Fix test error

Adding H(div) and H(curl) conforming NURBS FiniteElements

.gitignore

@@ -57,6 +57,8 @@ examples/ex2[0-9]
 examples/ex2[0-9]p
 examples/ex3[0-9]
 examples/ex3[0-9]p
+examples/ex4[0-9]
+examples/ex4[0-9]p
 
 examples/refined.mesh
 examples/displaced.mesh
@@ -270,16 +272,27 @@ miniapps/navier/*_output
 
 miniapps/nurbs/nurbs_ex1
 miniapps/nurbs/nurbs_ex1p
+miniapps/nurbs/nurbs_ex3
+miniapps/nurbs/nurbs_ex5
 miniapps/nurbs/nurbs_ex11p
+miniapps/nurbs/nurbs_ex24
+miniapps/nurbs/nurbs_solenoidal
 miniapps/nurbs/nurbs_printfunc
 miniapps/nurbs/nurbs_patch_ex1
 miniapps/nurbs/nurbs_curveint
 miniapps/nurbs/refined.mesh
 miniapps/nurbs/mesh.*
+miniapps/nurbs/sol_?.gf
 miniapps/nurbs/sol.*
 miniapps/nurbs/mode_*
 miniapps/nurbs/Example1*
+miniapps/nurbs/Example3*
+miniapps/nurbs/Example5*
+miniapps/nurbs/Solenoidal*
+miniapps/nurbs/ParaView
 miniapps/nurbs/sin-fit.mesh
+miniapps/nurbs/ex5.mesh
+miniapps/nurbs/exsol.mesh
 miniapps/nurbs/CurveInt
 miniapps/nurbs/nurbs_naca_cmesh
 miniapps/nurbs/naca-cmesh.mesh

.gitlab-ci.yml

@@ -13,6 +13,9 @@
 # at Lawrence Livermore National Laboratory (LLNL). This entire pipeline is
 # LLNL-specific!
 
+include:
+  - project: 'lc-templates/id_tokens'
+    file: 'id_tokens.yml'
 
 # The pipeline is divided into stages. Usually, jobs in a given stage wait for
 # the preceding stages to complete before to start. However, we sometimes use

.gitlab/configs/common.yml

@@ -9,6 +9,10 @@
 # terms of the BSD-3 license. We welcome feedback and contributions, see file
 # CONTRIBUTING.md for details.
 
+include:
+  - project: 'lc-templates/id_tokens'
+    file: 'id_tokens.yml'
+
 # We define the following GitLab pipeline variables:
 variables:
 

.gitlab/configs/lassen-config.yml

@@ -35,9 +35,8 @@ variables:
     - when: on_success
 
 # Lassen uses a different job scheduler (spectrum lsf) that does not allow
-# pre-allocation the same way slurm does.  We use pdebug queue on lassen
-# to speed-up the allocation.  However this would not be scalable to
-# multiple builds.
+# pre-allocation the same way slurm does.  We use the pci queue on lassen
+# to speed-up the allocation.
 .build_and_test_on_lassen:
   extends: [.on_lassen]
   stage: build_and_test
@@ -45,5 +44,5 @@ variables:
     - echo ${MFEM_DATA_DIR}
     - echo ${SPEC}
     # Next script uses 'THREADS': leaving it empty --> it uses 'make all -j'
-    - lalloc 1 -W 45 -q pdebug --atsdisable tests/gitlab/build_and_test --spec "${SPEC}" --data-dir "${MFEM_DATA_DIR}" --data
+    - lalloc 1 -W 45 -q pci --atsdisable tests/gitlab/build_and_test --spec "${SPEC}" --data-dir "${MFEM_DATA_DIR}" --data
   needs: [setup]

.gitlab/configs/quartz-config.yml

@@ -52,4 +52,4 @@ variables:
     - echo ${JOBID}
     - echo ${MFEM_DATA_DIR}
     - echo ${SPEC}
-    - srun $( [[ -n "${JOBID}" ]] && echo "--jobid=${JOBID}" ) -t 45 -N 1 tests/gitlab/build_and_test --spec "${SPEC}" --data-dir "${MFEM_DATA_DIR}" --data
+    - srun $( [[ -n "${JOBID}" ]] && echo "--jobid=${JOBID}" ) --reservation=ci -t 45 -N 1 tests/gitlab/build_and_test --spec "${SPEC}" --data-dir "${MFEM_DATA_DIR}" --data

.gitlab/lassen-build-and-test.yml

@@ -14,14 +14,14 @@ stages:
   - build_and_test
   - report
 
-opt_mpi_cuda_xl_16_1_1_12:
+opt_mpi_cuda_gcc:
   variables:
-    SPEC: "%xl@16.1.1.12 +mpi +cuda cuda_arch=70"
+    SPEC: "%gcc@8.3.1 +mpi +cuda cuda_arch=70"
   extends: .build_and_test_on_lassen
 
-opt_mpi_cuda_hypre_cuda_xl:
+opt_mpi_cuda_hypre_cuda_gcc:
   variables:
-    SPEC: "%xl@16.1.1.12 +mpi +cuda cuda_arch=70 ^hypre+cuda~shared cuda_arch=70"
+    SPEC: "%gcc@8.3.1 +mpi +cuda cuda_arch=70 ^hypre+cuda~shared cuda_arch=70"
   extends: .build_and_test_on_lassen
 
 # Jobs report

.gitlab/scripts/baseline

@@ -32,11 +32,11 @@ mkdir _${BASELINE_TEST} && cd _${BASELINE_TEST}
 
 # run
 if [[ "${MACHINE_NAME}" == "quartz" || "${MACHINE_NAME}" == "ruby" ]]; then
-  salloc --nodes=1 -p pdebug ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
+  salloc --nodes=1 --reservation=ci ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
 elif [[ ${MACHINE_NAME} == "corona" ]]; then
   salloc --nodes=1 -t 60 -p pbatch ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
 elif [[ ${MACHINE_NAME} == "lassen" ]]; then
-  lalloc 1 -q pdebug ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
+  lalloc 1 -q pci ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
 else
   echo "Unknown machine: MACHINE_NAME=$MACHINE_NAME"
   exit 1

CHANGELOG

@@ -11,6 +11,16 @@
 Version 4.7.1 (development)
 ===========================
 
+- Added NURBS-based H(div) and H(curl) elements in 2D and 3D. Only on single 
+  patch meshes. Only implemented for serial computations.
+
+- Added miniapps to demonstrate the H(div) and H(curl) NURBS elements.
+
+- Added an MFEM example for the eikonal equation. This new solver is based on
+  the proximal Galerkin method introduced by Keith and Surowiec.
+
+- API change: in class GridFunction, 'fec' was renamed to 'fec_owned'.
+
 
 Version 4.7, released on May 7, 2024
 ====================================

config/sample-runs.sh

@@ -32,7 +32,7 @@ groups_serial=(
 '"examples"
    "Examples:"
    "examples"
-   "ex{,1,2,3}[0-9].cpp"'
+   "ex{,[1-9]}[0-9].cpp"'
 #   "ex1.cpp"'
 '"sundials"
    "SUNDIALS examples:"
@@ -58,6 +58,10 @@ groups_serial=(
    "HiOp examples:"
    "examples/hiop"
    "ex9.cpp"'
+'"moonolith"
+   "Moonolith examples:"
+   "examples/moonolith"
+   "ex1.cpp"'
 '"pumi"
    "PUMI examples:"
    "examples/pumi"
@@ -66,25 +70,38 @@ groups_serial=(
 '"meshing"
    "Meshing miniapps:"
    "miniapps/meshing"
-   "mobius-strip.cpp klein-bottle.cpp extruder.cpp toroid.cpp
+   "mobius-strip.cpp klein-bottle.cpp extruder.cpp toroid.cpp mesh-quality.cpp
+    polar-nc.cpp reflector.cpp shaper.cpp trimmer.cpp twist.cpp
     mesh-optimizer.cpp minimal-surface.cpp"'
 '"adjoint"
    "Adjoint miniapps:"
    "miniapps/adjoint"
    "cvsRoberts_ASAi_dns.cpp"'
+'"autodiff"
+   "Autodiff miniapps:"
+   "miniapps/autodiff"
+   "seq_example.cpp seq_test.cpp"' # 'seq_test.cpp' has no sample runs
+'"dpg"
+   "DPG miniapps:"
+   "miniapps/dpg"
+   "{acoustics,convection-diffusion,diffusion,maxwell}.cpp"'
 '"gslib"
    "GSLIB miniapps:"
    "miniapps/gslib"
    "field-diff.cpp field-interp.cpp findpts.cpp schwarz_ex1.cpp "'
+# todo: miniapps/mtop
 '"nurbs"
    "NURBS miniapps:"
    "miniapps/nurbs"
    "nurbs_ex1.cpp"'
+# todo: add other nurbs miniapps
+# todo: miniapps/solvers (serial)
 '"tools"
    "Tools miniapps:"
    "miniapps/tools"
    "convert-dc.cpp display-basis.cpp get-values.cpp load-dc.cpp
     lor-transfer.cpp"'
+# todo: add other tools miniapps
 '"toys"
    "Toys miniapps:"
    "miniapps/toys"
@@ -100,7 +117,7 @@ groups_parallel=(
 '"examples"
    "Examples:"
    "examples"
-   "ex{,1,2,3}[0-9]p.cpp"'
+   "ex{,[1-9]}[0-9]p.cpp"'
 #   "ex1p.cpp"'
 '"sundials"
    "SUNDIALS examples:"
@@ -126,6 +143,10 @@ groups_parallel=(
    "HiOp examples:"
    "examples/hiop"
    "ex9p.cpp"'
+'"moonolith"
+   "Moonolith examples:"
+   "examples/moonolith"
+   "ex{1,2}p.cpp"'
 '"pumi"
    "PUMI examples:"
    "examples/pumi"
@@ -138,24 +159,41 @@ groups_parallel=(
 '"meshing"
    "Meshing miniapps:"
    "miniapps/meshing"
-   "pmesh-optimizer.cpp pmesh-fitting.cpp pminimal-surface.cpp"'
+   "pmesh-optimizer.cpp pmesh-fitting.cpp pminimal-surface.cpp
+    fit-node-position.cpp"'
 '"electromagnetics"
    "Electromagnetics miniapps:"
    "miniapps/electromagnetics"
    "joule.cpp"'
-#   "{volta,tesla,joule}.cpp"' # todo: multiline sample runs
+#   "{joule,maxwell,tesla,volta}.cpp"' # todo: multiline sample runs
 '"adjoint"
    "Adjoint miniapps:"
    "miniapps/adjoint"
    "adjoint_advection_diffusion.cpp"'
+'"autodiff"
+   "Autodiff miniapps:"
+   "miniapps/autodiff"
+   "par_example.cpp"'
+'"dpg"
+   "DPG miniapps:"
+   "miniapps/dpg"
+   "p{acoustics,convection-diffusion,diffusion,maxwell}.cpp"'
 '"gslib"
    "GSLIB miniapps:"
    "miniapps/gslib"
    "pfindpts.cpp schwarz_ex1p.cpp"'
+'"hdiv-linear-solver"
+   "H(div) linear solver miniapps:"
+   "miniapps/hdiv-linear-solver"
+   "grad_div.cpp darcy.cpp"'
+# 'miniapps/hooke/hooke.cpp' has no sample runs
+# todo: miniapps/mtop
+# todo: miniapps/multidomain
 '"navier"
    "Navier miniapps:"
    "miniapps/navier"
    "navier_cht.cpp"'
+# todo: add other navier miniapps
 '"nurbs"
    "NURBS miniapps:"
    "miniapps/nurbs"
@@ -164,14 +202,18 @@ groups_parallel=(
    "Shifted miniapps:"
    "miniapps/shifted"
    "distance.cpp"'
+# todo: add other shifted miniapps
 '"solvers"
    "Solvers miniapps:"
    "miniapps/solvers"
    "block-solvers.cpp"'
+# todo: add other solvers miniapps
+# todo: miniapps/spde
 '"tools"
    "Tools miniapps:"
    "miniapps/tools"
-   "convert-cd.cpp get-values.cpp load-dc.cpp"'
+   "convert-dc.cpp get-values.cpp load-dc.cpp"'
+# todo: add other tools miniapps
 '"convergence"
    "Convergence tests:"
    "tests/convergence"
@@ -186,7 +228,7 @@ groups_all=(
 '"examples"
    "Examples:"
    "examples"
-   "ex\"{,1,2,3}[0-9]\"{,p}.cpp"'
+   "ex\"{,[1-9]}[0-9]\"{,p}.cpp"'
 '"sundials"
    "SUNDIALS examples:"
    "examples/sundials"
@@ -215,10 +257,14 @@ groups_all=(
    "HiOp examples:"
    "examples/hiop"
    "ex9.cpp ex9p.cpp"'
+'"moonolith"
+   "Moonolith examples:"
+   "examples/moonolith"
+   "ex1.cpp ex{1,2}p.cpp"'
 '"pumi"
    "PUMI examples:"
    "examples/pumi"
-   "ex1.cpp ex1p.cpp ex2.cpp ex6p.cpp"'
+   "ex1.cpp ex2.cpp ex1p.cpp ex6p.cpp"'
 '"superlu"
    "Superlu examples:"
    "examples/superlu"
@@ -226,43 +272,67 @@ groups_all=(
 '"meshing"
    "Meshing miniapps:"
    "miniapps/meshing"
-   "mobius-strip.cpp klein-bottle.cpp extruder.cpp toroid.cpp
-    {,p}mesh-optimizer.cpp pmesh-fitting.cpp {,p}minimal-surface.cpp"'
+   "mobius-strip.cpp klein-bottle.cpp extruder.cpp toroid.cpp mesh-quality.cpp
+    polar-nc.cpp reflector.cpp shaper.cpp trimmer.cpp twist.cpp
+    {,p}mesh-optimizer.cpp pmesh-fitting.cpp {,p}minimal-surface.cpp
+    fit-node-position.cpp"'
 '"electromagnetics"
    "Electromagnetics miniapps:"
    "miniapps/electromagnetics"
    "joule.cpp"'
-#   "{volta,tesla,joule}.cpp"' # todo: multiline sample runs
+#   "{joule,maxwell,tesla,volta}.cpp"' # todo: multiline sample runs
 '"adjoint"
    "Adjoint miniapps:"
    "miniapps/adjoint"
-   "adjoint_advection_diffusion.cpp cvsRoberts_ASAi_dns.cpp"'
+   "cvsRoberts_ASAi_dns.cpp adjoint_advection_diffusion.cpp"'
+'"autodiff"
+   "Autodiff miniapps:"
+   "miniapps/autodiff"
+   "seq_example.cpp seq_test.cpp par_example.cpp"'
+# 'seq_test.cpp' has no sample runs
+'"dpg"
+   "DPG miniapps:"
+   "miniapps/dpg"
+   "{,p}{acoustics,convection-diffusion,diffusion,maxwell}.cpp"'
 '"gslib"
    "GSLIB miniapps:"
    "miniapps/gslib"
    "field-diff.cpp field-interp.cpp findpts.cpp schwarz_ex1.cpp pfindpts.cpp
     schwarz_ex1p.cpp"'
+'"hdiv-linear-solver"
+   "H(div) linear solver miniapps:"
+   "miniapps/hdiv-linear-solver"
+   "grad_div.cpp darcy.cpp"'
+# 'miniapps/hooke/hooke.cpp' has no sample runs
+# todo: miniapps/mtop
+# todo: miniapps/multidomain
 '"navier"
    "Navier miniapps:"
    "miniapps/navier"
    "navier_cht.cpp"'
+# todo: add other navier miniapps
 '"nurbs"
    "NURBS miniapps:"
    "miniapps/nurbs"
    "nurbs_ex1.cpp nurbs_ex1p.cpp nurbs_ex11p.cpp"'
+# todo: add other nurbs miniapps
 '"shifted"
    "Shifted miniapps:"
    "miniapps/shifted"
    "distance.cpp"'
+# todo: add other shifted miniapps
 '"solvers"
    "Solvers miniapps:"
    "miniapps/solvers"
    "block-solvers.cpp"'
+# todo: add other solvers miniapps
+# todo: miniapps/spde
 '"tools"
    "Tools miniapps:"
    "miniapps/tools"
    "convert-dc.cpp display-basis.cpp get-values.cpp load-dc.cpp
     lor-transfer.cpp"'
+# todo: add other tools miniapps
 '"toys"
    "Toys miniapps:"
    "miniapps/toys"
@@ -386,7 +456,7 @@ function help_message()
       mfem_config [${mfem_config}]
          Set MFEM configuration options
       make [${make}], mpiexec [${mpiexec}], mpiexec_np [${mpiexec_np}]
-         Their values can also set using the respective uppercase environment
+         Their values can also be set using the respective uppercase environment
          variable
       mfem_build_dir [${mfem_build_dir}]
          Same as '-d': set this variable to something different from <mfem_dir>

data/pipe-nurbs-2d.mesh

@@ -18,9 +18,9 @@ elements
 boundary
 4
 1 1 0 1
-1 1 2 3
-1 1 3 0
-1 1 1 2
+2 1 2 3
+3 1 3 0
+4 1 1 2
 
 edges
 4

doc/CodeDocumentation.conf.in

@@ -980,6 +980,7 @@ INPUT                  = @MFEM_SOURCE_DIR@/doc/CodeDocumentation.dox \
                          @MFEM_SOURCE_DIR@/miniapps/mtop \
                          @MFEM_SOURCE_DIR@/miniapps/multidomain \
                          @MFEM_SOURCE_DIR@/miniapps/navier \
+                         @MFEM_SOURCE_DIR@/miniapps/stabilized \
                          @MFEM_SOURCE_DIR@/miniapps/nurbs \
                          @MFEM_SOURCE_DIR@/miniapps/parelag \
                          @MFEM_SOURCE_DIR@/miniapps/performance \
@@ -1049,7 +1050,8 @@ RECURSIVE              = NO
 EXCLUDE                = @MFEM_SOURCE_DIR@/config/_config.hpp \
                          @MFEM_SOURCE_DIR@/config/get_hypre_version.cpp \
                          @MFEM_SOURCE_DIR@/general/tinyxml2.h \
-                         @MFEM_SOURCE_DIR@/general/tinyxml2.cpp
+                         @MFEM_SOURCE_DIR@/general/tinyxml2.cpp \
+                         @MFEM_SOURCE_DIR@/linalg/lapack.hpp
 
 # The EXCLUDE_SYMLINKS tag can be used to select whether or not files or
 # directories that are symbolic links (a Unix file system feature) are excluded

doc/CodeDocumentation.dox

@@ -114,7 +114,9 @@ namespace mfem {
  * - <a class="el" href="ex37p_8cpp_source.html">Example 37p</a>: parallel topology optimization
  * - <a class="el" href="ex38_8cpp_source.html">Example 38</a>: cut-surface and cut-volume integration
  * - <a class="el" href="ex39_8cpp_source.html">Example 39</a>: named mesh attributes
- * - <a class="el" href="ex39p_8cpp_source.html">Example 39</a>: parallel named mesh attributes
+ * - <a class="el" href="ex39p_8cpp_source.html">Example 39p</a>: parallel named mesh attributes
+ * - <a class="el" href="ex40_8cpp_source.html">Example 40</a>: eikonal equation
+ * - <a class="el" href="ex40p_8cpp_source.html">Example 40p</a>: parallel eikonal equation
  *
  * <H4>AmgX Examples</H4>
  * - Variants of Examples
@@ -180,6 +182,21 @@ namespace mfem {
  *   <a class="el" href="examples_2superlu_2ex1p_8cpp_source.html">1p</a>,
  *   demonstrating the use of MFEM's \link superlu.hpp SuperLU integration\endlink.
  *
+ * <H4>NURBS Examples</H4>
+ * - Variants of Examples
+ *  <a class="el" href="nurbs__ex1_8cpp_source.html">1</a>,
+ *  <a class="el" href="nurbs__ex1p_8cpp_source.html">1p</a>,
+ *  <a class="el" href="nurbs__ex3_8cpp_source.html">3</a>,
+ *  <a class="el" href="nurbs__ex5_8cpp_source.html">5</a>,
+ *  <a class="el" href="nurbs__ex11p_8cpp_source.html">11p</a>, and
+ *  <a class="el" href="nurbs__ex24_8cpp_source.html">24</a>,
+ *   demonstrating howto perform NURBS-based Isogeometric Analysis.
+ * - Variant of Example <a class="el" href="nurbs__patch__ex1_8cpp_source.html">1</a>: demonstrates the use of patch integration
+ * - <a class="el" href="nurbs__solenoidal_8cpp_source.html">NURBS Divergence-free</a>: solve a solenoidal vector projection with NURBS-based H(div) elements
+ * - <a class="el" href="nurbs__curveint_8cpp_source.html">NURBS Interpolation</a>: NURBS interpolation of given geometry
+ * - <a class="el" href="nurbs__naca__cmesh_8cpp_source.html">NURBS NACA Mesher</a>: generate NURBS based mesh around a NACA foil
+ * - <a class="el" href="nurbs__printfunc_8cpp_source.html">NURBS Printer</a>: print the NURBS-basis
+ *
  * <H3>Miniapps</H3>
  * - <a class="el" href="volta_8cpp_source.html">Volta</a>: simple electrostatics simulation code
  * - <a class="el" href="tesla_8cpp_source.html">Tesla</a>: simple magnetostatics simulation code

examples/CMakeLists.txt

@@ -45,6 +45,7 @@ list(APPEND ALL_EXE_SRCS
   ex37.cpp
   ex38.cpp
   ex39.cpp
+  ex40.cpp
   )
 
 if (MFEM_USE_MPI)
@@ -87,6 +88,7 @@ if (MFEM_USE_MPI)
     ex36p.cpp
     ex37p.cpp
     ex39p.cpp
+    ex40p.cpp
   )
 endif()
 
@@ -146,10 +148,10 @@ if (MFEM_ENABLE_TESTING)
   # Add CUDA/HIP tests.
   set(DEVICE_EXAMPLES
       # serial examples with device support:
-      ex1 ex3 ex4 ex5 ex6 ex9 ex22 ex24 ex25 ex26 ex34
+      ex1 ex3 ex4 ex5 ex6 ex9 ex14 ex22 ex24 ex25 ex26 ex34
       # parallel examples with device support:
-      ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex9p ex13p ex22p ex24p ex25p ex26p
-      ex34p ex35p)
+      ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex9p ex13p ex14p ex22p ex24p ex25p
+      ex26p ex34p ex35p)
   set(MFEM_TEST_DEVICE)
   if (MFEM_USE_CUDA)
     set(MFEM_TEST_DEVICE "cuda")
@@ -159,6 +161,11 @@ if (MFEM_ENABLE_TESTING)
   if (MFEM_TEST_DEVICE)
     foreach(TEST_NAME ${DEVICE_EXAMPLES})
       set(THIS_TEST_OPTIONS "-no-vis" "-d" "${MFEM_TEST_DEVICE}")
+      if (${TEST_NAME} MATCHES "ex14p")
+        list(APPEND THIS_TEST_OPTIONS "-rs" "2" "-rp" "0" "-pa")
+      elseif (${TEST_NAME} MATCHES "ex14")
+        list(APPEND THIS_TEST_OPTIONS "-r" "2" "-pa")
+      endif()
       if (NOT (${TEST_NAME} MATCHES ".*p$"))
         add_test(NAME ${TEST_NAME}_${MFEM_TEST_DEVICE}_ser
           COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})

examples/ex23.cpp

@@ -44,7 +44,7 @@ protected:
    BilinearForm *M;
    BilinearForm *K;
 
-   SparseMatrix Mmat, Kmat, Kmat0;
+   SparseMatrix Mmat, Kmat;
    SparseMatrix *T; // T = M + dt K
    real_t current_dt;
 
@@ -83,25 +83,24 @@ WaveOperator::WaveOperator(FiniteElementSpace &f,
    : SecondOrderTimeDependentOperator(f.GetTrueVSize(), (real_t) 0.0),
      fespace(f), M(NULL), K(NULL), T(NULL), current_dt(0.0), z(height)
 {
-   const real_t rel_tol = 1e-8;
-
-   fespace.GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
-
+   // Assemble Laplace matrix
    c2 = new ConstantCoefficient(speed*speed);
-
    K = new BilinearForm(&fespace);
    K->AddDomainIntegrator(new DiffusionIntegrator(*c2));
    K->Assemble();
 
-   Array<int> dummy;
-   K->FormSystemMatrix(dummy, Kmat0);
-   K->FormSystemMatrix(ess_tdof_list, Kmat);
-
+   // Assemble Mass matrix
    M = new BilinearForm(&fespace);
    M->AddDomainIntegrator(new MassIntegrator());
    M->Assemble();
+
+   // Apply Bcs
+   fespace.GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+   K->FormSystemMatrix(ess_tdof_list, Kmat);
    M->FormSystemMatrix(ess_tdof_list, Mmat);
 
+   // Configure preconditioner
+   const real_t rel_tol = 1e-8;
    M_solver.iterative_mode = false;
    M_solver.SetRelTol(rel_tol);
    M_solver.SetAbsTol(0.0);
@@ -110,14 +109,13 @@ WaveOperator::WaveOperator(FiniteElementSpace &f,
    M_solver.SetPreconditioner(M_prec);
    M_solver.SetOperator(Mmat);
 
+   // Configure solver
    T_solver.iterative_mode = false;
    T_solver.SetRelTol(rel_tol);
    T_solver.SetAbsTol(0.0);
    T_solver.SetMaxIter(100);
    T_solver.SetPrintLevel(0);
    T_solver.SetPreconditioner(T_prec);
-
-   T = NULL;
 }
 
 void WaveOperator::Mult(const Vector &u, const Vector &du_dt,
@@ -126,9 +124,11 @@ void WaveOperator::Mult(const Vector &u, const Vector &du_dt,
    // Compute:
    //    d2udt2 = M^{-1}*-K(u)
    // for d2udt2
-   Kmat.Mult(u, z);
+   K->FullMult(u, z);
    z.Neg(); // z = -z
+   z.SetSubVector(ess_tdof_list, 0.0);
    M_solver.Mult(z, d2udt2);
+   d2udt2.SetSubVector(ess_tdof_list, 0.0);
 }
 
 void WaveOperator::ImplicitSolve(const real_t fac0, const real_t fac1,
@@ -142,14 +142,11 @@ void WaveOperator::ImplicitSolve(const real_t fac0, const real_t fac1,
       T = Add(1.0, Mmat, fac0, Kmat);
       T_solver.SetOperator(*T);
    }
-   Kmat0.Mult(u, z);
+   K->FullMult(u, z);
    z.Neg();
-
-   for (int i = 0; i < ess_tdof_list.Size(); i++)
-   {
-      z[ess_tdof_list[i]] = 0.0;
-   }
+   z.SetSubVector(ess_tdof_list, 0.0);
    T_solver.Mult(z, d2udt2);
+   d2udt2.SetSubVector(ess_tdof_list, 0.0);
 }
 
 void WaveOperator::SetParameters(const Vector &u)
@@ -314,7 +311,6 @@ int main(int argc, char *argv[])
          ess_bdr = 0;
       }
    }
-
    WaveOperator oper(fespace, ess_bdr, speed);
 
    u_gf.SetFromTrueDofs(u);

examples/ex40.cpp

@@ -0,0 +1,374 @@
+//                                MFEM Example 40
+//
+// Compile with: make ex40
+//
+// Sample runs: ex40 -step 10 -gr 2.0
+//              ex40 -step 10 -gr 2.0 -o 3 -r 1
+//              ex40 -step 10 -gr 2.0 -r 4 -m ../data/l-shape.mesh
+//              ex40 -step 10 -gr 2.0 -r 2 -m ../data/fichera.mesh
+//
+// Description: This example code demonstrates how to use MFEM to solve the
+//              eikonal equation,
+//
+//                      |∇𝑢| = 1 in Ω,  𝑢 = g on ∂Ω.
+//
+//              The solution of this problem coincides with the unique optimum of
+//              the nonlinear program
+//
+//                   maximize ∫_Ω 𝑢 d𝑥 subject to |∇𝑢| ≤ 1,   𝑢 = g on Ω,      (⋆)
+//
+//              which is the foundation for method implemented below.
+//
+//              Following the proximal Galerkin methodology [1] (see also Example
+//              36), we construct a Legendre function for the unit ball
+//              𝐵₁ := {𝑥 ∈ Rⁿ | |𝑥| < 1}. Our choice is the Hellinger entropy,
+//
+//                    h(𝑥) = −( 1 − |𝑥|² )^{1/2},
+//
+//              although other choices are possible, each leading to a slightly
+//              different algorithm. We then adaptively regularize the optimization
+//              problem (⋆) with the Bregman divergence of the Hellinger entropy,
+//
+//                 maximize  ∫_Ω 𝑢 d𝑥 - αₖ⁻¹ Dₕ(∇𝑢,∇𝑢ₖ₋₁)  subject to  𝑢 = g on Ω.
+//
+//              This results in a sequence of functions ( 𝜓ₖ , 𝑢ₖ ),
+//
+//                      𝑢ₖ → 𝑢,    𝜓ₖ/|𝜓ₖ| → ∇𝑢    as k → \infty,
+//
+//              defined by the nonlinear saddle-point problems
+//
+//               Find 𝜓ₖ ∈ H(div,Ω) and 𝑢ₖ ∈ L²(Ω) such that
+//               ( Zₖ(𝜓ₖ) , τ ) + ( 𝑢ₖ , ∇⋅τ ) = ⟨ g , τ⋅n ⟩         ∀ τ ∈ H(div,Ω)
+//               ( ∇⋅𝜓ₖ , v )                 = ( ∇⋅𝜓ₖ₋₁ - 1 , v )   ∀ v ∈ L²(Ω)
+//
+//              where Zₖ(𝜓) := ∇h⁻¹(αₖ 𝜓) = 𝜓 / ( αₖ⁻² + |𝜓|² )^{1/2} and step size
+//              αₖ > 0. These saddle-point problems are solved using a damped Newton's
+//              method. This example assumes that g = 0 and allows the step size to
+//              grow geometrically, αₖ = α₀rᵏ, where r ≥ 1 is the growth rate.
+//
+//              [1] Keith, B. and Surowiec, T. (2023) Proximal Galerkin: A structure-
+//                  preserving finite element method for pointwise bound constraints.
+//                  arXiv:2307.12444 [math.NA]
+
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+
+using namespace std;
+using namespace mfem;
+
+class ZCoefficient : public VectorCoefficient
+{
+protected:
+   GridFunction *psi;
+   real_t alpha;
+
+public:
+   ZCoefficient(int vdim, GridFunction &psi_, real_t alpha_ = 1.0)
+      : VectorCoefficient(vdim), psi(&psi_), alpha(alpha_) { }
+
+   virtual void Eval(Vector &V, ElementTransformation &T,
+                     const IntegrationPoint &ip);
+   void SetAlpha(real_t alpha_) { alpha = alpha_; }
+};
+
+class DZCoefficient : public MatrixCoefficient
+{
+protected:
+   GridFunction *psi;
+   real_t alpha;
+
+public:
+   DZCoefficient(int height, GridFunction &psi_, real_t alpha_ = 1.0)
+      : MatrixCoefficient(height),  psi(&psi_), alpha(alpha_) { }
+
+   virtual void Eval(DenseMatrix &K, ElementTransformation &T,
+                     const IntegrationPoint &ip);
+   void SetAlpha(real_t alpha_) { alpha = alpha_; }
+};
+
+int main(int argc, char *argv[])
+{
+   // 1. Parse command-line options.
+   const char *mesh_file = "../data/star.mesh";
+   int order = 1;
+   int max_it = 5;
+   int ref_levels = 3;
+   real_t alpha = 1.0;
+   real_t growth_rate = 1.0;
+   real_t newton_scaling = 0.9;
+   real_t tichonov = 1e-1;
+   real_t tol = 1e-4;
+   bool visualization = true;
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree).");
+   args.AddOption(&ref_levels, "-r", "--refs",
+                  "Number of h-refinements.");
+   args.AddOption(&max_it, "-mi", "--max-it",
+                  "Maximum number of iterations");
+   args.AddOption(&tol, "-tol", "--tol",
+                  "Stopping criteria based on the difference between"
+                  "successive solution updates");
+   args.AddOption(&alpha, "-step", "--step",
+                  "Initial size alpha");
+   args.AddOption(&growth_rate, "-gr", "--growth-rate",
+                  "Growth rate of the step size alpha");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.Parse();
+   if (!args.Good())
+   {
+      args.PrintUsage(cout);
+      return 1;
+   }
+   args.PrintOptions(cout);
+
+   // 2. Read the mesh from the mesh file.
+   Mesh mesh(mesh_file, 1, 1);
+   int dim = mesh.Dimension();
+   int sdim = mesh.SpaceDimension();
+
+   MFEM_ASSERT(mesh.bdr_attributes.Size(),
+               "This example does not currently support meshes"
+               " without boundary attributes."
+              )
+
+   // 3. Postprocess the mesh.
+   // 3A. Refine the mesh to increase the resolution.
+   for (int l = 0; l < ref_levels; l++)
+   {
+      mesh.UniformRefinement();
+   }
+
+   // 3B. Interpolate the geometry after refinement to control geometry error.
+   // NOTE: Minimum second-order interpolation is used to improve the accuracy.
+   int curvature_order = max(order,2);
+   mesh.SetCurvature(curvature_order);
+
+   // 4. Define the necessary finite element spaces on the mesh.
+   RT_FECollection RTfec(order, dim);
+   FiniteElementSpace RTfes(&mesh, &RTfec);
+
+   L2_FECollection L2fec(order, dim);
+   FiniteElementSpace L2fes(&mesh, &L2fec);
+
+   cout << "Number of H(div) dofs: "
+        << RTfes.GetTrueVSize() << endl;
+   cout << "Number of L² dofs: "
+        << L2fes.GetTrueVSize() << endl;
+
+   // 5. Define the offsets for the block matrices
+   Array<int> offsets(3);
+   offsets[0] = 0;
+   offsets[1] = RTfes.GetVSize();
+   offsets[2] = L2fes.GetVSize();
+   offsets.PartialSum();
+
+   BlockVector x(offsets), rhs(offsets);
+   x = 0.0; rhs = 0.0;
+
+   // 6. Define the solution vectors as a finite element grid functions
+   //    corresponding to the fespaces.
+   GridFunction u_gf, delta_psi_gf;
+   delta_psi_gf.MakeRef(&RTfes,x,offsets[0]);
+   u_gf.MakeRef(&L2fes,x,offsets[1]);
+
+   GridFunction psi_old_gf(&RTfes);
+   GridFunction psi_gf(&RTfes);
+   GridFunction u_old_gf(&L2fes);
+
+   // 7. Define initial guesses for the solution variables.
+   delta_psi_gf = 0.0;
+   psi_gf = 0.0;
+   u_gf = 0.0;
+   psi_old_gf = psi_gf;
+   u_old_gf = u_gf;
+
+   // 8. Prepare for glvis output.
+   char vishost[] = "localhost";
+   int  visport   = 19916;
+   socketstream sol_sock;
+   if (visualization)
+   {
+      sol_sock.open(vishost,visport);
+      sol_sock.precision(8);
+   }
+
+   // 9. Coefficients to be used later.
+   ConstantCoefficient neg_one(-1.0);
+   ConstantCoefficient zero(0.0);
+   ConstantCoefficient tichonov_cf(tichonov);
+   ConstantCoefficient neg_tichonov_cf(-1.0*tichonov);
+   ZCoefficient Z(sdim, psi_gf, alpha);
+   DZCoefficient DZ(sdim, psi_gf, alpha);
+   ScalarVectorProductCoefficient neg_Z(-1.0, Z);
+   DivergenceGridFunctionCoefficient div_psi_cf(&psi_gf);
+   DivergenceGridFunctionCoefficient div_psi_old_cf(&psi_old_gf);
+   SumCoefficient psi_old_minus_psi(div_psi_old_cf, div_psi_cf, 1.0, -1.0);
+
+   // 10. Assemble constant matrices/vectors to avoid reassembly in the loop.
+   LinearForm b0, b1;
+   b0.MakeRef(&RTfes,rhs.GetBlock(0),0);
+   b1.MakeRef(&L2fes,rhs.GetBlock(1),0);
+
+   b0.AddDomainIntegrator(new VectorFEDomainLFIntegrator(neg_Z));
+   b1.AddDomainIntegrator(new DomainLFIntegrator(neg_one));
+   b1.AddDomainIntegrator(new DomainLFIntegrator(psi_old_minus_psi));
+
+   BilinearForm a00(&RTfes);
+   a00.AddDomainIntegrator(new VectorFEMassIntegrator(DZ));
+   a00.AddDomainIntegrator(new VectorFEMassIntegrator(tichonov_cf));
+
+   MixedBilinearForm a10(&RTfes,&L2fes);
+   a10.AddDomainIntegrator(new VectorFEDivergenceIntegrator());
+   a10.Assemble();
+   a10.Finalize();
+   SparseMatrix &A10 = a10.SpMat();
+   SparseMatrix *A01 = Transpose(A10);
+
+   BilinearForm a11(&L2fes);
+   a11.AddDomainIntegrator(new MassIntegrator(neg_tichonov_cf));
+   a11.Assemble();
+   a11.Finalize();
+   SparseMatrix &A11 = a11.SpMat();
+
+   // 11. Iterate.
+   int k;
+   int total_iterations = 0;
+   real_t increment_u = 0.1;
+   GridFunction u_tmp(&L2fes);
+   for (k = 0; k < max_it; k++)
+   {
+      u_tmp = u_old_gf;
+      Z.SetAlpha(alpha);
+      DZ.SetAlpha(alpha);
+
+      mfem::out << "\nOUTER ITERATION " << k+1 << endl;
+
+      int j;
+      for ( j = 0; j < 5; j++)
+      {
+         total_iterations++;
+
+         b0.Assemble();
+         b1.Assemble();
+
+         a00.Assemble(false);
+         a00.Finalize(false);
+         SparseMatrix &A00 = a00.SpMat();
+
+         // Construct Schur-complement preconditioner
+         Vector A00_diag(a00.Height());
+         A00.GetDiag(A00_diag);
+         A00_diag.Reciprocal();
+         SparseMatrix *S = Mult_AtDA(*A01, A00_diag);
+
+         BlockDiagonalPreconditioner prec(offsets);
+         prec.SetDiagonalBlock(0,new DSmoother(A00));
+#ifndef MFEM_USE_SUITESPARSE
+         prec.SetDiagonalBlock(1,new GSSmoother(*S));
+#else
+         prec.SetDiagonalBlock(1,new UMFPackSolver(*S));
+#endif
+         prec.owns_blocks = 1;
+
+         BlockOperator A(offsets);
+         A.SetBlock(0,0,&A00);
+         A.SetBlock(1,0,&A10);
+         A.SetBlock(0,1,A01);
+         A.SetBlock(1,1,&A11);
+
+         GMRES(A,prec,rhs,x,0,2000,500,1e-12,0.0);
+         delete S;
+
+         u_tmp -= u_gf;
+         real_t Newton_update_size = u_tmp.ComputeL2Error(zero);
+         u_tmp = u_gf;
+
+         // Damped Newton update
+         psi_gf.Add(newton_scaling, delta_psi_gf);
+         a00.Update();
+
+         if (visualization)
+         {
+            sol_sock << "solution\n" << mesh << u_gf << "window_title 'Discrete solution'"
+                     << flush;
+         }
+
+         mfem::out << "Newton_update_size = " << Newton_update_size << endl;
+
+         if (Newton_update_size < increment_u)
+         {
+            break;
+         }
+      }
+
+      u_tmp = u_gf;
+      u_tmp -= u_old_gf;
+      increment_u = u_tmp.ComputeL2Error(zero);
+
+      mfem::out << "Number of Newton iterations = " << j+1 << endl;
+      mfem::out << "Increment (|| uₕ - uₕ_prvs||) = " << increment_u << endl;
+
+      u_old_gf = u_gf;
+      psi_old_gf = psi_gf;
+
+      if (increment_u < tol || k == max_it-1)
+      {
+         break;
+      }
+
+      alpha *= max(growth_rate, 1_r);
+
+   }
+
+   mfem::out << "\n Outer iterations: " << k+1
+             << "\n Total iterations: " << total_iterations
+             << "\n Total dofs:       " << RTfes.GetTrueVSize() + L2fes.GetTrueVSize()
+             << endl;
+
+   delete A01;
+   return 0;
+}
+
+void ZCoefficient::Eval(Vector &V, ElementTransformation &T,
+                        const IntegrationPoint &ip)
+{
+   MFEM_ASSERT(psi != NULL, "grid function is not set");
+   MFEM_ASSERT(alpha > 0, "alpha is not positive");
+
+   Vector psi_vals(vdim);
+   psi->GetVectorValue(T, ip, psi_vals);
+   real_t norm = psi_vals.Norml2();
+   real_t phi = 1.0 / sqrt(1.0/(alpha*alpha) + norm*norm);
+
+   V = psi_vals;
+   V *= phi;
+}
+
+void DZCoefficient::Eval(DenseMatrix &K, ElementTransformation &T,
+                         const IntegrationPoint &ip)
+{
+   MFEM_ASSERT(psi != NULL, "grid function is not set");
+   MFEM_ASSERT(alpha > 0, "alpha is not positive");
+
+   Vector psi_vals(height);
+   psi->GetVectorValue(T, ip, psi_vals);
+   real_t norm = psi_vals.Norml2();
+   real_t phi = 1.0 / sqrt(1.0/(alpha*alpha) + norm*norm);
+
+   K = 0.0;
+   for (int i = 0; i < height; i++)
+   {
+      K(i,i) = phi;
+      for (int j = 0; j < height; j++)
+      {
+         K(i,j) -= psi_vals(i) * psi_vals(j) * pow(phi, 3);
+      }
+   }
+}

