 :         for (const auto l: elem.node_index_range()) // Contribution to Hessian from node l
     638             :         {
     639    57136176 :           for (const auto var_id1 : make_range(dim)) // Contribution from each x/y/z component of node l
     640             :           {
     641             :             // Build dS/dR_l, the derivative of the physical-to-target mapping Jacobin w.r.t.
     642             :             // the l-th node
     643    42782744 :             RealTensor dS_dR_l = RealTensor(0);
     644   170583792 :             for (const auto ii : make_range(dim))
     645   170401056 :               dS_dR_l(var_id1, ii) = dphi_maps[ii][l][qp];
     646    46347930 :             dS_dR_l *= target_jacobians[qp];
     647             : 
     648             :             // Jacobian is symmetric, only need to loop over lower triangular portion
     649   342899656 :             for (const auto p: make_range(l + 1)) // Contribution to Hessian from node p
     650             :             {
     651  1198855392 :               for (const auto var_id2 : make_range(dim)) // Contribution from each x/y/z component of node p
     652             :               {
     653             :                 // Build dS/dR_l, the derivative of the physical-to-target mapping Jacobin w.r.t.
     654             :                 // the p-th node
     655   898738480 :                 RealTensor dS_dR_p = RealTensor(0);
     656  3591744096 :                 for (const auto jj : make_range(dim))
     657  3590672448 :                   dS_dR_p(var_id2, jj) = dphi_maps[jj][p][qp];
     658   973633140 :                 dS_dR_p *= target_jacobians[qp];
     659             : 
     660    74894660 :                 Real d2E_dR2 = 0.;
     661             :                 // Perform tensor contraction
     662  3591744096 :                 for (const auto i : make_range(dim))
     663             :                 {
     664             : 
     665 10765617504 :                   for (const auto j : make_range(dim))
     666             :                   {
     667             : 
     668  8072611888 :                     const auto S_ij = S(i,j);
     669  8072611888 :                     const auto S_inv_ji = S_inv(j,i);
     670             : 
     671             :                     // Apply the stuff only depending on i and j before entering a, b loops
     672             :                     // beta
     673             :                     const std::vector<Real> d2beta_dS2_coefs_ij_applied{
     674  8072611888 :                       d2beta_dS2_coefs_times_distortion_weight[1] * S_ij,
     675  8745328692 :                       d2beta_dS2_coefs_times_distortion_weight[2] * S_ij,
     676  8745328692 :                       d2beta_dS2_coefs_times_distortion_weight[3] * S_inv_ji,
     677  8745328692 :                       d2beta_dS2_coefs_times_distortion_weight[4] * S_inv_ji,
     678  8072611888 :                     };
     679             :                     // mu
     680  8072611888 :                     const auto dalpha_dS_coef_ij_applied = dalpha_dS_coef_times_dilation_weight * S_inv_ji;
     681             : 
     682   672716804 :                     Real d2E_dSdR_l = 0.;
     683   672716804 :                     Real d2E_dSdR_p = 0.;
     684             : 
     685 24211438048 :                     for (const auto a : make_range(i + 1))
     686             :                     {
     687             : 
     688             :                       // If this condition is met, both the ijab and abij
     689             :                       // contributions to the Jacobian are zero due to the
     690             :                       // spasity patterns of dS_dR_l and dS_dR_p and this
     691             :                       // iteration may be skipped
     692 16138826160 :                       if (!(a == var_id1 && i == var_id2) &&
     693 13521380992 :                           !(a == var_id2 && i == var_id1))
     694 12325336352 :                         continue;
     695             : 
     696  2693005616 :                       const auto S_inv_ja = S_inv(j,a);
     697             : 
     698  2693005616 :                       const Real d2beta_dS2_coef_ia_applied = d2beta_dS2_coefs_times_distortion_weight[0] * I(i,a);
     699  2693005616 :                       const Real d2beta_dS2_coef_ja_applied = d2beta_dS2_coefs_times_distortion_weight[5] * S_inv_ja;
     700  2693005616 :                       const Real alpha_ja_applied = alpha_times_dilation_weight * S_inv_ja;
     701             : 
     702  2693005616 :                       const auto b_limit = (a == i) ? j + 1 : dim;
     703  9868483936 :                       for (const auto b : make_range(b_limit))
     704             :                       {
     705             : 
     706             :                         // Combine precomputed coefficients with tensor products
     707             :                         // to get d2(beta) / dS2