examples/ex40p.cpp

@@ -0,0 +1,436 @@
+//                       MFEM Example 40 - Parallel Version
+//
+// Compile with: make ex40p
+//
+// Sample runs: mpirun -np 4 ex40p -step 10 -gr 2.0
+//              mpirun -np 4 ex40p -step 10 -gr 2.0 -o 3 -r 1
+//              mpirun -np 4 ex40p -step 10 -gr 2.0 -r 4 -m ../data/l-shape.mesh
+//              mpirun -np 4 ex40p -step 10 -gr 2.0 -r 2 -m ../data/fichera.mesh
+//
+// Description: This example code demonstrates how to use MFEM to solve the
+//              eikonal equation,
+//
+//                      |∇𝑢| = 1 in Ω,  𝑢 = g on ∂Ω.
+//
+//              The solution of this problem coincides with the unique optimum of
+//              the nonlinear program
+//
+//                   maximize ∫_Ω 𝑢 d𝑥 subject to |∇𝑢| ≤ 1,   𝑢 = g on Ω,      (⋆)
+//
+//              which is the foundation for method implemented below.
+//
+//              Following the proximal Galerkin methodology [1] (see also Example
+//              36), we construct a Legendre function for the unit ball
+//              𝐵₁ := {𝑥 ∈ Rⁿ | |𝑥| < 1}. Our choice is the Hellinger entropy,
+//
+//                    h(𝑥) = −( 1 − |𝑥|² )^{1/2},
+//
+//              although other choices are possible, each leading to a slightly
+//              different algorithm. We then adaptively regularize the optimization
+//              problem (⋆) with the Bregman divergence of the Hellinger entropy,
+//
+//                 maximize  ∫_Ω 𝑢 d𝑥 - αₖ⁻¹ Dₕ(∇𝑢,∇𝑢ₖ₋₁)  subject to  𝑢 = g on Ω.
+//
+//              This results in a sequence of functions ( 𝜓ₖ , 𝑢ₖ ),
+//
+//                      𝑢ₖ → 𝑢,    𝜓ₖ/|𝜓ₖ| → ∇𝑢    as k → \infty,
+//
+//              defined by the nonlinear saddle-point problems
+//
+//               Find 𝜓ₖ ∈ H(div,Ω) and 𝑢ₖ ∈ L²(Ω) such that
+//               ( Zₖ(𝜓ₖ) , τ ) + ( 𝑢ₖ , ∇⋅τ ) = ⟨ g , τ⋅n ⟩         ∀ τ ∈ H(div,Ω)
+//               ( ∇⋅𝜓ₖ , v )                 = ( ∇⋅𝜓ₖ₋₁ - 1 , v )   ∀ v ∈ L²(Ω)
+//
+//              where Zₖ(𝜓) := ∇h⁻¹(αₖ 𝜓) = 𝜓 / ( αₖ⁻² + |𝜓|² )^{1/2} and step size
+//              αₖ > 0. These saddle-point problems are solved using a damped Newton's
+//              method. This example assumes that g = 0 and allows the step size to
+//              grow geometrically, αₖ = α₀rᵏ, where r ≥ 1 is the growth rate.
+//
+//              [1] Keith, B. and Surowiec, T. (2023) Proximal Galerkin: A structure-
+//                  preserving finite element method for pointwise bound constraints.
+//                  arXiv:2307.12444 [math.NA]
+
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+
+using namespace std;
+using namespace mfem;
+
+class ZCoefficient : public VectorCoefficient
+{
+protected:
+   ParGridFunction *psi;
+   real_t alpha;
+
+public:
+   ZCoefficient(int vdim, ParGridFunction &psi_, real_t alpha_ = 1.0)
+      : VectorCoefficient(vdim), psi(&psi_), alpha(alpha_) { }
+
+   virtual void Eval(Vector &V, ElementTransformation &T,
+                     const IntegrationPoint &ip);
+   void SetAlpha(real_t alpha_) { alpha = alpha_; }
+};
+
+class DZCoefficient : public MatrixCoefficient
+{
+protected:
+   ParGridFunction *psi;
+   real_t alpha;
+
+public:
+   DZCoefficient(int height, ParGridFunction &psi_, real_t alpha_ = 1.0)
+      : MatrixCoefficient(height),  psi(&psi_), alpha(alpha_) { }
+
+   virtual void Eval(DenseMatrix &K, ElementTransformation &T,
+                     const IntegrationPoint &ip);
+   void SetAlpha(real_t alpha_) { alpha = alpha_; }
+};
+
+int main(int argc, char *argv[])
+{
+   // 0. Initialize MPI and HYPRE.
+   Mpi::Init();
+   int num_procs = Mpi::WorldSize();
+   int myid = Mpi::WorldRank();
+   Hypre::Init();
+
+   // 1. Parse command-line options.
+   const char *mesh_file = "../data/star.mesh";
+   int order = 1;
+   int max_it = 5;
+   int ref_levels = 3;
+   real_t alpha = 1.0;
+   real_t growth_rate = 1.0;
+   real_t newton_scaling = 0.9;
+   real_t tichonov = 1e-1;
+   real_t tol = 1e-4;
+   bool visualization = true;
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree).");
+   args.AddOption(&ref_levels, "-r", "--refs",
+                  "Number of h-refinements.");
+   args.AddOption(&max_it, "-mi", "--max-it",
+                  "Maximum number of iterations");
+   args.AddOption(&tol, "-tol", "--tol",
+                  "Stopping criteria based on the difference between"
+                  "successive solution updates");
+   args.AddOption(&alpha, "-step", "--step",
+                  "Initial size alpha");
+   args.AddOption(&growth_rate, "-gr", "--growth-rate",
+                  "Growth rate of the step size alpha");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.Parse();
+   if (!args.Good())
+   {
+      if (myid == 0)
+      {
+         args.PrintUsage(cout);
+      }
+      return 1;
+   }
+   if (myid == 0)
+   {
+      args.PrintOptions(cout);
+   }
+
+   // 2. Read the mesh from the mesh file.
+   Mesh mesh(mesh_file, 1, 1);
+   int dim = mesh.Dimension();
+   int sdim = mesh.SpaceDimension();
+
+   MFEM_ASSERT(mesh.bdr_attributes.Size(),
+               "This example does not currently support meshes"
+               " without boundary attributes."
+              )
+
+   // 3. Postprocess the mesh.
+   // 3A. Refine the mesh to increase the resolution.
+   for (int l = 0; l < ref_levels; l++)
+   {
+      mesh.UniformRefinement();
+   }
+
+   // 3B. Interpolate the geometry after refinement to control geometry error.
+   // NOTE: Minimum second-order interpolation is used to improve the accuracy.
+   int curvature_order = max(order,2);
+   mesh.SetCurvature(curvature_order);
+
+   ParMesh pmesh(MPI_COMM_WORLD, mesh);
+   mesh.Clear();
+
+   // 4. Define the necessary finite element spaces on the mesh.
+   RT_FECollection RTfec(order, dim);
+   ParFiniteElementSpace RTfes(&pmesh, &RTfec);
+
+   L2_FECollection L2fec(order, dim);
+   ParFiniteElementSpace L2fes(&pmesh, &L2fec);
+
+   int num_dofs_RT = RTfes.GlobalTrueVSize();
+   int num_dofs_L2 = L2fes.GlobalTrueVSize();
+   if (myid == 0)
+   {
+      cout << "Number of H(div) dofs: "
+           << num_dofs_RT << endl;
+      cout << "Number of L² dofs: "
+           << num_dofs_L2 << endl;
+   }
+
+   // 5. Define the offsets for the block matrices
+   Array<int> offsets(3);
+   offsets[0] = 0;
+   offsets[1] = RTfes.GetVSize();
+   offsets[2] = L2fes.GetVSize();
+   offsets.PartialSum();
+
+   Array<int> toffsets(3);
+   toffsets[0] = 0;
+   toffsets[1] = RTfes.GetTrueVSize();
+   toffsets[2] = L2fes.GetTrueVSize();
+   toffsets.PartialSum();
+
+   BlockVector x(offsets), rhs(offsets);
+   x = 0.0; rhs = 0.0;
+
+   BlockVector tx(toffsets), trhs(toffsets);
+   tx = 0.0; trhs = 0.0;
+
+   // 6. Define the solution vectors as a finite element grid functions
+   //    corresponding to the fespaces.
+   ParGridFunction u_gf, delta_psi_gf;
+   delta_psi_gf.MakeRef(&RTfes,x,offsets[0]);
+   u_gf.MakeRef(&L2fes,x,offsets[1]);
+
+   ParGridFunction psi_old_gf(&RTfes);
+   ParGridFunction psi_gf(&RTfes);
+   ParGridFunction u_old_gf(&L2fes);
+
+   // 7. Define initial guesses for the solution variables.
+   delta_psi_gf = 0.0;
+   psi_gf = 0.0;
+   u_gf = 0.0;
+   psi_old_gf = psi_gf;
+   u_old_gf = u_gf;
+
+   // 8. Prepare for glvis output.
+   char vishost[] = "localhost";
+   int  visport   = 19916;
+   socketstream sol_sock;
+   if (visualization)
+   {
+      sol_sock.open(vishost,visport);
+      sol_sock.precision(8);
+   }
+
+   // 9. Coefficients to be used later.
+   ConstantCoefficient neg_one(-1.0);
+   ConstantCoefficient zero(0.0);
+   ConstantCoefficient tichonov_cf(tichonov);
+   ConstantCoefficient neg_tichonov_cf(-1.0*tichonov);
+   ZCoefficient Z(sdim, psi_gf, alpha);
+   DZCoefficient DZ(sdim, psi_gf, alpha);
+   ScalarVectorProductCoefficient neg_Z(-1.0, Z);
+   DivergenceGridFunctionCoefficient div_psi_cf(&psi_gf);
+   DivergenceGridFunctionCoefficient div_psi_old_cf(&psi_old_gf);
+   SumCoefficient psi_old_minus_psi(div_psi_old_cf, div_psi_cf, 1.0, -1.0);
+
+   // 10. Assemble constant matrices/vectors to avoid reassembly in the loop.
+   ParLinearForm b0, b1;
+   b0.MakeRef(&RTfes,rhs.GetBlock(0),0);
+   b1.MakeRef(&L2fes,rhs.GetBlock(1),0);
+
+   b0.AddDomainIntegrator(new VectorFEDomainLFIntegrator(neg_Z));
+   b1.AddDomainIntegrator(new DomainLFIntegrator(neg_one));
+   b1.AddDomainIntegrator(new DomainLFIntegrator(psi_old_minus_psi));
+
+   ParBilinearForm a00(&RTfes);
+   a00.AddDomainIntegrator(new VectorFEMassIntegrator(DZ));
+   a00.AddDomainIntegrator(new VectorFEMassIntegrator(tichonov_cf));
+
+   ParMixedBilinearForm a10(&RTfes,&L2fes);
+   a10.AddDomainIntegrator(new VectorFEDivergenceIntegrator());
+   a10.Assemble();
+   a10.Finalize();
+   HypreParMatrix *A10 = a10.ParallelAssemble();
+
+   HypreParMatrix *A01 = A10->Transpose();
+
+   ParBilinearForm a11(&L2fes);
+   a11.AddDomainIntegrator(new MassIntegrator(neg_tichonov_cf));
+   a11.Assemble();
+   a11.Finalize();
+   HypreParMatrix *A11 = a11.ParallelAssemble();
+
+   // 11. Iterate.
+   int k;
+   int total_iterations = 0;
+   real_t increment_u = 0.1;
+   ParGridFunction u_tmp(&L2fes);
+   for (k = 0; k < max_it; k++)
+   {
+      u_tmp = u_old_gf;
+      Z.SetAlpha(alpha);
+      DZ.SetAlpha(alpha);
+
+      if (myid == 0)
+      {
+         mfem::out << "\nOUTER ITERATION " << k+1 << endl;
+      }
+
+      int j;
+      for ( j = 0; j < 5; j++)
+      {
+         total_iterations++;
+
+         b0.Assemble();
+         b0.ParallelAssemble(trhs.GetBlock(0));
+
+         b1.Assemble();
+         b1.ParallelAssemble(trhs.GetBlock(1));
+
+         a00.Assemble(false);
+         a00.Finalize(false);
+         HypreParMatrix *A00 = a00.ParallelAssemble();
+
+         // Construct Schur-complement preconditioner
+         HypreParVector A00_diag(MPI_COMM_WORLD, A00->GetGlobalNumRows(),
+                                 A00->GetRowStarts());
+         A00->GetDiag(A00_diag);
+         HypreParMatrix S_tmp(*A01);
+         S_tmp.InvScaleRows(A00_diag);
+         HypreParMatrix *S = ParMult(A10, &S_tmp, true);
+
+         BlockDiagonalPreconditioner prec(toffsets);
+         HypreBoomerAMG P00(*A00);
+         P00.SetPrintLevel(0);
+         HypreBoomerAMG P11(*S);
+         P11.SetPrintLevel(0);
+         prec.SetDiagonalBlock(0,&P00);
+         prec.SetDiagonalBlock(1,&P11);
+
+         BlockOperator A(toffsets);
+         A.SetBlock(0,0,A00);
+         A.SetBlock(1,0,A10);
+         A.SetBlock(0,1,A01);
+         A.SetBlock(1,1,A11);
+
+         GMRESSolver gmres(MPI_COMM_WORLD);
+         gmres.SetPrintLevel(-1);
+         gmres.SetRelTol(1e-8);
+         gmres.SetMaxIter(2000);
+         gmres.SetKDim(500);
+         gmres.SetOperator(A);
+         gmres.SetPreconditioner(prec);
+         gmres.Mult(trhs,tx);
+         delete S;
+         delete A00;
+
+         delta_psi_gf.SetFromTrueDofs(tx.GetBlock(0));
+         u_gf.SetFromTrueDofs(tx.GetBlock(1));
+
+         u_tmp -= u_gf;
+         real_t Newton_update_size = u_tmp.ComputeL2Error(zero);
+         u_tmp = u_gf;
+
+         // Damped Newton update
+         psi_gf.Add(newton_scaling, delta_psi_gf);
+         a00.Update();
+
+         if (visualization)
+         {
+            sol_sock << "parallel " << num_procs << " " << myid << "\n";
+            sol_sock << "solution\n" << pmesh << u_gf << "window_title 'Discrete solution'"
+                     << flush;
+         }
+
+         if (myid == 0)
+         {
+            mfem::out << "Newton_update_size = " << Newton_update_size << endl;
+         }
+
+         if (Newton_update_size < increment_u)
+         {
+            break;
+         }
+      }
+
+      u_tmp = u_gf;
+      u_tmp -= u_old_gf;
+      increment_u = u_tmp.ComputeL2Error(zero);
+
+      if (myid == 0)
+      {
+         mfem::out << "Number of Newton iterations = " << j+1 << endl;
+         mfem::out << "Increment (|| uₕ - uₕ_prvs||) = " << increment_u << endl;
+      }
+
+      u_old_gf = u_gf;
+      psi_old_gf = psi_gf;
+
+      if (increment_u < tol || k == max_it-1)
+      {
+         break;
+      }
+
+      alpha *= max(growth_rate, 1_r);
+
+   }
+
+   // 12. Print stats.
+   if (myid == 0)
+   {
+      mfem::out << "\n Outer iterations: " << k+1
+                << "\n Total iterations: " << total_iterations
+                << "\n Total dofs:       " << RTfes.GetTrueVSize() + L2fes.GetTrueVSize()
+                << endl;
+   }
+
+   // 13. Free the used memory.
+   delete A01;
+   delete A10;
+   delete A11;
+   return 0;
+}
+
+void ZCoefficient::Eval(Vector &V, ElementTransformation &T,
+                        const IntegrationPoint &ip)
+{
+   MFEM_ASSERT(psi != NULL, "grid function is not set");
+   MFEM_ASSERT(alpha > 0, "alpha is not positive");
+
+   Vector psi_vals(vdim);
+   psi->GetVectorValue(T, ip, psi_vals);
+   real_t norm = psi_vals.Norml2();
+   real_t phi = 1.0 / sqrt(1.0/(alpha*alpha) + norm*norm);
+
+   V = psi_vals;
+   V *= phi;
+}
+
+void DZCoefficient::Eval(DenseMatrix &K, ElementTransformation &T,
+                         const IntegrationPoint &ip)
+{
+   MFEM_ASSERT(psi != NULL, "grid function is not set");
+   MFEM_ASSERT(alpha > 0, "alpha is not positive");
+
+   Vector psi_vals(height);
+   psi->GetVectorValue(T, ip, psi_vals);
+   real_t norm = psi_vals.Norml2();
+   real_t phi = 1.0 / sqrt(1.0/(alpha*alpha) + norm*norm);
+
+   K = 0.0;
+   for (int i = 0; i < height; i++)
+   {
+      K(i,i) = phi;
+      for (int j = 0; j < height; j++)
+      {
+         K(i,j) -= psi_vals(i) * psi_vals(j) * pow(phi, 3);
+      }
+   }
+}

examples/makefile

@@ -23,14 +23,14 @@ MFEM_LIB_FILE = mfem_is_not_built
 
 SEQ_EXAMPLES = ex0 ex1 ex2 ex3 ex4 ex5 ex6 ex7 ex8 ex9 ex10 ex14 ex15 ex16 \
    ex17 ex18 ex19 ex20 ex21 ex22 ex23 ex24 ex25 ex26 ex27 ex28 ex29 ex30 \
-   ex31 ex33 ex34 ex36 ex37 ex38 ex39
+   ex31 ex33 ex34 ex36 ex37 ex38 ex39 ex40
 PAR_EXAMPLES = ex0p ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex8p ex9p ex10p ex11p \
    ex12p ex13p ex14p ex15p ex16p ex17p ex18p ex19p ex20p ex21p ex22p ex24p \
    ex25p ex26p ex27p ex28p ex29p ex30p ex31p ex32p ex33p ex34p ex35p ex36p \
-   ex37p ex39p
-SEQ_DEVICE_EXAMPLES = ex1 ex3 ex4 ex5 ex6 ex9 ex22 ex24 ex25 ex26 ex34
-PAR_DEVICE_EXAMPLES = ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex9p ex13p ex22p \
-   ex24p ex25p ex26p ex34p ex35p
+   ex37p ex39p ex40p
+SEQ_DEVICE_EXAMPLES = ex1 ex3 ex4 ex5 ex6 ex9 ex14 ex22 ex24 ex25 ex26 ex34
+PAR_DEVICE_EXAMPLES = ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex9p ex13p ex14p \
+   ex22p ex24p ex25p ex26p ex34p ex35p
 
 ifeq ($(MFEM_USE_LAPACK),YES)
    SEQ_EXAMPLES += ex38
@@ -138,6 +138,14 @@ ex10-test-seq: ex10
 	@$(call mfem-test,$<,, Serial example,-tf 5)
 ex10p-test-par: ex10p
 	@$(call mfem-test,$<, $(RUN_MPI), Parallel example,-tf 5)
+ex14-test-seq-cuda: ex14
+	@$(call mfem-test,$<,, Serial CUDA example,-r 2 -pa -d cuda)
+ex14p-test-par-cuda: ex14p
+	@$(call mfem-test,$<, $(RUN_MPI), Parallel CUDA example,-rs 2 -rp 0 -pa -d cuda)
+ex14-test-seq-hip: ex14
+	@$(call mfem-test,$<,, Serial HIP example,-r 2 -pa -d hip)
+ex14p-test-par-hip: ex14p
+	@$(call mfem-test,$<, $(RUN_MPI), Parallel HIP example,-rs 2 -rp 0 -pa -d hip)
 ex15-test-seq: ex15
 	@$(call mfem-test,$<,, Serial example,-e 1)
 ex15p-test-par: ex15p

examples/pumi/ex1.cpp

@@ -66,7 +66,6 @@ int main(int argc, char *argv[])
 {
    // 1. Initialize MPI (required by PUMI) and HYPRE.
    Mpi::Init(argc, argv);
-   int num_procs = Mpi::WorldSize();
    int myid = Mpi::WorldRank();
    Hypre::Init();
 

examples/pumi/ex2.cpp

@@ -80,8 +80,6 @@ int main(int argc, char *argv[])
 {
    // 1. Initialize MPI (required by PUMI) and HYPRE.
    Mpi::Init(argc, argv);
-   int num_proc = Mpi::WorldSize();
-   int myId = Mpi::WorldRank();
    Hypre::Init();
 
    // 2. Parse command-line options.

fem/bilininteg.cpp

@@ -1539,6 +1539,351 @@ const IntegrationRule &ConvectionIntegrator::GetRule(
    return GetRule(el,el,Trans);
 }
 
+
+void LaplaceIntegrator::AssembleElementMatrix(const FiniteElement &el,
+                                              ElementTransformation &Trans,
+                                              DenseMatrix &elmat )
+{
+   int nd = el.GetDof();
+   real_t w;
+
+   elmat.SetSize(nd);
+   shape.SetSize(nd);
+   laplace.SetSize(nd);
+
+   const IntegrationRule *ir = IntRule ? IntRule : &GetRule(el, el, Trans);
+
+   elmat = 0.0;
+   for (int i = 0; i < ir->GetNPoints(); i++)
+   {
+      const IntegrationPoint &ip = ir->IntPoint(i);
+      Trans.SetIntPoint (&ip);
+
+      el.CalcPhysShape(Trans, shape);
+      el.CalcPhysLaplacian(Trans, laplace);
+
+      w = Trans.Weight() * ip.weight * alpha;
+      if (Q)
+      {
+         w *= Q -> Eval(Trans, ip);
+      }
+      shape *= w;
+      AddMultVWt(shape, laplace, elmat);
+   }
+}
+
+void LaplaceIntegrator::AssembleElementMatrix2(const FiniteElement &trial_fe,
+                                               const FiniteElement &test_fe,
+                                               ElementTransformation &Trans,
+                                               DenseMatrix &elmat)
+{
+   int tr_nd = trial_fe.GetDof();
+   int te_nd = test_fe.GetDof();
+
+   elmat.SetSize(te_nd, tr_nd);
+   laplace.SetSize(tr_nd);
+   shape.SetSize(te_nd);
+
+   const IntegrationRule *ir = IntRule ? IntRule : &GetRule(trial_fe, test_fe,
+                                                            Trans);
+
+   elmat = 0.0;
+   for (int i = 0; i < ir->GetNPoints(); i++)
+   {
+      const IntegrationPoint &ip = ir->IntPoint(i);
+      Trans.SetIntPoint (&ip);
+
+      test_fe.CalcPhysShape(Trans, shape);
+      trial_fe.CalcPhysLaplacian(Trans, laplace);
+
+      real_t w = Trans.Weight() * ip.weight * alpha;
+      if (Q)
+      {
+         w *= Q -> Eval(Trans, ip);
+      }
+      AddMult_a_VWt(w, shape, laplace, elmat);
+   }
+}
+
+const IntegrationRule &LaplaceIntegrator::GetRule(
+   const FiniteElement &trial_fe,
+   const FiniteElement &test_fe,
+   ElementTransformation &Trans)
+{
+   int order = trial_fe.GetOrder() + test_fe.GetOrder();
+   return IntRules.Get(trial_fe.GetGeomType(), order);
+}
+
+void LaplaceGradIntegrator::AssembleElementMatrix(const FiniteElement &el,
+                                                  ElementTransformation &Trans,
+                                                  DenseMatrix &elmat )
+{
+   int nd = el.GetDof();
+   dim = el.GetDim();
+
+   elmat.SetSize(nd);
+   dshape.SetSize(nd,dim);
+   adjJ.SetSize(dim);
+   laplace.SetSize(nd);
+   vec2.SetSize(dim);
+   BdFidxT.SetSize(nd);
+
+   Vector vec1;
+
+   const IntegrationRule *ir = IntRule ? IntRule : &GetRule(el, el, Trans);
+
+   Q->Eval(Q_ir, Trans, *ir);
+
+   elmat = 0.0;
+   for (int i = 0; i < ir->GetNPoints(); i++)
+   {
+      const IntegrationPoint &ip = ir->IntPoint(i);
+      el.CalcDShape(ip, dshape);
+      el.CalcPhysLaplacian(Trans, laplace);
+
+      Trans.SetIntPoint(&ip);
+      CalcAdjugate(Trans.Jacobian(), adjJ);
+      Q_ir.GetColumnReference(i, vec1);
+      vec1 *= alpha * ip.weight;
+
+      adjJ.Mult(vec1, vec2);
+      dshape.Mult(vec2, BdFidxT);
+
+      AddMultVWt(BdFidxT, laplace, elmat);
+   }
+}
+
+void LaplaceGradIntegrator::AssembleElementMatrix2(const FiniteElement
+                                                   &trial_fe,
+                                                   const FiniteElement &test_fe,
+                                                   ElementTransformation &Trans,
+                                                   DenseMatrix &elmat)
+{
+   dim = trial_fe.GetDim();
+   int tr_nd = trial_fe.GetDof();
+   int te_nd = test_fe.GetDof();
+
+   elmat.SetSize(te_nd, tr_nd);
+   laplace.SetSize(tr_nd);
+   dshape.SetSize(te_nd,dim);
+   adjJ.SetSize(dim);
+   vec2.SetSize(dim);
+   BdFidxT.SetSize(te_nd);
+
+   Vector vec1;
+
+   const IntegrationRule *ir = IntRule ? IntRule : &GetRule(trial_fe, test_fe,
+                                                            Trans);
+
+   Q->Eval(Q_ir, Trans, *ir);
+
+   elmat = 0.0;
+   for (int i = 0; i < ir->GetNPoints(); i++)
+   {
+      const IntegrationPoint &ip = ir->IntPoint(i);
+      test_fe.CalcDShape(ip, dshape);
+      trial_fe.CalcPhysLaplacian(Trans, laplace);
+
+      Trans.SetIntPoint(&ip);
+      CalcAdjugate(Trans.Jacobian(), adjJ);
+      Q_ir.GetColumnReference(i, vec1);
+      vec1 *= alpha * ip.weight;
+
+      adjJ.Mult(vec1, vec2);
+      dshape.Mult(vec2, BdFidxT);
+
+      AddMultVWt(BdFidxT, laplace,elmat);
+   }
+}
+
+const IntegrationRule &LaplaceGradIntegrator::GetRule(
+   const FiniteElement &trial_fe,
+   const FiniteElement &test_fe,
+   ElementTransformation &Trans)
+{
+   int order = trial_fe.GetOrder() + test_fe.GetOrder();
+   return IntRules.Get(trial_fe.GetGeomType(), order);
+}
+
+void LaplaceLaplaceIntegrator::AssembleElementMatrix(const FiniteElement &el,
+                                                     ElementTransformation &Trans,
+                                                     DenseMatrix &elmat )
+{
+   int nd = el.GetDof();
+   real_t w;
+
+   elmat.SetSize(nd);
+   laplace.SetSize(nd);
+
+   const IntegrationRule *ir = IntRule ? IntRule : &GetRule(el, el, Trans);
+
+   elmat = 0.0;
+   for (int i = 0; i < ir->GetNPoints(); i++)
+   {
+      const IntegrationPoint &ip = ir->IntPoint(i);
+      Trans.SetIntPoint (&ip);
+
+      el.CalcPhysLaplacian(Trans, laplace);
+
+      w = Trans.Weight() * ip.weight * alpha;
+      if (Q)
+      {
+         w *= Q -> Eval(Trans, ip);
+      }
+      AddMult_a_VVt(w, laplace, elmat);
+   }
+}
+
+void LaplaceLaplaceIntegrator::AssembleElementMatrix2(const FiniteElement
+                                                      &trial_fe,
+                                                      const FiniteElement &test_fe,
+                                                      ElementTransformation &Trans,
+                                                      DenseMatrix &elmat)
+{
+   int
+   dim = trial_fe.GetDim();
+   int tr_nd = trial_fe.GetDof();
+   int te_nd = test_fe.GetDof();
+   real_t w;
+
+   elmat.SetSize(te_nd, tr_nd);
+   laplace.SetSize(tr_nd);
+   te_laplace.SetSize(te_nd);
+
+   const IntegrationRule *ir = IntRule ? IntRule : &GetRule(trial_fe, test_fe,
+                                                            Trans);
+
+   elmat = 0.0;
+   for (int i = 0; i < ir->GetNPoints(); i++)
+   {
+      const IntegrationPoint &ip = ir->IntPoint(i);
+      Trans.SetIntPoint (&ip);
+
+      trial_fe.CalcPhysLaplacian(Trans, laplace);
+      test_fe.CalcPhysLaplacian(Trans, te_laplace);
+
+      w = Trans.Weight() * ip.weight * alpha;
+      if (Q)
+      {
+         w *= Q -> Eval(Trans, ip);
+      }
+      AddMult_a_VWt(w, te_laplace, laplace, elmat);
+   }
+}
+
+const IntegrationRule &LaplaceLaplaceIntegrator::GetRule(
+   const FiniteElement &trial_fe,
+   const FiniteElement &test_fe,
+   ElementTransformation &Trans)
+{
+   int order = trial_fe.GetOrder() + test_fe.GetOrder();
+   return IntRules.Get(trial_fe.GetGeomType(), order);
+}
+
+void InverseEstimateIntegrator::AssembleElementMatrix(const FiniteElement &el,
+                                                      ElementTransformation &Trans,
+                                                      DenseMatrix &elmat )
+{
+   elmat = 0.0;
+
+   int nd = el.GetDof();
+   int dim = el.GetDim();
+
+   shape.SetSize(nd);
+   dshape.SetSize(nd,dim);
+   laplace.SetSize(nd);
+
+   lapmat.SetSize(nd,nd);
+   bimat.SetSize(nd,nd);
+   ovec.SetSize(nd);
+
+   real_t w,q;
+
+   const IntegrationRule *ir = IntRule;
+   if (ir == NULL)
+   {
+      int order = Trans.OrderGrad(&el) + Trans.Order() + el.GetOrder();
+      ir = &IntRules.Get(el.GetGeomType(), order);
+   }
+
+   bimat = 0.0;
+   lapmat = 0.0;
+   ovec = 0.0;
+   for (int i = 0; i < ir->GetNPoints(); i++)
+   {
+      const IntegrationPoint &ip = ir->IntPoint(i);
+      Trans.SetIntPoint(&ip);
+      w = Trans.Weight()*ip.weight;
+      if (Q)
+      {
+         q = Q->Eval(Trans, ip);
+      }
+
+      el.CalcPhysDShape(Trans, dshape);
+      AddMult_a_AAt(w*q, dshape, lapmat);
+
+      el.CalcPhysLaplacian(Trans, laplace);
+      AddMult_a_VVt(w*q*q, laplace, bimat);
+
+      el.CalcPhysShape(Trans, shape);
+      ovec.Add(w, shape);
+   }
+
+   // Power method
+   Vector x(nd);
+   x.Randomize(696383532);
+
+   // Correct nullspace + inverse
+   AddMult_a_VVt(1.0, ovec, lapmat);
+   DenseMatrixInverse L_inv(lapmat);
+
+   // DenseMatrix M_i, Q_i;
+   real_t alpha= 0.0, eval_i = 0.0, eval_prev = 0.0;
+
+   // Inverse power method
+   Vector x_tmp(nd);
+   int iter = 0;
+   const real_t rel_tol = 1e-4;
+
+   alpha = ovec*ovec;
+   ovec *= 1.0/sqrt(alpha);
+   do
+   {
+      // Othogonalize
+      alpha = x*ovec;
+      x.Add(-alpha, ovec);
+
+      // MatVec (2x)
+      bimat.Mult(x, x_tmp);
+      L_inv.Mult(x_tmp, x);
+
+      eval_prev = eval_i;
+      eval_i = x.Norml2();
+      x *= 1.0/eval_i;
+      ++iter;
+   }
+   while ((iter < 10000) && (fabs(eval_i - eval_prev)/fabs(eval_i) > rel_tol));
+   MFEM_VERIFY(fabs(eval_i - eval_prev)/fabs(eval_i) <= rel_tol,
+               "Inverse power method did not converge."
+               << "\n\t iter      = " << iter
+               << "\n\t eval_i    = " << eval_i
+               << "\n\t eval_prev = " << eval_prev
+               << "\n\t fabs(eval_i - eval_prev)/fabs(eval_i) = "
+               << fabs(eval_i - eval_prev)/fabs(eval_i));
+   cout<<"evev = "<<eval_i<<" "<<iter<<endl;
+}
+
+const IntegrationRule &InverseEstimateIntegrator::GetRule(
+   const FiniteElement &trial_fe,
+   const FiniteElement &test_fe,
+   ElementTransformation &Trans)
+{
+   // int order = Trans.OrderGrad(&trial_fe) + Trans.Order() + test_fe.GetOrder() - 2;
+   int order = trial_fe.GetOrder() + test_fe.GetOrder();
+   return IntRules.Get(trial_fe.GetGeomType(), order);
+}
+
+
 void VectorMassIntegrator::AssembleElementMatrix
 ( const FiniteElement &el, ElementTransformation &Trans,
   DenseMatrix &elmat )

fem/bilininteg.hpp

@@ -1741,7 +1741,7 @@ public:
    { vector_fe.CalcPhysDShape(Trans, shape); }
 };
 
-/** Class for integrating the bilinear form $a(u,v) := (-\hat{V} \cdot  \nabla \cdot u, \nabla \cdot v)$ in 2D
+/** Class for integrating the bilinear form $a(u,v) := (-\hat{V} \cdot \nabla u, \nabla \cdot v)$ in 2D
     or 3D and where $\hat{V}$ is a vector coefficient, $u$ is in $H^1$ and $v$ is in $H(div)$. */
 class MixedGradDivIntegrator : public MixedScalarVectorIntegrator
 {
@@ -1780,7 +1780,7 @@ public:
    { scalar_fe.CalcPhysDivShape(Trans, shape); }
 };
 