     708             :                         Real d2beta_dS2_times_distortion_weight = (
     709  7773434100 :                           d2beta_dS2_coef_ia_applied     * I(j,b) +
     710  7175478320 :                           d2beta_dS2_coefs_ij_applied[0] * S(a,b)     +
     711  7175478320 :                           d2beta_dS2_coefs_ij_applied[1] * S_inv(b,a) +
     712  7773434100 :                           d2beta_dS2_coefs_ij_applied[2] * S(a,b)     +
     713  7175478320 :                           d2beta_dS2_coefs_ij_applied[3] * S_inv(b,a) +
     714  7175478320 :                           d2beta_dS2_coef_ja_applied * S_inv(b,i)
     715  7175478320 :                         );
     716             : 
     717             :                         // Incorporate tensor product portion to get d2(mu) /
     718             :                         // dS2
     719  7175478320 :                         const Real d2mu_dS2_times_dilation_weight = dalpha_dS_coef_ij_applied * S_inv(b,a) - alpha_ja_applied * S_inv(b,i);
     720             : 
     721             : 
     722             :                         // Chain rule to change d/dS to d/dR
     723  7175478320 :                         const auto d2E_dS2 =
     724             :                             d2beta_dS2_times_distortion_weight +
     725             :                             d2mu_dS2_times_dilation_weight;
     726             : 
     727             :                         // if !(a == var_id1 (next line) && i == var_id2
     728             :                         // (outside 'a' loop)), dS_dR_l(p) multiplier is zero
     729  7175478320 :                         d2E_dSdR_l += d2E_dS2 * dS_dR_l(a, b);
     730             : 
     731  7175478320 :                         if (!(i == a && j == b))
     732             :                           // if !(a == var_id2 (next line) && i == var_id1
     733             :                           // (outside 'a' loop)), dS_dR_p(l) multiplier is zero
     734  6276739840 :                           d2E_dSdR_p += d2E_dS2 * dS_dR_p(a, b);
     735             : 
     736             :                       } // for b
     737             :                     } // for a
     738  8072611888 :                     d2E_dR2 +=
     739  8072611888 :                         d2E_dSdR_l * dS_dR_p(i, j) + d2E_dSdR_p * dS_dR_l(i, j);
     740             :                   }// for j
     741             :                 }// for i, end tensor contraction
     742             : 
     743             :                 // Jacobian contribution
     744   898738480 :                 const Real jacobian_contribution = quad_weight * d2E_dR2;
     745   898738480 :                 K[var_id1][var_id2](l, p) += jacobian_contribution;
     746             :                 // Jacobian is symmetric, add contribution to p,l entry
     747             :                 // Don't get the diagonal twice!
     748   898738480 :                 if (p < l)
     749             :                   // Note the transposition of var_id1 and var_id2 as these are also jacobian indices
     750   835182090 :                   K[var_id2][var_id1](p, l) += jacobian_contribution;
     751             : 
     752             :               }// for var_id2
     753             :             }// for p
     754             :           }// for var_id1
     755             :         }// for l
     756             :       }
     757     2689272 :     } // end of the quadrature point qp-loop
     758             : 
     759      377160 :   return request_jacobian;
     760     2689272 : }
     761             : 
     762        2556 : const MeshQualityInfo & VariationalSmootherSystem::get_mesh_info()
     763             : {
     764        2556 :   if (!_mesh_info.initialized)
     765        2414 :     compute_mesh_quality_info();
     766             : 
     767        2556 :   return _mesh_info;
     768             : }
     769             : 
     770       33434 : void VariationalSmootherSystem::compute_mesh_quality_info()
     771             : {
     772             :   // If the reference volume has not yet been computed, compute it.
     773       34370 :   if (std::abs(_ref_vol) < TOLERANCE * TOLERANCE)
     774           0 :     prepare_for_smoothing();
     775             : 
     776       34370 :   std::unique_ptr<DiffContext> con = this->build_context();
     777         936 :   FEMContext & femcontext = cast_ref<FEMContext &>(*con);
     778       33434 :   this->init_context(femcontext);
     779             : 
     780        1872 :   const auto & mesh = this->get_mesh();
     781       33434 :   const auto dim = mesh.mesh_dimension();
     782       33434 :   const Real half_dim = 0.5 * dim;
     783       33434 :   const auto distortion_weight = 1. - _dilation_weight;
     784             : 
     785             :   // Make pre-requests before reinit() for efficiency in
     786             :   // --enable-deprecated builds, and to avoid errors in
     787             :   // --disable-deprecated builds.