-/** Class for integrating the bilinear form $a(u,v) := (-\hat{V} \nabla \cdot u, \nabla \cdot v)$ in 2D
+/** Class for integrating the bilinear form $a(u,v) := (-\hat{V} \nabla \cdot u, \nabla v)$ in 2D
     or 3D and where $\hat{V}$ is a vector coefficient, $u$ is in $H(div)$ and $v$ is in $H^1$. */
 class MixedDivGradIntegrator : public MixedScalarVectorIntegrator
 {
@@ -1820,7 +1820,7 @@ public:
    { scalar_fe.CalcPhysDivShape(Trans, shape); }
 };
 
-/** Class for integrating the bilinear form $a(u,v) := (-\hat{V} u, \nabla \cdot v)$ in 2D or 3D
+/** Class for integrating the bilinear form $a(u,v) := (-\hat{V} u, \nabla v)$ in 2D or 3D
     and where $\hat{V}$ is a vector coefficient, $u$ is in $H^1$ or $L_2$ and $v$ is in $H^1$. */
 class MixedScalarWeakDivergenceIntegrator : public MixedScalarVectorIntegrator
 {
@@ -2450,6 +2450,135 @@ public:
                                       DenseMatrix &);
 };
 
+
+/// $\alpha (Q \Delta u, v)$
+class LaplaceIntegrator : public BilinearFormIntegrator
+{
+protected:
+   Coefficient *Q;
+   real_t alpha;
+
+private:
+   Vector laplace, shape;
+
+public:
+   LaplaceIntegrator(Coefficient &q, real_t a = 1.0)
+      : Q(&q) { alpha = a; }
+
+   virtual void AssembleElementMatrix(const FiniteElement &,
+                                      ElementTransformation &,
+                                      DenseMatrix &);
+
+   virtual void AssembleElementMatrix2(const FiniteElement &trial_fe,
+                                       const FiniteElement &test_fe,
+                                       ElementTransformation &Trans,
+                                       DenseMatrix &elmat);
+
+   static const IntegrationRule &GetRule(const FiniteElement &trial_fe,
+                                         const FiniteElement &test_fe,
+                                         ElementTransformation &Trans);
+};
+
+/// $\alpha (u, Q \Delta v)$
+class TransposeLaplaceIntegrator : public TransposeIntegrator
+{
+public:
+   TransposeLaplaceIntegrator (Coefficient &q, real_t a = 1.0)
+      : TransposeIntegrator(new LaplaceIntegrator(q, a)) { }
+};
+
+/// $\alpha (\Delta u, Q \cdot \nabla v)$
+class LaplaceGradIntegrator : public BilinearFormIntegrator
+{
+protected:
+   VectorCoefficient *Q;
+   real_t alpha;
+   int dim;
+
+private:
+   Vector laplace, vec2, BdFidxT;
+   DenseMatrix dshape, adjJ, Q_ir;
+
+public:
+   LaplaceGradIntegrator(VectorCoefficient &q, real_t a = 1.0)
+      : Q(&q) { alpha = a; }
+
+   virtual void AssembleElementMatrix(const FiniteElement &,
+                                      ElementTransformation &,
+                                      DenseMatrix &);
+
+   virtual void AssembleElementMatrix2(const FiniteElement &trial_fe,
+                                       const FiniteElement &test_fe,
+                                       ElementTransformation &Trans,
+                                       DenseMatrix &elmat);
+
+   static const IntegrationRule &GetRule(const FiniteElement &trial_fe,
+                                         const FiniteElement &test_fe,
+                                         ElementTransformation &Trans);
+};
+
+/// $\alpha (Q \cdot \nabla u,  \Delta v)$
+class GradLaplaceIntegrator : public TransposeIntegrator
+{
+public:
+   GradLaplaceIntegrator(VectorCoefficient &q, real_t a = 1.0)
+      : TransposeIntegrator(new LaplaceGradIntegrator(q, a)) { }
+};
+
+/// $\alpha (Q \Delta u, \Delta v)$
+class  LaplaceLaplaceIntegrator : public BilinearFormIntegrator
+{
+protected:
+   Coefficient *Q;
+   real_t alpha;
+
+private:
+   Vector laplace, te_laplace;
+
+public:
+   LaplaceLaplaceIntegrator(Coefficient &q, real_t a = 1.0)
+      : Q(&q) { alpha = a; }
+
+   virtual void AssembleElementMatrix(const FiniteElement &,
+                                      ElementTransformation &,
+                                      DenseMatrix &);
+
+   virtual void AssembleElementMatrix2(const FiniteElement &trial_fe,
+                                       const FiniteElement &test_fe,
+                                       ElementTransformation &Trans,
+                                       DenseMatrix &elmat);
+
+   static const IntegrationRule &GetRule(const FiniteElement &trial_fe,
+                                         const FiniteElement &test_fe,
+                                         ElementTransformation &Trans);
+};
+
+// Alias for @LaplaceLaplaceIntegrator.
+using BiHarmonicIntegrator = LaplaceLaplaceIntegrator;
+
+/// Get the inverse estimate
+class InverseEstimateIntegrator : public BilinearFormIntegrator
+{
+protected:
+   Coefficient *Q;
+
+private:
+   Vector laplace, shape, ovec;//, vec2, BdFidxT;
+   DenseMatrix dshape, lapmat, bimat;//, adjJ, Q_ir;
+
+public:
+   InverseEstimateIntegrator(Coefficient &q)
+      : Q(&q) { }
+
+   virtual void AssembleElementMatrix(const FiniteElement &,
+                                      ElementTransformation &,
+                                      DenseMatrix &);
+
+   static const IntegrationRule &GetRule(const FiniteElement &trial_fe,
+                                         const FiniteElement &test_fe,
+                                         ElementTransformation &Trans);
+};
+
 /** Class for integrating the bilinear form $a(u,v) := (Q u, v)$,
     where $u=(u_1,\dots,u_n)$ and $v=(v_1,\dots,v_n)$, $u_i$ and $v_i$ are defined
     by scalar FE through standard transformation. */

fem/coefficient.cpp

@@ -807,6 +807,7 @@ void SymmetricMatrixCoefficient::ProjectSymmetric(QuadratureFunction &qf)
 
    QuadratureSpaceBase &qspace = *qf.GetSpace();
    const int ne = qspace.GetNE();
+   qf.HostWrite();
    DenseMatrix values;
    DenseSymmetricMatrix matrix;
    for (int iel = 0; iel < ne; ++iel)
@@ -818,7 +819,7 @@ void SymmetricMatrixCoefficient::ProjectSymmetric(QuadratureFunction &qf)
       {
          const IntegrationPoint &ip = ir[iq];
          T.SetIntPoint(&ip);
-         matrix.UseExternalData(&values(0, iq), vdim);
+         matrix.UseExternalData(&values(0, iq), height);
          Eval(matrix, T, ip);
       }
    }
@@ -828,13 +829,12 @@ void SymmetricMatrixCoefficient::ProjectSymmetric(QuadratureFunction &qf)
 void SymmetricMatrixCoefficient::Eval(DenseMatrix &K, ElementTransformation &T,
                                       const IntegrationPoint &ip)
 {
-   mat.SetSize(height);
-   Eval(mat, T, ip);
+   Eval(mat_aux, T, ip);
    for (int j = 0; j < width; ++j)
    {
       for (int i = 0; i < height; ++ i)
       {
-         K(i, j) = mat(i, j);
+         K(i, j) = mat_aux(i, j);
       }
    }
 }
@@ -924,6 +924,75 @@ void MatrixArrayCoefficient::Eval(DenseMatrix &K, ElementTransformation &T,
    }
 }
 
+MatrixArrayVectorCoefficient::MatrixArrayVectorCoefficient (int dim)
+   : MatrixCoefficient (dim)
+{
+   Coeff.SetSize(height);
+   ownCoeff.SetSize(height);
+   for (int i = 0; i < height; i++)
+   {
+      Coeff[i] = NULL;
+      ownCoeff[i] = true;
+   }
+}
+
+void MatrixArrayVectorCoefficient::SetTime(real_t t)
+{
+   for (int i=0; i < height; i++)
+   {
+      if (Coeff[i]) { Coeff[i]->SetTime(t); }
+   }
+   this->MatrixCoefficient::SetTime(t);
+}
+
+void MatrixArrayVectorCoefficient::Set(int i, VectorCoefficient * c, bool own)
+{
+   MFEM_ASSERT(i < height && i >= 0, "Row "
+               << i << " does not exist. " <<
+               "Matrix height = " << height << ".");
+   if (ownCoeff[i]) { delete Coeff[i]; }
+   Coeff[i] = c;
+   ownCoeff[i] = own;
+}
+
+MatrixArrayVectorCoefficient::~MatrixArrayVectorCoefficient ()
+{
+   for (int i=0; i < height; i++)
+   {
+      if (ownCoeff[i]) { delete Coeff[i]; }
+   }
+}
+
+void MatrixArrayVectorCoefficient::Eval(int i, Vector &V,
+                                        ElementTransformation &T,
+                                        const IntegrationPoint &ip)
+{
+   MFEM_ASSERT(i < height && i >= 0, "Row "
+               << i << " does not exist. " <<
+               "Matrix height = " << height << ".");
+   if (Coeff[i])
+   {
+      Coeff[i] -> Eval(V, T, ip);
+   }
+   else
+   {
+      V = 0.0;
+   }
+}
+
+void MatrixArrayVectorCoefficient::Eval(DenseMatrix &K,
+                                        ElementTransformation &T,
+                                        const IntegrationPoint &ip)
+{
+   K.SetSize(height, width);
+   Vector V(width);
+   for (int i = 0; i < height; i++)
+   {
+      this->Eval(i, V, T, ip);
+      K.SetRow(i, V);
+   }
+}
+
 void MatrixRestrictedCoefficient::SetTime(real_t t)
 {
    if (c) { c->SetTime(t); }
@@ -1041,6 +1110,27 @@ real_t DeterminantCoefficient::Eval(ElementTransformation &T,
    return ma.Det();
 }
 
+TraceCoefficient::TraceCoefficient(MatrixCoefficient &A)
+   : a(&A), ma(A.GetHeight(), A.GetWidth())
+{
+   MFEM_ASSERT(A.GetHeight() == A.GetWidth(),
+               "TraceCoefficient:  "
+               "Argument must be a square matrix.");
+}
+
+void TraceCoefficient::SetTime(real_t t)
+{
+   if (a) { a->SetTime(t); }
+   this->Coefficient::SetTime(t);
+}
+
+real_t TraceCoefficient::Eval(ElementTransformation &T,
+                              const IntegrationPoint &ip)
+{
+   a->Eval(ma, T, ip);
+   return ma.Trace();
+}
+
 VectorSumCoefficient::VectorSumCoefficient(int dim)
    : VectorCoefficient(dim),
      ACoef(NULL), BCoef(NULL),
@@ -1326,6 +1416,30 @@ void InverseMatrixCoefficient::Eval(DenseMatrix &M,
    M.Invert();
 }
 
+ExponentialMatrixCoefficient::ExponentialMatrixCoefficient(MatrixCoefficient &A)
+   : MatrixCoefficient(A.GetHeight(), A.GetWidth()), a(&A)
+{
+   MFEM_ASSERT(A.GetHeight() == A.GetWidth() && A.GetHeight() == 2,
+               "ExponentialMatrixCoefficient:  "
+               << "Argument must be a square 2x2 matrix."
+               << "  Height = " << A.GetHeight()
+               << ", Width = " << A.GetWidth());
+}
+
+void ExponentialMatrixCoefficient::SetTime(real_t t)
+{
+   if (a) { a->SetTime(t); }
+   this->MatrixCoefficient::SetTime(t);
+}
+
+void ExponentialMatrixCoefficient::Eval(DenseMatrix &M,
+                                        ElementTransformation &T,
+                                        const IntegrationPoint &ip)
+{
+   a->Eval(M, T, ip);
+   M.Exponential();
+}
+
 OuterProductCoefficient::OuterProductCoefficient(VectorCoefficient &A,
                                                  VectorCoefficient &B)
    : MatrixCoefficient(A.GetVDim(), B.GetVDim()), a(&A), b(&B),
@@ -1390,6 +1504,295 @@ void CrossCrossCoefficient::Eval(DenseMatrix &M, ElementTransformation &T,
    M *= ((a == NULL ) ? aConst : a->Eval(T, ip) );
 }
 
+
+InverseEstimateCoefficient::InverseEstimateCoefficient(FiniteElementSpace *f)
+   : fes(f), Q(NULL), ir(NULL)
+{
+   ComputeInverseEstimates();
+}
+
+InverseEstimateCoefficient::InverseEstimateCoefficient(FiniteElementSpace *f,
+                                                       Coefficient &q)
+   : fes(f), Q(&q), ir(NULL)
+{
+   ComputeInverseEstimates();
+}
+
+GridFunction *InverseEstimateCoefficient::GetGridFunction()
+{
+   FiniteElementCollection* fec_ec = new L2_FECollection(0,
+                                                         fes ->GetMesh()->Dimension());
+   FiniteElementSpace *fes_ec = new FiniteElementSpace(fes ->GetMesh(), fec_ec);
+   GridFunction *gf = new GridFunction(fes_ec, elemInvEst.GetData());
+   gf->MakeOwner(fec_ec);
+   return gf;
+}
+
+void InverseEstimateCoefficient::ComputeInverseEstimates()
+{
+   elemInvEst.SetSize(fes -> GetNE());
+   SetIntRule(*fes->GetFE(0));
+   for (int i = 0; i < fes -> GetNE(); i++)
+   {
+      elemInvEst[i] = ElementInverseEstimate(*fes->GetFE(i),
+                                             *fes->GetElementTransformation(i));
+   }
+}
+
+void InverseEstimateCoefficient::SetIntRule(const FiniteElement &el)
+{
+   ir = &IntRules.Get(el.GetGeomType(), 2*el.GetOrder());
+}
+
+real_t InverseEstimateCoefficient::ElementInverseEstimate(
+   const FiniteElement &el,
+   ElementTransformation &Trans)
+{
+   if (el.GetOrder() < 2)
+   {
+      return std::numeric_limits<real_t>::min();
+   }
+
+   int nd = el.GetDof();
+   int dim = el.GetDim();
+
+   shape.SetSize(nd);
+   dshape.SetSize(nd,dim);
+   laplace.SetSize(nd);
+
+   lapmat.SetSize(nd,nd);
+   bimat.SetSize(nd,nd);
+   ovec.SetSize(nd);
+
+   real_t w,q = 1.0;
+
+   bimat = 0.0;
+   lapmat = 0.0;
+   ovec = 0.0;
+   for (int i = 0; i < ir->GetNPoints(); i++)
+   {
+      const IntegrationPoint &ip = ir->IntPoint(i);
+      Trans.SetIntPoint(&ip);
+      w = Trans.Weight()*ip.weight;
+      if (Q)
+      {
+         q = Q->Eval(Trans, ip);
+      }
+
+      el.CalcPhysDShape(Trans, dshape);
+      AddMult_a_AAt(w*q, dshape, lapmat);
+
+      el.CalcPhysLaplacian(Trans, laplace);
+      AddMult_a_VVt(w*q*q, laplace, bimat);
+
+      el.CalcPhysShape(Trans, shape);
+      ovec.Add(w, shape);
+   }
+   ovec *= 1.0/ovec.Norml2();
+
+   // Correct nullspace
+   AddMultVVt(ovec, lapmat);
+
+   // Return largest eigenvalue
+   return bimat.Eigenvalue(lapmat);
+}
+
+ElasticInverseEstimateCoefficient
+::ElasticInverseEstimateCoefficient(FiniteElementSpace *f)
+   : fes(f), Q(NULL), ir(NULL)
+{
+   ComputeInverseEstimates();
+}
+
+ElasticInverseEstimateCoefficient
+::ElasticInverseEstimateCoefficient(FiniteElementSpace *f,
+                                    Coefficient &q)
+   : fes(f), Q(&q), ir(NULL)
+{
+   ComputeInverseEstimates();
+}
+
+GridFunction *ElasticInverseEstimateCoefficient::GetGridFunction()
+{
+   FiniteElementCollection* fec_ec = new L2_FECollection(0,
+                                                         fes ->GetMesh()->Dimension());
+   FiniteElementSpace *fes_ec = new FiniteElementSpace(fes ->GetMesh(), fec_ec);
+   GridFunction *gf = new GridFunction(fes_ec, elemInvEst.GetData());
+   gf->MakeOwner(fec_ec);
+   return gf;
+}
+
+void ElasticInverseEstimateCoefficient::ComputeInverseEstimates()
+{
+   elemInvEst.SetSize(fes -> GetNE());
+   SetIntRule(*fes->GetFE(0));
+   int dim = fes->GetFE(0)->GetDim();
+
+   emat.SetSize(dim,dim);
+   divmat.SetSize(dim,dim);
+   for (int i = 0; i < dim; i++)
+   {
+      for (int j = 0; j < dim; j++)
+      {
+         emat(i,j)= new DenseMatrix();
+         divmat(i,j)= new DenseMatrix();
+      }
+   }
+
+   hmap.SetSize(dim,dim);
+
+   if (dim == 2)
+   {
+      hmap(0,0) = 0;
+      hmap(0,1) =  hmap(1,0) =  1;
+      hmap(1,1) = 2;
+   }
+   else if (dim == 2)
+   {
+      hmap(0,0) = 0;
+      hmap(0,1) = hmap(1,0) = 1;
+      hmap(0,2) = hmap(2,0) = 2;
+      hmap(1,1) = 3;
+      hmap(1,2) = hmap(2,1) = 4;
+      hmap(2,2) = 5;
+   }
+   else
+   {
+      mfem_error("Only implemented for 2D and 3D");
+   }
+
+   for (int i = 0; i < fes -> GetNE(); i++)
+   {
+      elemInvEst[i] = ElementInverseEstimate(*fes->GetFE(i),
+                                             *fes->GetElementTransformation(i));
+   }
+}
+
+void ElasticInverseEstimateCoefficient::SetIntRule(const FiniteElement &el)
+{
+   ir = &IntRules.Get(el.GetGeomType(), 2*el.GetOrder());
+}
+
+real_t ElasticInverseEstimateCoefficient::ElementInverseEstimate(
+   const FiniteElement &el,
+   ElementTransformation &Trans)
+{
+   // if (el.GetDerivType() != (int) FiniteElement::HESS)
+   //  {
+   //     return std::numeric_limits<real_t>::min();
+   //  }
+
+   int nd = el.GetDof();
+   int dim = el.GetDim();
+
+   shape.SetSize(nd);
+   dshape.SetSize(nd,dim);
+   hshape.SetSize(nd,dim*(dim+1)/2);
+
+   for (int i = 0; i < dim; i++)
+   {
+      for (int j = 0; j < dim; j++)
+      {
+         emat(i,j)->SetSize(nd,nd);
+         *emat(i,j) = 0.0;
+         divmat(i,j)->SetSize(nd,nd);
+         *divmat(i,j) = 0.0;
+      }
+   }
+
+   real_t w,q = 1.0;
+   for (int ii = 0; ii < ir->GetNPoints(); ii++)
+   {
+      const IntegrationPoint &ip = ir->IntPoint(ii);
+      Trans.SetIntPoint(&ip);
+      w = Trans.Weight()*ip.weight;
+      if (Q)
+      {
+         q = Q->Eval(Trans, ip);
+      }
+
+      el.CalcPhysDShape(Trans, dshape);
+      for (int i = 0; i < dim; i++)
+      {
+         for (int j = 0; j < dim; j++)
+         {
+            AddMult_a_VVt(w*q, Vector(dshape.GetColumn(i),nd), *emat(j,j));
+
+            AddMult_a_VWt(w*q, Vector(dshape.GetColumn(i),nd),
+                          Vector(dshape.GetColumn(j),nd), *emat(j,i));
+
+            AddMult_a_VWt(w*q, Vector(dshape.GetColumn(j),nd),
+                          Vector(dshape.GetColumn(i),nd), *emat(i,j));
+
+            AddMult_a_VVt(w*q, Vector(dshape.GetColumn(j),nd), *emat(i,i));
+         }
+      }
+
+      el.CalcPhysHessian(Trans, hshape);
+      for (int i = 0; i < dim; i++)
+      {
+         for (int j = 0; j < dim; j++)
+         {
+            for (int k = 0; k < dim; k++)
+            {
+               AddMult_a_VWt(w*q*q, Vector(hshape.GetColumn(hmap(i,i)),nd),
+                             Vector(hshape.GetColumn(hmap(k,k)),nd), *divmat(j,j));
+
+               AddMult_a_VWt(w*q*q, Vector(hshape.GetColumn(hmap(i,i)),nd),
+                             Vector(hshape.GetColumn(hmap(k,j)),nd), *divmat(j,k));
+
+               AddMult_a_VWt(w*q*q, Vector(hshape.GetColumn(hmap(i,j)),nd),
+                             Vector(hshape.GetColumn(hmap(k,k)),nd), *divmat(i,j));
+
+               AddMult_a_VWt(w*q*q, Vector(hshape.GetColumn(hmap(i,j)),nd),
+                             Vector(hshape.GetColumn(hmap(k,j)),nd), *divmat(i,k));
+            }
+         }
+      }
+   }
+
+   // Collect matrices
+   emat_tot.SetSize(nd*dim,nd*dim);
+   divmat_tot.SetSize(nd*dim,nd*dim);
+   for (int i = 0; i < dim; i++)
+   {
+      for (int j = 0; j < dim; j++)
+      {
+         emat_tot  .SetSubMatrix(i*nd, j*nd, *emat(i,j));
+         divmat_tot.SetSubMatrix(i*nd, j*nd, *divmat(i,j));
+      }
+   }
+
+   // Correct nullspace
+   DenseMatrix ns;
+   emat_tot.NullSpace(ns, 1e-10);
+   for (int i = 0; i < ns.Width(); i++)
+   {
+      AddMultVVt(Vector(ns.GetColumn(i),nd*dim), emat_tot);
+   }
+
+   // Return largest eigenvalue
+   return divmat_tot.Eigenvalue(emat_tot);
+}
+
+ElasticInverseEstimateCoefficient::~ElasticInverseEstimateCoefficient()
+{
+   for (int i = 0; i < emat.NumRows(); i++)
+   {
+      for (int j = 0; j < emat.NumCols(); j++)
+      {
+         delete emat(i,j);
+      }
+   }
+   for (int i = 0; i < divmat.NumRows(); i++)
+   {
+      for (int j = 0; j < divmat.NumCols(); j++)
+      {
+         delete divmat(i,j);
+      }
+   }
+}
+
 real_t LpNormLoop(real_t p, Coefficient &coeff, Mesh &mesh,
                   const IntegrationRule *irs[])
 {

fem/coefficient.hpp

@@ -1334,6 +1334,46 @@ public:
    virtual ~MatrixArrayCoefficient();
 };
 
+/** @brief Matrix coefficient defined row-wise by an array of vector
+    coefficients. Rows that are not set will evaluate to zero. The
+    matrix coefficient is stored as an array indexing the rows of
+    the matrix. */
+class MatrixArrayVectorCoefficient : public MatrixCoefficient
+{
+private:
+   Array<VectorCoefficient *> Coeff;
+   Array<bool> ownCoeff;
+
+public:
+   /** @brief Construct a coefficient matrix of dimensions @a dim * @a dim. The
+       actual coefficients still need to be added with Set(). */
+   explicit MatrixArrayVectorCoefficient (int dim);
+
+   /// Set the time for internally stored coefficients
+   void SetTime(real_t t) override;
+
+   /// Get the vector coefficient located at the i-th row of the matrix
+   VectorCoefficient* GetCoeff (int i) { return Coeff[i]; }
+
+   /** @brief Set the coefficient located at the i-th row of the matrix.
+       By this will take ownership of the Coefficient passed in, but this
+       can be overridden with the @a own parameter. */
+   void Set(int i, VectorCoefficient * c, bool own=true);
+
+   using MatrixCoefficient::Eval;
+
+   /// Evaluate coefficient located at the i-th row of the matrix using integration
+   /// point @a ip.
+   void Eval(int i, Vector &V, ElementTransformation &T,
+             const IntegrationPoint &ip);
+
+   /// Evaluate the matrix coefficient @a ip.
+   void Eval(DenseMatrix &K, ElementTransformation &T,
+             const IntegrationPoint &ip) override;
+
+   virtual ~MatrixArrayVectorCoefficient();
+};
+
 
 /** @brief Derived matrix coefficient that has the value of the parent matrix
     coefficient where it is active and is zero otherwise. */
@@ -1426,12 +1466,13 @@ public:
 class SymmetricMatrixCoefficient : public MatrixCoefficient
 {
 protected:
+
    /// Internal matrix used when evaluating this coefficient as a DenseMatrix.
-   DenseSymmetricMatrix mat;
+   mutable DenseSymmetricMatrix mat_aux;
 public:
    /// Construct a dim x dim matrix coefficient.
    explicit SymmetricMatrixCoefficient(int dimension)
-      : MatrixCoefficient(dimension, true) { }
+      : MatrixCoefficient(dimension, true), mat_aux(height) { }
 
    /// Get the size of the matrix.
    int GetSize() const { return height; }
@@ -1464,8 +1505,9 @@ public:
    virtual void Eval(DenseMatrix &K, ElementTransformation &T,
                      const IntegrationPoint &ip);
 
-   /// Return a reference to the constant matrix.
-   const DenseSymmetricMatrix& GetMatrix() { return mat; }
+
+   /// @deprecated Return a reference to the internal matrix used when evaluating this coefficient as a DenseMatrix.
+   MFEM_DEPRECATED const DenseSymmetricMatrix& GetMatrix() { return mat_aux; }
 
    virtual ~SymmetricMatrixCoefficient() { }
 };
@@ -1485,6 +1527,10 @@ public:
    /// Evaluate the matrix coefficient at @a ip.
    virtual void Eval(DenseSymmetricMatrix &M, ElementTransformation &T,
                      const IntegrationPoint &ip) { M = mat; }
+
+   /// Return a reference to the constant matrix.
+   const DenseSymmetricMatrix& GetMatrix() { return mat; }
+
 };
 
 
@@ -1761,6 +1807,31 @@ public:
                        const IntegrationPoint &ip);
 };
 
+/// Scalar coefficient defined as the trace of a matrix coefficient
+class TraceCoefficient : public Coefficient
+{
+private:
+   MatrixCoefficient * a;
+
+   mutable DenseMatrix ma;
+
+public:
+   /// Construct with the matrix.
+   TraceCoefficient(MatrixCoefficient &A);
+
+   /// Set the time for internally stored coefficients
+   void SetTime(real_t t);
+
+   /// Reset the matrix coefficient
+   void SetACoef(MatrixCoefficient &A) { a = &A; }
+   /// Return the matrix coefficient
+   MatrixCoefficient * GetACoef() const { return a; }
+
+   /// Evaluate the trace coefficient at @a ip.
+   virtual real_t Eval(ElementTransformation &T,
+                       const IntegrationPoint &ip);
+};
+
 /// Vector coefficient defined as the linear combination of two vectors
 class VectorSumCoefficient : public VectorCoefficient
 {
@@ -2112,7 +2183,7 @@ public:
                      const IntegrationPoint &ip);
 };
 
-/// Matrix coefficient defined as the transpose a matrix coefficient
+/// Matrix coefficient defined as the transpose of a matrix coefficient
 class TransposeMatrixCoefficient : public MatrixCoefficient
 {
 private:
@@ -2135,7 +2206,7 @@ public:
                      const IntegrationPoint &ip);
 };
 
-/// Matrix coefficient defined as the inverse a matrix coefficient.
+/// Matrix coefficient defined as the inverse of a matrix coefficient.
 class InverseMatrixCoefficient : public MatrixCoefficient
 {
 private:
@@ -2158,6 +2229,29 @@ public:
                      const IntegrationPoint &ip);
 };
 
+/// Matrix coefficient defined as the exponential of a matrix coefficient.
+class ExponentialMatrixCoefficient : public MatrixCoefficient
+{
+private:
+   MatrixCoefficient * a;
+
+public:
+   /// Construct the matrix coefficient.  Result is $ \exp(A) $.
+   ExponentialMatrixCoefficient(MatrixCoefficient &A);
+
+   /// Set the time for internally stored coefficients
+   void SetTime(real_t t);
+
+   /// Reset the matrix coefficient
+   void SetACoef(MatrixCoefficient &A) { a = &A; }
+   /// Return the matrix coefficient
+   MatrixCoefficient * GetACoef() const { return a; }
+
+   /// Evaluate the matrix coefficient at @a ip.
+   virtual void Eval(DenseMatrix &M, ElementTransformation &T,
+                     const IntegrationPoint &ip);
+};
+
 /// Matrix coefficient defined as the outer product of two vector coefficients.
 class OuterProductCoefficient : public MatrixCoefficient
 {
@@ -2234,6 +2328,127 @@ public:
 };
 ///@}
 
+/** @brief
+*/
+class InverseEstimateCoefficient : public Coefficient
+{
+private:
+   ///
+   Vector elemInvEst;
+   /// FE space on which the grid function lives. Owned if #fec is not NULL.
+   FiniteElementSpace *fes;
+
+   ///
+   const IntegrationRule *ir;
+
+   ///
+   Coefficient *Q;
+   Vector laplace, shape, ovec, evec;
+   DenseMatrix dshape, lapmat, bimat;
+
+   ///
+   void SetIntRule(const FiniteElement &el);
+
+   ///
+   void ComputeInverseEstimates();
+
+   real_t ElementInverseEstimate(const FiniteElement &el,
+                                 ElementTransformation &Trans);
+
+public:
+   ///
+   InverseEstimateCoefficient(FiniteElementSpace *f);
+   InverseEstimateCoefficient(FiniteElementSpace *f, Coefficient &q);
+
+   /// Caller gets owner ship of GridFunction and
+   GridFunction *GetGridFunction();
+
+   /// Reset the scalar factor
+   void SetDiffusion(Coefficient &q)
+   {
+      if (Q != &q)
+      {
+         Q = &q;
+         ComputeInverseEstimates();
+      }
+   }
+   /// Return the scalar factor
+   Coefficient * GetDiffusion() const { return Q; }
+
+   /// Evaluate the coefficient at @a ip.
+   virtual real_t Eval(ElementTransformation &T,
+                       const IntegrationPoint &ip)
+   { return elemInvEst[T.ElementNo]; }
+
+};
+
+class ElasticInverseEstimateCoefficient : public Coefficient
+{
+private:
+   ///
+   Vector elemInvEst;
+   /// FE space on which the grid function lives. Owned if #fec is not NULL.
+   FiniteElementSpace *fes;
+
+   ///
+   const IntegrationRule *ir;
+
+   ///
+   Coefficient *Q;
+   Vector shape, ovec, evec;
+   DenseMatrix dshape, hshape, emat_tot, divmat_tot;
+   Array2D<DenseMatrix*> emat,divmat;
+   Array2D<int> hmap;
+
+   ///
+   void SetIntRule(const FiniteElement &el);
+
+   ///
+   void ComputeInverseEstimates();
+
+   ///
+   real_t ElementInverseEstimate(const FiniteElement &el,
+                                 ElementTransformation &Trans);
+
+public:
+   ///
+   ElasticInverseEstimateCoefficient(FiniteElementSpace *f);
+   ElasticInverseEstimateCoefficient(FiniteElementSpace *f, Coefficient &q);
+
+   /// Caller gets owner ship of GridFunction and
+   GridFunction *GetGridFunction();
+
+   /// Reset the scalar factor
+   void SetDiffusion(Coefficient &q)
+   {
+      if (Q != &q)
+      {
+         Q = &q;
+         ComputeInverseEstimates();
+      }
+   }
+   void SetShearModulus(Coefficient &q) { SetDiffusion(q);}
+
+   /// Return the scalar factor
+   Coefficient * GetDiffusion() const { return Q; }
+   Coefficient * GetModulus() const { return GetDiffusion(); }
+
+   /// Evaluate the coefficient at @a ip.
+   virtual real_t Eval(ElementTransformation &T,
+                       const IntegrationPoint &ip)
+   { return elemInvEst[T.ElementNo]; }
+
+   // Destructor
+   ~ElasticInverseEstimateCoefficient();
+
+};
+
+
+///@}
+
+
+
+
 /** @brief Vector quadrature function coefficient which requires that the
     quadrature rules used for this vector coefficient be the same as those that
     live within the supplied QuadratureFunction. */

fem/eltrans.hpp

@@ -52,6 +52,15 @@ protected:
    const DenseMatrix &EvalTransAdjugateJ();
    const DenseMatrix &EvalInverseJ();
 
+   /// @name Tolerance used for point comparisons
+   ///@{
+#ifdef MFEM_USE_DOUBLE
+   static constexpr real_t tol_0 = 1e-15;
+#elif defined(MFEM_USE_SINGLE)
+   static constexpr real_t tol_0 = 1e-7;
+#endif
+   ///@}
+
 public:
 
    /** This enumeration declares the values stored in
@@ -176,7 +185,7 @@ public:
        returned. This method is not 100 percent reliable for non-linear
        transformations. */
    virtual int TransformBack(const Vector &pt, IntegrationPoint &ip,
-                             const real_t phys_tol = 1e-15) = 0;
+                             const real_t phys_tol = tol_0) = 0;
 
    virtual ~ElementTransformation() { }
 };
@@ -281,9 +290,15 @@ public:
         rel_qpts_order(-1),
         solver_type(NewtonElementProject),
         max_iter(16),
+#ifdef MFEM_USE_DOUBLE
         ref_tol(1e-15),
         phys_rtol(1e-15),
         ip_tol(1e-8),
+#elif defined(MFEM_USE_SINGLE)
+        ref_tol(1e-7),
+        phys_rtol(1e-7),
+        ip_tol(1e-4),
+#endif
         print_level(-1)
    { }
 
@@ -449,7 +464,7 @@ public:
        returned. This method is not 100 percent reliable for non-linear
        transformations. */
    virtual int TransformBack(const Vector & v, IntegrationPoint & ip,
-                             const real_t phys_rel_tol = 1e-15)
+                             const real_t phys_rel_tol = tol_0)
    {
       InverseElementTransformation inv_tr(this);
       inv_tr.SetPhysicalRelTol(phys_rel_tol);

fem/fe/fe_base.cpp

@@ -221,7 +221,7 @@ void FiniteElement::CalcPhysLaplacian(ElementTransformation &Trans,
    {
       for (int nd = 0; nd < dof; nd++)
       {
-         Laplacian[nd] = hess(nd,0) + hess(nd,4) + hess(nd,5);
+         Laplacian[nd] = hess(nd,0) + hess(nd,3) + hess(nd,5);
       }
    }
    else if (dim == 2)
@@ -259,10 +259,10 @@ void FiniteElement::CalcPhysLinLaplacian(ElementTransformation &Trans,
       scale[1] = 2*Gij(0,1);
       scale[2] = 2*Gij(0,2);
 
-      scale[3] = 2*Gij(1,2);
-      scale[4] =   Gij(2,2);
+      scale[3] =   Gij(1,1);
+      scale[4] = 2*Gij(1,2);
 
-      scale[5] =   Gij(1,1);
+      scale[5] =   Gij(2,2);
    }
    else if (dim == 2)
    {
@@ -299,12 +299,12 @@ void  FiniteElement::CalcPhysHessian(ElementTransformation &Trans,
       map[2] = 2;
 
       map[3] = 1;
-      map[4] = 5;
-      map[5] = 3;
+      map[4] = 3;
+      map[5] = 4;
 
       map[6] = 2;
-      map[7] = 3;
-      map[8] = 4;
+      map[7] = 4;
+      map[8] = 5;
    }
    else if (dim == 2)
    {

fem/fe/fe_base.hpp

@@ -299,7 +299,8 @@ public:
       NONE, ///< No derivatives implemented
       GRAD, ///< Implements CalcDShape methods
       DIV,  ///< Implements CalcDivShape methods
-      CURL  ///< Implements CalcCurlShape methods
+      CURL, ///< Implements CalcCurlShape methods
+      HESS  ///< Implements CalcHessian & CalcDShape methods
    };
 
    /** @brief Construct FiniteElement with given
@@ -356,7 +357,7 @@ public:
 
    /** @brief Returns the FiniteElement::DerivType of the element describing the
        spatial derivative method implemented, one of {NONE, GRAD,
-       DIV, CURL}. */
+       DIV, CURL, HESS}. */
    int GetDerivType() const { return deriv_type; }
 
    /** @brief Returns the FiniteElement::DerivType of the element describing how
@@ -394,7 +395,32 @@ public:
    /// Get a const reference to the nodes of the element
    const IntegrationRule & GetNodes() const { return Nodes; }
 
-   // virtual functions for finite elements on vector spaces
+   /** @brief Evaluate the Hessians of all shape functions of a scalar finite
+       element in reference space at the given point @a ip. */
+   /** Each row of the result DenseMatrix @a Hessian contains upper triangular
+       part of the Hessian of one shape function.
+       The order in 2D is {u_xx, u_xy, u_yy}.
+       The size (#dof x (#dim (#dim+1)/2) of @a Hessian must be set in advance.*/
+   virtual void CalcHessian(const IntegrationPoint &ip,
+                            DenseMatrix &Hessian) const;
+
+   /** @brief Evaluate the Hessian of all shape functions of a scalar finite
+       element in physical space at the given point @a ip. */
+   /** The size (#dof, #dim*(#dim+1)/2) of @a Hessian must be set in advance. */
+   void CalcPhysHessian(ElementTransformation &Trans,
+                        DenseMatrix& Hessian) const;
+
+   /** @brief Evaluate the Laplacian of all shape functions of a scalar finite
+       element in physical space at the given point @a ip. */
+   /** The size (#dof) of @a Laplacian must be set in advance. */
+   void CalcPhysLaplacian(ElementTransformation &Trans,
+                          Vector& Laplacian) const;
+
+   /** @brief Evaluate the Laplacian of all shape functions of a scalar finite
+       element in physical space at the given point @a ip. */
+   /** The size (#dof) of @a Laplacian must be set in advance. */
+   void CalcPhysLinLaplacian(ElementTransformation &Trans,
+                             Vector& Laplacian) const;
 
    /** @brief Evaluate the values of all shape functions of a *vector* finite
        element in reference space at the given point @a ip. */
@@ -454,30 +480,6 @@ public:
    */
    virtual void GetFaceDofs(int face, int **dofs, int *ndofs) const;
 
-   /** @brief Evaluate the Hessians of all shape functions of a scalar finite
-       element in reference space at the given point @a ip. */
-   /** Each row of the result DenseMatrix @a Hessian contains upper triangular
-       part of the Hessian of one shape function.
-       The order in 2D is {u_xx, u_xy, u_yy}.
-       The size (#dof x (#dim (#dim+1)/2) of @a Hessian must be set in advance.*/
-   virtual void CalcHessian(const IntegrationPoint &ip,
-                            DenseMatrix &Hessian) const;
-
-   /** @brief Evaluate the Hessian of all shape functions of a scalar finite
-       element in reference space at the given point @a ip. */
-   /** The size (#dof, #dim*(#dim+1)/2) of @a Hessian must be set in advance. */
-   virtual void CalcPhysHessian(ElementTransformation &Trans,
-                                DenseMatrix& Hessian) const;
-
-   /** @brief Evaluate the Laplacian of all shape functions of a scalar finite
-       element in reference space at the given point @a ip. */
-   /** The size (#dof) of @a Laplacian must be set in advance. */
-   virtual void CalcPhysLaplacian(ElementTransformation &Trans,
-                                  Vector& Laplacian) const;
-
-   virtual void CalcPhysLinLaplacian(ElementTransformation &Trans,
-                                     Vector& Laplacian) const;
-
    /** @brief Return the local interpolation matrix @a I (Dof x Dof) where the
        fine element is the image of the base geometry under the given
        transformation. */

fem/fe/fe_nurbs.cpp

@@ -349,10 +349,10 @@ void NURBS3DFiniteElement::CalcHessian (const IntegrationPoint &ip,
             d2sum[1] += ( hessian(o,1) = dsx*dsy*sz*weights(o) );
             d2sum[2] += ( hessian(o,2) = dsx*sy*dsz*weights(o) );
 
-            d2sum[3] += ( hessian(o,3) = sx*dsy*dsz*weights(o) );
+            d2sum[3] += ( hessian(o,3) = sx*d2sy*sz*weights(o) );
+            d2sum[4] += ( hessian(o,4) = sx*dsy*dsz*weights(o) );
 
-            d2sum[4] += ( hessian(o,4) = sx*sy*d2sz*weights(o) );
-            d2sum[5] += ( hessian(o,5) = sx*d2sy*sz*weights(o) );
+            d2sum[5] += ( hessian(o,5) = sx*sy*d2sz*weights(o) );
          }
       }
    }
@@ -387,19 +387,632 @@ void NURBS3DFiniteElement::CalcHessian (const IntegrationPoint &ip,
                      + u[o]*sum*(2*dsum[0]*dsum[2] - d2sum[2]);
 
       hessian(o,3) = hessian(o,3)*sum
-                     - du(o,1)*sum*dsum[2]
-                     - du(o,2)*sum*dsum[1]
-                     + u[o]*sum*(2*dsum[1]*dsum[2] - d2sum[3]);
+                     - 2*du(o,1)*sum*dsum[1]
+                     + u[o]*sum*(2*dsum[1]*dsum[1] - d2sum[3]);
 
       hessian(o,4) = hessian(o,4)*sum
-                     - 2*du(o,2)*sum*dsum[2]
-                     + u[o]*sum*(2*dsum[2]*dsum[2] - d2sum[4]);
+                     - du(o,1)*sum*dsum[2]
+                     - du(o,2)*sum*dsum[1]
+                     + u[o]*sum*(2*dsum[1]*dsum[2] - d2sum[4]);
 
       hessian(o,5) = hessian(o,5)*sum
-                     - 2*du(o,1)*sum*dsum[1]
-                     + u[o]*sum*(2*dsum[1]*dsum[1] - d2sum[5]);
-
+                     - 2*du(o,2)*sum*dsum[2]
+                     + u[o]*sum*(2*dsum[2]*dsum[2] - d2sum[5]);
    }
 }
 
+
+void NURBS_HDiv2DFiniteElement::SetOrder() const
+{
+   orders[0] = kv[0]->GetOrder();
+   orders[1] = kv[1]->GetOrder();
+
+   if (kv1[0]) { delete kv1[0]; }
+   if (kv1[1]) { delete kv1[1]; }
+
+   kv1[0] = kv[0]->DegreeElevate(1);
+   kv1[1] = kv[1]->DegreeElevate(1);
+
+   shape_x.SetSize(orders[0]+1);
+   shape_y.SetSize(orders[1]+1);
+
+   dshape_x.SetSize(orders[0]+1);
+   dshape_y.SetSize(orders[1]+1);
+
+   d2shape_x.SetSize(orders[0]+1);
+   d2shape_y.SetSize(orders[1]+1);
+
+   shape1_x.SetSize(orders[0]+2);
+   shape1_y.SetSize(orders[1]+2);
+
+   dshape1_x.SetSize(orders[0]+2);
+   dshape1_y.SetSize(orders[1]+2);
+
+   d2shape1_x.SetSize(orders[0]+2);
+   d2shape1_y.SetSize(orders[1]+2);
+
+   order = max(orders[0]+1, orders[1]+1);
+   dof = (orders[0] + 2)*(orders[1] + 1)
+         + (orders[1] + 1)*(orders[1] + 2);
+   u.SetSize(dof);
+   du.SetSize(dof);
+   weights.SetSize(dof);
+}
+
+void NURBS_HDiv2DFiniteElement::CalcVShape(const IntegrationPoint &ip,
+                                           DenseMatrix &shape) const
+{
+   kv[0]->CalcShape(shape_x, ijk[0], ip.x);
+   kv[1]->CalcShape(shape_y, ijk[1], ip.y);
+
+   kv1[0]->CalcShape(shape1_x, ijk[0], ip.x);
+   kv1[1]->CalcShape(shape1_y, ijk[1], ip.y);
+
+   int o = 0;
+   for (int j = 0; j <= orders[1]; j++)
+   {
+      const real_t sy = shape_y(j);
+      for (int i = 0; i <= orders[0]+1; i++, o++)
+      {
+         shape(o,0) = shape1_x(i)*sy;
+         shape(o,1) = 0.0;
+      }
+   }
+
+   for (int j = 0; j <= orders[1]+1; j++)
+   {
+      const real_t sy1 = shape1_y(j);
+      for (int i = 0; i <= orders[0]; i++, o++)
+      {
+         shape(o,0) = 0.0;
+         shape(o,1) = shape_x(i)*sy1;
+      }
+   }
+}
+
+void NURBS_HDiv2DFiniteElement::CalcVShape(ElementTransformation &Trans,
+                                           DenseMatrix &shape) const
+{
+   CalcVShape(Trans.GetIntPoint(), shape);
+   const DenseMatrix & J = Trans.Jacobian();
+   MFEM_ASSERT(J.Width() == 2 && J.Height() == 2,
+               "NURBS_HDiv2DFiniteElement cannot be embedded in "
+               "3 dimensional spaces");
+   for (int i=0; i<dof; i++)
+   {
+      real_t sx = shape(i, 0);
+      real_t sy = shape(i, 1);
+      shape(i, 0) = sx * J(0, 0) + sy * J(0, 1);
+      shape(i, 1) = sx * J(1, 0) + sy * J(1, 1);
+   }
+   shape *= (1.0 / Trans.Weight());
+}
+
+void NURBS_HDiv2DFiniteElement::CalcDivShape(const IntegrationPoint &ip,
+                                             Vector &divshape) const
+{
+   kv[0]->CalcShape ( shape_x, ijk[0], ip.x);
+   kv[1]->CalcShape ( shape_y, ijk[1], ip.y);
+
+   kv1[0]->CalcDShape(dshape1_x, ijk[0], ip.x);
+   kv1[1]->CalcDShape(dshape1_y, ijk[1], ip.y);
+
+   int o = 0;
+   for (int j = 0; j <= orders[1]; j++)
+   {
+      const real_t sy = shape_y(j);
+      for (int i = 0; i <= orders[0]+1; i++, o++)
+      {
+         divshape(o) = dshape1_x(i)*sy;
+      }
+   }
+
+   for (int j = 0; j <= orders[1]+1; j++)
+   {
+      const real_t dsy1 = dshape1_y(j);
+      for (int i = 0; i <= orders[0]; i++, o++)
+      {
+         divshape(o) = shape_x(i)*dsy1;
+      }
+   }
+}
+
+NURBS_HDiv2DFiniteElement::~NURBS_HDiv2DFiniteElement()
+{
+   if (kv1[0]) { delete kv1[0]; }
+   if (kv1[1]) { delete kv1[1]; }
+}
+
+
+void NURBS_HDiv3DFiniteElement::SetOrder() const
+{
+   orders[0] = kv[0]->GetOrder();
+   orders[1] = kv[1]->GetOrder();
+   orders[2] = kv[2]->GetOrder();
+
+   if (kv1[0]) { delete kv1[0]; }
+   if (kv1[1]) { delete kv1[1]; }
+   if (kv1[2]) { delete kv1[2]; }
+
+   kv1[0] = kv[0]->DegreeElevate(1);
+   kv1[1] = kv[1]->DegreeElevate(1);
+   kv1[2] = kv[2]->DegreeElevate(1);
+
+   shape_x.SetSize(orders[0]+1);
+   shape_y.SetSize(orders[1]+1);
+   shape_z.SetSize(orders[2]+1);
+
+   dshape_x.SetSize(orders[0]+1);
+   dshape_y.SetSize(orders[1]+1);
+   dshape_z.SetSize(orders[2]+1);
+
+   d2shape_x.SetSize(orders[0]+1);
+   d2shape_y.SetSize(orders[1]+1);
+   d2shape_z.SetSize(orders[2]+1);
+
+   shape1_x.SetSize(orders[0]+2);
+   shape1_y.SetSize(orders[1]+2);
+   shape1_z.SetSize(orders[2]+2);
+
+   dshape1_x.SetSize(orders[0]+2);
+   dshape1_y.SetSize(orders[1]+2);
+   dshape1_z.SetSize(orders[2]+2);
+
+   d2shape1_x.SetSize(orders[0]+2);
+   d2shape1_y.SetSize(orders[1]+2);
+   d2shape1_z.SetSize(orders[2]+2);
+
+   order = max(orders[0]+1, max( orders[1]+1, orders[2]+1));
+   dof = (orders[0] + 2)*(orders[1] + 1)*(orders[2] + 1) +
+         (orders[0] + 1)*(orders[1] + 2)*(orders[2] + 1) +
+         (orders[0] + 1)*(orders[1] + 1)*(orders[2] + 2);
+   u.SetSize(dof);
+   du.SetSize(dof);
+   weights.SetSize(dof);
+}
+
+void NURBS_HDiv3DFiniteElement::CalcVShape(const IntegrationPoint &ip,
+                                           DenseMatrix &shape) const
+{
+   kv[0]->CalcShape(shape_x, ijk[0], ip.x);
+   kv[1]->CalcShape(shape_y, ijk[1], ip.y);
+   kv[2]->CalcShape(shape_z, ijk[2], ip.z);
+
+   kv1[0]->CalcShape(shape1_x, ijk[0], ip.x);
+   kv1[1]->CalcShape(shape1_y, ijk[1], ip.y);
+   kv1[2]->CalcShape(shape1_z, ijk[2], ip.z);
+
+   shape = 0.0;
+   int o = 0;
+   for (int k = 0; k <= orders[2]; k++)
+   {
+      const real_t sz = shape_z(k);
+      for (int j = 0; j <= orders[1]; j++)
+      {
+         const real_t sy_sz = shape_y(j)*sz;
+         for (int i = 0; i <= orders[0]+1; i++, o++)
+         {
+            shape(o,0) = shape1_x(i)*sy_sz;
+         }
+      }
+   }
+
+   for (int  k = 0; k <= orders[2]; k++)
+   {
+      const real_t sz = shape_z(k);
+      for (int j = 0; j <= orders[1]+1; j++)
+      {
+         const real_t sy1_sz = shape1_y(j)*sz;
+         for (int i = 0; i <= orders[0]; i++, o++)
+         {
+            shape(o,1) = shape_x(i)*sy1_sz;
+         }
+      }
+   }
+
+   for (int  k = 0; k <= orders[2]+1; k++)
+   {
+      const real_t sz1 = shape1_z(k);
+      for (int j = 0; j <= orders[1]; j++)
+      {
+         const real_t sy_sz1 = shape_y(j)*sz1;
+         for (int i = 0; i <= orders[0]; i++, o++)
+         {
+            shape(o,2) = shape_x(i)*sy_sz1;
+         }
+      }
+   }
+}
+
+void NURBS_HDiv3DFiniteElement::CalcVShape(ElementTransformation &Trans,
+                                           DenseMatrix &shape) const
+{
+   CalcVShape(Trans.GetIntPoint(), shape);
+   const DenseMatrix & J = Trans.Jacobian();
+   MFEM_ASSERT(J.Width() == 3 && J.Height() == 3,
+               "RT_R2D_FiniteElement cannot be embedded in "
+               "3 dimensional spaces");
+   for (int i=0; i<dof; i++)
+   {
+      real_t sx = shape(i, 0);
+      real_t sy = shape(i, 1);
+      real_t sz = shape(i, 2);
+      shape(i, 0) = sx * J(0, 0) + sy * J(0, 1) + sz * J(0, 2);
+      shape(i, 1) = sx * J(1, 0) + sy * J(1, 1) + sz * J(1, 2);
+      shape(i, 2) = sx * J(2, 0) + sy * J(2, 1) + sz * J(2, 2);
+   }
+   shape *= (1.0 / Trans.Weight());
+}
+
+void NURBS_HDiv3DFiniteElement::CalcDivShape(const IntegrationPoint &ip,
+                                             Vector &divshape) const
+{
+   kv[0]->CalcShape ( shape_x, ijk[0], ip.x);
+   kv[1]->CalcShape ( shape_y, ijk[1], ip.y);
+   kv[2]->CalcShape ( shape_z, ijk[2], ip.z);
+
+   kv1[0]->CalcDShape(dshape1_x, ijk[0], ip.x);
+   kv1[1]->CalcDShape(dshape1_y, ijk[1], ip.y);
+   kv1[2]->CalcDShape(dshape1_z, ijk[2], ip.z);
+
+   int o = 0;
+   for (int  k = 0; k <= orders[2]; k++)
+   {
+      const real_t sz = shape_z(k);
+      for (int j = 0; j <= orders[1]; j++)
+      {
+         const real_t sy_sz = shape_y(j)*sz;
+         for (int i = 0; i <= orders[0]+1; i++, o++)
+         {
+            divshape(o) = dshape1_x(i)*sy_sz;
+         }
+      }
+   }
+
+   for (int  k = 0; k <= orders[2]; k++)
+   {
+      const real_t sz = shape_z(k);
+      for (int j = 0; j <= orders[1]+1; j++)
+      {
+         const real_t dy1_sz = dshape1_y(j)*sz;
+         for (int i = 0; i <= orders[0]; i++, o++)
+         {
+            divshape(o) = shape_x(i)*dy1_sz;
+         }
+      }
+   }
+
+   for (int  k = 0; k <= orders[2]+1; k++)
+   {
+      const real_t dz1 = dshape1_z(k);
+      for (int j = 0; j <= orders[1]; j++)
+      {
+         const real_t sy_dz1 = shape_y(j)*dz1;
+         for (int i = 0; i <= orders[0]; i++, o++)
+         {
+            divshape(o) = shape_x(i)*sy_dz1;
+         }
+      }
+   }
+}
+
+NURBS_HDiv3DFiniteElement::~NURBS_HDiv3DFiniteElement()
+{
+   if (kv1[0]) { delete kv1[0]; }
+   if (kv1[1]) { delete kv1[1]; }
+   if (kv1[2]) { delete kv1[2]; }
+}
+
+void NURBS_HCurl2DFiniteElement::SetOrder() const
+{
+   orders[0] = kv[0]->GetOrder();
+   orders[1] = kv[1]->GetOrder();
+
+   if (kv1[0]) { delete kv1[0]; }
+   if (kv1[1]) { delete kv1[1]; }
+
+   kv1[0] = kv[0]->DegreeElevate(1);
+   kv1[1] = kv[1]->DegreeElevate(1);
+
+   shape_x.SetSize(orders[0]+1);
+   shape_y.SetSize(orders[1]+1);
+
+   dshape_x.SetSize(orders[0]+1);
+   dshape_y.SetSize(orders[1]+1);
+
+   d2shape_x.SetSize(orders[0]+1);
+   d2shape_y.SetSize(orders[1]+1);
+
+   shape1_x.SetSize(orders[0]+2);
+   shape1_y.SetSize(orders[1]+2);
+
+   dshape1_x.SetSize(orders[0]+2);
+   dshape1_y.SetSize(orders[1]+2);
+
+   d2shape1_x.SetSize(orders[0]+2);
+   d2shape1_y.SetSize(orders[1]+2);
+
+   order = max(orders[0]+1, orders[1]+1);
+   dof = (orders[0] + 1)*(orders[1] + 2)
+         + (orders[1] + 2)*(orders[1] + 1);
+   u.SetSize(dof);
+   du.SetSize(dof);
+   weights.SetSize(dof);
+}
+
+void NURBS_HCurl2DFiniteElement::CalcVShape(const IntegrationPoint &ip,
+                                            DenseMatrix &shape) const
+{
+   kv[0]->CalcShape(shape_x, ijk[0], ip.x);
+   kv[1]->CalcShape(shape_y, ijk[1], ip.y);
+
+   kv1[0]->CalcShape(shape1_x, ijk[0], ip.x);
+   kv1[1]->CalcShape(shape1_y, ijk[1], ip.y);
+
+   int o = 0;
+   for (int j = 0; j <= orders[1]+1; j++)
+   {
+      const real_t sy1 = shape1_y(j);
+      for (int i = 0; i <= orders[0]; i++, o++)
+      {
+         shape(o,0) = shape_x(i)*sy1;
+         shape(o,1) = 0.0;
+      }
+   }
+
+   for (int j = 0; j <= orders[1]; j++)
+   {
+      const real_t sy = shape_y(j);
+      for (int i = 0; i <= orders[0]+1; i++, o++)
+      {
+         shape(o,0) = 0.0;
+         shape(o,1) = shape1_x(i)*sy;
+      }
+   }
+}
+
+void NURBS_HCurl2DFiniteElement::CalcVShape(ElementTransformation &Trans,
+                                            DenseMatrix &shape) const
+{
+   CalcVShape(Trans.GetIntPoint(), shape);
+   const DenseMatrix & JI = Trans.InverseJacobian();
+   MFEM_ASSERT(JI.Width() == 2 && JI.Height() == 2,
+               "NURBS_HCurl2DFiniteElement cannot be embedded in "
+               "3 dimensional spaces");
+   for (int i=0; i<dof; i++)
+   {
+      real_t sx = shape(i, 0);
+      real_t sy = shape(i, 1);
+      shape(i, 0) = sx * JI(0, 0) + sy * JI(1, 0);
+      shape(i, 1) = sx * JI(0, 1) + sy * JI(1, 1);
+   }
+}
+
+void NURBS_HCurl2DFiniteElement::CalcCurlShape(const IntegrationPoint &ip,
+                                               DenseMatrix &curl_shape) const
+{
+   kv[0]->CalcShape ( shape_x, ijk[0], ip.x);
+   kv[1]->CalcShape ( shape_y, ijk[1], ip.y);
+
+   kv1[0]->CalcDShape(dshape1_x, ijk[0], ip.x);
+   kv1[1]->CalcDShape(dshape1_y, ijk[1], ip.y);
+
+   int o = 0;
+   for (int j = 0; j <= orders[1]+1; j++)
+   {
+      const real_t dsy1 = dshape1_y(j);
+      for (int i = 0; i <= orders[0]; i++, o++)
+      {
+         curl_shape(o,0) = -shape_x(i)*dsy1;
+      }
+   }
+
+   for (int j = 0; j <= orders[1]; j++)
+   {
+      const real_t sy = shape_y(j);
+      for (int i = 0; i <= orders[0]+1; i++, o++)
+      {
+         curl_shape(o,0) = dshape1_x(i)*sy;
+      }
+   }
+}
+
+NURBS_HCurl2DFiniteElement::~NURBS_HCurl2DFiniteElement()
+{
+   if (kv1[0]) { delete kv1[0]; }
+   if (kv1[1]) { delete kv1[1]; }
+}
+
+
+void NURBS_HCurl3DFiniteElement::SetOrder() const
+{
+   orders[0] = kv[0]->GetOrder();
+   orders[1] = kv[1]->GetOrder();
+   orders[2] = kv[2]->GetOrder();
+
+   if (kv1[0]) { delete kv1[0]; }
+   if (kv1[1]) { delete kv1[1]; }
+   if (kv1[2]) { delete kv1[2]; }
+
+   kv1[0] = kv[0]->DegreeElevate(1);
+   kv1[1] = kv[1]->DegreeElevate(1);
+   kv1[2] = kv[2]->DegreeElevate(1);
+
+   shape_x.SetSize(orders[0]+1);
+   shape_y.SetSize(orders[1]+1);
+   shape_z.SetSize(orders[2]+1);
+
+   dshape_x.SetSize(orders[0]+1);
+   dshape_y.SetSize(orders[1]+1);
+   dshape_z.SetSize(orders[2]+1);
+
+   d2shape_x.SetSize(orders[0]+1);
+   d2shape_y.SetSize(orders[1]+1);
+   d2shape_z.SetSize(orders[2]+1);
+
+   shape1_x.SetSize(orders[0]+2);
+   shape1_y.SetSize(orders[1]+2);
+   shape1_z.SetSize(orders[2]+2);
+
+   dshape1_x.SetSize(orders[0]+2);
+   dshape1_y.SetSize(orders[1]+2);
+   dshape1_z.SetSize(orders[2]+2);
+
+   d2shape1_x.SetSize(orders[0]+2);
+   d2shape1_y.SetSize(orders[1]+2);
+   d2shape1_z.SetSize(orders[2]+2);
+
+   order = max(orders[0]+1, max( orders[1]+1, orders[2]+1));
+   dof = (orders[0] + 1)*(orders[1] + 2)*(orders[2] + 2) +
+         (orders[0] + 2)*(orders[1] + 1)*(orders[2] + 2) +
+         (orders[0] + 2)*(orders[1] + 2)*(orders[2] + 1);
+   u.SetSize(dof);
+   du.SetSize(dof);
+   weights.SetSize(dof);
+}
+
+void NURBS_HCurl3DFiniteElement::CalcVShape(const IntegrationPoint &ip,
+                                            DenseMatrix &shape) const
+{
+   kv[0]->CalcShape(shape_x, ijk[0], ip.x);
+   kv[1]->CalcShape(shape_y, ijk[1], ip.y);
+   kv[2]->CalcShape(shape_z, ijk[2], ip.z);
+
+   kv1[0]->CalcShape(shape1_x, ijk[0], ip.x);
+   kv1[1]->CalcShape(shape1_y, ijk[1], ip.y);
+   kv1[2]->CalcShape(shape1_z, ijk[2], ip.z);
+
+   shape = 0.0;
+   int o = 0;
+   for (int  k = 0; k <= orders[2]+1; k++)
+   {
+      const real_t sz1 = shape1_z(k);
+      for (int j = 0; j <= orders[1]+1; j++)
+      {
+         const real_t sy1_sz1 = shape1_y(j)*sz1;
+         for (int i = 0; i <= orders[0]; i++, o++)
+         {
+            shape(o,0) = shape_x(i)*sy1_sz1;
+         }
+      }
+   }
+
+   for (int  k = 0; k <= orders[2]+1; k++)
+   {
+      const real_t sz1 = shape1_z(k);
+      for (int j = 0; j <= orders[1]; j++)
+      {
+         const real_t sy_sz1 = shape_y(j)*sz1;
+         for (int i = 0; i <= orders[0]+1; i++, o++)
+         {
+            shape(o,1) = shape1_x(i)*sy_sz1;
+         }
+      }
+   }
+
+   for (int  k = 0; k <= orders[2]; k++)
+   {
+      const real_t sz = shape_z(k);
+      for (int j = 0; j <= orders[1]+1; j++)
+      {
+         const real_t sy1_sz = shape1_y(j)*sz;
+         for (int i = 0; i <= orders[0]+1; i++, o++)
+         {
+            shape(o,2) = shape1_x(i)*sy1_sz;
+         }
+      }
+   }
+}
+
+void NURBS_HCurl3DFiniteElement::CalcVShape(ElementTransformation &Trans,
+                                            DenseMatrix &shape) const
+{
+   CalcVShape(Trans.GetIntPoint(), shape);
+   const DenseMatrix & JI = Trans.InverseJacobian();
+   MFEM_ASSERT(JI.Width() == 3 && JI.Height() == 3,
+               "NURBS_HCurl3DFiniteElement must be in a"
+               "3 dimensional spaces");
+   for (int i=0; i<dof; i++)
+   {
+      real_t sx = shape(i, 0);
+      real_t sy = shape(i, 1);
+      real_t sz = shape(i, 2);
+      shape(i, 0) = sx * JI(0, 0) + sy * JI(1, 0) + sz * JI(2, 0);
+      shape(i, 1) = sx * JI(0, 1) + sy * JI(1, 1) + sz * JI(2, 1);
+      shape(i, 2) = sx * JI(0, 2) + sy * JI(1, 2) + sz * JI(2, 2);
+   }
+}
+
+void NURBS_HCurl3DFiniteElement::CalcCurlShape(const IntegrationPoint &ip,
+                                               DenseMatrix &curl_shape) const
+{
+   kv[0]->CalcShape ( shape_x, ijk[0], ip.x);
+   kv[1]->CalcShape ( shape_y, ijk[1], ip.y);
+   kv[2]->CalcShape ( shape_z, ijk[2], ip.z);
+
+   kv1[0]->CalcShape(shape1_x, ijk[0], ip.x);
+   kv1[1]->CalcShape(shape1_y, ijk[1], ip.y);
+   kv1[2]->CalcShape(shape1_z, ijk[2], ip.z);
+
+   kv1[0]->CalcDShape(dshape1_x, ijk[0], ip.x);
+   kv1[1]->CalcDShape(dshape1_y, ijk[1], ip.y);
+   kv1[2]->CalcDShape(dshape1_z, ijk[2], ip.z);
+
+   int o = 0;
+   for (int  k = 0; k <= orders[2]+1; k++)
+   {
+      const real_t sz1 = shape1_z(k), dsz1 = dshape1_z(k);
+      for (int j = 0; j <= orders[1]+1; j++)
+      {
+         const real_t sy1_dsz1 = shape1_y(j)*dsz1,
+                      dsy1_sz1 = dshape1_y(j)*sz1;
+         for (int i = 0; i <= orders[0]; i++, o++)
+         {
+            curl_shape(o,0) = 0.0;
+            curl_shape(o,1) =  shape_x(i)*sy1_dsz1;
+            curl_shape(o,2) = -shape_x(i)*dsy1_sz1;
+         }
+      }
+   }
+
+   for (int  k = 0; k <= orders[2]+1; k++)
+   {
+      const real_t sz1 = shape1_z(k), dsz1 = dshape1_z(k);
+      for (int j = 0; j <= orders[1]; j++)
+      {
+         const real_t sy_dsz1 = shape_y(j)*dsz1,
+                      sy_sz1 = shape_y(j)*sz1;
+         for (int i = 0; i <= orders[0]+1; i++, o++)
+         {
+            curl_shape(o,0) = -shape1_x(i)*sy_dsz1;
+            curl_shape(o,1) = 0.0;
+            curl_shape(o,2) = dshape1_x(i)*sy_sz1;
+         }
+      }
+   }
+
+   for (int  k = 0; k <= orders[2]; k++)
+   {
+      const real_t sz = shape_z(k);
+      for (int j = 0; j <= orders[1]+1; j++)
+      {
+         const real_t sy1_sz = shape1_y(j)*sz,
+                      dsy1_sz = dshape1_y(j)*sz;
+         for (int i = 0; i <= orders[0]+1; i++, o++)
+         {
+            curl_shape(o,0) = shape1_x(i)*dsy1_sz;
+            curl_shape(o,1) = -dshape1_x(i)*sy1_sz;
+            curl_shape(o,2) = 0.0;
+         }
+      }
+   }
+}
+
+NURBS_HCurl3DFiniteElement::~NURBS_HCurl3DFiniteElement()
+{
+   if (kv1[0]) { delete kv1[0]; }
+   if (kv1[1]) { delete kv1[1]; }
+   if (kv1[2]) { delete kv1[2]; }
+}
+
 }

fem/fe/fe_nurbs.hpp

@@ -20,7 +20,7 @@ namespace mfem
 class KnotVector;
 
 /// An arbitrary order and dimension NURBS element
-class NURBSFiniteElement : public ScalarFiniteElement
+class NURBSFiniteElement
 {
 protected:
    mutable Array <const KnotVector*> kv;
@@ -30,31 +30,34 @@ protected:
 
 public:
    /** @brief Construct NURBSFiniteElement with given
-       @param D    Reference space dimension
-       @param G    Geometry type (of type Geometry::Type)
-       @param Do   Number of degrees of freedom in the FiniteElement
-       @param O    Order/degree of the FiniteElement
-       @param F    FunctionSpace type of the FiniteElement
+       @param dim    Reference space dimension
     */
-   NURBSFiniteElement(int D, Geometry::Type G, int Do, int O, int F)
-      : ScalarFiniteElement(D, G, Do, O, F)
+   NURBSFiniteElement(int dim)
    {
       ijk = NULL;
       patch = elem = -1;
       kv.SetSize(dim);
-      weights.SetSize(dof);
-      weights = 1.0;
    }
 
+   /// Resets the patch and element data stored in the element
    void                 Reset      ()         const { patch = elem = -1; }
+   /// Set which IJK in patch should be evaluated
    void                 SetIJK     (const int *IJK) const { ijk = IJK; }
+   /// Get which patch is currently considered
    int                  GetPatch   ()         const { return patch; }
+   /// Set which patch should be evaluated
    void                 SetPatch   (int p)    const { patch = p; }
+   /// Set which elemenet should be evaluated
    int                  GetElement ()         const { return elem; }
+   /// Get which element is currently considered
    void                 SetElement (int e)    const { elem = e; }
+   /// Get the KnotVectors
    Array <const KnotVector*> &KnotVectors()   const { return kv; }
+   /// Get the Weights
    Vector              &Weights    ()         const { return weights; }
-   /// Update the NURBSFiniteElement according to the currently set knot vectors
+   /// Update the polynomial order according to the currently set knotvectors
+   /// Resizes all internal data members to have the correct size
+   /// related to the polynomial order
    virtual void         SetOrder   ()         const { }
 
    /// Returns the indices (i,j) in 2D or (i,j,k) in 3D of this element in the
@@ -64,7 +67,8 @@ public:
 
 
 /// An arbitrary order 1D NURBS element on a segment
-class NURBS1DFiniteElement : public NURBSFiniteElement
+class NURBS1DFiniteElement : public ScalarFiniteElement,
+   public NURBSFiniteElement
 {
 protected:
    mutable Vector shape_x;
@@ -72,7 +76,8 @@ protected:
 public:
    /// Construct the NURBS1DFiniteElement of order @a p
    NURBS1DFiniteElement(int p)
-      : NURBSFiniteElement(1, Geometry::SEGMENT, p + 1, p, FunctionSpace::Qk),
+      : ScalarFiniteElement(1, Geometry::SEGMENT, p + 1, p, FunctionSpace::Qk),
+        NURBSFiniteElement(1),
         shape_x(p + 1) { }
 
    virtual void SetOrder() const;
@@ -84,7 +89,8 @@ public:
 };
 
 /// An arbitrary order 2D NURBS element on a square
-class NURBS2DFiniteElement : public NURBSFiniteElement
+class NURBS2DFiniteElement : public ScalarFiniteElement,
+   public NURBSFiniteElement
 {
 protected:
    mutable Vector u, shape_x, shape_y, dshape_x, dshape_y, d2shape_x, d2shape_y;
@@ -93,16 +99,18 @@ protected:
 public:
    /// Construct the NURBS2DFiniteElement of order @a p
    NURBS2DFiniteElement(int p)
-      : NURBSFiniteElement(2, Geometry::SQUARE, (p + 1)*(p + 1), p,
-                           FunctionSpace::Qk),
+      : ScalarFiniteElement(2, Geometry::SQUARE, (p + 1)*(p + 1), p,
+                            FunctionSpace::Qk),
+        NURBSFiniteElement(2),
         u(dof), shape_x(p + 1), shape_y(p + 1), dshape_x(p + 1),
         dshape_y(p + 1), d2shape_x(p + 1), d2shape_y(p + 1), du(dof,2)
    { orders[0] = orders[1] = p; }
 
    /// Construct the NURBS2DFiniteElement with x-order @a px and y-order @a py
    NURBS2DFiniteElement(int px, int py)
-      : NURBSFiniteElement(2, Geometry::SQUARE, (px + 1)*(py + 1),
-                           std::max(px, py), FunctionSpace::Qk),
+      : ScalarFiniteElement(2, Geometry::SQUARE, (px + 1)*(py + 1),
+                            std::max(px, py), FunctionSpace::Qk),
+        NURBSFiniteElement(2),
         u(dof), shape_x(px + 1), shape_y(py + 1), dshape_x(px + 1),
         dshape_y(py + 1), d2shape_x(px + 1), d2shape_y(py + 1), du(dof,2)
    { orders[0] = px; orders[1] = py; }
@@ -116,7 +124,8 @@ public:
 };
 
 /// An arbitrary order 3D NURBS element on a cube
-class NURBS3DFiniteElement : public NURBSFiniteElement
+class NURBS3DFiniteElement : public ScalarFiniteElement,
+   public NURBSFiniteElement
 {
 protected:
    mutable Vector u, shape_x, shape_y, shape_z;
@@ -127,8 +136,9 @@ protected:
 public:
    /// Construct the NURBS3DFiniteElement of order @a p
    NURBS3DFiniteElement(int p)
-      : NURBSFiniteElement(3, Geometry::CUBE, (p + 1)*(p + 1)*(p + 1), p,
-                           FunctionSpace::Qk),
+      : ScalarFiniteElement(3, Geometry::CUBE, (p + 1)*(p + 1)*(p + 1), p,
+                            FunctionSpace::Qk),
+        NURBSFiniteElement(3),
         u(dof), shape_x(p + 1), shape_y(p + 1), shape_z(p + 1),
         dshape_x(p + 1), dshape_y(p + 1), dshape_z(p + 1),
         d2shape_x(p + 1), d2shape_y(p + 1), d2shape_z(p + 1), du(dof,3)
@@ -137,8 +147,9 @@ public:
    /// Construct the NURBS3DFiniteElement with x-order @a px and y-order @a py
    /// and z-order @a pz
    NURBS3DFiniteElement(int px, int py, int pz)
-      : NURBSFiniteElement(3, Geometry::CUBE, (px + 1)*(py + 1)*(pz + 1),
-                           std::max(std::max(px,py),pz), FunctionSpace::Qk),
+      : ScalarFiniteElement(3, Geometry::CUBE, (px + 1)*(py + 1)*(pz + 1),
+                            std::max(std::max(px,py),pz), FunctionSpace::Qk),
+        NURBSFiniteElement(2),
         u(dof), shape_x(px + 1), shape_y(py + 1), shape_z(pz + 1),
         dshape_x(px + 1), dshape_y(py + 1), dshape_z(pz + 1),
         d2shape_x(px + 1), d2shape_y(py + 1), d2shape_z(pz + 1), du(dof,3)
@@ -152,6 +163,352 @@ public:
                              DenseMatrix &hessian) const;
 };
 