     788         936 :   const auto & fe_map = femcontext.get_element_fe(0)->get_fe_map();
     789         936 :   const auto & quad_weights = femcontext.get_element_qrule().get_weights();
     790         936 :   const auto & JxW = fe_map.get_JxW();
     791         936 :   fe_map.get_dxyzdxi();
     792         936 :   fe_map.get_dxyzdeta();
     793         936 :   fe_map.get_dxyzdzeta();
     794             : 
     795       33434 :   MeshQualityInfo info;
     796             : 
     797      822120 :   for (const auto * elem : mesh.active_local_element_ptr_range())
     798             :     {
     799      412440 :       femcontext.pre_fe_reinit(*this, elem);
     800      412440 :       femcontext.elem_fe_reinit();
     801             : 
     802             :       // Element-integrated quantities
     803       34346 :       Real det_S_int = 0.;
     804       34346 :       Real beta_int = 0.;
     805       34346 :       Real mu_int = 0.;
     806       34346 :       Real combined_int = 0.;
     807             : 
     808      412440 :       const auto & target_jacobian_dets = _target_jacobian_dets[elem->type()];
     809             : 
     810     3938520 :       for (const auto qp : index_range(JxW))
     811             :         {
     812     3819824 :           det_S_int += JxW[qp] / _target_jacobian_dets[elem->type()][qp];
     813     3819824 :           const auto quad_weight = quad_weights[qp] / target_jacobian_dets[qp];
     814             : 
     815             :           // Grab the physical-to-reference mapping Jacobian matrix (i.e., "S") at this qp
     816     3526080 :           RealTensor S = get_jacobian_at_qp(fe_map, dim, qp);
     817             : 
     818             :           // Apply target element transformation to get the physical-to-target jacobian
     819     3526080 :           S *= _target_jacobians[elem->type()][qp];
     820             : 
     821             :           // Determinant of S
     822     3526080 :           const auto det = S.det();
     823     3526080 :           const auto det_sq = det * det;
     824             : 
     825     3526080 :           if (det > info.max_qp_det_S)
     826      107138 :             info.max_qp_det_S = det;
     827     3418942 :           else if (det < info.min_qp_det_S)
     828      125286 :             info.min_qp_det_S = det;
     829             : 
     830     3526080 :           if (det < TOLERANCE * TOLERANCE)
     831        1332 :             info.mesh_is_tangled = true;
     832             : 
     833             :           // trace of S^T * S
     834     3526080 :           const auto tr = trace(S.transpose() * S, dim);
     835             : 
     836             :           // The chi function allows us to handle degenerate elements
     837     3526080 :           const auto chi = chi_epsilon(det, _epsilon_squared_assembly);
     838             : 
     839             :           // distortion
     840     3526080 :           const Real beta = std::pow(tr / dim, half_dim) / chi;
     841     3526080 :           beta_int += beta * quad_weight;
     842             : 
     843             :           // dilation
     844     3526080 :           const Real mu = 0.5 * (_ref_vol + det_sq / _ref_vol) / chi;
     845     3526080 :           mu_int += mu * quad_weight;
     846             : 
     847             :           // combined
     848     3526080 :           const Real E = distortion_weight * beta + _dilation_weight * mu;
     849     3526080 :           combined_int += E * quad_weight;
     850             :         }
     851             : 
     852      412440 :       info.total_det_S += det_S_int;
     853      412440 :       if (det_S_int > info.max_elem_det_S.first)
     854        5209 :         info.max_elem_det_S = std::make_pair(det_S_int, elem->id());
     855      412440 :       if (det_S_int < info.min_elem_det_S.first)
     856        6414 :         info.min_elem_det_S = std::make_pair(det_S_int, elem->id());
     857             : 
     858      412440 :       info.total_distortion += beta_int;
     859      412440 :       if (beta_int > info.max_elem_distortion.first)
     860        5234 :         info.max_elem_distortion = std::make_pair(beta_int, elem->id());
     861      412440 :       if (beta_int < info.min_elem_distortion.first)
     862        5955 :         info.min_elem_distortion = std::make_pair(beta_int, elem->id());
     863             : 
     864      412440 :       info.total_dilation += mu_int;
     865      412440 :       if (mu_int > info.max_elem_dilation.first)
     866        4631 :         info.max_elem_dilation = std::make_pair(mu_int, elem->id());
     867      412440 :       if (mu_int < info.min_elem_dilation.first)
     868        5710 :         info.min_elem_dilation = std::make_pair(mu_int, elem->id());
     869             : 
     870      412440 :       info.total_combined += combined_int;
     871      412440 :       if (combined_int > info.max_elem_combined.first)
     872        5032 :         info.max_elem_combined = std::make_pair(combined_int, elem->id());
     873      412440 :       if (combined_int < info.min_elem_combined.first)
     874        5940 :         info.min_elem_combined = std::make_pair(combined_int, elem->id());
     875             : 
     876      412440 :       if (_verbosity > 90)
     877             :         {
     878           0 :           libMesh::out << "Elem " << elem->id() << " quality:" << std::endl
     879           0 :                        << "  distortion-dilation metric: " << combined_int << std::endl
     880           0 :                        << "  distortion metric: " << beta_int << std::endl
     881           0 :                        << "  dilation metric: " << mu_int << std::endl
     882           0 :                        << "  det(S): " << det_S_int << std::endl;
     883             :         }
     884             : 
     885       31562 :     } // for elem
     886             : 
     887             :   // Get contributions from elements on other processors