+
+/** An arbitrary order H(div)-conforming 2D NURBS element on a square.
+    More details in the following papers:
+
+[1] Annalisa Buffa, Carlo De Falco, Giancarlo Sangalli
+"Isogeometric analysis: stable elements for the 2D Stokes equation."
+International Journal for Numerical Methods in Fluids 65 (11‐12) 1407-1422
+
+[2] John A Evans, Thomas JR Hughes
+"Isogeometric divergence-conforming B-splines for the unsteady Navier–Stokes equations."
+Journal of Computational Physics (241) 141-167
+*/
+class NURBS_HDiv2DFiniteElement : public VectorFiniteElement,
+   public NURBSFiniteElement
+{
+protected:
+   mutable Vector shape_x, shape_y, dshape_x, dshape_y, d2shape_x, d2shape_y;
+   mutable Vector shape1_x, shape1_y, dshape1_x, dshape1_y, d2shape1_x, d2shape1_y;
+   mutable Vector u;
+   mutable DenseMatrix du;
+   mutable Array <const KnotVector*> kv1;
+
+public:
+   /// Construct the NURBS_HDiv2DFiniteElement of order @a p
+   NURBS_HDiv2DFiniteElement(int p)
+      : VectorFiniteElement(2, Geometry::SQUARE, 2*(p + 1)*(p + 2), p,
+                            H_DIV,FunctionSpace::Qk),
+        NURBSFiniteElement(2),
+        shape_x(p + 1), shape_y(p + 1), dshape_x(p + 1),
+        dshape_y(p + 1), d2shape_x(p + 1), d2shape_y(p + 1),
+        shape1_x(p + 2), shape1_y(p + 2), dshape1_x(p + 2),
+        dshape1_y(p + 2), d2shape1_x(p + 2), d2shape1_y(p + 2),
+        u(dof), du(dof,2)
+   {
+      orders[0] = orders[1] = p;
+      kv1.SetSize(dim);
+      kv1[0] = nullptr;
+      kv1[1] = nullptr;
+   }
+
+   /// Construct the NURBS_HDiv2DFiniteElement with x-order @a px and y-order @a py
+   NURBS_HDiv2DFiniteElement(int px, int py)
+      : VectorFiniteElement(2, Geometry::SQUARE,
+                            (px + 2)*(py + 1)+(px + 1)*(py + 2),
+                            std::max(px, py), H_DIV, FunctionSpace::Qk),
+        NURBSFiniteElement(2),
+        shape_x(px + 1), shape_y(py + 1), dshape_x(px + 1),
+        dshape_y(py + 1), d2shape_x(px + 1), d2shape_y(py + 1),
+        shape1_x(px + 2), shape1_y(py + 2), dshape1_x(px + 2),
+        dshape1_y(py + 2), d2shape1_x(px + 2), d2shape1_y(py + 2),
+        u(dof), du(dof,2)
+   {
+      orders[0] = px; orders[1] = py;
+      kv1.SetSize(dim);
+      kv1[0] = nullptr;
+      kv1[1] = nullptr;
+   }
+
+   virtual void SetOrder() const;
+
+   virtual void CalcVShape(const IntegrationPoint &ip,
+                           DenseMatrix &shape) const;
+
+   /** @brief Evaluate the values of all shape functions of a *vector* finite
+       element in physical space at the point described by @a Trans. */
+   /** Each row of the result DenseMatrix @a shape contains the components of
+       one vector shape function. The size (#dof x SDim) of @a shape must be set
+       in advance, where SDim >= #dim is the physical space dimension as
+       described by @a Trans. */
+   virtual void CalcVShape(ElementTransformation &Trans,
+                           DenseMatrix &shape) const;
+
+   /** @brief Evaluate the divergence of all shape functions of a *vector*
+       finite element in reference space at the given point @a ip. */
+   /** The size (#dof) of the result Vector @a divshape must be set in advance.
+    */
+   virtual void CalcDivShape(const IntegrationPoint &ip,
+                             Vector &divshape) const;
+
+   ~NURBS_HDiv2DFiniteElement();
+};
+
+
+/** An arbitrary order H(div)-conforming 3D NURBS element on a cube
+    More details in the following papers:
+
+    [1] Annalisa Buffa, Carlo De Falco, Giancarlo Sangalli
+    "Isogeometric analysis: stable elements for the 2D Stokes equation."
+    International Journal for Numerical Methods in Fluids 65 (11‐12) 1407-1422
+
+    [2] John A Evans, Thomas JR Hughes
+    "Isogeometric divergence-conforming B-splines for the unsteady
+    Navier–Stokes equations."
+    Journal of Computational Physics (241) 141-167 */
+class NURBS_HDiv3DFiniteElement : public VectorFiniteElement,
+   public NURBSFiniteElement
+{
+protected:
+   mutable Vector shape_x, shape_y, shape_z;
+   mutable Vector dshape_x, dshape_y, dshape_z;
+   mutable Vector d2shape_x, d2shape_y, d2shape_z;
+   mutable Vector shape1_x, shape1_y, shape1_z;
+   mutable Vector dshape1_x, dshape1_y, dshape1_z;
+   mutable Vector d2shape1_x, d2shape1_y, d2shape1_z;
+   mutable Vector u;
+   mutable DenseMatrix du;
+   mutable Array <const KnotVector*> kv1;
+
+public:
+   /// Construct the NURBS_HDiv3DFiniteElement of order @a p
+   NURBS_HDiv3DFiniteElement(int p)
+      : VectorFiniteElement(3, Geometry::CUBE, 3*(p + 1)*(p + 1)*(p + 2),
+                            p, H_DIV,FunctionSpace::Qk),
+        NURBSFiniteElement(3),
+        shape_x(p + 1), shape_y(p + 1), shape_z(p + 1),
+        dshape_x(p + 1), dshape_y(p + 1), dshape_z(p + 1),
+        d2shape_x(p + 1), d2shape_y(p + 1), d2shape_z(p + 1),
+        shape1_x(p + 2), shape1_y(p + 2), shape1_z(p + 2),
+        dshape1_x(p + 2), dshape1_y(p + 2),dshape1_z(p + 2),
+        d2shape1_x(p + 2), d2shape1_y(p + 2), d2shape1_z(p + 2),
+        u(dof), du(dof,3)
+   {
+      orders[0] = orders[1] = orders[2] = p;
+      kv1.SetSize(dim);
+      kv1[0] = nullptr;
+      kv1[1] = nullptr;
+      kv1[2] = nullptr;
+   }
+
+   /// Construct the NURBS_HDiv3DFiniteElement with x-order @a px, y-order @a py and z-order @a pz
+   NURBS_HDiv3DFiniteElement(int px, int py, int pz)
+      : VectorFiniteElement(3, Geometry::CUBE,
+                            (px + 2)*(py + 1)*(pz + 1) +
+                            (px + 1)*(py + 2)*(pz + 1) +
+                            (px + 1)*(py + 1)*(pz + 2),
+                            std::max(px, py), H_DIV, FunctionSpace::Qk),
+        NURBSFiniteElement(3),
+        shape_x(px + 1), shape_y(py + 1), shape_z(pz + 1),
+        dshape_x(px + 1), dshape_y(py + 1), dshape_z(pz + 1),
+        d2shape_x(px + 1), d2shape_y(py + 1), d2shape_z(pz + 1),
+        shape1_x(px + 2), shape1_y(py + 2), shape1_z(pz + 2),
+        dshape1_x(px + 2), dshape1_y(py + 2),dshape1_z(pz + 2),
+        d2shape1_x(px + 2), d2shape1_y(py + 2), d2shape1_z(pz + 2),
+        u(dof), du(dof,3)
+   {
+      orders[0] = px; orders[1] = py; orders[2] = pz;
+      kv1.SetSize(dim);
+      kv1[0] = nullptr;
+      kv1[1] = nullptr;
+      kv1[2] = nullptr;
+   }
+
+   virtual void SetOrder() const;
+
+   virtual void CalcVShape(const IntegrationPoint &ip,
+                           DenseMatrix &shape) const;
+
+   /** @brief Evaluate the values of all shape functions of a *vector* finite
+       element in physical space at the point described by @a Trans. */
+   /** Each row of the result DenseMatrix @a shape contains the components of
+       one vector shape function. The size (#dof x SDim) of @a shape must be set
+       in advance, where SDim >= #dim is the physical space dimension as
+       described by @a Trans. */
+   virtual void CalcVShape(ElementTransformation &Trans,
+                           DenseMatrix &shape) const;
+
+   /** @brief Evaluate the divergence of all shape functions of a *vector*
+       finite element in reference space at the given point @a ip. */
+   /** The size (#dof) of the result Vector @a divshape must be set in advance.
+    */
+   virtual void CalcDivShape(const IntegrationPoint &ip,
+                             Vector &divshape) const;
+
+   ~NURBS_HDiv3DFiniteElement();
+};
+
+
+/** An arbitrary order H(curl)-conforming 2D NURBS element on a square
+    More details in the following paper:
+
+    [1] Annalisa Buffa, Giancarlo Sangalli, Rafael Vázquez
+    "Isogeometric analysis in electromagnetics: B-splines approximation."
+    Computer Methods in Applied Mechanics and Engineering (199) 1143-1152 */
+class NURBS_HCurl2DFiniteElement : public VectorFiniteElement,
+   public NURBSFiniteElement
+{
+protected:
+   mutable Vector shape_x, shape_y, dshape_x, dshape_y, d2shape_x, d2shape_y;
+   mutable Vector shape1_x, shape1_y, dshape1_x, dshape1_y, d2shape1_x, d2shape1_y;
+   mutable Vector u;
+   mutable DenseMatrix du;
+   mutable Array <const KnotVector*> kv1;
+
+public:
+   /// Construct the NURBS_HCurl2DFiniteElement of order @a p
+   NURBS_HCurl2DFiniteElement(int p)
+      : VectorFiniteElement(2, Geometry::SQUARE, 2*(p + 1)*(p + 2), p,
+                            H_CURL,FunctionSpace::Qk),
+        NURBSFiniteElement(2),
+        shape_x(p + 1), shape_y(p + 1), dshape_x(p + 1),
+        dshape_y(p + 1), d2shape_x(p + 1), d2shape_y(p + 1),
+        shape1_x(p + 2), shape1_y(p + 2), dshape1_x(p + 2),
+        dshape1_y(p + 2), d2shape1_x(p + 2), d2shape1_y(p + 2),
+        u(dof), du(dof,2)
+   {
+      orders[0] = orders[1] = p;
+      kv1.SetSize(dim);
+      kv1[0] = nullptr;
+      kv1[1] = nullptr;
+   }
+
+   /// Construct the NURBS_HCurl2DFiniteElement with x-order @a px and y-order @a py
+   NURBS_HCurl2DFiniteElement(int px, int py)
+      : VectorFiniteElement(2, Geometry::SQUARE,
+                            (px + 1)*(py + 2)+(px + 2)*(py + 1),
+                            std::max(px, py), H_CURL, FunctionSpace::Qk),
+        NURBSFiniteElement(2),
+        shape_x(px + 1), shape_y(py + 1), dshape_x(px + 1),
+        dshape_y(py + 1), d2shape_x(px + 1), d2shape_y(py + 1),
+        shape1_x(px + 2), shape1_y(py + 2), dshape1_x(px + 2),
+        dshape1_y(py + 2), d2shape1_x(px + 2), d2shape1_y(py + 2),
+        u(dof), du(dof,2)
+   {
+      orders[0] = px; orders[1] = py;
+      kv1.SetSize(dim);
+      kv1[0] = nullptr;
+      kv1[1] = nullptr;
+   }
+
+   virtual void SetOrder() const;
+
+   virtual void CalcVShape(const IntegrationPoint &ip,
+                           DenseMatrix &shape) const;
+
+   /** @brief Evaluate the values of all shape functions of a *vector* finite
+       element in physical space at the point described by @a Trans. */
+   /** Each row of the result DenseMatrix @a shape contains the components of
+       one vector shape function. The size (#dof x SDim) of @a shape must be set
+       in advance, where SDim >= #dim is the physical space dimension as
+       described by @a Trans. */
+   virtual void CalcVShape(ElementTransformation &Trans,
+                           DenseMatrix &shape) const;
+
+   /** @brief Evaluate the curl of all shape functions of a *vector* finite
+       element in reference space at the given point @a ip. */
+   /** Each row of the result DenseMatrix @a curl_shape contains the components
+       of the curl of one vector shape function. The size (#dof x CDim) of
+       @a curl_shape must be set in advance, where CDim = 3 for #dim = 3 and
+       CDim = 1 for #dim = 2. */
+   virtual void CalcCurlShape(const IntegrationPoint &ip,
+                              DenseMatrix &curl_shape) const;
+
+   ~NURBS_HCurl2DFiniteElement();
+};
+
+
+/** An arbitrary order H(curl)-conforming 3D NURBS element on a cube
+    More details in the following paper:
+
+    [1] Annalisa Buffa, Giancarlo Sangalli, Rafael Vázquez
+    "Isogeometric analysis in electromagnetics: B-splines approximation."
+    Computer Methods in Applied Mechanics and Engineering (199) 1143-1152 */
+class NURBS_HCurl3DFiniteElement : public VectorFiniteElement,
+   public NURBSFiniteElement
+{
+protected:
+   mutable Vector shape_x, shape_y, shape_z;
+   mutable Vector dshape_x, dshape_y, dshape_z;
+   mutable Vector d2shape_x, d2shape_y, d2shape_z;
+   mutable Vector shape1_x, shape1_y, shape1_z;
+   mutable Vector dshape1_x, dshape1_y, dshape1_z;
+   mutable Vector d2shape1_x, d2shape1_y, d2shape1_z;
+   mutable Vector u;
+   mutable DenseMatrix du;
+   mutable Array <const KnotVector*> kv1;
+
+public:
+   /// Construct the NURBS_HCurl3DFiniteElement of order @a p
+   NURBS_HCurl3DFiniteElement(int p)
+      : VectorFiniteElement(3, Geometry::CUBE, 3*(p + 1)*(p + 2)*(p + 2), p,
+                            H_CURL,FunctionSpace::Qk),
+        NURBSFiniteElement(3),
+        shape_x(p + 1), shape_y(p + 1), shape_z(p + 1),
+        dshape_x(p + 1), dshape_y(p + 1), dshape_z(p + 1),
+        d2shape_x(p + 1), d2shape_y(p + 1), d2shape_z(p + 1),
+        shape1_x(p + 2), shape1_y(p + 2), shape1_z(p + 2),
+        dshape1_x(p + 2), dshape1_y(p + 2),dshape1_z(p + 2),
+        d2shape1_x(p + 2), d2shape1_y(p + 2), d2shape1_z(p + 2),
+        u(dof), du(dof,3)
+   {
+      orders[0] = orders[1] = orders[2] = p;
+      kv1.SetSize(dim);
+      kv1[0] = nullptr;
+      kv1[1] = nullptr;
+      kv1[2] = nullptr;
+   }
+
+   /// Construct the NURBS_HCurl3DFiniteElement with x-order @a px, y-order @a py and z-order @a pz
+   NURBS_HCurl3DFiniteElement(int px, int py, int pz)
+      : VectorFiniteElement(3, Geometry::CUBE,
+                            (px + 1)*(py + 2)*(pz + 2) +
+                            (px + 2)*(py + 1)*(pz + 2) +
+                            (px + 2)*(py + 2)*(pz + 1),
+                            std::max(std::max(px, py), pz), H_CURL, FunctionSpace::Qk),
+        NURBSFiniteElement(3),
+        shape_x(px + 1), shape_y(py + 1), shape_z(pz + 1),
+        dshape_x(px + 1), dshape_y(py + 1), dshape_z(pz + 1),
+        d2shape_x(px + 1), d2shape_y(py + 1), d2shape_z(pz + 1),
+        shape1_x(px + 2), shape1_y(py + 2), shape1_z(pz + 2),
+        dshape1_x(px + 2), dshape1_y(py + 2),dshape1_z(pz + 2),
+        d2shape1_x(px + 2), d2shape1_y(py + 2), d2shape1_z(pz + 2),
+        u(dof), du(dof,3)
+   {
+      orders[0] = px; orders[1] = py; orders[2] = pz;
+      kv1.SetSize(dim);
+      kv1[0] = nullptr;
+      kv1[1] = nullptr;
+      kv1[2] = nullptr;
+   }
+
+   virtual void SetOrder() const;
+
+   virtual void CalcVShape(const IntegrationPoint &ip,
+                           DenseMatrix &shape) const;
+
+   /** @brief Evaluate the values of all shape functions of a *vector* finite
+       element in physical space at the point described by @a Trans. */
+   /** Each row of the result DenseMatrix @a shape contains the components of
+       one vector shape function. The size (#dof x SDim) of @a shape must be set
+       in advance, where SDim >= #dim is the physical space dimension as
+       described by @a Trans. */
+   virtual void CalcVShape(ElementTransformation &Trans,
+                           DenseMatrix &shape) const;
+
+   /** @brief Evaluate the curl of all shape functions of a *vector* finite
+       element in reference space at the given point @a ip. */
+   /** Each row of the result DenseMatrix @a curl_shape contains the components
+       of the curl of one vector shape function. The size (#dof x CDim) of
+       @a curl_shape must be set in advance, where CDim = 3 for #dim = 3 and
+       CDim = 1 for #dim = 2. */
+   virtual void CalcCurlShape(const IntegrationPoint &ip,
+                              DenseMatrix &curl_shape) const;
+
+   ~NURBS_HCurl3DFiniteElement();
+};
+
 } // namespace mfem
 
 #endif

fem/fe_coll.cpp

@@ -344,6 +344,32 @@ FiniteElementCollection *FiniteElementCollection::New(const char *name)
    {
       fec = new Local_FECollection(name + 6);
    }
+   else if (!strncmp(name, "NURBS_HDiv", 10))
+   {
+      if (name[10] != '\0')
+      {
+         // "NURBS" + "number" --> fixed order nurbs collection
+         fec = new NURBS_HDivFECollection(atoi(name + 10));
+      }
+      else
+      {
+         // "NURBS" --> variable order nurbs collection
+         fec = new NURBS_HDivFECollection();
+      }
+   }
+   else if (!strncmp(name, "NURBS_HCurl", 11))
+   {
+      if (name[11] != '\0')
+      {
+         // "NURBS" + "number" --> fixed order nurbs collection
+         fec = new NURBS_HCurlFECollection(atoi(name + 11));
+      }
+      else
+      {
+         // "NURBS" --> variable order nurbs collection
+         fec = new NURBS_HCurlFECollection();
+      }
+   }
    else if (!strncmp(name, "NURBS", 5))
    {
       if (name[5] != '\0')
@@ -3533,4 +3559,192 @@ FiniteElementCollection *NURBSFECollection::GetTraceCollection() const
    return NULL;
 }
 
+
+NURBS_HDivFECollection::NURBS_HDivFECollection(int Order, const int dim)
+   : NURBSFECollection((Order == VariableOrder) ? 1 : Order)
+{
+   const int order = (Order == VariableOrder) ? 1 : Order;
+
+   SegmentFE       = new NURBS1DFiniteElement(order);
+   QuadrilateralFE = new NURBS2DFiniteElement(order);
+
+   QuadrilateralVFE  = new NURBS_HDiv2DFiniteElement(order);
+   ParallelepipedVFE = new NURBS_HDiv3DFiniteElement(order);
+
+   if (dim != -1) { SetDim(dim); }
+   SetOrder(Order);
+}
+
+void NURBS_HDivFECollection::SetDim(int dim)
+{
+   if (dim == 2)
+   {
+      sFE = SegmentFE;
+      qFE = QuadrilateralVFE;
+      hFE = nullptr;
+   }
+   else if (dim == 3)
+   {
+      sFE = nullptr;
+      qFE = QuadrilateralFE;
+      hFE = ParallelepipedVFE;
+   }
+   else
+   {
+      mfem::err<<"Dimension = "<<dim<<endl;
+      mfem_error ("NURBS_HDivFECollection: wrong dimension!");
+   }
+}
+
+NURBS_HDivFECollection::~NURBS_HDivFECollection()
+{
+   delete SegmentFE;
+   delete QuadrilateralFE;
+   delete QuadrilateralVFE;
+   delete ParallelepipedVFE;
+}
+
+const FiniteElement *
+NURBS_HDivFECollection::FiniteElementForGeometry(Geometry::Type GeomType) const
+{
+   switch (GeomType)
+   {
+      case Geometry::SEGMENT:     return sFE;
+      case Geometry::SQUARE:      return qFE;
+      case Geometry::CUBE:        return hFE;
+      default:
+         if (error_mode == RETURN_NULL) { return nullptr; }
+         mfem_error ("NURBS_HDivFECollection: unknown geometry type.");
+   }
+   return QuadrilateralFE; // Make some compilers happy
+}
+
+void NURBS_HDivFECollection::SetOrder(int Order) const
+{
+   mOrder = Order;
+   if (Order != VariableOrder)
+   {
+      snprintf(name, 16, "NURBS_HDiv%i", Order);
+   }
+   else
+   {
+      snprintf(name, 16, "NURBS_HDiv");
+   }
+}
+
+int NURBS_HDivFECollection::DofForGeometry(Geometry::Type GeomType) const
+{
+   mfem_error("NURBS_HDivFECollection::DofForGeometry");
+   return 0; // Make some compilers happy
+}
+
+const int *NURBS_HDivFECollection::DofOrderForOrientation(
+   Geometry::Type GeomType,
+   int Or) const
+{
+   mfem_error("NURBS_HDivFECollection::DofOrderForOrientation");
+   return NULL;
+}
+
+FiniteElementCollection *NURBS_HDivFECollection::GetTraceCollection() const
+{
+   MFEM_ABORT("NURBS finite elements can not be statically condensed!");
+   return NULL;
+}
+
+NURBS_HCurlFECollection::NURBS_HCurlFECollection(int Order, const int dim)
+   : NURBSFECollection((Order == VariableOrder) ? 1 : Order)
+{
+   const int order = (Order == VariableOrder) ? 1 : Order;
+
+   SegmentFE       = new NURBS1DFiniteElement(order+1);
+   QuadrilateralFE = new NURBS2DFiniteElement(order+1);
+
+   QuadrilateralVFE  = new NURBS_HCurl2DFiniteElement(order);
+   ParallelepipedVFE = new NURBS_HCurl3DFiniteElement(order);
+   if (dim != -1) { SetDim(dim); }
+   SetOrder(Order);
+}
+
+void NURBS_HCurlFECollection::SetDim(int dim)
+{
+   if (dim == 2)
+   {
+      sFE = SegmentFE;
+      qFE = QuadrilateralVFE;
+      hFE = nullptr;
+   }
+   else if (dim == 3)
+   {
+      sFE = nullptr;
+      qFE = QuadrilateralFE;
+      hFE = ParallelepipedVFE;
+   }
+   else
+   {
+      mfem::err<<"Dimension = "<<dim<<endl;
+      mfem_error ("NURBS_HCurlFECollection: wrong dimension!");
+   }
+}
+
+
+
+NURBS_HCurlFECollection::~NURBS_HCurlFECollection()
+{
+   delete SegmentFE;
+   delete QuadrilateralFE;
+   delete QuadrilateralVFE;
+   delete ParallelepipedVFE;
+}
+
+const FiniteElement *
+NURBS_HCurlFECollection::FiniteElementForGeometry(Geometry::Type GeomType) const
+{
+   switch (GeomType)
+   {
+      case Geometry::SEGMENT:     return sFE;
+      case Geometry::SQUARE:      return qFE;
+      case Geometry::CUBE:        return hFE;
+      default:
+         if (error_mode == RETURN_NULL) { return nullptr; }
+         mfem_error ("NURBS_HCurlFECollection: unknown geometry type.");
+   }
+   return QuadrilateralFE; // Make some compilers happy
+}
+
+void NURBS_HCurlFECollection::SetOrder(int Order) const
+{
+   mOrder = Order;
+   if (Order != VariableOrder)
+   {
+      snprintf(name, 16, "NURBS_HCurl%i", Order);
+   }
+   else
+   {
+      snprintf(name, 16, "NURBS_HCurl");
+   }
+}
+
+int NURBS_HCurlFECollection::DofForGeometry(Geometry::Type GeomType) const
+{
+   mfem_error("NURBS_HCurlFECollection::DofForGeometry");
+   return 0; // Make some compilers happy
+}
+
+const int *NURBS_HCurlFECollection::DofOrderForOrientation(
+   Geometry::Type GeomType,
+   int Or) const
+{
+   mfem_error("NURBS_HCurlFECollection::DofOrderForOrientation");
+   return NULL;
+}
+
+FiniteElementCollection *NURBS_HCurlFECollection::GetTraceCollection() const
+{
+   MFEM_ABORT("NURBS finite elements can not be statically condensed!");
+   return NULL;
+}
+
+
+
 }

fem/fe_coll.hpp

@@ -680,8 +680,8 @@ public:
 /// Arbitrary order non-uniform rational B-splines (NURBS) finite elements.
 class NURBSFECollection : public FiniteElementCollection
 {
-private:
-   PointFiniteElement   *PointFE;
+protected:
+   PointFiniteElement *PointFE;
    NURBS1DFiniteElement *SegmentFE;
    NURBS2DFiniteElement *QuadrilateralFE;
    NURBS3DFiniteElement *ParallelepipedFE;
@@ -701,13 +701,15 @@ public:
       order, or VariableOrder (default). */
    explicit NURBSFECollection(int Order = VariableOrder);
 
-   void Reset() const
+   virtual void Reset() const
    {
       SegmentFE->Reset();
       QuadrilateralFE->Reset();
       ParallelepipedFE->Reset();
    }
 
+   virtual void SetDim(const int dim) {};
+
    /** @brief Get the order of the NURBS collection: either a positive number,
        when using fixed order, or VariableOrder. */
    /** @note Not to be confused with FiniteElementCollection::GetOrder(). */
@@ -715,7 +717,7 @@ public:
 
    /** @brief Set the order and the name, based on the given @a Order: either a
        positive number for fixed order, or VariableOrder. */
-   void SetOrder(int Order) const;
+   virtual void SetOrder(int Order) const;
 
    const FiniteElement *
    FiniteElementForGeometry(Geometry::Type GeomType) const override;
@@ -734,6 +736,109 @@ public:
    virtual ~NURBSFECollection();
 };
 
+/// Arbitrary order H(div) NURBS finite elements.
+class NURBS_HDivFECollection : public NURBSFECollection
+{
+private:
+
+   NURBS1DFiniteElement *SegmentFE;
+   NURBS2DFiniteElement *QuadrilateralFE;
+
+   NURBS_HDiv2DFiniteElement *QuadrilateralVFE;
+   NURBS_HDiv3DFiniteElement *ParallelepipedVFE;
+
+   FiniteElement *sFE;
+   FiniteElement *qFE;
+   FiniteElement *hFE;
+
+public:
+
+   /** @brief The parameter @a Order must be either a positive number, for fixed
+      order, or VariableOrder (default). */
+   explicit NURBS_HDivFECollection(int Order = VariableOrder, const int vdim = -1);
+
+   virtual void Reset() const override
+   {
+      SegmentFE->Reset();
+      QuadrilateralFE->Reset();
+      QuadrilateralVFE->Reset();
+      ParallelepipedVFE->Reset();
+   }
+
+   virtual void SetDim(const int dim) override;
+
+   /** @brief Set the order and the name, based on the given @a Order: either a
+       positive number for fixed order, or VariableOrder. */
+   virtual void SetOrder(int Order) const override;
+
+   const FiniteElement *
+   FiniteElementForGeometry(Geometry::Type GeomType) const override;
+
+   int DofForGeometry(Geometry::Type GeomType) const override;
+
+   const int *DofOrderForOrientation(Geometry::Type GeomType,
+                                     int Or) const override;
+
+   const char *Name() const override { return name; }
+
+   int GetContType() const override { return CONTINUOUS; }
+
+   FiniteElementCollection *GetTraceCollection() const override;
+
+   virtual ~NURBS_HDivFECollection();
+};
+
+/// Arbitrary order H(curl) NURBS finite elements.
+class NURBS_HCurlFECollection : public NURBSFECollection
+{
+private:
+   NURBS1DFiniteElement *SegmentFE;
+   NURBS2DFiniteElement *QuadrilateralFE;
+
+   NURBS_HCurl2DFiniteElement *QuadrilateralVFE;
+   NURBS_HCurl3DFiniteElement *ParallelepipedVFE;
+
+   FiniteElement *sFE;
+   FiniteElement *qFE;
+   FiniteElement *hFE;
+public:
+
+   /** @brief The parameter @a Order must be either a positive number, for fixed
+      order, or VariableOrder (default). */
+   explicit NURBS_HCurlFECollection(int Order = VariableOrder,
+                                    const int vdim = -1);
+
+   virtual void Reset() const override
+   {
+      SegmentFE->Reset();
+      QuadrilateralFE->Reset();
+      QuadrilateralVFE->Reset();
+      ParallelepipedVFE->Reset();
+   }
+
+   virtual void SetDim(const int dim) override;
+
+   /** @brief Set the order and the name, based on the given @a Order: either a
+       positive number for fixed order, or VariableOrder. */
+   virtual void SetOrder(int Order) const override;
+
+   const FiniteElement *
+   FiniteElementForGeometry(Geometry::Type GeomType) const override;
+
+   int DofForGeometry(Geometry::Type GeomType) const override;
+
+   const int *DofOrderForOrientation(Geometry::Type GeomType,
+                                     int Or) const override;
+
+   const char *Name() const override { return name; }
+
+   int GetContType() const override { return CONTINUOUS; }
+
+   FiniteElementCollection *GetTraceCollection() const override;
+
+   virtual ~NURBS_HCurlFECollection();
+};
+
 /// Piecewise-(bi/tri)linear continuous finite elements.
 class LinearFECollection : public FiniteElementCollection
 {

fem/fespace.cpp

@@ -1525,6 +1525,67 @@ SparseMatrix *FiniteElementSpace::RefinementMatrix_main(
    return P;
 }
 
+SparseMatrix *FiniteElementSpace::VariableOrderRefinementMatrix(
+   const int coarse_ndofs, const Table &coarse_elem_dof) const
+{
+   MFEM_VERIFY(mesh->GetLastOperation() == Mesh::REFINE, "");
+
+   Array<int> dofs, coarse_dofs, coarse_vdofs;
+   Vector row;
+
+   Mesh::GeometryList elem_geoms(*mesh);
+
+   SparseMatrix *P = new SparseMatrix(GetVSize(), coarse_ndofs*vdim);
+
+   Array<int> mark(P->Height());
+   mark = 0;
+
+   const CoarseFineTransformations &rtrans = mesh->GetRefinementTransforms();
+   DenseMatrix lP;
+   IsoparametricTransformation isotr;
+   for (int k = 0; k < mesh->GetNE(); k++)
+   {
+      const Embedding &emb = rtrans.embeddings[k];
+      const Geometry::Type geom = mesh->GetElementBaseGeometry(k);
+
+      const FiniteElement *fe = GetFE(k);
+      isotr.SetIdentityTransformation(geom);
+      const int ldof = fe->GetDof();
+      lP.SetSize(ldof, ldof);
+      const DenseTensor &pmats = rtrans.point_matrices[geom];
+      isotr.SetPointMat(pmats(emb.matrix));
+      fe->GetLocalInterpolation(isotr, lP);
+
+      const int fine_ldof = lP.Height();
+
+      elem_dof->GetRow(k, dofs);
+      coarse_elem_dof.GetRow(emb.parent, coarse_dofs);
+
+      for (int vd = 0; vd < vdim; vd++)
+      {
+         coarse_dofs.Copy(coarse_vdofs);
+         DofsToVDofs(vd, coarse_vdofs, coarse_ndofs);
+
+         for (int i = 0; i < fine_ldof; i++)
+         {
+            const int r = DofToVDof(dofs[i], vd);
+            int m = (r >= 0) ? r : (-1 - r);
+
+            if (!mark[m])
+            {
+               lP.GetRow(i, row);
+               P->SetRow(r, coarse_vdofs, row);
+               mark[m] = 1;
+            }
+         }
+      }
+   }
+
+   MFEM_VERIFY(mark.Sum() == P->Height(), "Not all rows of P set.");
+   P->Finalize();
+   return P;
+}
+
 void FiniteElementSpace::GetLocalRefinementMatrices(
    Geometry::Type geom, DenseTensor &localP) const
 {
@@ -1556,15 +1617,20 @@ SparseMatrix* FiniteElementSpace::RefinementMatrix(int old_ndofs,
                "Previous mesh is not coarser.");
 
    Mesh::GeometryList elem_geoms(*mesh);
-
-   DenseTensor localP[Geometry::NumGeom];
-   for (int i = 0; i < elem_geoms.Size(); i++)
+   if (!IsVariableOrder())
    {
-      GetLocalRefinementMatrices(elem_geoms[i], localP[elem_geoms[i]]);
+      DenseTensor localP[Geometry::NumGeom];
+      for (int i = 0; i < elem_geoms.Size(); i++)
+      {
+         GetLocalRefinementMatrices(elem_geoms[i], localP[elem_geoms[i]]);
+      }
+      return RefinementMatrix_main(old_ndofs, *old_elem_dof, old_elem_fos,
+                                   localP);
+   }
+   else
+   {
+      return VariableOrderRefinementMatrix(old_ndofs, *old_elem_dof);
    }
-
-   return RefinementMatrix_main(old_ndofs, *old_elem_dof, old_elem_fos,
-                                localP);
 }
 
 FiniteElementSpace::RefinementOperator::RefinementOperator(
@@ -1582,9 +1648,12 @@ FiniteElementSpace::RefinementOperator::RefinementOperator(
 
    Mesh::GeometryList elem_geoms(*fespace->GetMesh());
 
-   for (int i = 0; i < elem_geoms.Size(); i++)
+   if (!fespace->IsVariableOrder())
    {
-      fespace->GetLocalRefinementMatrices(elem_geoms[i], localP[elem_geoms[i]]);
+      for (int i = 0; i < elem_geoms.Size(); i++)
+      {
+         fespace->GetLocalRefinementMatrices(elem_geoms[i], localP[elem_geoms[i]]);
+      }
    }
 
    ConstructDoFTransArray();
@@ -1597,10 +1666,13 @@ FiniteElementSpace::RefinementOperator::RefinementOperator(
 {
    Mesh::GeometryList elem_geoms(*fespace->GetMesh());
 
-   for (int i = 0; i < elem_geoms.Size(); i++)
+   if (!fespace->IsVariableOrder())
    {
-      fespace->GetLocalRefinementMatrices(*coarse_fes, elem_geoms[i],
-                                          localP[elem_geoms[i]]);
+      for (int i = 0; i < elem_geoms.Size(); i++)
+      {
+         fespace->GetLocalRefinementMatrices(*coarse_fes, elem_geoms[i],
+                                             localP[elem_geoms[i]]);
+      }
    }
 
    // Make a copy of the coarse elem_dof Table.
@@ -1676,11 +1748,25 @@ void FiniteElementSpace::RefinementOperator::Mult(const Vector &x,
 
    Vector subY, subX;
 
+   DenseMatrix eP;
+   IsoparametricTransformation isotr;
+
    for (int k = 0; k < mesh_ref->GetNE(); k++)
    {
       const Embedding &emb = trans_ref.embeddings[k];
       const Geometry::Type geom = mesh_ref->GetElementBaseGeometry(k);
-      const DenseMatrix &lP = localP[geom](emb.matrix);
+      if (fespace->IsVariableOrder())
+      {
+         const FiniteElement *fe = fespace->GetFE(k);
+         isotr.SetIdentityTransformation(geom);
+         const int ldof = fe->GetDof();
+         eP.SetSize(ldof, ldof);
+         const DenseTensor &pmats = trans_ref.point_matrices[geom];
+         isotr.SetPointMat(pmats(emb.matrix));
+         fe->GetLocalInterpolation(isotr, eP);
+      }
+      const DenseMatrix &lP = (fespace->IsVariableOrder()) ? eP : localP[geom](
+                                 emb.matrix);
 
       subY.SetSize(lP.Height());
 
@@ -1745,11 +1831,28 @@ void FiniteElementSpace::RefinementOperator::MultTranspose(const Vector &x,
 
    Vector subY, subX, subYt;
 
+   DenseMatrix eP;
+   IsoparametricTransformation isotr;
+   const FiniteElement *fe = nullptr;
+
    for (int k = 0; k < mesh_ref->GetNE(); k++)
    {
       const Embedding &emb = trans_ref.embeddings[k];
       const Geometry::Type geom = mesh_ref->GetElementBaseGeometry(k);
-      const DenseMatrix &lP = localP[geom](emb.matrix);
+
+      if (fespace->IsVariableOrder())
+      {
+         fe = fespace->GetFE(k);
+         isotr.SetIdentityTransformation(geom);
+         const int ldof = fe->GetDof();
+         eP.SetSize(ldof);
+         const DenseTensor &pmats = trans_ref.point_matrices[geom];
+         isotr.SetPointMat(pmats(emb.matrix));
+         fe->GetLocalInterpolation(isotr, eP);
+      }
+
+      const DenseMatrix &lP = (fespace->IsVariableOrder()) ? eP : localP[geom](
+                                 emb.matrix);
 
       DofTransformation *doftrans = fespace->GetElementDofs(k, f_dofs);
       old_elem_dof->GetRow(emb.parent, c_dofs);
@@ -2108,9 +2211,12 @@ SparseMatrix* FiniteElementSpace::DerefinementMatrix(int old_ndofs,
    Mesh::GeometryList elem_geoms(*mesh);
 
    DenseTensor localR[Geometry::NumGeom];
-   for (int i = 0; i < elem_geoms.Size(); i++)
+   if (!IsVariableOrder())
    {
-      GetLocalDerefinementMatrices(elem_geoms[i], localR[elem_geoms[i]]);
+      for (int i = 0; i < elem_geoms.Size(); i++)
+      {
+         GetLocalDerefinementMatrices(elem_geoms[i], localR[elem_geoms[i]]);
+      }
    }
 
    SparseMatrix *R = new SparseMatrix(ndofs*vdim, old_ndofs*vdim);
@@ -2125,14 +2231,34 @@ SparseMatrix* FiniteElementSpace::DerefinementMatrix(int old_ndofs,
 
    bool is_dg = FEColl()->GetContType() == FiniteElementCollection::DISCONTINUOUS;
    int num_marked = 0;
+   const FiniteElement *fe = nullptr;
+   DenseMatrix localRVO; //for variable order only
    for (int k = 0; k < dtrans.embeddings.Size(); k++)
    {
       const Embedding &emb = dtrans.embeddings[k];
       Geometry::Type geom = mesh->GetElementBaseGeometry(emb.parent);
-      DenseMatrix &lR = localR[geom](emb.matrix);
+
+      if (IsVariableOrder())
+      {
+         fe = GetFE(emb.parent);
+         const DenseTensor &pmats = dtrans.point_matrices[geom];
+         const int ldof = fe->GetDof();
+
+         IsoparametricTransformation isotr;
+         isotr.SetIdentityTransformation(geom);
+
+         localRVO.SetSize(ldof, ldof);
+         isotr.SetPointMat(pmats(emb.matrix));
+         // Local restriction is size ldofxldof assuming that the parent and
+         // child are of same polynomial order.
+         fe->GetLocalRestriction(isotr, localRVO);
+      }
+      DenseMatrix &lR = IsVariableOrder() ? localRVO : localR[geom](emb.matrix);
 
       elem_dof->GetRow(emb.parent, dofs);
       old_elem_dof->GetRow(k, old_dofs);
+      MFEM_VERIFY(old_dofs.Size() == dofs.Size(),
+                  "Parent and child must have same #dofs.");
 
       for (int vd = 0; vd < vdim; vd++)
       {
@@ -2158,7 +2284,7 @@ SparseMatrix* FiniteElementSpace::DerefinementMatrix(int old_ndofs,
       }
    }
 
-   if (!is_dg)
+   if (!is_dg && !IsVariableOrder())
    {
       MFEM_VERIFY(num_marked == R->Height(),
                   "internal error: not all rows of R were set.");
@@ -2216,6 +2342,7 @@ void FiniteElementSpace::Constructor(Mesh *mesh_, NURBSExtension *NURBSext_,
 
    const NURBSFECollection *nurbs_fec =
       dynamic_cast<const NURBSFECollection *>(fec_);
+
    if (nurbs_fec)
    {
       MFEM_VERIFY(mesh_->NURBSext, "NURBS FE space requires a NURBS mesh.");
@@ -2312,12 +2439,63 @@ void FiniteElementSpace::UpdateNURBS()
    face_dof = NULL;
    face_to_be.DeleteAll();
 
+   // Depending on the element type create the appropriate extensions
+   // for the individual components.
    dynamic_cast<const NURBSFECollection *>(fec)->Reset();
 
-   ndofs = NURBSext->GetNDof();
-   elem_dof = NURBSext->GetElementDofTable();
-   bdr_elem_dof = NURBSext->GetBdrElementDofTable();
+   if (dynamic_cast<const NURBS_HDivFECollection *>(fec))
+   {
+      VNURBSext.SetSize(mesh->Dimension());
+      for (int d = 0; d < mesh->Dimension(); d++)
+      {
+         VNURBSext[d] = NURBSext->GetDivExtension(d);
+      }
+   }
 
+   if (dynamic_cast<const NURBS_HCurlFECollection *>(fec))
+   {
+      VNURBSext.SetSize(mesh->Dimension());
+      for (int d = 0; d < mesh->Dimension(); d++)
+      {
+         VNURBSext[d] = NURBSext->GetCurlExtension(d);
+      }
+   }
+
+   // If required: concatenate the dof tables of the individual components into
+   // one dof table for the vector fespace.
+   if (VNURBSext.Size() == 2)
+   {
+      int offset1 = VNURBSext[0]->GetNDof();
+      ndofs = VNURBSext[0]->GetNDof() + VNURBSext[1]->GetNDof();
+
+      // Merge Tables
+      elem_dof = new Table(*VNURBSext[0]->GetElementDofTable(),
+                           *VNURBSext[1]->GetElementDofTable(),offset1 );
+
+      bdr_elem_dof = new Table(*VNURBSext[0]->GetBdrElementDofTable(),
+                               *VNURBSext[1]->GetBdrElementDofTable(),offset1);
+   }
+   else if (VNURBSext.Size() == 3)
+   {
+      int offset1 = VNURBSext[0]->GetNDof();
+      int offset2 = offset1 + VNURBSext[1]->GetNDof();
+      ndofs = offset2 + VNURBSext[2]->GetNDof();
+
+      // Merge Tables
+      elem_dof = new Table(*VNURBSext[0]->GetElementDofTable(),
+                           *VNURBSext[1]->GetElementDofTable(),offset1,
+                           *VNURBSext[2]->GetElementDofTable(),offset2);
+
+      bdr_elem_dof = new Table(*VNURBSext[0]->GetBdrElementDofTable(),
+                               *VNURBSext[1]->GetBdrElementDofTable(),offset1,
+                               *VNURBSext[2]->GetBdrElementDofTable(),offset2);
+   }
+   else
+   {
+      ndofs = NURBSext->GetNDof();
+      elem_dof = NURBSext->GetElementDofTable();
+      bdr_elem_dof = NURBSext->GetBdrElementDofTable();
+   }
    mesh_sequence = mesh->GetSequence();
    sequence++;
 }
@@ -3319,11 +3497,21 @@ void FiniteElementSpace::Destroy()
    dof_elem_array.DeleteAll();
    dof_ldof_array.DeleteAll();
 
+   for (int i = 0; i < VNURBSext.Size(); i++)
+   {
+      delete VNURBSext[i];
+   }
+
    if (NURBSext)
    {
       if (own_ext) { delete NURBSext; }
       delete face_dof;
       face_to_be.DeleteAll();
+      if (VNURBSext.Size() > 0 )
+      {
+         delete elem_dof;
+         delete bdr_elem_dof;
+      }
    }
    else
    {
@@ -3335,6 +3523,8 @@ void FiniteElementSpace::Destroy()
       delete [] bdofs;
    }
    ceed::RemoveBasisAndRestriction(this);
+
+
 }
 
 void FiniteElementSpace::DestroyDoFTransArray()
@@ -3353,19 +3543,27 @@ void FiniteElementSpace::GetTransferOperator(
 
    if (T.Type() == Operator::MFEM_SPARSEMAT)
    {
-      Mesh::GeometryList elem_geoms(*mesh);
-
-      DenseTensor localP[Geometry::NumGeom];
-      for (int i = 0; i < elem_geoms.Size(); i++)
+      if (!IsVariableOrder())
       {
-         GetLocalRefinementMatrices(coarse_fes, elem_geoms[i],
-                                    localP[elem_geoms[i]]);
+         Mesh::GeometryList elem_geoms(*mesh);
+
+         DenseTensor localP[Geometry::NumGeom];
+         for (int i = 0; i < elem_geoms.Size(); i++)
+         {
+            GetLocalRefinementMatrices(coarse_fes, elem_geoms[i],
+                                       localP[elem_geoms[i]]);
+         }
+         T.Reset(RefinementMatrix_main(coarse_fes.GetNDofs(),
+                                       coarse_fes.GetElementToDofTable(),
+                                       coarse_fes.
+                                       GetElementToFaceOrientationTable(),
+                                       localP));
+      }
+      else
+      {
+         T.Reset(VariableOrderRefinementMatrix(coarse_fes.GetNDofs(),
+                                               coarse_fes.GetElementToDofTable()));
       }
-      T.Reset(RefinementMatrix_main(coarse_fes.GetNDofs(),
-                                    coarse_fes.GetElementToDofTable(),
-                                    coarse_fes.
-                                    GetElementToFaceOrientationTable(),
-                                    localP));
    }
    else
    {
@@ -3416,19 +3614,33 @@ void FiniteElementSpace::GetTrueTransferOperator(
 
 void FiniteElementSpace::UpdateElementOrders()
 {
-   const CoarseFineTransformations &cf_tr = mesh->GetRefinementTransforms();
-
    Array<char> new_order(mesh->GetNE());
    switch (mesh->GetLastOperation())
    {
       case Mesh::REFINE:
       {
+         const CoarseFineTransformations &cf_tr = mesh->GetRefinementTransforms();
          for (int i = 0; i < mesh->GetNE(); i++)
          {
             new_order[i] = elem_order[cf_tr.embeddings[i].parent];
          }
          break;
       }
+      case Mesh::DEREFINE:
+      {
+         const CoarseFineTransformations &cf_tr =
+            mesh->ncmesh->GetDerefinementTransforms();
+         Table coarse_to_fine;
+         cf_tr.MakeCoarseToFineTable(coarse_to_fine);
+         Array<int> tabrow;
+         for (int i = 0; i < coarse_to_fine.Size(); i++)
+         {
+            coarse_to_fine.GetRow(i, tabrow);
+            //For now we require that all children are of same polynomial order.
+            new_order[i] = elem_order[tabrow[0]];
+         }
+         break;
+      }
       default:
          MFEM_ABORT("not implemented yet");
    }
@@ -3523,11 +3735,23 @@ void FiniteElementSpace::Update(bool want_transform)
          {
             BuildConformingInterpolation();
             Th.Reset(DerefinementMatrix(old_ndofs, old_elem_dof, old_elem_fos));
-            if (cP && cR)
+            if (IsVariableOrder())
             {
-               Th.SetOperatorOwner(false);
-               Th.Reset(new TripleProductOperator(cP.get(), cR.get(), Th.Ptr(),
-                                                  false, false, true));
+               if (cP && cR_hp)
+               {
+                  Th.SetOperatorOwner(false);
+                  Th.Reset(new TripleProductOperator(cP.get(), cR_hp.get(), Th.Ptr(),
+                                                     false, false, true));
+               }
+            }
+            else
+            {
+               if (cP && cR)
+               {
+                  Th.SetOperatorOwner(false);
+                  Th.Reset(new TripleProductOperator(cP.get(), cR.get(), Th.Ptr(),
+                                                     false, false, true));
+               }
             }
             break;
          }
@@ -3640,6 +3864,7 @@ FiniteElementCollection *FiniteElementSpace::Load(Mesh *m, std::istream &input)
    input >> ord;
 
    NURBSFECollection *nurbs_fec = dynamic_cast<NURBSFECollection*>(r_fec);
+   if (nurbs_fec) { nurbs_fec->SetDim(m->Dimension()); }
    NURBSExtension *nurbs_ext = NULL;
    if (fes_format == 90) // original format, v0.9
    {

fem/fespace.hpp

@@ -268,6 +268,10 @@ protected:
    Array<int> dof_elem_array, dof_ldof_array;
 
    NURBSExtension *NURBSext;
+   /** array of NURBS extension for H(div) and H(curl) vector elements.
+       For each direction an extension is created from the base NURBSext,
+       with an increase in order in the appropriate direction. */
+   Array<NURBSExtension*> VNURBSext;
    int own_ext;
    mutable Array<int> face_to_be; // NURBS FE space only
 
@@ -469,6 +473,11 @@ protected:
                                        const Table *coarse_elem_fos,
                                        const DenseTensor localP[]) const;
 
+   /* This method returns the Refinement matrix (i.e., the embedding)
+      from a coarse variable-order fes to a fine fes (after a geometric refinement) */
+   SparseMatrix *VariableOrderRefinementMatrix(const int coarse_ndofs,
+                                               const Table &coarse_elem_dof) const;
+
    void GetLocalRefinementMatrices(Geometry::Type geom,
                                    DenseTensor &localP) const;
    void GetLocalDerefinementMatrices(Geometry::Type geom,
@@ -517,6 +526,8 @@ protected:
                                                const Array<int> *perm);
 
 public:
+
+
    /** @brief Default constructor: the object is invalid until initialized using
        the method Load(). */
    FiniteElementSpace();

fem/gridfunc.cpp

@@ -12,6 +12,8 @@
 // Implementation of GridFunction
 
 #include "gridfunc.hpp"
+#include "linearform.hpp"
+#include "bilinearform.hpp"
 #include "quadinterpolator.hpp"
 #include "../mesh/nurbs.hpp"
 #include "../general/text.hpp"
@@ -39,7 +41,7 @@ GridFunction::GridFunction(Mesh *m, std::istream &input)
    UseDevice(true);
 
    fes = new FiniteElementSpace;
-   fec = fes->Load(m, input);
+   fec_owned = fes->Load(m, input);
 
    skip_comment_lines(input, '#');
    istream::int_type next_char = input.peek();
@@ -81,10 +83,10 @@ GridFunction::GridFunction(Mesh *m, GridFunction *gf_array[], int num_pieces)
    int vdim, ordering;
 
    fes = gf_array[0]->FESpace();
-   fec = FiniteElementCollection::New(fes->FEColl()->Name());
+   fec_owned = FiniteElementCollection::New(fes->FEColl()->Name());
    vdim = fes->GetVDim();
    ordering = fes->GetOrdering();
-   fes = new FiniteElementSpace(m, fec, vdim, ordering);
+   fes = new FiniteElementSpace(m, fec_owned, vdim, ordering);
    SetSize(fes->GetVSize());
 
    if (m->NURBSext)
@@ -153,11 +155,11 @@ GridFunction::GridFunction(Mesh *m, GridFunction *gf_array[], int num_pieces)
 
 void GridFunction::Destroy()
 {
-   if (fec)
+   if (fec_owned)
    {
       delete fes;
-      delete fec;
-      fec = NULL;
+      delete fec_owned;
+      fec_owned = NULL;
    }
 }
 
@@ -325,10 +327,9 @@ int GridFunction::VectorDim() const
    const FiniteElement *fe;
    if (!fes->GetNE())
    {
-      const FiniteElementCollection *fe_coll = fes->FEColl();
       static const Geometry::Type geoms[3] =
       { Geometry::SEGMENT, Geometry::TRIANGLE, Geometry::TETRAHEDRON };
-      fe = fe_coll->
+      fe = fes->FEColl()->
            FiniteElementForGeometry(geoms[fes->GetMesh()->Dimension()-1]);
    }
    else
@@ -350,7 +351,8 @@ int GridFunction::CurlDim() const
    {
       static const Geometry::Type geoms[3] =
       { Geometry::SEGMENT, Geometry::TRIANGLE, Geometry::TETRAHEDRON };
-      fe = fec->FiniteElementForGeometry(geoms[fes->GetMesh()->Dimension()-1]);
+      fe = fes->FEColl()->
+           FiniteElementForGeometry(geoms[fes->GetMesh()->Dimension()-1]);
    }
    else
    {
@@ -1321,9 +1323,9 @@ void GridFunction::ProjectVectorFieldOn(GridFunction &vec_field, int comp)
    }
 }
 
-void GridFunction::AccumulateAndCountDerivativeValues(int comp, int der_comp,
-                                                      GridFunction &der,
-                                                      Array<int> &zones_per_dof)
+void GridFunction::AccumulateAndCountDerivativeValues(
+   int comp, int der_comp, GridFunction &der,
+   Array<int> &zones_per_dof) const
 {
    FiniteElementSpace * der_fes = der.FESpace();
    ElementTransformation * transf;
@@ -1374,7 +1376,8 @@ void GridFunction::AccumulateAndCountDerivativeValues(int comp, int der_comp,
    }
 }
 
-void GridFunction::GetDerivative(int comp, int der_comp, GridFunction &der)
+void GridFunction::GetDerivative(int comp, int der_comp,
+                                 GridFunction &der) const
 {
    Array<int> overlap;
    AccumulateAndCountDerivativeValues(comp, der_comp, der, overlap);
@@ -2061,41 +2064,37 @@ void GridFunction::AccumulateAndCountBdrValues(
    Coefficient *coeff[], VectorCoefficient *vcoeff, const Array<int> &attr,
    Array<int> &values_counter)
 {
-   int i, j, fdof, d, ind, vdim;
-   real_t val;
-   const FiniteElement *fe;
-   ElementTransformation *transf;
    Array<int> vdofs;
    Vector vc;
 
    values_counter.SetSize(Size());
    values_counter = 0;
 
-   vdim = fes->GetVDim();
-
+   const int vdim = fes->GetVDim();
    HostReadWrite();
 
-   for (i = 0; i < fes->GetNBE(); i++)
+   for (int i = 0; i < fes->GetNBE(); i++)
    {
       if (attr[fes->GetBdrAttribute(i) - 1] == 0) { continue; }
 
-      fe = fes->GetBE(i);
-      fdof = fe->GetDof();
-      transf = fes->GetBdrElementTransformation(i);
+      const FiniteElement *fe = fes->GetBE(i);
+      const int fdof = fe->GetDof();
+      ElementTransformation *transf = fes->GetBdrElementTransformation(i);
       const IntegrationRule &ir = fe->GetNodes();
       fes->GetBdrElementVDofs(i, vdofs);
 
-      for (j = 0; j < fdof; j++)
+      for (int j = 0; j < fdof; j++)
       {
          const IntegrationPoint &ip = ir.IntPoint(j);
          transf->SetIntPoint(&ip);
          if (vcoeff) { vcoeff->Eval(vc, *transf, ip); }
-         for (d = 0; d < vdim; d++)
+         for (int d = 0; d < vdim; d++)
          {
             if (!vcoeff && !coeff[d]) { continue; }
 
-            val = vcoeff ? vc(d) : coeff[d]->Eval(*transf, ip);
-            if ( (ind = vdofs[fdof*d+j]) < 0 )
+            real_t val = vcoeff ? vc(d) : coeff[d]->Eval(*transf, ip);
+            int ind = vdofs[fdof*d+j];
+            if ( ind < 0 )
             {
                val = -val, ind = -1-ind;
             }
@@ -2117,10 +2116,11 @@ void GridFunction::AccumulateAndCountBdrValues(
    // iff A_ij != 0. It is sufficient to resolve just the first level of
    // dependency, since A is a projection matrix: A^n = A due to cR.cP = I.
    // Cases like these arise in 3D when boundary edges are constrained by
-   // (depend on) internal faces/elements. We use the virtual method
-   // GetBoundaryClosure from NCMesh to resolve the dependencies.
-
-   if (fes->Nonconforming() && fes->GetMesh()->Dimension() == 3)
+   // (depend on) internal faces/elements, or for internal boundaries in 2 or
+   // 3D. We use the virtual method GetBoundaryClosure from NCMesh to resolve
+   // the dependencies.
+   if (fes->Nonconforming() && (fes->GetMesh()->Dimension() == 2 ||
+                                fes->GetMesh()->Dimension() == 3))
    {
       Vector vals;
       Mesh *mesh = fes->GetMesh();
@@ -2128,26 +2128,19 @@ void GridFunction::AccumulateAndCountBdrValues(
       Array<int> bdr_edges, bdr_vertices, bdr_faces;
       ncmesh->GetBoundaryClosure(attr, bdr_vertices, bdr_edges, bdr_faces);
 
-      for (i = 0; i < bdr_edges.Size(); i++)
+      auto mark_dofs = [&](ElementTransformation &transf, const FiniteElement &fe)
       {
-         int edge = bdr_edges[i];
-         fes->GetEdgeVDofs(edge, vdofs);
-         if (vdofs.Size() == 0) { continue; }
-
-         transf = mesh->GetEdgeTransformation(edge);
-         transf->Attribute = -1; // TODO: set the boundary attribute
-         fe = fes->GetEdgeElement(edge);
          if (!vcoeff)
          {
-            vals.SetSize(fe->GetDof());
-            for (d = 0; d < vdim; d++)
+            vals.SetSize(fe.GetDof());
+            for (int d = 0; d < vdim; d++)
             {
                if (!coeff[d]) { continue; }
 
-               fe->Project(*coeff[d], *transf, vals);
+               fe.Project(*coeff[d], transf, vals);
                for (int k = 0; k < vals.Size(); k++)
                {
-                  ind = vdofs[d*vals.Size()+k];
+                  const int ind = vdofs[d*vals.Size()+k];
                   if (++values_counter[ind] == 1)
                   {
                      (*this)(ind) = vals(k);
@@ -2161,11 +2154,11 @@ void GridFunction::AccumulateAndCountBdrValues(
          }
          else // vcoeff != NULL
          {
-            vals.SetSize(vdim*fe->GetDof());
-            fe->Project(*vcoeff, *transf, vals);
+            vals.SetSize(vdim*fe.GetDof());
+            fe.Project(*vcoeff, transf, vals);
             for (int k = 0; k < vals.Size(); k++)
             {
-               ind = vdofs[k];
+               const int ind = vdofs[k];
                if (++values_counter[ind] == 1)
                {
                   (*this)(ind) = vals(k);
@@ -2176,6 +2169,26 @@ void GridFunction::AccumulateAndCountBdrValues(
                }
             }
          }
+      };
+
+      for (auto edge : bdr_edges)
+      {
+         fes->GetEdgeVDofs(edge, vdofs);
+         if (vdofs.Size() == 0) { continue; }
+
+         ElementTransformation *transf = mesh->GetEdgeTransformation(edge);
+         const FiniteElement *fe = fes->GetEdgeElement(edge);
+         mark_dofs(*transf, *fe);
+      }
+
+      for (auto face : bdr_faces)
+      {
+         fes->GetFaceVDofs(face, vdofs);
+         if (vdofs.Size() == 0) { continue; }
+
+         ElementTransformation *transf = mesh->GetFaceTransformation(face);
+         const FiniteElement *fe = fes->GetFaceElement(face);
+         mark_dofs(*transf, *fe);
       }
    }
 }
@@ -2228,26 +2241,37 @@ void GridFunction::AccumulateAndCountBdrTangentValues(
       accumulate_dofs(dofs, lvec, *this, values_counter);
    }
 
-   if (fes->Nonconforming() && fes->GetMesh()->Dimension() == 3)
+   if (fes->Nonconforming() && (fes->GetMesh()->Dimension() == 2 ||
+                                fes->GetMesh()->Dimension() == 3))
    {
       Mesh *mesh = fes->GetMesh();
       NCMesh *ncmesh = mesh->ncmesh;
       Array<int> bdr_edges, bdr_vertices, bdr_faces;
       ncmesh->GetBoundaryClosure(bdr_attr, bdr_vertices, bdr_edges, bdr_faces);
 
-      for (int i = 0; i < bdr_edges.Size(); i++)
+      for (auto edge : bdr_edges)
       {
-         int edge = bdr_edges[i];
          fes->GetEdgeDofs(edge, dofs);
          if (dofs.Size() == 0) { continue; }
 
          T = mesh->GetEdgeTransformation(edge);
-         T->Attribute = -1; // TODO: set the boundary attribute
          fe = fes->GetEdgeElement(edge);
          lvec.SetSize(fe->GetDof());
          fe->Project(vcoeff, *T, lvec);
          accumulate_dofs(dofs, lvec, *this, values_counter);
       }
+
+      for (auto face : bdr_faces)
+      {
+         fes->GetFaceDofs(face, dofs);
+         if (dofs.Size() == 0) { continue; }
+
+         T = mesh->GetFaceTransformation(face);
+         fe = fes->GetFaceElement(face);
+         lvec.SetSize(fe->GetDof());
+         fe->Project(vcoeff, *T, lvec);
+         accumulate_dofs(dofs, lvec, *this, values_counter);
+      }
    }
 }
 
@@ -2350,19 +2374,48 @@ void GridFunction::ProjectCoefficient(Coefficient &coeff)
 
    if (delta_c == NULL)
    {
-      Array<int> vdofs;
-      Vector vals;
-
-      for (int i = 0; i < fes->GetNE(); i++)
+      if (fes->GetNURBSext() == NULL)
       {
-         doftrans = fes->GetElementVDofs(i, vdofs);
-         vals.SetSize(vdofs.Size());
-         fes->GetFE(i)->Project(coeff, *fes->GetElementTransformation(i), vals);
-         if (doftrans)
+         Array<int> vdofs;
+         Vector vals;
+
+         for (int i = 0; i < fes->GetNE(); i++)
          {
-            doftrans->TransformPrimal(vals);
+            doftrans = fes->GetElementVDofs(i, vdofs);
+            vals.SetSize(vdofs.Size());
+            fes->GetFE(i)->Project(coeff, *fes->GetElementTransformation(i), vals);
+            if (doftrans)
+            {
+               doftrans->TransformPrimal(vals);
+            }
+            SetSubVector(vdofs, vals);
          }
-         SetSubVector(vdofs, vals);
+      }
+      else
+      {
+         // Define and assemble linear form
+         LinearForm b(fes);
+         b.AddDomainIntegrator(new DomainLFIntegrator(coeff));
+         b.Assemble();
+
+         // Define and assemble bilinear form
+         BilinearForm a(fes);
+         a.AddDomainIntegrator(new MassIntegrator());
+         a.Assemble();
+
+         // Set solver and preconditioner
+         SparseMatrix A(a.SpMat());
+         GSSmoother  prec(A);
+         CGSolver cg;
+         cg.SetOperator(A);
+         cg.SetPreconditioner(prec);
+         cg.SetRelTol(1e-12);
+         cg.SetMaxIter(1000);
+         cg.SetPrintLevel(0);
+
+         // Solve and get solution
+         *this = 0.0;
+         cg.Mult(b,*this);
       }
    }
    else
@@ -2403,22 +2456,54 @@ void GridFunction::ProjectCoefficient(
 
 void GridFunction::ProjectCoefficient(VectorCoefficient &vcoeff)
 {
-   int i;
-   Array<int> vdofs;
-   Vector vals;
-
-   DofTransformation * doftrans = NULL;
-
-   for (i = 0; i < fes->GetNE(); i++)
+   if (fes->GetNURBSext() == NULL)
    {
-      doftrans = fes->GetElementVDofs(i, vdofs);
-      vals.SetSize(vdofs.Size());
-      fes->GetFE(i)->Project(vcoeff, *fes->GetElementTransformation(i), vals);
-      if (doftrans)
+
+      int i;
+      Array<int> vdofs;
+      Vector vals;
+
+      DofTransformation * doftrans = NULL;
+
+      for (i = 0; i < fes->GetNE(); i++)
       {
-         doftrans->TransformPrimal(vals);
+         doftrans = fes->GetElementVDofs(i, vdofs);
+         vals.SetSize(vdofs.Size());
+         fes->GetFE(i)->Project(vcoeff, *fes->GetElementTransformation(i), vals);
+         if (doftrans)
+         {
+            doftrans->TransformPrimal(vals);
+         }
+         SetSubVector(vdofs, vals);
       }
-      SetSubVector(vdofs, vals);
+
+   }
+
+   else
+   {
+      // Define and assemble linear form
+      LinearForm b(fes);
+      b.AddDomainIntegrator(new VectorFEDomainLFIntegrator(vcoeff));
+      b.Assemble();
+
+      // Define and assemble bilinear form
+      BilinearForm a(fes);
+      a.AddDomainIntegrator(new VectorFEMassIntegrator());
+      a.Assemble();
+
+      // Set solver and preconditioner
+      SparseMatrix A(a.SpMat());
+      GSSmoother  prec(A);
+      CGSolver cg;
+      cg.SetOperator(A);
+      cg.SetPreconditioner(prec);
+      cg.SetRelTol(1e-12);
+      cg.SetMaxIter(1000);
+      cg.SetPrintLevel(0);
+
+      // Solve and get solution
+      *this = 0.0;
+      cg.Mult(b,*this);
    }
 }
 
@@ -2966,6 +3051,57 @@ real_t GridFunction::ComputeDivError(
    return (error < 0.0) ? -sqrt(-error) : sqrt(error);
 }
 
+real_t GridFunction::ComputeLaplaceError(
+   Coefficient *exlap, const IntegrationRule *irs[]) const
+{
+   real_t error = 0.0, a;
+   const FiniteElement *fe;
+   ElementTransformation *Tr;
+   Array<int> dofs;
+   int intorder, fdof;
+   Vector laplace;
+
+   for (int i = 0; i < fes->GetNE(); i++)
+   {
+      laplace.SetSize(fdof);
+      fe = fes->GetFE(i);
+      Tr = fes->GetElementTransformation(i);
+      intorder = 2*fe->GetOrder() + 3;
+      const IntegrationRule *ir;
+      if (irs)
+      {
+         ir = irs[fe->GetGeomType()];
+      }
+      else
+      {
+         ir = &(IntRules.Get(fe->GetGeomType(), intorder));
+      }
+      fes->GetElementDofs(i, dofs);
+      fdof = fe->GetDof();
+      laplace.SetSize(fdof);
+      for (int j = 0; j < ir->GetNPoints(); j++)
+      {
+         const IntegrationPoint &ip = ir->IntPoint(j);
+         Tr->SetIntPoint(&ip);
+         fe->CalcPhysLaplacian(*Tr, laplace);
+         a = 0;
+         for (int k = 0; k < fdof; k++)
+            if (dofs[k] >= 0)
+            {
+               a += (*this)(dofs[k]) * laplace(k);
+            }
+            else
+            {
+               a -= (*this)(-1-dofs[k]) * laplace(k);
+            }
+         a -= exlap->Eval(*Tr, ip);
+         error += ip.weight * Tr->Weight() * a * a;
+      }
+   }
+
+   return (error < 0.0) ? -sqrt(-error) : sqrt(error);
+}
+
 real_t GridFunction::ComputeDGFaceJumpError(Coefficient *exsol,
                                             Coefficient *ell_coeff,
                                             class JumpScaling jump_scaling,
@@ -3904,7 +4040,7 @@ void GridFunction::LegacyNCReorder()
       mesh->GetEdgeVertices(i, ev);
       if (old_vertex[ev[0]] > old_vertex[ev[1]])
       {
-         const int *ind = fec->DofOrderForOrientation(Geometry::SEGMENT, -1);
+         const int *ind = fes->FEColl()->DofOrderForOrientation(Geometry::SEGMENT, -1);
 
          fes->GetEdgeInteriorDofs(i, dofs);
          for (int k = 0; k < dofs.Size(); k++)

fem/gridfunc.hpp

@@ -30,14 +30,14 @@ namespace mfem
 class GridFunction : public Vector
 {
 protected:
-   /// FE space on which the grid function lives. Owned if #fec is not NULL.
+   /// FE space on which the grid function lives. Owned if #fec_owned is not NULL.
    FiniteElementSpace *fes;
 
    /** @brief Used when the grid function is read from a file. It can also be
        set explicitly, see MakeOwner().
 
        If not NULL, this pointer is owned by the GridFunction. */
-   FiniteElementCollection *fec;
+   FiniteElementCollection *fec_owned;
 
    long fes_sequence; // see FiniteElementSpace::sequence, Mesh::sequence
 
@@ -72,16 +72,16 @@ protected:
 
 public:
 
-   GridFunction() { fes = NULL; fec = NULL; fes_sequence = 0; UseDevice(true); }
+   GridFunction() { fes = NULL; fec_owned = NULL; fes_sequence = 0; UseDevice(true); }
 
    /// Copy constructor. The internal true-dof vector #t_vec is not copied.
    GridFunction(const GridFunction &orig)
-      : Vector(orig), fes(orig.fes), fec(NULL), fes_sequence(orig.fes_sequence)
+      : Vector(orig), fes(orig.fes), fec_owned(NULL), fes_sequence(orig.fes_sequence)
    { UseDevice(true); }
 
    /// Construct a GridFunction associated with the FiniteElementSpace @a *f.
    GridFunction(FiniteElementSpace *f) : Vector(f->GetVSize())
-   { fes = f; fec = NULL; fes_sequence = f->GetSequence(); UseDevice(true); }
+   { fes = f; fec_owned = NULL; fes_sequence = f->GetSequence(); UseDevice(true); }
 
    /// Construct a GridFunction using previously allocated array @a data.
    /** The GridFunction does not assume ownership of @a data which is assumed to
@@ -91,13 +91,13 @@ public:
     */
    GridFunction(FiniteElementSpace *f, real_t *data)
       : Vector(data, f->GetVSize())
-   { fes = f; fec = NULL; fes_sequence = f->GetSequence(); UseDevice(true); }
+   { fes = f; fec_owned = NULL; fes_sequence = f->GetSequence(); UseDevice(true); }
 
    /** @brief Construct a GridFunction using previously allocated Vector @a base
        starting at the given offset, @a base_offset. */
    GridFunction(FiniteElementSpace *f, Vector &base, int base_offset = 0)
       : Vector(base, base_offset, f->GetVSize())
-   { fes = f; fec = NULL; fes_sequence = f->GetSequence(); UseDevice(true); }
+   { fes = f; fec_owned = NULL; fes_sequence = f->GetSequence(); UseDevice(true); }
 
    /// Construct a GridFunction on the given Mesh, using the data from @a input.
    /** The content of @a input should be in the format created by the method
@@ -116,12 +116,12 @@ public:
    GridFunction &operator=(const GridFunction &rhs)
    { return operator=((const Vector &)rhs); }
 
-   /// Make the GridFunction the owner of #fec and #fes.
-   /** If the new FiniteElementCollection, @a fec_, is NULL, ownership of #fec
+   /// Make the GridFunction the owner of #fec_owned and #fes.
+   /** If the new FiniteElementCollection, @a fec_, is NULL, ownership of #fec_owned
        and #fes is taken away. */
-   void MakeOwner(FiniteElementCollection *fec_) { fec = fec_; }
+   void MakeOwner(FiniteElementCollection *fec_) { fec_owned = fec_; }
 
-   FiniteElementCollection *OwnFEC() { return fec; }
+   FiniteElementCollection *OwnFEC() { return fec_owned; }
 
    int VectorDim() const;
    int CurlDim() const;
@@ -321,7 +321,7 @@ public:
        @param[out] der       The resulting derivative (scalar function). The
                              FiniteElementSpace of this function must be set
                              before the call. */
-   void GetDerivative(int comp, int der_comp, GridFunction &der);
+   void GetDerivative(int comp, int der_comp, GridFunction &der) const;
 
    real_t GetDivergence(ElementTransformation &tr) const;
 
@@ -387,7 +387,8 @@ public:
    /** @brief Project @a coeff Coefficient to @a this GridFunction. The
        projection computation depends on the choice of the FiniteElementSpace
        #fes. Note that this is usually interpolation at the degrees of freedom
-       in each element (not L2 projection). */
+       in each element (not L2 projection). For NURBS spaces these degrees of
+       freedom are not available and L2 projection is resorted to as fallback. */
    virtual void ProjectCoefficient(Coefficient &coeff);
 
    /** @brief Project @a coeff Coefficient to @a this GridFunction, using one
@@ -398,7 +399,8 @@ public:
    /** @brief Project @a vcoeff VectorCoefficient to @a this GridFunction. The
        projection computation depends on the choice of the FiniteElementSpace
        #fes. Note that this is usually interpolation at the degrees of freedom
-       in each element (not L2 projection).*/
+       in each element (not L2 projection). For NURBS spaces these degrees of
+       freedom are not available and L2 projection is resorted to as fallback. */
    void ProjectCoefficient(VectorCoefficient &vcoeff);
 
    /** @brief Project @a vcoeff VectorCoefficient to @a this GridFunction, using
@@ -443,7 +445,7 @@ protected:
        GetDerivative() method; see its documentation. */
    void AccumulateAndCountDerivativeValues(int comp, int der_comp,
                                            GridFunction &der,
-                                           Array<int> &zones_per_dof);
+                                           Array<int> &zones_per_dof) const;
 
    void AccumulateAndCountBdrValues(Coefficient *coeff[],
                                     VectorCoefficient *vcoeff,
@@ -531,6 +533,10 @@ public:
    virtual real_t ComputeDivError(Coefficient *exdiv,
                                   const IntegrationRule *irs[] = NULL) const;
 
+   /// Returns ||lap u_ex - lap u_h||_L2 for H1 elements
+   virtual real_t ComputeLaplaceError(Coefficient *exlap,
+                                      const IntegrationRule *irs[] = NULL) const;
+
    /// Returns the Face Jumps error for L2 elements. The error can be weighted
    /// by a constant nu, by nu/h, or nu*p^2/h, depending on the value of
    /// @a jump_scaling.

fem/gslib.cpp

@@ -1393,8 +1393,10 @@ GSOPGSLIB::~GSOPGSLIB()
 void GSOPGSLIB::UpdateIdentifiers(const Array<long long> &ids)
 {
    long long minval = ids.Min();
+#ifdef MFEM_USE_MPI
    MPI_Allreduce(MPI_IN_PLACE, &minval, 1, MPI_LONG_LONG_INT,
                  MPI_MIN, gsl_comm->c);
+#endif
    MFEM_VERIFY(minval >= 0, "Unique identifier cannot be negative.");
    if (gsl_data != NULL) { gslib_gs_free(gsl_data); }
    num_ids = ids.Size();

fem/intrules_cut.cpp

@@ -220,8 +220,8 @@ void MomentFittingIntRules::ComputeSurfaceWeights1D(ElementTransformation& Tr)
    {
       IntegrationPoint ip2;
       ip2.x = .5;
-      while (LvlSet->Eval(Tr, ip2) > 1e-12
-             || LvlSet->Eval(Tr, ip2) < -1e-12)
+      while (LvlSet->Eval(Tr, ip2) > tol_1
+             || LvlSet->Eval(Tr, ip2) < -tol_1)
       {
          if (LvlSet->Eval(Tr, ip0) * LvlSet->Eval(Tr, ip2) < 0.)
          {
@@ -237,12 +237,12 @@ void MomentFittingIntRules::ComputeSurfaceWeights1D(ElementTransformation& Tr)
       intp.x = ip2.x;
       intp.weight = 1. / Tr.Weight();
    }
-   else if (LvlSet->Eval(Tr, ip0) > 0. && LvlSet->Eval(Tr, ip1) <= 1e-12)
+   else if (LvlSet->Eval(Tr, ip0) > 0. && LvlSet->Eval(Tr, ip1) <= tol_1)
    {
       intp.x = 1.;
       intp.weight = 1. / Tr.Weight();
    }
-   else if (LvlSet->Eval(Tr, ip1) > 0. && LvlSet->Eval(Tr, ip0) <= 1e-12)
+   else if (LvlSet->Eval(Tr, ip1) > 0. && LvlSet->Eval(Tr, ip0) <= tol_1)
    {
       intp.x = 0.;
       intp.weight = 1. / Tr.Weight();
@@ -290,8 +290,8 @@ void MomentFittingIntRules::ComputeVolumeWeights1D(ElementTransformation& Tr,
          }
       }
    }
-   else if (LvlSet->Eval(Tr, ip0) <= -1e-12
-            || LvlSet->Eval(Tr, ip1) <= -1e-12)
+   else if (LvlSet->Eval(Tr, ip0) <= -tol_1
+            || LvlSet->Eval(Tr, ip1) <= -tol_1)
    {
       for (int ip = 0; ip < ir.GetNPoints(); ip++)
       {
@@ -356,24 +356,24 @@ void MomentFittingIntRules::ComputeSurfaceWeights2D(ElementTransformation& Tr)
       IntegrationPoint ipB;
       Trafo.TransformBack(pointB, ipB);
 
-      if (LvlSet->Eval(Trafo, ipA) < -1e-12
-          || LvlSet->Eval(Trafo, ipB) < -1e-12)
+      if (LvlSet->Eval(Trafo, ipA) < -tol_1
+          || LvlSet->Eval(Trafo, ipB) < -tol_1)
       {
          interior = false;
       }
 
-      if (LvlSet->Eval(Trafo, ipA) > -1e-12
-          && LvlSet->Eval(Trafo, ipB) > -1e-12)
+      if (LvlSet->Eval(Trafo, ipA) > -tol_1
+          && LvlSet->Eval(Trafo, ipB) > -tol_1)
       {
          layout = Layout::inside;
       }
-      else if (LvlSet->Eval(Trafo, ipA) > 1e-15
+      else if (LvlSet->Eval(Trafo, ipA) > tol_2
                && LvlSet->Eval(Trafo, ipB) <= 0.)
       {
          layout = Layout::intersected;
       }
       else if (LvlSet->Eval(Trafo, ipA) <= 0.
-               && LvlSet->Eval(Trafo, ipB) > 1e-15)
+               && LvlSet->Eval(Trafo, ipB) > tol_2)
       {
          layout = Layout::intersected;
          Vector temp(pointA.Size());
@@ -399,10 +399,10 @@ void MomentFittingIntRules::ComputeSurfaceWeights2D(ElementTransformation& Tr)
          IntegrationPoint ip;
          Trafo.TransformBack(mid, ip);
 