     888       33434 :   communicate_pair_max(info.max_elem_det_S, mesh.comm());
     889       33434 :   communicate_pair_min(info.min_elem_det_S, mesh.comm());
     890       33434 :   mesh.comm().max(info.max_qp_det_S);
     891       33434 :   mesh.comm().min(info.min_qp_det_S);
     892       33434 :   mesh.comm().sum(info.total_det_S);
     893             : 
     894       33434 :   communicate_pair_max(info.max_elem_distortion, mesh.comm());
     895       33434 :   communicate_pair_min(info.min_elem_distortion, mesh.comm());
     896       33434 :   mesh.comm().sum(info.total_distortion);
     897             : 
     898       33434 :   communicate_pair_max(info.max_elem_dilation, mesh.comm());
     899       33434 :   communicate_pair_min(info.min_elem_dilation, mesh.comm());
     900       33434 :   mesh.comm().sum(info.total_dilation);
     901             : 
     902       33434 :   communicate_pair_max(info.max_elem_combined, mesh.comm());
     903       33434 :   communicate_pair_min(info.min_elem_combined, mesh.comm());
     904       33434 :   mesh.comm().sum(info.total_combined);
     905             : 
     906       33434 :   mesh.comm().max(info.mesh_is_tangled);
     907             : 
     908       33434 :   info.initialized = true;
     909             : 
     910       33434 :   _mesh_info = info;
     911             : 
     912       33434 :   if (_verbosity > 50)
     913           0 :     libMesh::out << info;
     914       33434 : }
     915             : 
     916             : std::pair<std::unique_ptr<Elem>, std::vector<std::unique_ptr<Node>>>
     917        1878 : VariationalSmootherSystem::get_target_elem(const ElemType & type)
     918             : {
     919             :   // Build target element
     920        1948 :   auto target_elem = Elem::build(type);
     921             : 
     922             :   // Volume of reference element
     923        1878 :   const auto ref_vol = target_elem->reference_elem()->volume();
     924             : 
     925             :   // Update the nodes of the target element, depending on type
     926          70 :   const Real sqrt_2 = std::sqrt(Real(2));
     927          70 :   const Real sqrt_3 = std::sqrt(Real(3));
     928         210 :   std::vector<std::unique_ptr<Node>> owned_nodes;
     929             : 
     930        1948 :   const auto type_str = Utility::enum_to_string(type);
     931             : 
     932             :   // Elems deriving from Tri
     933        1878 :   if (type_str.compare(0, 3, "TRI") == 0)
     934             :     {
     935             : 
     936             :       // The target element will be an equilateral triangle with area equal to
     937             :       // the area of the reference element.
     938             : 
     939             :       // Equilateral triangle side length preserving area of the reference element
     940         245 :       const auto side_length = std::sqrt(4. / sqrt_3 * ref_vol);
     941             : 
     942             :       // Define the nodal locations of the vertices
     943           8 :       const auto & s = side_length;
     944             :       //                                         x        y                  node_id
     945         253 :       owned_nodes.emplace_back(Node::build(Point(0.,      0.),               0));
     946         245 :       owned_nodes.emplace_back(Node::build(Point(s,       0.),               1));
     947         253 :       owned_nodes.emplace_back(Node::build(Point(0.5 * s, 0.5 * sqrt_3 * s), 2));
     948             : 
     949         245 :       switch (type)
     950             :         {
     951           6 :             case TRI3: {
     952             :               // Nothing to do here, vertices already added above
     953           6 :               break;
     954             :             }
     955             : 
     956          55 :             case TRI6: {
     957             :               // Define the midpoint nodes of the equilateral triangle
     958             :               //                                         x         y                   node_id
     959          55 :               owned_nodes.emplace_back(Node::build(Point(0.50 * s, 0.00),              3));
     960          57 :               owned_nodes.emplace_back(Node::build(Point(0.75 * s, 0.25 * sqrt_3 * s), 4));
     961          57 :               owned_nodes.emplace_back(Node::build(Point(0.25 * s, 0.25 * sqrt_3 * s), 5));
     962             : 
     963          55 :               break;
     964             :             }
     965             : 
     966           0 :           default:
     967           0 :             libmesh_error_msg("Unsupported triangular element: " << type_str);
     968             :             break;
     969             :         }
     970             :     } // if Tri
     971             : 
     972             :   // Elems deriving from Prism
     973        1633 :   else if (type_str.compare(0, 5, "PRISM") == 0)
     974             :     {
     975             : 
     976             :       // The target element will be a prism with an equilateral triangular
     977             :       // base with volume equal to the volume of the reference element.
     978             : 
     979             :       // For an equilateral triangular base with side length s, the
     980             :       // base area is s^2 * sqrt(3) / 4.
     981             :       // The prism height that will result in equal face areas is
     982             :       // s * sqrt(3) / 4. We choose s such that the target element has
     983             :       // the same volume as the reference element:
     984             :       // v = (s^2 * sqrt(3) / 4) * (s * sqrt(3) / 4) = 3 * s^3 / 4
     985             :       // --> s = (16 * v / 3)^(1/3)
     986             :       // I have no particular motivation for imposing equal face areas,
     987             :       // so this can be updated if a more `optimal` target prism is
     988             :       // identified.