-         while (LvlSet->Eval(Trafo, ip) > 1e-12
-                || LvlSet->Eval(Trafo, ip) < -1e-12)
+         while (LvlSet->Eval(Trafo, ip) > tol_1
+                || LvlSet->Eval(Trafo, ip) < -tol_1)
          {
-            if (LvlSet->Eval(Trafo, ip) > 1e-12)
+            if (LvlSet->Eval(Trafo, ip) > tol_1)
             {
                pointC = mid;
             }
@@ -539,7 +539,7 @@ void MomentFittingIntRules::ComputeSurfaceWeights2D(ElementTransformation& Tr)
       temp2 = 0.;
       for (int i = 0; i < nBasis; i++)
       {
-         if (SVD.Singularvalue(i) > 1e-12)
+         if (SVD.Singularvalue(i) > tol_1)
          {
             temp2(i) = temp(i) / SVD.Singularvalue(i);
          }
@@ -606,24 +606,24 @@ void MomentFittingIntRules::ComputeVolumeWeights2D(ElementTransformation& Tr,
       IntegrationPoint ipB;
       Trafo.TransformBack(pointB, ipB);
 
-      if (LvlSet->Eval(Trafo, ipA) < -1e-12
-          || LvlSet->Eval(Trafo, ipB) < -1e-12)
+      if (LvlSet->Eval(Trafo, ipA) < -tol_1
+          || LvlSet->Eval(Trafo, ipB) < -tol_1)
       {
          interior = false;
       }
 
-      if (LvlSet->Eval(Trafo, ipA) > -1e-12
-          && LvlSet->Eval(Trafo, ipB) > -1e-12)
+      if (LvlSet->Eval(Trafo, ipA) > -tol_1
+          && LvlSet->Eval(Trafo, ipB) > -tol_1)
       {
          layout = Layout::inside;
       }
-      else if (LvlSet->Eval(Trafo, ipA) > 1e-15
+      else if (LvlSet->Eval(Trafo, ipA) > tol_2
                && LvlSet->Eval(Trafo, ipB) <= 0.)
       {
          layout = Layout::intersected;
       }
       else if (LvlSet->Eval(Trafo, ipA) <= 0.
-               && LvlSet->Eval(Trafo, ipB) > 1e-15)
+               && LvlSet->Eval(Trafo, ipB) > tol_2)
       {
          layout = Layout::intersected;
          Vector temp(pointA.Size());
@@ -648,10 +648,10 @@ void MomentFittingIntRules::ComputeVolumeWeights2D(ElementTransformation& Tr,
          IntegrationPoint ip;
          Trafo.TransformBack(mid, ip);
 
-         while (LvlSet->Eval(Trafo, ip) > 1e-12
-                || LvlSet->Eval(Trafo, ip) < -1e-12)
+         while (LvlSet->Eval(Trafo, ip) > tol_1
+                || LvlSet->Eval(Trafo, ip) < -tol_1)
          {
-            if (LvlSet->Eval(Trafo, ip) > 1e-12)
+            if (LvlSet->Eval(Trafo, ip) > tol_1)
             {
                pointC = mid;
             }
@@ -786,7 +786,7 @@ void MomentFittingIntRules::ComputeVolumeWeights2D(ElementTransformation& Tr,
       VolumeSVD->LeftSingularvectors().MultTranspose(RHS, temp);
       for (int i = 0; i < nBasisVolume; i++)
       {
-         if (VolumeSVD->Singularvalue(i) > 1e-12)
+         if (VolumeSVD->Singularvalue(i) > tol_1)
          {
             temp2(i) = temp(i) / VolumeSVD->Singularvalue(i);
          }
@@ -865,18 +865,18 @@ void MomentFittingIntRules::ComputeSurfaceWeights3D(ElementTransformation& Tr)
       IntegrationPoint ipD;
       Trafo.TransformBack(pointD, ipD);
 
-      if (LvlSet->Eval(Trafo, ipA) < -1e-12
-          || LvlSet->Eval(Trafo, ipB) < -1e-12
-          || LvlSet->Eval(Trafo, ipC) < -1e-12
-          || LvlSet->Eval(Trafo, ipD) < -1e-12)
+      if (LvlSet->Eval(Trafo, ipA) < -tol_1
+          || LvlSet->Eval(Trafo, ipB) < -tol_1
+          || LvlSet->Eval(Trafo, ipC) < -tol_1
+          || LvlSet->Eval(Trafo, ipD) < -tol_1)
       {
          interior = false;
       }
 
-      if (LvlSet->Eval(Trafo, ipA) > -1e-12
-          || LvlSet->Eval(Trafo, ipB) > -1e-12
-          || LvlSet->Eval(Trafo, ipC) > -1e-12
-          || LvlSet->Eval(Trafo, ipD) > -1e-12)
+      if (LvlSet->Eval(Trafo, ipA) > -tol_1
+          || LvlSet->Eval(Trafo, ipB) > -tol_1
+          || LvlSet->Eval(Trafo, ipC) > -tol_1
+          || LvlSet->Eval(Trafo, ipD) > -tol_1)
       {
          element_int = true;
       }
@@ -978,7 +978,7 @@ void MomentFittingIntRules::ComputeSurfaceWeights3D(ElementTransformation& Tr)
       temp2 = 0.;
       for (int i = 0; i < nBasis; i++)
       {
-         if (SVD.Singularvalue(i) > 1e-12)
+         if (SVD.Singularvalue(i) > tol_1)
          {
             temp2(i) = temp(i) / SVD.Singularvalue(i);
          }
@@ -1047,18 +1047,18 @@ void MomentFittingIntRules::ComputeVolumeWeights3D(ElementTransformation& Tr,
       IntegrationPoint ipD;
       Trafo.TransformBack(pointD, ipD);
 
-      if (LvlSet->Eval(Trafo, ipA) < -1e-12
-          || LvlSet->Eval(Trafo, ipB) < -1e-12
-          || LvlSet->Eval(Trafo, ipC) < -1e-12
-          || LvlSet->Eval(Trafo, ipD) < -1e-12)
+      if (LvlSet->Eval(Trafo, ipA) < -tol_1
+          || LvlSet->Eval(Trafo, ipB) < -tol_1
+          || LvlSet->Eval(Trafo, ipC) < -tol_1
+          || LvlSet->Eval(Trafo, ipD) < -tol_1)
       {
          interior = false;
       }
 
-      if (LvlSet->Eval(Trafo, ipA) > -1e-12
-          || LvlSet->Eval(Trafo, ipB) > -1e-12
-          || LvlSet->Eval(Trafo, ipC) > -1e-12
-          || LvlSet->Eval(Trafo, ipD) > -1e-12)
+      if (LvlSet->Eval(Trafo, ipA) > -tol_1
+          || LvlSet->Eval(Trafo, ipB) > -tol_1
+          || LvlSet->Eval(Trafo, ipC) > -tol_1
+          || LvlSet->Eval(Trafo, ipD) > -tol_1)
       {
          element_int = true;
       }
@@ -1159,7 +1159,7 @@ void MomentFittingIntRules::ComputeVolumeWeights3D(ElementTransformation& Tr,
       VolumeSVD->LeftSingularvectors().MultTranspose(RHS, temp);
       temp2 = 0.;
       for (int i = 0; i < nBasisVolume; i++)
-         if (VolumeSVD->Singularvalue(i) > 1e-12)
+         if (VolumeSVD->Singularvalue(i) > tol_1)
          {
             temp2(i) = temp(i) / VolumeSVD->Singularvalue(i);
          }

fem/intrules_cut.hpp

@@ -36,6 +36,17 @@ protected:
    /// Space order for the LS projection.
    int lsOrder;
 
+   /// @name Tolerances used for point comparisons
+   ///@{
+#ifdef MFEM_USE_DOUBLE
+   static constexpr real_t tol_1 = 1e-12;
+   static constexpr real_t tol_2 = 1e-15;
+#elif defined(MFEM_USE_SINGLE)
+   static constexpr real_t tol_1 = 1e-5;
+   static constexpr real_t tol_2 = 1e-7;
+#endif
+   ///@}
+
    /** @brief Constructor to set up the generated cut IntegrationRules.
 
        @param [in] order  Order of the constructed IntegrationRule.

fem/lininteg.cpp

@@ -123,6 +123,35 @@ void DomainLFGradIntegrator::AssembleDeltaElementVect(
    dshape.Mult(Qvec, elvect);
 }
 
+void DomainLFLaplaceIntegrator::AssembleRHSElementVect(const FiniteElement &el,
+                                                       ElementTransformation &Tr,
+                                                       Vector &elvect)
+{
+   int dof = el.GetDof();
+
+   laplace.SetSize(dof);       // vector of size dof
+   elvect.SetSize(dof);
+   elvect = 0.0;
+
+   const IntegrationRule *ir = NULL;//IntRule;
+   if (ir == NULL)
+   {
+      ir = &IntRules.Get(el.GetGeomType(), oa * el.GetOrder() + ob + 4);
+   }
+
+   for (int i = 0; i < ir->GetNPoints(); i++)
+   {
+      const IntegrationPoint &ip = ir->IntPoint(i);
+
+      Tr.SetIntPoint (&ip);
+      real_t val = Tr.Weight() * Q.Eval(Tr, ip) * alpha;
+
+      el.CalcPhysLaplacian(Tr, laplace);
+
+      add(elvect, ip.weight * val, laplace, elvect);
+   }
+}
+
 void BoundaryLFIntegrator::AssembleRHSElementVect(
    const FiniteElement &el, ElementTransformation &Tr, Vector &elvect)
 {

fem/lininteg.hpp

@@ -174,6 +174,29 @@ public:
    using LinearFormIntegrator::AssembleRHSElementVect;
 };
 
+/// Class for domain integrator $ L(v) := (f, \Delta v) $
+class DomainLFLaplaceIntegrator : public LinearFormIntegrator
+{
+private:
+   Vector laplace;
+   Coefficient &Q;
+   real_t alpha;
+   int oa, ob;
+public:
+   /// Constructs the domain integrator $ (Q, \nabla v) $
+   DomainLFLaplaceIntegrator(Coefficient &QF, real_t alp = 1.0, int a = 2,
+                             int b = 0)
+      : Q(QF), oa(a), ob(b)   { alpha = alp; }
+
+   /** Given a particular Finite Element and a transformation (Tr)
+       computes the element right hand side element vector, elvect. */
+   virtual void AssembleRHSElementVect(const FiniteElement &el,
+                                       ElementTransformation &Tr,
+                                       Vector &elvect);
+
+   using LinearFormIntegrator::AssembleRHSElementVect;
+};
+
 
 /// Class for boundary integration $ L(v) := (g, v) $
 class BoundaryLFIntegrator : public LinearFormIntegrator

fem/pgridfunc.cpp

@@ -39,9 +39,10 @@ ParGridFunction::ParGridFunction(ParMesh *pmesh, const GridFunction *gf,
 {
    const FiniteElementSpace *glob_fes = gf->FESpace();
    // duplicate the FiniteElementCollection from 'gf'
-   fec = FiniteElementCollection::New(glob_fes->FEColl()->Name());
+   fec_owned = FiniteElementCollection::New(glob_fes->FEColl()->Name());
    // create a local ParFiniteElementSpace from the global one:
-   fes = pfes = new ParFiniteElementSpace(pmesh, glob_fes, partitioning, fec);
+   fes = pfes = new ParFiniteElementSpace(pmesh, glob_fes, partitioning,
+                                          fec_owned);
    SetSize(pfes->GetVSize());
 
    if (partitioning)
@@ -81,7 +82,7 @@ ParGridFunction::ParGridFunction(ParMesh *pmesh, std::istream &input)
    : GridFunction(pmesh, input)
 {
    // Convert the FiniteElementSpace, fes, to a ParFiniteElementSpace:
-   pfes = new ParFiniteElementSpace(pmesh, fec, fes->GetVDim(),
+   pfes = new ParFiniteElementSpace(pmesh, fec_owned, fes->GetVDim(),
                                     fes->GetOrdering());
    delete fes;
    fes = pfes;
@@ -520,7 +521,7 @@ void ParGridFunction::CountElementsPerVDof(Array<int> &elem_per_vdof) const
 }
 
 void ParGridFunction::GetDerivative(int comp, int der_comp,
-                                    ParGridFunction &der)
+                                    ParGridFunction &der) const
 {
    Array<int> overlap;
    AccumulateAndCountDerivativeValues(comp, der_comp, der, overlap);

fem/pgridfunc.hpp

@@ -231,7 +231,7 @@ public:
    void CountElementsPerVDof(Array<int> &elem_per_vdof) const override;
 
    /// Parallel version of GridFunction::GetDerivative(); see its documentation.
-   void GetDerivative(int comp, int der_comp, ParGridFunction &der);
+   void GetDerivative(int comp, int der_comp, ParGridFunction &der) const;
 
    /** Sets the output vector @a dof_vals to the values of the degrees of
        freedom of element @a el. If @a el is greater than or equal to the number

fem/tmop.cpp

@@ -2949,6 +2949,15 @@ void TMOP_Integrator::EnableSurfaceFitting(const GridFunction &s0,
    MFEM_VERIFY(surf_fit_pos == NULL,
                "Using both fitting approaches is not supported.");
 
+   const int dim = s0.FESpace()->GetMesh()->Dimension();
+   Mesh *mesh = s0.FESpace()->GetMesh();
+   MFEM_VERIFY(mesh->GetNodes()->Size() == dim*s0.Size(),
+               "Mesh and level-set polynomial order must be the same.");
+   const H1_FECollection *fec = dynamic_cast<const H1_FECollection *>
+                                (s0.FESpace()->FEColl());
+   MFEM_VERIFY(fec, "Only H1_FECollection is supported for the surface fitting "
+               "grid function.");
+
    delete surf_fit_gf;
    surf_fit_gf = new GridFunction(s0);
    surf_fit_gf->CountElementsPerVDof(surf_fit_dof_count);
@@ -2987,12 +2996,24 @@ void TMOP_Integrator::EnableSurfaceFitting(const GridFunction &pos,
 void TMOP_Integrator::EnableSurfaceFitting(const ParGridFunction &s0,
                                            const Array<bool> &smarker,
                                            Coefficient &coeff,
-                                           AdaptivityEvaluator &ae)
+                                           AdaptivityEvaluator &ae,
+                                           AdaptivityEvaluator *aegrad,
+                                           AdaptivityEvaluator *aehess)
 {
    // To have both we must duplicate the markers.
    MFEM_VERIFY(surf_fit_pos == NULL,
                "Using both fitting approaches is not supported.");
 
+   const int dim = s0.FESpace()->GetMesh()->Dimension();
+   ParMesh *pmesh = s0.ParFESpace()->GetParMesh();
+   MFEM_VERIFY(pmesh->GetNodes()->Size() == dim*s0.Size(),
+               "Mesh and level-set polynomial order must be the same.");
+   const H1_FECollection *fec = dynamic_cast<const H1_FECollection *>
+                                (s0.FESpace()->FEColl());
+   MFEM_VERIFY(fec, "Only H1_FECollection is supported for the surface fitting "
+               "grid function.");
+
+
    delete surf_fit_gf;
    surf_fit_gf = new GridFunction(s0);
    s0.CountElementsPerVDof(surf_fit_dof_count);
@@ -3000,11 +3021,80 @@ void TMOP_Integrator::EnableSurfaceFitting(const ParGridFunction &s0,
    surf_fit_coeff = &coeff;
    surf_fit_eval = &ae;
 
-   surf_fit_eval->SetParMetaInfo(*s0.ParFESpace()->GetParMesh(),
-                                 *s0.ParFESpace());
+   surf_fit_eval->SetParMetaInfo(*pmesh, *s0.ParFESpace());
    surf_fit_eval->SetInitialField
    (*surf_fit_gf->FESpace()->GetMesh()->GetNodes(), *surf_fit_gf);
-   surf_fit_gf_bg = false;
+
+   if (!aegrad) { return; }
+
+   MFEM_VERIFY(aehess, "AdaptivityEvaluator for Hessians must be provided too.");
+
+   ParFiniteElementSpace *fes = s0.ParFESpace();
+
+   // FE space for gradients.
+   delete surf_fit_grad;
+   H1_FECollection *fec_grad = new H1_FECollection(fec->GetOrder(), dim,
+                                                   fec->GetBasisType());
+   ParFiniteElementSpace *fes_grad = new ParFiniteElementSpace(pmesh, fec_grad,
+                                                               dim);
+   // Initial gradients.
+   surf_fit_grad = new GridFunction(fes_grad);
+   surf_fit_grad->MakeOwner(fec_grad);
+   for (int d = 0; d < dim; d++)
+   {
+      ParGridFunction surf_fit_grad_comp(fes, surf_fit_grad->GetData()+d*s0.Size());
+      s0.GetDerivative(1, d, surf_fit_grad_comp);
+   }
+   surf_fit_eval_grad = aegrad;
+   surf_fit_eval_grad->SetParMetaInfo(*pmesh, *fes_grad);
+   surf_fit_eval_grad->SetInitialField(*pmesh->GetNodes(), *surf_fit_grad);
+
+   // FE space for Hessians.
+   delete surf_fit_hess;
+   H1_FECollection *fec_hess = new H1_FECollection(fec->GetOrder(), dim,
+                                                   fec->GetBasisType());
+   ParFiniteElementSpace *fes_hess = new ParFiniteElementSpace(pmesh, fec_hess,
+                                                               dim*dim);
+   // Initial Hessians.
+   surf_fit_hess = new GridFunction(fes_hess);
+   surf_fit_hess->MakeOwner(fec_hess);
+   int id = 0;
+   for (int d = 0; d < dim; d++)
+   {
+      for (int idir = 0; idir < dim; idir++)
+      {
+         ParGridFunction surf_fit_grad_comp(fes,
+                                            surf_fit_grad->GetData()+d*s0.Size());
+         ParGridFunction surf_fit_hess_comp(fes,
+                                            surf_fit_hess->GetData()+id*s0.Size());
+         surf_fit_grad_comp.GetDerivative(1, idir, surf_fit_hess_comp);
+         id++;
+      }
+   }
+   surf_fit_eval_hess = aehess;
+   surf_fit_eval_hess->SetParMetaInfo(*pmesh, *fes_hess);
+   surf_fit_eval_hess->SetInitialField(*pmesh->GetNodes(), *surf_fit_hess);
+
+   // Store DOF indices that are marked for fitting. Used to reduce work for
+   // transferring information between source/background and current mesh.
+   surf_fit_marker_dof_index.SetSize(0);
+#ifdef MFEM_USE_GSLIB
+   if (dynamic_cast<InterpolatorFP *>(surf_fit_eval) &&
+       dynamic_cast<InterpolatorFP *>(surf_fit_eval_grad) &&
+       dynamic_cast<InterpolatorFP *>(surf_fit_eval_hess))
+   {
+      for (int i = 0; i < surf_fit_marker->Size(); i++)
+      {
+         if ((*surf_fit_marker)[i] == true)
+         {
+            surf_fit_marker_dof_index.Append(i);
+         }
+      }
+   }
+#endif
+
+   *surf_fit_grad = 0.0;
+   *surf_fit_hess = 0.0;
 }
 
 void TMOP_Integrator::EnableSurfaceFittingFromSource(
@@ -3022,16 +3112,17 @@ void TMOP_Integrator::EnableSurfaceFittingFromSource(
    // Setup for level set function
    delete surf_fit_gf;
    surf_fit_gf = new GridFunction(s0);
-   *surf_fit_gf = 0.0;
    surf_fit_marker = &smarker;
    surf_fit_coeff = &coeff;
    surf_fit_eval = &ae;
 
-   surf_fit_gf_bg = true;
    surf_fit_eval->SetParMetaInfo(*s_bg.ParFESpace()->GetParMesh(),
                                  *s_bg.ParFESpace());
    surf_fit_eval->SetInitialField
    (*s_bg.FESpace()->GetMesh()->GetNodes(), s_bg);
+   GridFunction *nodes = s0.FESpace()->GetMesh()->GetNodes();
+   surf_fit_eval->ComputeAtNewPosition(*nodes, *surf_fit_gf,
+                                       nodes->FESpace()->GetOrdering());
 
    // Setup for gradient on background mesh
    MFEM_VERIFY(s_bg_grad.ParFESpace()->GetOrdering() ==
@@ -3041,11 +3132,11 @@ void TMOP_Integrator::EnableSurfaceFittingFromSource(
    delete surf_fit_grad;
    surf_fit_grad = new GridFunction(s0_grad);
    *surf_fit_grad = 0.0;
-   surf_fit_eval_bg_grad = &age;
-   surf_fit_eval_bg_hess = &ahe;
-   surf_fit_eval_bg_grad->SetParMetaInfo(*s_bg_grad.ParFESpace()->GetParMesh(),
-                                         *s_bg_grad.ParFESpace());
-   surf_fit_eval_bg_grad->SetInitialField
+   surf_fit_eval_grad = &age;
+   surf_fit_eval_hess = &ahe;
+   surf_fit_eval_grad->SetParMetaInfo(*s_bg_grad.ParFESpace()->GetParMesh(),
+                                      *s_bg_grad.ParFESpace());
+   surf_fit_eval_grad->SetInitialField
    (*s_bg_grad.FESpace()->GetMesh()->GetNodes(), s_bg_grad);
 
    // Setup for Hessian on background mesh
@@ -3056,9 +3147,9 @@ void TMOP_Integrator::EnableSurfaceFittingFromSource(
    delete surf_fit_hess;
    surf_fit_hess = new GridFunction(s0_hess);
    *surf_fit_hess = 0.0;
-   surf_fit_eval_bg_hess->SetParMetaInfo(*s_bg_hess.ParFESpace()->GetParMesh(),
-                                         *s_bg_hess.ParFESpace());
-   surf_fit_eval_bg_hess->SetInitialField
+   surf_fit_eval_hess->SetParMetaInfo(*s_bg_hess.ParFESpace()->GetParMesh(),
+                                      *s_bg_hess.ParFESpace());
+   surf_fit_eval_hess->SetInitialField
    (*s_bg_hess.FESpace()->GetMesh()->GetNodes(), s_bg_hess);
 
    // Count number of zones that share each of the DOFs
@@ -3863,7 +3954,7 @@ void TMOP_Integrator::AssembleElemVecSurfFit(const FiniteElement &el_x,
 
    Vector sigma_e(dof_s);
    DenseMatrix surf_fit_grad_e(dof_s, dim);
-   if (surf_fit_gf || surf_fit_gf_bg)
+   if (surf_fit_gf)
    {
       surf_fit_gf->GetSubVector(vdofs, sigma_e);
 
@@ -3871,7 +3962,7 @@ void TMOP_Integrator::AssembleElemVecSurfFit(const FiniteElement &el_x,
       // The FE coefficients of the gradient go in surf_fit_grad_e.
       Vector grad_ptr(surf_fit_grad_e.GetData(), dof_s * dim);
       DenseMatrix grad_phys; // This will be (dof x dim, dof).
-      if (surf_fit_gf_bg)
+      if (surf_fit_grad)
       {
          surf_fit_grad->FESpace()->GetElementVDofs(el_id, dofs);
          surf_fit_grad->GetSubVector(dofs, grad_ptr);
@@ -3945,7 +4036,7 @@ void TMOP_Integrator::AssembleElemGradSurfFit(const FiniteElement &el_x,
    Vector sigma_e(dof_s);
    DenseMatrix surf_fit_grad_e(dof_s, dim);
    DenseMatrix surf_fit_hess_e(dof_s, dim*dim);
-   if (surf_fit_gf || surf_fit_gf_bg)
+   if (surf_fit_gf)
    {
       surf_fit_gf->GetSubVector(vdofs, sigma_e);
 
@@ -3953,7 +4044,7 @@ void TMOP_Integrator::AssembleElemGradSurfFit(const FiniteElement &el_x,
       // The FE coefficients of the gradient go in surf_fit_grad_e.
       Vector grad_ptr(surf_fit_grad_e.GetData(), dof_s * dim);
       DenseMatrix grad_phys; // This will be (dof x dim, dof).
-      if (surf_fit_gf_bg)
+      if (surf_fit_grad)
       {
          surf_fit_grad->FESpace()->GetElementVDofs(el_id, dofs);
          surf_fit_grad->GetSubVector(dofs, grad_ptr);
@@ -3967,7 +4058,7 @@ void TMOP_Integrator::AssembleElemGradSurfFit(const FiniteElement &el_x,
       // Project the Hessian of sigma in the same space.
       // The FE coefficients of the Hessian go in surf_fit_hess_e.
       Vector hess_ptr(surf_fit_hess_e.GetData(), dof_s*dim*dim);
-      if (surf_fit_gf_bg)
+      if (surf_fit_hess)
       {
          surf_fit_hess->FESpace()->GetElementVDofs(el_id, dofs);
          surf_fit_hess->GetSubVector(dofs, hess_ptr);
@@ -3994,7 +4085,7 @@ void TMOP_Integrator::AssembleElemGradSurfFit(const FiniteElement &el_x,
       Tpr.SetIntPoint(&ip);
       real_t w = surf_fit_normal * surf_fit_coeff->Eval(Tpr, ip);
 
-      if (surf_fit_gf || surf_fit_gf_bg)
+      if (surf_fit_gf)
       {
          Vector gg_ptr(surf_fit_hess_s.GetData(), dim * dim);
          surf_fit_hess_e.GetRow(s, gg_ptr);
@@ -4376,6 +4467,130 @@ void TMOP_Integrator::ComputeMinJac(const Vector &x,
    dx = detv_avg_min / dxscale;
 }
 
+void TMOP_Integrator::RemapSurfaceFittingLevelSetAtNodes(const Vector &new_x,
+                                                         int new_x_ordering)
+{
+   if (!surf_fit_gf) { return; }
+
+   if (surf_fit_marker_dof_index.Size())
+   {
+      // Interpolate information only at DOFs marked for fitting.
+      const int dim = surf_fit_gf->FESpace()->GetMesh()->Dimension();
+      const int cnt = surf_fit_marker_dof_index.Size();
+      const int total_cnt = new_x.Size()/dim;
+      Vector new_x_sorted(cnt*dim);
+      if (new_x_ordering == 0)
+      {
+         for (int d = 0; d < dim; d++)
+         {
+            for (int i = 0; i < cnt; i++)
+            {
+               int dof_index = surf_fit_marker_dof_index[i];
+               new_x_sorted(i + d*cnt) = new_x(dof_index + d*total_cnt);
+            }
+         }
+      }
+      else
+      {
+         for (int i = 0; i < cnt; i++)
+         {
+            int dof_index = surf_fit_marker_dof_index[i];
+            for (int d = 0; d < dim; d++)
+            {
+               new_x_sorted(d + i*dim) = new_x(d + dof_index*dim);
+            }
+         }
+      }
+
+      // Interpolate values of the LS.
+      Vector surf_fit_gf_int, surf_fit_grad_int, surf_fit_hess_int;
+      surf_fit_eval->ComputeAtNewPosition(new_x_sorted, surf_fit_gf_int,
+                                          new_x_ordering);
+      for (int i = 0; i < cnt; i++)
+      {
+         int dof_index = surf_fit_marker_dof_index[i];
+         (*surf_fit_gf)[dof_index] = surf_fit_gf_int(i);
+      }
+
+      // Interpolate gradients of the LS.
+      surf_fit_eval_grad->ComputeAtNewPosition(new_x_sorted, surf_fit_grad_int,
+                                               new_x_ordering);
+      // Assumes surf_fit_grad and surf_fit_gf share the same space
+      const int grad_dim = surf_fit_grad->VectorDim();
+      const int grad_cnt = surf_fit_grad->Size()/grad_dim;
+      if (surf_fit_grad->FESpace()->GetOrdering() == Ordering::byNODES)
+      {
+         for (int d = 0; d < grad_dim; d++)
+         {
+            for (int i = 0; i < cnt; i++)
+            {
+               int dof_index = surf_fit_marker_dof_index[i];
+               (*surf_fit_grad)[dof_index + d*grad_cnt] =
+                  surf_fit_grad_int(i + d*cnt);
+            }
+         }
+      }
+      else
+      {
+         for (int i = 0; i < cnt; i++)
+         {
+            int dof_index = surf_fit_marker_dof_index[i];
+            for (int d = 0; d < grad_dim; d++)
+            {
+               (*surf_fit_grad)[dof_index*grad_dim + d] =
+                  surf_fit_grad_int(i*grad_dim + d);
+            }
+         }
+      }
+
+      // Interpolate Hessians of the LS.
+      surf_fit_eval_hess->ComputeAtNewPosition(new_x_sorted, surf_fit_hess_int,
+                                               new_x_ordering);
+      // Assumes surf_fit_hess and surf_fit_gf share the same space
+      const int hess_dim = surf_fit_hess->VectorDim();
+      const int hess_cnt = surf_fit_hess->Size()/hess_dim;
+      if (surf_fit_hess->FESpace()->GetOrdering() == Ordering::byNODES)
+      {
+         for (int d = 0; d < hess_dim; d++)
+         {
+            for (int i = 0; i < cnt; i++)
+            {
+               int dof_index = surf_fit_marker_dof_index[i];
+               (*surf_fit_hess)[dof_index + d*hess_cnt] =
+                  surf_fit_hess_int(i + d*cnt);
+            }
+         }
+      }
+      else
+      {
+         for (int i = 0; i < cnt; i++)
+         {
+            int dof_index = surf_fit_marker_dof_index[i];
+            for (int d = 0; d < hess_dim; d++)
+            {
+               (*surf_fit_hess)[dof_index*hess_dim + d] =
+                  surf_fit_hess_int(i*hess_dim + d);
+            }
+         }
+      }
+
+   }
+   else
+   {
+      surf_fit_eval->ComputeAtNewPosition(new_x, *surf_fit_gf, new_x_ordering);
+      if (surf_fit_eval_grad)
+      {
+         surf_fit_eval_grad->ComputeAtNewPosition(new_x, *surf_fit_grad,
+                                                  new_x_ordering);
+      }
+      if (surf_fit_eval_hess)
+      {
+         surf_fit_eval_hess->ComputeAtNewPosition(new_x, *surf_fit_hess,
+                                                  new_x_ordering);
+      }
+   }
+}
+
 void TMOP_Integrator::
 UpdateAfterMeshPositionChange(const Vector &x_new,
                               const FiniteElementSpace &x_fes)
@@ -4406,112 +4621,11 @@ UpdateAfterMeshPositionChange(const Vector &x_new,
       adapt_lim_eval->ComputeAtNewPosition(x_new, *adapt_lim_gf, ordering);
    }
 
-   // Update surf_fit_gf if surface fitting is enabled.
+   // Update surf_fit_gf (and optionally its gradients) if surface
+   // fitting is enabled.
    if (surf_fit_gf)
    {
-      if (surf_fit_gf_bg)
-      {
-         // Interpolate information for only DOFs marked for fitting.
-         const int dim = surf_fit_gf->FESpace()->GetMesh()->Dimension();
-         const int cnt = surf_fit_marker_dof_index.Size();
-         const int total_cnt = x_new.Size()/dim;
-         Vector new_x_sorted(cnt*dim);
-         if (ordering == 0)
-         {
-            for (int d = 0; d < dim; d++)
-            {
-               for (int i = 0; i < cnt; i++)
-               {
-                  int dof_index = surf_fit_marker_dof_index[i];
-                  new_x_sorted(i + d*cnt) = x_new(dof_index + d*total_cnt);
-               }
-            }
-         }
-         else
-         {
-            for (int i = 0; i < cnt; i++)
-            {
-               int dof_index = surf_fit_marker_dof_index[i];
-               for (int d = 0; d < dim; d++)
-               {
-                  new_x_sorted(d + i*dim) = x_new(d + dof_index*dim);
-               }
-            }
-         }
-
-         Vector surf_fit_gf_int, surf_fit_grad_int, surf_fit_hess_int;
-         surf_fit_eval->ComputeAtNewPosition(
-            new_x_sorted, surf_fit_gf_int, ordering);
-         for (int i = 0; i < cnt; i++)
-         {
-            int dof_index = surf_fit_marker_dof_index[i];
-            (*surf_fit_gf)[dof_index] = surf_fit_gf_int(i);
-         }
-
-         surf_fit_eval_bg_grad->ComputeAtNewPosition(
-            new_x_sorted, surf_fit_grad_int, ordering);
-         // Assumes surf_fit_grad and surf_fit_gf share the same space
-         const int grad_dim = surf_fit_grad->VectorDim();
-         const int grad_cnt = surf_fit_grad->Size()/grad_dim;
-         if (surf_fit_grad->FESpace()->GetOrdering() == Ordering::byNODES)
-         {
-            for (int d = 0; d < grad_dim; d++)
-            {
-               for (int i = 0; i < cnt; i++)
-               {
-                  int dof_index = surf_fit_marker_dof_index[i];
-                  (*surf_fit_grad)[dof_index + d*grad_cnt] =
-                     surf_fit_grad_int(i + d*cnt);
-               }
-            }
-         }
-         else
-         {
-            for (int i = 0; i < cnt; i++)
-            {
-               int dof_index = surf_fit_marker_dof_index[i];
-               for (int d = 0; d < grad_dim; d++)
-               {
-                  (*surf_fit_grad)[dof_index*dim + d] =
-                     surf_fit_grad_int(i*dim + d);
-               }
-            }
-         }
-
-         surf_fit_eval_bg_hess->ComputeAtNewPosition(
-            new_x_sorted, surf_fit_hess_int, ordering);
-         // Assumes surf_fit_hess and surf_fit_gf share the same space
-         const int hess_dim = surf_fit_hess->VectorDim();
-         const int hess_cnt = surf_fit_hess->Size()/hess_dim;
-         if (surf_fit_hess->FESpace()->GetOrdering() == Ordering::byNODES)
-         {
-            for (int d = 0; d < hess_dim; d++)
-            {
-               for (int i = 0; i < cnt; i++)
-               {
-                  int dof_index = surf_fit_marker_dof_index[i];
-                  (*surf_fit_hess)[dof_index + d*hess_cnt] =
-                     surf_fit_hess_int(i + d*cnt);
-               }
-            }
-         }
-         else
-         {
-            for (int i = 0; i < cnt; i++)
-            {
-               int dof_index = surf_fit_marker_dof_index[i];
-               for (int d = 0; d < hess_dim; d++)
-               {
-                  (*surf_fit_hess)[dof_index*dim + d] =
-                     surf_fit_hess_int(i*dim + d);
-               }
-            }
-         }
-      }
-      else
-      {
-         surf_fit_eval->ComputeAtNewPosition(x_new, *surf_fit_gf, ordering);
-      }
+      RemapSurfaceFittingLevelSetAtNodes(x_new, ordering);
    }
 }
 

fem/tmop.hpp

@@ -1784,12 +1784,11 @@ protected:
    // Fitting to given physical positions.
    TMOP_QuadraticLimiter *surf_fit_limiter; // Owned. Created internally.
    const GridFunction *surf_fit_pos;        // Not owned. Positions to fit.
-   real_t surf_fit_normal;
-   bool surf_fit_gf_bg;
-   GridFunction *surf_fit_grad, *surf_fit_hess;
-   AdaptivityEvaluator *surf_fit_eval_bg_grad, *surf_fit_eval_bg_hess;
-   Array<int> surf_fit_dof_count;
-   Array<int> surf_fit_marker_dof_index;
+   real_t surf_fit_normal;                  // Normalization factor.
+   GridFunction *surf_fit_grad, *surf_fit_hess; // Owned. Created internally.
+   AdaptivityEvaluator *surf_fit_eval_grad, *surf_fit_eval_hess; // Not owned.
+   Array<int> surf_fit_dof_count;            // Number of dofs per node.
+   Array<int> surf_fit_marker_dof_index;     // Indices of nodes to fit.
 