     989             : 
     990             :       // Side length that preserves the volume of the reference element
     991         336 :       const auto side_length = std::cbrt(16. * ref_vol / 3.);
     992             :       // Prism height with the property that all faces have equal area
     993         336 :       const auto target_height = 0.25 * side_length * sqrt_3;
     994             : 
     995          14 :       const auto & s = side_length;
     996          14 :       const auto & h = target_height;
     997             :       //                                         x        y                 z    node_id
     998         350 :       owned_nodes.emplace_back(Node::build(Point(0.,      0.,               0.), 0));
     999         336 :       owned_nodes.emplace_back(Node::build(Point(s,       0.,               0.), 1));
    1000         364 :       owned_nodes.emplace_back(Node::build(Point(0.5 * s, 0.5 * sqrt_3 * s, 0.), 2));
    1001         350 :       owned_nodes.emplace_back(Node::build(Point(0.,      0.,               h),  3));
    1002         350 :       owned_nodes.emplace_back(Node::build(Point(s,       0.,               h),  4));
    1003         336 :       owned_nodes.emplace_back(Node::build(Point(0.5 * s, 0.5 * sqrt_3 * s, h),  5));
    1004             : 
    1005         336 :       if (type == PRISM15 || type == PRISM18 || type == PRISM20 || type == PRISM21)
    1006             :         {
    1007             :           // Define the edge midpoint nodes of the prism
    1008          10 :           const auto & on = owned_nodes;
    1009         217 :           owned_nodes.emplace_back(Node::build(Point((*on[0] + *on[1]) / 2.), 6));
    1010         217 :           owned_nodes.emplace_back(Node::build(Point((*on[1] + *on[2]) / 2.), 7));
    1011         217 :           owned_nodes.emplace_back(Node::build(Point((*on[2] + *on[0]) / 2.), 8));
    1012         217 :           owned_nodes.emplace_back(Node::build(Point((*on[0] + *on[3]) / 2.), 9));
    1013         217 :           owned_nodes.emplace_back(Node::build(Point((*on[1] + *on[4]) / 2.), 10));
    1014         217 :           owned_nodes.emplace_back(Node::build(Point((*on[2] + *on[5]) / 2.), 11));
    1015         217 :           owned_nodes.emplace_back(Node::build(Point((*on[3] + *on[4]) / 2.), 12));
    1016         217 :           owned_nodes.emplace_back(Node::build(Point((*on[4] + *on[5]) / 2.), 13));
    1017         217 :           owned_nodes.emplace_back(Node::build(Point((*on[5] + *on[3]) / 2.), 14));
    1018             : 
    1019         207 :           if (type == PRISM18 || type == PRISM20 || type == PRISM21)
    1020             :             {
    1021             :               // Define the rectangular face midpoint nodes of the prism
    1022         145 :               owned_nodes.emplace_back(Node::build(Point((*on[0] + *on[1] + *on[3] + *on[4]) / 4.), 15));
    1023         145 :               owned_nodes.emplace_back(Node::build(Point((*on[1] + *on[2] + *on[4] + *on[5]) / 4.), 16));
    1024         145 :               owned_nodes.emplace_back(Node::build(Point((*on[0] + *on[2] + *on[3] + *on[5]) / 4.), 17));
    1025             : 
    1026         137 :               if (type == PRISM20 || type == PRISM21)
    1027             :                 {
    1028             :                   // Define the triangular face midpoint nodes of the prism
    1029          72 :                   owned_nodes.emplace_back(Node::build(Point((*on[0] + *on[1] + *on[2]) / 3.), 18));
    1030          72 :                   owned_nodes.emplace_back(Node::build(Point((*on[3] + *on[4] + *on[5]) / 3.), 19));
    1031             : 
    1032          66 :                   if (type == PRISM21)
    1033             :                     // Define the interior point of the prism
    1034          52 :                     owned_nodes.emplace_back(Node::build(Point((*on[9] + *on[10] + *on[11]) / 3.), 20));
    1035             : 
    1036             :                 }
    1037          10 :             }
    1038             :         }
    1039             : 
    1040         129 :       else if (type != PRISM6)
    1041           0 :         libmesh_error_msg("Unsupported prism element: " << type_str);
    1042             : 
    1043             :     } // if Prism
    1044             : 
    1045             :   // Elems deriving from Pyramid
    1046        1297 :   else if (type_str.compare(0, 7, "PYRAMID") == 0)
    1047             :     {
    1048             : 
    1049             :       // The target element is a pyramid with an square base and
    1050             :       // equilateral triangular sides with volume equal to the volume of the
    1051             :       // reference element.
    1052             : 
    1053             :       // A pyramid with square base sidelength s and equilateral triangular
    1054             :       // sides has height h = s / sqrt(2).
    1055             :       // The volume is v = s^2 h / 3 = s^3 / ( 3 sqrt(2)).
    1056             :       // Solving for s: s = (3 sqrt(2) v)^(1/3), where v is the volume of the
    1057             :       // non-optimal reference element.
    1058             : 
    1059             :       // Side length that preserves the volume of the reference element
    1060         305 :       const auto side_length = std::cbrt(3. * sqrt_2 * ref_vol);
    1061             :       // Pyramid height with the property that all faces are equilateral triangles
    1062         305 :       const auto target_height = side_length / sqrt_2;
    1063             : 
    1064          10 :       const auto & s = side_length;
    1065          10 :       const auto & h = target_height;
    1066             : 
    1067             :       //                                         x        y        z    node_id
    1068         315 :       owned_nodes.emplace_back(Node::build(Point(0.,      0.,      0.), 0));
    1069         315 :       owned_nodes.emplace_back(Node::build(Point(s,       0.,      0.), 1));
    1070         315 :       owned_nodes.emplace_back(Node::build(Point(s,       s,       0.), 2));
    1071         305 :       owned_nodes.emplace_back(Node::build(Point(0.,      s,       0.), 3));
    1072         315 :       owned_nodes.emplace_back(Node::build(Point(0.5 * s, 0.5 * s, h),  4));
    1073             : 
    1074         305 :       if (type == PYRAMID13 || type == PYRAMID14 || type == PYRAMID18)
    1075             :         {
    1076           8 :           const auto & on = owned_nodes;
    1077             :           // Define the edge midpoint nodes of the pyramid
    1078             : 
    1079             :           // Base node to base node midpoint nodes
    1080         242 :           owned_nodes.emplace_back(Node::build((*on[0] + *on[1]) / 2., 5));
    1081         242 :           owned_nodes.emplace_back(Node::build((*on[1] + *on[2]) / 2., 6));
    1082         242 :           owned_nodes.emplace_back(Node::build((*on[2] + *on[3]) / 2., 7));
    1083         242 :           owned_nodes.emplace_back(Node::build((*on[3] + *on[0]) / 2., 8));
    1084             : 
    1085             :           // Base node to apex node midpoint nodes
    1086         242 :           owned_nodes.emplace_back(Node::build(Point((*on[0] + *on[4]) / 2.), 9));
    1087         242 :           owned_nodes.emplace_back(Node::build(Point((*on[1] + *on[4]) / 2.), 10));
    1088         242 :           owned_nodes.emplace_back(Node::build(Point((*on[2] + *on[4]) / 2.), 11));
    1089         242 :           owned_nodes.emplace_back(Node::build(Point((*on[3] + *on[4]) / 2.), 12));
    1090             : 
    1091         234 :           if (type == PYRAMID14 || type == PYRAMID18)
    1092             :             {
    1093             :               // Define the square face midpoint node of the pyramid
    1094         151 :               owned_nodes.emplace_back(
    1095         169 :                   Node::build(Point((*on[0] + *on[1] + *on[2] + *on[3]) / 4.), 13));
    1096             : 
    1097         163 :               if (type == PYRAMID18)
    1098             :                 {
    1099             :                   // Define the triangular face nodes
    1100          96 :                   owned_nodes.emplace_back(Node::build(Point((*on[0] + *on[1] + *on[4]) / 3.), 14));
    1101          96 :                   owned_nodes.emplace_back(Node::build(Point((*on[1] + *on[2] + *on[4]) / 3.), 15));
    1102          96 :                   owned_nodes.emplace_back(Node::build(Point((*on[2] + *on[3] + *on[4]) / 3.), 16));
    1103         100 :                   owned_nodes.emplace_back(Node::build(Point((*on[3] + *on[0] + *on[4]) / 3.), 17));
    1104             :                 }
    1105           8 :             }
    1106             :         }
    1107             : 
    1108          71 :       else if (type != PYRAMID5)
    1109           0 :         libmesh_error_msg("Unsupported pyramid element: " << type_str);
    1110             : 
    1111             :     } // if Pyramid
    1112             : 
    1113             :   // Elems deriving from Tet
    1114         992 :   else if (type_str.compare(0, 3, "TET") == 0)
    1115             :     {
    1116             : 
    1117             :       // The ideal target element is a a regular tet with equilateral
    1118             :       // triangles for all faces, with volume equal to the volume of the
    1119             :       // reference element.
    1120             : 
    1121             :       // The volume of a tet is given by v = b * h / 3, where b is the area of
    1122             :       // the base face and h is the height of the apex node. The area of an
    1123             :       // equilateral triangle with side length s is b = sqrt(3) s^2 / 4.
    1124             :       // For all faces to have side length s, the height of the apex node is
    1125             :       // h = sqrt(2/3) * s. Then the volume is v = sqrt(2) * s^3 / 12.
    1126             :       // Solving for s, the side length that will preserve the volume of the
    1127             :       // reference element is s = (6 * sqrt(2) * v)^(1/3), where v is the volume
    1128             :       // of the non-optimal reference element (i.e., a right tet).
    1129             : 
    1130             :       // Side length that preserves the volume of the reference element
    1131         278 :       const auto side_length = std::cbrt(6. * sqrt_2 * ref_vol);
    1132             :       // tet height with the property that all faces are equilateral triangles
    1133         278 :       const auto target_height = sqrt_2 / sqrt_3 * side_length;
    1134             : 
    1135           8 :       const auto & s = side_length;
    1136           8 :       const auto & h = target_height;
    1137             : 
    1138             :       // For regular tet
    1139             :       //                                         x        y                z     node_id
    1140         286 :       owned_nodes.emplace_back(Node::build(Point(0.,      0.,               0.), 0));
    1141         278 :       owned_nodes.emplace_back(Node::build(Point(s,       0.,               0.), 1));
    1142         286 :       owned_nodes.emplace_back(Node::build(Point(0.5 * s, 0.5 * sqrt_3 * s, 0.), 2));
    1143         286 :       owned_nodes.emplace_back(Node::build(Point(0.5 * s, sqrt_3 / 6. * s,  h),  3));
    1144             : 
    1145         278 :       if (type == TET10 || type == TET14)
    1146             :         {
    1147           6 :           const auto & on = owned_nodes;
    1148             :           // Define the edge midpoint nodes of the tet
    1149             : 
    1150             :           // Base node to base node midpoint nodes
    1151         213 :           owned_nodes.emplace_back(Node::build(Point((*on[0] + *on[1]) / 2.), 4));
    1152         213 :           owned_nodes.emplace_back(Node::build(Point((*on[1] + *on[2]) / 2.), 5));
    1153         213 :           owned_nodes.emplace_back(Node::build(Point((*on[2] + *on[0]) / 2.), 6));
    1154             :           // Base node to apex node midpoint nodes
    1155         213 :           owned_nodes.emplace_back(Node::build(Point((*on[0] + *on[3]) / 2.), 7));
    1156         213 :           owned_nodes.emplace_back(Node::build(Point((*on[1] + *on[3]) / 2.), 8));
    1157         213 :           owned_nodes.emplace_back(Node::build(Point((*on[2] + *on[3]) / 2.), 9));
    1158             : 
    1159         207 :           if (type == TET14)
    1160             :             {
    1161             :               // Define the face midpoint nodes of the tet
    1162         140 :               owned_nodes.emplace_back(Node::build(Point((*on[0] + *on[1] + *on[2]) / 3.), 10));
    1163         140 :               owned_nodes.emplace_back(Node::build(Point((*on[0] + *on[1] + *on[3]) / 3.), 11));
    1164         140 :               owned_nodes.emplace_back(Node::build(Point((*on[1] + *on[2] + *on[3]) / 3.), 12));
    1165         144 :               owned_nodes.emplace_back(Node::build(Point((*on[0] + *on[2] + *on[3]) / 3.), 13));
    1166           6 :             }
    1167             :         }
    1168             : 
    1169          71 :       else if (type != TET4)
    1170           0 :         libmesh_error_msg("Unsupported tet element: " << type_str);
    1171             : 
    1172             :     } // if Tet
    1173             : 
    1174             :   // Set the target_elem equal to the reference elem
    1175             :   else
    1176        9087 :     for (const auto & node : target_elem->reference_elem()->node_ref_range())
    1177        8987 :       owned_nodes.emplace_back(Node::build(node, node.id()));
    1178             : 
    1179             :   // Set nodes of target element
    1180       22413 :   for (const auto & node_ptr : owned_nodes)
    1181       21321 :     target_elem->set_node(node_ptr->id(), node_ptr.get());
    1182             : 
    1183          70 :   libmesh_assert(relative_fuzzy_equals(target_elem->volume(), ref_vol, TOLERANCE));
    1184             : 
    1185        1948 :   return std::make_pair(std::move(target_elem), std::move(owned_nodes));
    1186        1738 : }
    1187             : 
    1188        1878 : void VariationalSmootherSystem::get_target_to_reference_jacobian(
    1189             :     const Elem * const target_elem,
    1190             :     const FEMContext & femcontext,
    1191             :     std::vector<RealTensor> & jacobians,
    1192             :     std::vector<Real> & jacobian_dets)
    1193             : {
    1194             : 
    1195        1878 :   const auto dim = target_elem->dim();
    1196             : 
    1197          70 :   const auto & qrule_points = femcontext.get_element_qrule().get_points();
    1198          70 :   const auto & qrule_weights = femcontext.get_element_qrule().get_weights();
    1199          70 :   const auto nq_points = femcontext.get_element_qrule().n_points();
    1200             : 
    1201             :   // If the target element is the reference element, Jacobian matrix is
    1202             :   // identity, det of inverse is 1. These will only be overwritten if a
    1203             :   // different target element is explicitly specified.
    1204        1948 :   jacobians = std::vector<RealTensor>(nq_points, RealTensor(
    1205             :         1., 0., 0.,
    1206             :         0., 1., 0.,
    1207          70 :         0., 0., 1.));
    1208        1878 :   jacobian_dets = std::vector<Real>(nq_points, 1.0);
    1209             : 
    1210             :   // Don't use "if (*target_elem == *(target_elem->reference_elem()))" here, it
    1211             :   // only compares global node ids, not the node locations themselves.
    1212          70 :   bool target_equals_reference = true;
    1213        1878 :   const auto * ref_elem = target_elem->reference_elem();
    1214       22413 :   for (const auto local_id : make_range(target_elem->n_nodes()))
    1215       20535 :     target_equals_reference &= target_elem->node_ref(local_id) == ref_elem->node_ref(local_id);
    1216        1878 :   if (target_equals_reference)
    1217         714 :     return;
    1218             : 
    1219             :   // Create FEMap to compute target_element mapping information
    1220        1244 :   FEMap fe_map_target;
    1221             : 
    1222             :   // pre-request mapping derivatives
    1223          40 :   fe_map_target.get_dxyzdxi();
    1224          40 :   fe_map_target.get_dxyzdeta();
    1225          40 :   fe_map_target.get_dxyzdzeta();
    1226             : 
    1227             :   // build map
    1228        1164 :   fe_map_target.init_reference_to_physical_map(dim, qrule_points, target_elem);
    1229        1164 :   fe_map_target.compute_map(dim, qrule_weights, target_elem, /*d2phi=*/false);
    1230             : 
    1231       27229 :   for (const auto qp : make_range(nq_points))
    1232             :     {
    1233             :       // We use Larisa's H notation to denote the reference-to-target jacobian
    1234       26065 :       RealTensor H = get_jacobian_at_qp(fe_map_target, dim, qp);
    1235             : 
    1236             :       // The target-to-reference jacobian is the inverse of the
    1237             :       // reference-to-target jacobian
    1238       26065 :       jacobians[qp] = H.inverse();
    1239       27105 :       jacobian_dets[qp] = jacobians[qp].det();
    1240             :     }
    1241        1084 : }
    1242             : 
    1243           0 : std::ostream &operator<<(std::ostream &os, const MeshQualityInfo & info)
    1244             : {
    1245           0 :   os << "Mesh quality info:" << std::endl
    1246           0 :      << "  Mesh distortion-dilation metric: "
    1247           0 :      << info.total_combined << std::endl
    1248           0 :      << "  Mesh distortion metric: "
    1249           0 :      << info.total_distortion << std::endl
    1250           0 :      << "  Mesh dilation metric: "
    1251           0 :      << info.total_dilation << std::endl
    1252           0 :      << "  Max distortion-dilation is in elem "
    1253           0 :      << info.max_elem_combined.second << ": "
    1254           0 :      << info.max_elem_combined.first << std::endl
    1255           0 :      << "  Max distortion is in elem "
    1256           0 :      << info.max_elem_distortion.second << ": "
    1257           0 :      << info.max_elem_distortion.first << std::endl
    1258           0 :      << "  Max dilation is in elem "
    1259           0 :      << info.max_elem_dilation.second << ": "
    1260           0 :      << info.max_elem_dilation.first << std::endl
    1261           0 :      << "  Max det(S) is in elem "
    1262           0 :      << info.max_elem_det_S.second << ": "
    1263           0 :      << info.max_elem_det_S.first << std::endl
    1264           0 :      << "  Min distortion-dilation is in elem "
    1265           0 :      << info.min_elem_combined.second << ": "
    1266           0 :      << info.min_elem_combined.first << std::endl
    1267           0 :      << "  Min distortion is in elem "
    1268           0 :      << info.min_elem_distortion.second << ": "
    1269           0 :      << info.min_elem_distortion.first << std::endl
    1270           0 :      << "  Min dilation is in elem "
    1271           0 :      << info.min_elem_dilation.second << ": "
    1272           0 :      << info.min_elem_dilation.first << std::endl
    1273           0 :      << "  Min det(S) is in elem "
    1274           0 :      << info.min_elem_det_S.second << ": "
    1275           0 :      << info.min_elem_det_S.first << std::endl
    1276           0 :      << "  Max qp det(S): " << info.max_qp_det_S << std::endl
    1277           0 :      << "  Min qp det(S): " << info.min_qp_det_S << std::endl
    1278           0 :      << "  Mesh-integrated det(S): " << info.total_det_S << std::endl
    1279           0 :      << "  Tangled: " << info.mesh_is_tangled << std::endl;
    1280             : 
    1281           0 :   return os;
    1282             : }
    1283             : 
    1284             : } // namespace libMesh
 |