    DiscreteAdaptTC *discr_tc;
 
@@ -1985,6 +1984,10 @@ protected:
    real_t ComputeUntanglerMaxMuBarrier(const Vector &x,
                                        const FiniteElementSpace &fes);
 
+   // Remaps the internal surface fitting gridfunction object at provided
+   // locations.
+   void RemapSurfaceFittingLevelSetAtNodes(const Vector &new_x,
+                                           int new_x_ordering);
 public:
    /** @param[in] m    TMOP_QualityMetric for r-adaptivity (not owned).
        @param[in] tc   Target-matrix construction algorithm to use (not owned).
@@ -2000,9 +2003,8 @@ public:
         surf_fit_marker(NULL), surf_fit_coeff(NULL),
         surf_fit_gf(NULL), surf_fit_eval(NULL),
         surf_fit_limiter(NULL), surf_fit_pos(NULL),
-        surf_fit_normal(1.0),
-        surf_fit_gf_bg(false), surf_fit_grad(NULL), surf_fit_hess(NULL),
-        surf_fit_eval_bg_grad(NULL), surf_fit_eval_bg_hess(NULL),
+        surf_fit_normal(1.0), surf_fit_grad(NULL), surf_fit_hess(NULL),
+        surf_fit_eval_grad(NULL), surf_fit_eval_hess(NULL),
         discr_tc(dynamic_cast<DiscreteAdaptTC *>(tc)),
         fdflag(false), dxscale(1.0e3), fd_call_flag(false), exact_action(false)
    { PA.enabled = false; }
@@ -2103,9 +2105,15 @@ public:
 
 #ifdef MFEM_USE_MPI
    /// Parallel support for surface fitting to the zero level set of a function.
+   /// Here, we add two optional inputs: @a aegrad and @a aehess. When provided,
+   /// the first and second derivative of the input level set are computed on
+   /// the initial mesh, and @a aegrad and @a aehess are used to remap grad_s(x)
+   /// from grad_s0(x0) and hess_s(x) from hess_s0(x0), respectively.
    void EnableSurfaceFitting(const ParGridFunction &s0,
                              const Array<bool> &smarker, Coefficient &coeff,
-                             AdaptivityEvaluator &ae);
+                             AdaptivityEvaluator &ae,
+                             AdaptivityEvaluator *aegrad = NULL,
+                             AdaptivityEvaluator *aehess = NULL);
 
    /** @brief Fitting of certain DOFs in the current mesh to the zero level set
        of a function defined on another (finer) source mesh.

fem/tmop_tools.cpp

@@ -429,11 +429,13 @@ real_t TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
 #endif
 
    real_t scale = 1.0;
-   real_t avg_surf_fit_err, max_surf_fit_err = 0.0;
-   if (surf_fit_max_threshold > 0.0)
+   bool fitting = IsSurfaceFittingEnabled();
+   real_t init_fit_avg_err, init_fit_max_err = 0.0;
+   if (fitting && surf_fit_converge_error)
    {
-      GetSurfaceFittingError(x_out_loc, avg_surf_fit_err, max_surf_fit_err);
-      if (max_surf_fit_err < surf_fit_max_threshold)
+      GetSurfaceFittingError(x_out_loc, init_fit_avg_err, init_fit_max_err);
+      // Check for convergence
+      if (init_fit_max_err < surf_fit_max_err_limit)
       {
          if (print_options.iterations)
          {
@@ -444,11 +446,12 @@ real_t TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
          return scale;
       }
    }
-   if (adapt_inc_count >= max_adapt_inc_count)
+
+   if (surf_fit_adapt_count >= surf_fit_adapt_count_limit)
    {
       if (print_options.iterations)
       {
-         mfem::out << "TMOPNewtonSolver converged "
+         mfem::out << "TMOPNewtonSolver terminated "
                    "based on max number of times surface fitting weight can"
                    "be increased. \n";
       }
@@ -467,7 +470,7 @@ real_t TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
       // reference to detect deteriorations.
       MFEM_VERIFY(min_det_ptr != NULL, " Initial mesh was valid, but"
                   " intermediate mesh is invalid. Contact TMOP Developers.");
-      MFEM_VERIFY(min_detJ_threshold == 0.0,
+      MFEM_VERIFY(min_detJ_limit == 0.0,
                   "This setup is not supported. Contact TMOP Developers.");
       *min_det_ptr = untangle_factor * min_detT_in;
    }
@@ -478,6 +481,7 @@ real_t TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
    bool x_out_ok = false;
    real_t energy_out = 0.0, min_detT_out;
    const real_t norm_in = Norm(r);
+   real_t avg_fit_err, max_fit_err = 0.0;
 
    const real_t detJ_factor = (solver_type == 1) ? 0.25 : 0.5;
    compute_metric_quantile_flag = false;
@@ -488,6 +492,9 @@ real_t TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
    // Perform the line search.
    for (int i = 0; i < 12; i++)
    {
+      avg_fit_err = 0.0;
+      max_fit_err = 0.0;
+
       // Update the mesh and get the L-vector in x_out_loc.
       add(x, -scale, c, x_out);
       if (serial)
@@ -502,7 +509,7 @@ real_t TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
 
       // Check the changes in detJ.
       min_detT_out = ComputeMinDet(x_out_loc, *fes);
-      if (untangling == false && min_detT_out <= min_detJ_threshold)
+      if (untangling == false && min_detT_out <= min_detJ_limit)
       {
          // No untangling, and detJ got negative (or small) -- no good.
          if (print_options.iterations)
@@ -529,18 +536,19 @@ real_t TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
       // Check the changes in total energy.
       ProcessNewState(x_out);
 
-      real_t avg_fit_err, max_fit_err = 0.0;
-      if (surf_fit_max_threshold > 0.0)
+      // Ensure sufficient decrease in fitting error if we are trying to
+      // converge based on error.
+      if (fitting && surf_fit_converge_error)
       {
          GetSurfaceFittingError(x_out_loc, avg_fit_err, max_fit_err);
-      }
-      if (surf_fit_max_threshold > 0.0 && max_fit_err >= 1.2*max_surf_fit_err)
-      {
-         if (print_options.iterations)
+         if (max_fit_err >= 1.2*init_fit_max_err)
          {
-            mfem::out << "Scale = " << scale << " Surf fit err increased.\n";
+            if (print_options.iterations)
+            {
+               mfem::out << "Scale = " << scale << " Surf fit err increased.\n";
+            }
+            scale *= 0.5; continue;
          }
-         scale *= 0.5; continue;
       }
 
       if (serial)
@@ -614,7 +622,7 @@ real_t TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
 
    if (x_out_ok == false) { scale = 0.0; }
 
-   if (surf_fit_scale_factor > 0.0) { update_surf_fit_coeff = true; }
+   if (surf_fit_scale_factor > 0.0) { surf_fit_coeff_update = true; }
    compute_metric_quantile_flag = true;
 
    return scale;
@@ -657,7 +665,7 @@ void TMOPNewtonSolver::GetSurfaceFittingWeight(Array<real_t> &weights) const
    for (int i = 0; i < integs.Size(); i++)
    {
       ti = dynamic_cast<TMOP_Integrator *>(integs[i]);
-      if (ti)
+      if (ti && ti->IsSurfaceFittingEnabled())
       {
          weight = ti->GetSurfaceFittingWeight();
          weights.Append(weight);
@@ -668,8 +676,11 @@ void TMOPNewtonSolver::GetSurfaceFittingWeight(Array<real_t> &weights) const
          Array<TMOP_Integrator *> ati = co->GetTMOPIntegrators();
          for (int j = 0; j < ati.Size(); j++)
          {
-            weight = ati[j]->GetSurfaceFittingWeight();
-            weights.Append(weight);
+            if (ati[j]->IsSurfaceFittingEnabled())
+            {
+               weight = ati[j]->GetSurfaceFittingWeight();
+               weights.Append(weight);
+            }
          }
       }
    }
@@ -716,6 +727,39 @@ void TMOPNewtonSolver::GetSurfaceFittingError(const Vector &x_loc,
    }
 }
 
+bool TMOPNewtonSolver::IsSurfaceFittingEnabled() const
+{
+   const NonlinearForm *nlf = dynamic_cast<const NonlinearForm *>(oper);
+   const Array<NonlinearFormIntegrator*> &integs = *nlf->GetDNFI();
+   TMOP_Integrator *ti  = NULL;
+   TMOPComboIntegrator *co = NULL;
+
+   for (int i = 0; i < integs.Size(); i++)
+   {
+      ti = dynamic_cast<TMOP_Integrator *>(integs[i]);
+      if (ti)
+      {
+         if (ti->IsSurfaceFittingEnabled())
+         {
+            return true;
+         }
+      }
+      co = dynamic_cast<TMOPComboIntegrator *>(integs[i]);
+      if (co)
+      {
+         Array<TMOP_Integrator *> ati = co->GetTMOPIntegrators();
+         for (int j = 0; j < ati.Size(); j++)
+         {
+            if (ati[j]->IsSurfaceFittingEnabled())
+            {
+               return true;
+            }
+         }
+      }
+   }
+   return false;
+}
+
 void TMOPNewtonSolver::ProcessNewState(const Vector &x) const
 {
    const NonlinearForm *nlf = dynamic_cast<const NonlinearForm *>(oper);
@@ -801,38 +845,46 @@ void TMOPNewtonSolver::ProcessNewState(const Vector &x) const
    // adaptive surface fitting is enabled. The idea is to increase the
    // coefficient if the surface fitting error does not sufficiently
    // decrease between subsequent TMOPNewtonSolver iterations.
-   if (update_surf_fit_coeff)
+   if (surf_fit_coeff_update)
    {
       // Get surface fitting errors.
-      GetSurfaceFittingError(x_loc, surf_fit_err_avg, surf_fit_err_max);
+      GetSurfaceFittingError(x_loc, surf_fit_avg_err, surf_fit_max_err);
       // Get array with surface fitting weights.
-      Array<real_t> weights;
-      GetSurfaceFittingWeight(weights);
+      Array<real_t> fitweights;
+      GetSurfaceFittingWeight(fitweights);
 
       if (print_options.iterations)
       {
          mfem::out << "Avg/Max surface fitting error: " <<
-                   surf_fit_err_avg << " " <<
-                   surf_fit_err_max << "\n";
+                   surf_fit_avg_err << " " <<
+                   surf_fit_max_err << "\n";
          mfem::out << "Min/Max surface fitting weight: " <<
-                   weights.Min() << " " << weights.Max() << "\n";
+                   fitweights.Min() << " " << fitweights.Max() << "\n";
       }
 
-      real_t change_surf_fit_err = surf_fit_err_avg_prvs-surf_fit_err_avg;
-      real_t rel_change_surf_fit_err = change_surf_fit_err/surf_fit_err_avg_prvs;
+      real_t change_surf_fit_err = surf_fit_avg_err_prvs-surf_fit_avg_err;
+      real_t rel_change_surf_fit_err = change_surf_fit_err/surf_fit_avg_err_prvs;
+
       // Increase the surface fitting coefficient if the surface fitting error
-      // does not decrease sufficiently.
-      if (rel_change_surf_fit_err < surf_fit_rel_change_threshold)
+      // does not decrease sufficiently. If we are converging based on residual,
+      // also make sure we have not reached the maximum fitting weight and
+      // error threshold.
+      if (rel_change_surf_fit_err < surf_fit_err_rel_change_limit &&
+          (surf_fit_converge_error ||
+           (fitweights.Max() < surf_fit_weight_limit &&
+            surf_fit_max_err > surf_fit_max_err_limit)))
       {
-         UpdateSurfaceFittingWeight(surf_fit_scale_factor);
-         adapt_inc_count += 1;
+         real_t scale_factor = std::min(surf_fit_scale_factor,
+                                        surf_fit_weight_limit/fitweights.Max());
+         UpdateSurfaceFittingWeight(scale_factor);
+         surf_fit_adapt_count += 1;
       }
       else
       {
-         adapt_inc_count = 0;
+         surf_fit_adapt_count = 0;
       }
-      surf_fit_err_avg_prvs = surf_fit_err_avg;
-      update_surf_fit_coeff = false;
+      surf_fit_avg_err_prvs = surf_fit_avg_err;
+      surf_fit_coeff_update = false;
    }
 }
 

fem/tmop_tools.hpp

@@ -134,18 +134,20 @@ protected:
    int solver_type;
    bool parallel;
 
-   // Line search step is rejected if min(detJ) <= min_detJ_threshold.
-   real_t min_detJ_threshold = 0.0;
+   // Line search step is rejected if min(detJ) <= min_detJ_limit.
+   real_t min_detJ_limit = 0.0;
 
    // Surface fitting variables.
-   mutable real_t surf_fit_err_avg_prvs = 10000.0;
-   mutable real_t surf_fit_err_avg, surf_fit_err_max;
-   mutable bool update_surf_fit_coeff = false;
-   real_t surf_fit_max_threshold = -1.0;
-   real_t surf_fit_rel_change_threshold = 0.001;
+   mutable real_t surf_fit_avg_err_prvs = 10000.0;
+   mutable real_t surf_fit_avg_err, surf_fit_max_err;
+   mutable bool surf_fit_coeff_update = false;
+   real_t surf_fit_max_err_limit = -1.0;
+   real_t surf_fit_err_rel_change_limit = 0.001;
    real_t surf_fit_scale_factor = 0.0;
-   mutable int adapt_inc_count = 0;
-   mutable int max_adapt_inc_count = 10;
+   mutable int surf_fit_adapt_count = 0;
+   mutable int surf_fit_adapt_count_limit = 10;
+   mutable real_t surf_fit_weight_limit = 1e10;
+   bool surf_fit_converge_error = false;
 
    // Minimum determinant over the whole mesh. Used for mesh untangling.
    real_t *min_det_ptr = nullptr;
@@ -191,6 +193,9 @@ protected:
    void GetSurfaceFittingWeight(Array<real_t> &weights) const;
    ///@}
 
+   /// Check if surface fitting is enabled.
+   bool IsSurfaceFittingEnabled() const;
+
 public:
 #ifdef MFEM_USE_MPI
    TMOPNewtonSolver(MPI_Comm comm, const IntegrationRule &irule, int type = 0)
@@ -224,38 +229,94 @@ public:
    /// (ii) surface fitting weight.
    virtual void ProcessNewState(const Vector &x) const;
 
-   /** @name Methods for adaptive surface fitting weight. (Experimental) */
-   /// Enable/Disable adaptive surface fitting weight.
-   /// The weight is modified after each TMOPNewtonSolver iteration as:
-   /// w_{k+1} = w_{k} * @a surf_fit_scale_factor if relative change in
-   /// max surface fitting error < @a surf_fit_rel_change_threshold.
-   /// The solver terminates if the maximum surface fitting error does
-   /// not sufficiently decrease for @a max_adapt_inc_count consecutive
-   /// solver iterations or if the max error falls below @a surf_fit_max_threshold.
-   void EnableAdaptiveSurfaceFitting()
-   {
-      surf_fit_scale_factor = 10.0;
-      surf_fit_rel_change_threshold = 0.001;
-   }
+   /** @name Methods for adaptive surface fitting.
+       \brief These methods control the behavior of the weight and the
+       termination of the solver. (Experimental)
+
+       Adaptive fitting weight: The weight is modified after each
+       TMOPNewtonSolver iteration as:
+       w_{k+1} = w_{k} * \ref surf_fit_scale_factor if the relative
+       change in average fitting error < \ref surf_fit_err_rel_change_limit.
+       When converging based on the residual, we enforce the fitting weight
+       to be at-most \ref surf_fit_weight_limit, and increase it only if the
+       fitting error is below user prescribed threshold
+       (\ref surf_fit_max_err_limit).
+       See \ref SetAdaptiveSurfaceFittingScalingFactor and
+       \ref SetAdaptiveSurfaceFittingRelativeChangeThreshold.
+
+       Note that the solver stops if the maximum surface fitting error
+       does not sufficiently decrease for \ref surf_fit_adapt_count_limit (default 10)
+       consecutive increments of the fitting weight during weight adaptation.
+       This typically occurs when the mesh cannot align with the level-set
+       without degrading element quality.
+       See \ref SetMaxNumberofIncrementsForAdaptiveFitting.
+
+       Convergence criterion: There are two modes, residual- and error-based,
+       which can be toggled using \ref SetSurfaceFittingConvergenceBasedOnError.
+
+       (i) Residual based (default): Stop when the norm of the gradient of the
+       TMOP objective reaches the prescribed tolerance. This method is best used
+       with a reasonable value for \ref surf_fit_weight_limit when the
+       adaptive surface fitting scheme is used. See method
+       \ref SetSurfaceFittingWeightLimit.
+
+       (ii) Error based: Stop when the maximum fitting error
+       reaches the user-prescribed threshold, \ref surf_fit_max_err_limit.
+       In this case, \ref surf_fit_weight_limit is ignored during weight
+       adaptation.
+   */
+   ///@{
    void SetAdaptiveSurfaceFittingScalingFactor(real_t factor)
    {
+      MFEM_VERIFY(factor > 1.0, "Scaling factor must be greater than 1.");
       surf_fit_scale_factor = factor;
    }
    void SetAdaptiveSurfaceFittingRelativeChangeThreshold(real_t threshold)
    {
-      surf_fit_rel_change_threshold = threshold;
+      surf_fit_err_rel_change_limit = threshold;
    }
+   /// Used for stopping based on the number of consecutive failed weight
+   /// adaptation iterations.
+   // TODO: Rename to SetMaxNumberofIncrementsForAdaptiveSurfaceFitting
+   // in future.
    void SetMaxNumberofIncrementsForAdaptiveFitting(int count)
    {
-      max_adapt_inc_count = count;
+      surf_fit_adapt_count_limit = count;
    }
+   /// Used for error-based surface fitting termination.
    void SetTerminationWithMaxSurfaceFittingError(real_t max_error)
    {
-      surf_fit_max_threshold = max_error;
+      surf_fit_max_err_limit = max_error;
+      surf_fit_converge_error = true;
    }
+   /// Could be used with both error-based or residual-based convergence.
+   void SetSurfaceFittingMaxErrorLimit(real_t max_error)
+   {
+      surf_fit_max_err_limit = max_error;
+   }
+   /// Used for residual-based surface fitting termination.
+   void SetSurfaceFittingWeightLimit(real_t weight)
+   {
+      surf_fit_weight_limit = weight;
+   }
+   /// Toggle convergence based on residual or error.
+   void SetSurfaceFittingConvergenceBasedOnError(bool mode)
+   {
+      surf_fit_converge_error = mode;
+      if (surf_fit_converge_error)
+      {
+         MFEM_VERIFY(surf_fit_max_err_limit >= 0,
+                     "Fitting error based convergence requires the user to "
+                     "first set the error threshold."
+                     "See SetTerminationWithMaxSurfaceFittingError");
+      }
+   }
+   ///@}
+
+   /// Set minimum determinant enforced during line-search.
    void SetMinimumDeterminantThreshold(real_t threshold)
    {
-      min_detJ_threshold = threshold;
+      min_detJ_limit = threshold;
    }
 
    virtual void Mult(const Vector &b, Vector &x) const

fem/transfer.cpp

@@ -1233,6 +1233,8 @@ void PRefinementTransferOperator::Mult(const Vector& x, Vector& y) const
 
    int vdim = lFESpace.GetVDim();
 
+   y = 0.0;
+
    for (int i = 0; i < mesh->GetNE(); i++)
    {
       DofTransformation * doftrans_h = hFESpace.GetElementDofs(i, h_dofs);

general/backends.hpp

@@ -63,9 +63,9 @@
 #define MFEM_FOREACH_THREAD(i,k,N) for(int i=0; i<N; i++)
 #endif
 
-// 'double' atomicAdd implementation for previous versions of CUDA
+// 'double' and 'float' atomicAdd implementation for previous versions of CUDA
 #if defined(MFEM_USE_CUDA) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 600
-MFEM_DEVICE inline real_t atomicAdd(real_t *add, real_t val)
+MFEM_DEVICE inline mfem::real_t atomicAdd(mfem::real_t *add, mfem::real_t val)
 {
    unsigned long long int *ptr = (unsigned long long int *) add;
    unsigned long long int old = *ptr, reg;

general/isockstream.cpp

@@ -51,7 +51,7 @@ int isockstream::establish()
 {
    // char myname[129];
    char   myname[] = "localhost";
-   int    sfd;
+   int    sfd = -1;
    struct addrinfo hints, *res, *rp;
 
    memset(&hints, 0, sizeof(hints));

general/socketstream.cpp

@@ -134,7 +134,7 @@ int socketbuf::open(const char hostname[], int port)
       {
          closesocket(socket_descriptor);
          socket_descriptor = -2;
-         return -1;
+         continue;
       }
 #endif
 
@@ -148,7 +148,7 @@ int socketbuf::open(const char hostname[], int port)
    }
 
    freeaddrinfo(res);
-   return 0;
+   return (socket_descriptor < 0) ? -1 : 0;
 }
 
 int socketbuf::close()

general/table.cpp

@@ -37,6 +37,78 @@ Table::Table(const Table &table)
    }
 }
 
+Table::Table(const Table &table1,
+             const Table &table2, int offset)
+{
+   MFEM_ASSERT(table1.size == table2.size,
+               "Tables have different sizes can not merge.");
+   size = table1.size;
+
+   const int nnz = table1.I[size] + table2.I[size];
+   I.New(size+1, table1.I.GetMemoryType());
+   J.New(nnz, table1.J.GetMemoryType());
+
+   I[0] = 0;
+   Array<int> row;
+   for (int i = 0; i < size; i++)
+   {
+      I[i+1] = I[i];
+
+      table1.GetRow(i, row);
+      for (int r = 0; r < row.Size(); r++,  I[i+1] ++)
+      {
+         J[ I[i+1] ] = row[r];
+      }
+
+      table2.GetRow(i, row);
+      for (int r = 0; r < row.Size(); r++,  I[i+1] ++)
+      {
+         J[ I[i+1] ] = (row[r] < 0) ? row[r] - offset : row[r] + offset;
+      }
+
+   }
+}
+
+Table::Table(const Table &table1,
+             const Table &table2, int offset2,
+             const Table &table3, int offset3)
+{
+   MFEM_ASSERT(table1.size == table2.size,
+               "Tables have different sizes can not merge.");
+   MFEM_ASSERT(table1.size == table3.size,
+               "Tables have different sizes can not merge.");
+   size = table1.size;
+
+   const int nnz = table1.I[size] + table2.I[size] + table3.I[size];
+   I.New(size+1, table1.I.GetMemoryType());
+   J.New(nnz, table1.J.GetMemoryType());
+
+   I[0] = 0;
+   Array<int> row;
+   for (int i = 0; i < size; i++)
+   {
+      I[i+1] = I[i];
+
+      table1.GetRow(i, row);
+      for (int r = 0; r < row.Size(); r++,  I[i+1] ++)
+      {
+         J[ I[i+1] ] = row[r];
+      }
+
+      table2.GetRow(i, row);
+      for (int r = 0; r < row.Size(); r++,  I[i+1] ++)
+      {
+         J[ I[i+1] ] = (row[r] < 0) ? row[r] - offset2 : row[r] + offset2;
+      }
+
+      table3.GetRow(i, row);
+      for (int r = 0; r < row.Size(); r++,  I[i+1] ++)
+      {
+         J[ I[i+1] ] = (row[r] < 0) ? row[r] - offset3 : row[r] + offset3;
+      }
+   }
+}
+
 Table& Table::operator=(const Table &rhs)
 {
    Clear();

general/table.hpp

@@ -58,6 +58,14 @@ public:
    /// Copy constructor
    Table(const Table &);
 
+   /** Merge constructors
+       This is used to combine two or three tables into one table.*/
+   Table(const Table &table1,
+         const Table &table2, int offset2);
+   Table(const Table &table1,
+         const Table &table2, int offset2,
+         const Table &table3, int offset3);
+
    /// Assignment operator: deep copy
    Table& operator=(const Table &rhs);
 

linalg/CMakeLists.txt

@@ -44,6 +44,7 @@ list(APPEND HDRS
   handle.hpp
   invariants.hpp
   kernels.hpp
+  lapack.hpp
   linalg.hpp
   matrix.hpp
   ode.hpp

linalg/complex_densemat.cpp

@@ -10,60 +10,9 @@
 // CONTRIBUTING.md for details.
 
 #include "complex_densemat.hpp"
+#include "lapack.hpp"
 #include <complex>
 
-#ifdef MFEM_USE_LAPACK
-#ifdef MFEM_USE_SINGLE
-extern "C" void
-cgetrf_(int *, int *, std::complex<float> *, int *, int *, int *);
-extern "C" void
-cgetrs_(char *, int *, int *, std::complex<float> *, int *, int *,
-        std::complex<float> *, int *, int *);
-extern "C" void
-cgetri_(int *, std::complex<float> *, int *, int *,
-        std::complex<float> *, int *, int *);
-extern "C" void
-ctrsm_(char *, char *, char *, char *, int *, int *, std::complex<float> *,
-       std::complex<float> *, int *, std::complex<float> *, int *);
-extern "C" void
-cpotrf_(char *, int *, std::complex<float> *, int *, int *);
-
-extern "C" void
-ctrtrs_(char *, char*, char *, int *, int *, std::complex<float> *, int *,
-        std::complex<float> *, int *, int *);
-extern "C" void
-cpotri_(char *, int *, std::complex<float> *, int*, int *);
-
-extern "C" void
-cpotrs_(char *, int *, int *, std::complex<float> *, int *,
-        std::complex<float> *, int *, int *);
-#elif defined MFEM_USE_DOUBLE
-extern "C" void
-zgetrf_(int *, int *, std::complex<double> *, int *, int *, int *);
-extern "C" void
-zgetrs_(char *, int *, int *, std::complex<double> *, int *, int *,
-        std::complex<double> *, int *, int *);
-extern "C" void
-zgetri_(int *, std::complex<double> *, int *, int *,
-        std::complex<double> *, int *, int *);
-extern "C" void
-ztrsm_(char *, char *, char *, char *, int *, int *, std::complex<double> *,
-       std::complex<double> *, int *, std::complex<double> *, int *);
-extern "C" void
-zpotrf_(char *, int *, std::complex<double> *, int *, int *);
-
-extern "C" void
-ztrtrs_(char *, char*, char *, int *, int *, std::complex<double> *, int *,
-        std::complex<double> *, int *, int *);
-extern "C" void
-zpotri_(char *, int *, std::complex<double> *, int*, int *);
-
-extern "C" void
-zpotrs_(char *, int *, int *, std::complex<double> *, int *,
-        std::complex<double> *, int *, int *);
-#endif
-#endif
-
 namespace mfem
 {
 
@@ -175,35 +124,17 @@ ComplexDenseMatrix * ComplexDenseMatrix::ComputeInverse()
    std::complex<real_t> qwork, *work;
    int    info;
 
-#ifdef MFEM_USE_SINGLE
-   cgetrf_(&w, &w, data, &w, ipiv, &info);
-#elif defined MFEM_USE_DOUBLE
-   zgetrf_(&w, &w, data, &w, ipiv, &info);
-#else
-   MFEM_ABORT("Floating point type undefined");
-#endif
+   MFEM_LAPACK_COMPLEX(getrf_)(&w, &w, data, &w, ipiv, &info);
    if (info)
    {
       mfem_error("DenseMatrix::Invert() : Error in ZGETRF");
    }
 
-#ifdef MFEM_USE_SINGLE
-   cgetri_(&w, data, &w, ipiv, &qwork, &lwork, &info);
-#elif defined MFEM_USE_DOUBLE
-   zgetri_(&w, data, &w, ipiv, &qwork, &lwork, &info);
-#else
-   MFEM_ABORT("Floating point type undefined");
-#endif
+   MFEM_LAPACK_COMPLEX(getri_)(&w, data, &w, ipiv, &qwork, &lwork, &info);
    lwork = (int) qwork.real();
    work = new std::complex<real_t>[lwork];
 
-#ifdef MFEM_USE_SINGLE
-   cgetri_(&w, data, &w, ipiv, work, &lwork, &info);
-#elif defined MFEM_USE_DOUBLE
-   zgetri_(&w, data, &w, ipiv, work, &lwork, &info);
-#else
-   MFEM_ABORT("Floating point type undefined");
-#endif
+   MFEM_LAPACK_COMPLEX(getri_)(&w, data, &w, ipiv, work, &lwork, &info);
    if (info)
    {
       mfem_error("DenseMatrix::Invert() : Error in ZGETRI");
@@ -493,11 +424,7 @@ bool ComplexLUFactors::Factor(int m, real_t TOL)
 #ifdef MFEM_USE_LAPACK
    int info = 0;
    MFEM_VERIFY(data, "Matrix data not set");
-#ifdef MFEM_USE_SINGLE
-   if (m) { cgetrf_(&m, &m, data, &m, ipiv, &info); }
-#elif defined MFEM_USE_DOUBLE
-   if (m) { zgetrf_(&m, &m, data, &m, ipiv, &info); }
-#endif
+   if (m) { MFEM_LAPACK_COMPLEX(getrf_)(&m, &m, data, &m, ipiv, &info); }
    return info == 0;
 #else
    // compiling without LAPACK
@@ -659,13 +586,10 @@ void ComplexLUFactors::Solve(int m, int n, real_t *X_r, real_t * X_i) const
    std::complex<real_t> * x = ComplexFactors::RealToComplex(m*n,X_r,X_i);
    char trans = 'N';
    int  info = 0;
-#ifdef MFEM_USE_SINGLE
-   if (m > 0 && n > 0) { cgetrs_(&trans, &m, &n, data, &m, ipiv, x, &m, &info); }
-#elif defined MFEM_USE_DOUBLE
-   if (m > 0 && n > 0) { zgetrs_(&trans, &m, &n, data, &m, ipiv, x, &m, &info); }
-#else
-   MFEM_ABORT("Floating point type undefined");
-#endif
+   if (m > 0 && n > 0)
+   {
+      MFEM_LAPACK_COMPLEX(getrs_)(&trans, &m, &n, data, &m, ipiv, x, &m, &info);
+   }
    MFEM_VERIFY(!info, "LAPACK: error in ZGETRS");
    ComplexFactors::ComplexToReal(m*n,x,X_r,X_i);
    delete [] x;
@@ -685,15 +609,8 @@ void ComplexLUFactors::RightSolve(int m, int n, real_t *X_r, real_t * X_i) const
    if (m > 0 && n > 0)
    {
       std::complex<real_t> alpha(1.0,0.0);
-#ifdef MFEM_USE_SINGLE
-      ctrsm_(&side,&u_ch,&n_ch,&n_ch,&n,&m,&alpha,data,&m,X,&n);
-      ctrsm_(&side,&l_ch,&n_ch,&u_ch,&n,&m,&alpha,data,&m,X,&n);
-#elif defined MFEM_USE_DOUBLE
-      ztrsm_(&side,&u_ch,&n_ch,&n_ch,&n,&m,&alpha,data,&m,X,&n);
-      ztrsm_(&side,&l_ch,&n_ch,&u_ch,&n,&m,&alpha,data,&m,X,&n);
-#else
-      MFEM_ABORT("Floating point type undefined");
-#endif
+      MFEM_LAPACK_COMPLEX(trsm_)(&side,&u_ch,&n_ch,&n_ch,&n,&m,&alpha,data,&m,X,&n);
+      MFEM_LAPACK_COMPLEX(trsm_)(&side,&l_ch,&n_ch,&u_ch,&n,&m,&alpha,data,&m,X,&n);
    }
 #else
    // compiling without LAPACK
@@ -815,13 +732,7 @@ bool ComplexCholeskyFactors::Factor(int m, real_t TOL)
    int info = 0;
    char uplo = 'L';
    MFEM_VERIFY(data, "Matrix data not set");
-#ifdef MFEM_USE_SINGLE
-   if (m) {cpotrf_(&uplo, &m, data, &m, &info);}
-#elif defined MFEM_USE_DOUBLE
-   if (m) {zpotrf_(&uplo, &m, data, &m, &info);}
-#else
-   MFEM_ABORT("Floating point type undefined");
-#endif
+   if (m) { MFEM_LAPACK_COMPLEX(potrf_)(&uplo, &m, data, &m, &info); }
    return info == 0;
 #else
    // Cholesky–Crout algorithm
@@ -921,13 +832,8 @@ void ComplexCholeskyFactors::LSolve(int m, int n, real_t * X_r,
    char diag = 'N';
    int info = 0;
 
-#ifdef MFEM_USE_SINGLE
-   ctrtrs_(&uplo, &trans, &diag, &m, &n, data, &m, x, &m, &info);
-#elif defined MFEM_USE_DOUBLE
-   ztrtrs_(&uplo, &trans, &diag, &m, &n, data, &m, x, &m, &info);
-#else
-   MFEM_ABORT("Floating point type undefined");
-#endif
+   MFEM_LAPACK_COMPLEX(trtrs_)(&uplo, &trans, &diag, &m, &n, data, &m, x, &m,
+                               &info);
    MFEM_VERIFY(!info, "ComplexCholeskyFactors:LSolve:: info");
 #else
    for (int k = 0; k < n; k++)
@@ -960,13 +866,8 @@ void ComplexCholeskyFactors::USolve(int m, int n, real_t * X_r,
    char diag = 'N';
    int info = 0;
 
-#ifdef MFEM_USE_SINGLE
-   ctrtrs_(&uplo, &trans, &diag, &m, &n, data, &m, x, &m, &info);
-#elif defined MFEM_USE_DOUBLE
-   ztrtrs_(&uplo, &trans, &diag, &m, &n, data, &m, x, &m, &info);
-#else
-   MFEM_ABORT("Floating point type undefined");
-#endif
+   MFEM_LAPACK_COMPLEX(trtrs_)(&uplo, &trans, &diag, &m, &n, data, &m, x, &m,
+                               &info);
    MFEM_VERIFY(!info, "ComplexCholeskyFactors:USolve:: info");
 #else
    // X <- L^{-t} X
@@ -994,13 +895,7 @@ void ComplexCholeskyFactors::Solve(int m, int n, real_t * X_r,
    char uplo = 'L';
    int info = 0;
    std::complex<real_t> *x = ComplexFactors::RealToComplex(m*n,X_r,X_i);
-#ifdef MFEM_USE_SINGLE
-   cpotrs_(&uplo, &m, &n, data, &m, x, &m, &info);
-#elif defined MFEM_USE_DOUBLE
-   zpotrs_(&uplo, &m, &n, data, &m, x, &m, &info);
-#else
-   MFEM_ABORT("Floating point type undefined");
-#endif
+   MFEM_LAPACK_COMPLEX(potrs_)(&uplo, &m, &n, data, &m, x, &m, &info);
    MFEM_VERIFY(!info, "ComplexCholeskyFactors:Solve:: info");
    ComplexFactors::ComplexToReal(m*n,x,X_r,X_i);
    delete x;
@@ -1026,15 +921,8 @@ void ComplexCholeskyFactors::RightSolve(int m, int n, real_t * X_r,
    std::complex<real_t> alpha(1.0,0.0);
    if (m > 0 && n > 0)
    {
-#ifdef MFEM_USE_SINGLE
-      ctrsm_(&side,&uplo,&transt,&diag,&n,&m,&alpha,data,&m,x,&n);
-      ctrsm_(&side,&uplo,&trans,&diag,&n,&m,&alpha,data,&m,x,&n);
-#elif defined MFEM_USE_DOUBLE
-      ztrsm_(&side,&uplo,&transt,&diag,&n,&m,&alpha,data,&m,x,&n);
-      ztrsm_(&side,&uplo,&trans,&diag,&n,&m,&alpha,data,&m,x,&n);
-#else
-      MFEM_ABORT("Floating point type undefined");
-#endif
+      MFEM_LAPACK_COMPLEX(trsm_)(&side,&uplo,&transt,&diag,&n,&m,&alpha,data,&m,x,&n);
+      MFEM_LAPACK_COMPLEX(trsm_)(&side,&uplo,&trans,&diag,&n,&m,&alpha,data,&m,x,&n);
    }
 #else
    // X <- X L^{-H}
@@ -1085,13 +973,7 @@ void ComplexCholeskyFactors::GetInverseMatrix(int m, real_t * X_r,
    }
    char uplo = 'L';
    int info = 0;
-#ifdef MFEM_USE_SINGLE
-   cpotri_(&uplo, &m, X, &m, &info);
-#elif defined MFEM_USE_DOUBLE
-   zpotri_(&uplo, &m, X, &m, &info);
-#else
-   MFEM_ABORT("Floating point type undefined");
-#endif
+   MFEM_LAPACK_COMPLEX(potri_)(&uplo, &m, X, &m, &info);
    MFEM_VERIFY(!info, "ComplexCholeskyFactors:GetInverseMatrix:: info");
    // fill in the upper triangular part
    for (int i = 0; i<m; i++)

linalg/densemat.hpp

@@ -207,6 +207,10 @@ public:
    /// Replaces the current matrix with its square root inverse
    void SquareRootInverse();
 
+   /// Replaces the current matrix with its exponential
+   /// (currently only supports 2x2 matrices)
+   void Exponential();
+
    /// Calculates the determinant of the matrix
    /// (optimized for 2x2, 3x3, and 4x4 matrices)
    real_t Det() const;
@@ -268,35 +272,51 @@ public:
    /// Compute the square of the Frobenius norm of the matrix
    real_t FNorm2() const { real_t s, n2; FNorm(s, n2); return s*s*n2; }
 
-   /// Compute eigenvalues of A x = ev x where A = *this
+   /** Compute eigenvalues of A x = ev x where A = *this
+       A is assumed to be symmetric. */
    void Eigenvalues(Vector &ev)
    { Eigensystem(ev); }
 
-   /// Compute eigenvalues and eigenvectors of A x = ev x where A = *this
+   /** Compute ith eigenvalue of A x = ev x where A = *this
+       A is assumed to be symmetric. */
+   real_t Eigenvalue(int i = -1);
+
+   /** Compute eigenvalues and eigenvectors of A x = ev x where A = *this
+       A is assumed to be symmetric. */
    void Eigenvalues(Vector &ev, DenseMatrix &evect)
    { Eigensystem(ev, &evect); }
 
-   /// Compute eigenvalues and eigenvectors of A x = ev x where A = *this
+   /** Compute eigenvalues and eigenvectors of A x = ev x where A = *this
+       A is assumed to be symmetric. */
    void Eigensystem(Vector &ev, DenseMatrix &evect)
    { Eigensystem(ev, &evect); }
 
-   /** Compute generalized eigenvalues and eigenvectors of A x = ev B x,
-       where A = *this */
+   /** Compute generalized eigenvalues of A x = ev B x, where A = *this
+       A and B are assumed to be symmetric. */
    void Eigenvalues(DenseMatrix &b, Vector &ev)
    { Eigensystem(b, ev); }
 
-   /// Compute generalized eigenvalues of A x = ev B x, where A = *this
+   /** Compute ith eigenvalue of A x = ev B x where A = *this
+       A and B are assumed to be symmetric. */
+   real_t Eigenvalue(DenseMatrix &b, int i = -1);
+
+   /** Compute generalized eigenvalues and eigenvectors of A x = ev B x,
+       where A = *this. A and B are assumed to be symmetric.*/
    void Eigenvalues(DenseMatrix &b, Vector &ev, DenseMatrix &evect)
    { Eigensystem(b, ev, &evect); }
 
    /** Compute generalized eigenvalues and eigenvectors of A x = ev B x,
-       where A = *this */
+       where A = *this. A and B are assumed to be symmetric.*/
    void Eigensystem(DenseMatrix &b, Vector &ev, DenseMatrix &evect)
    { Eigensystem(b, ev, &evect); }
 
    void SingularValues(Vector &sv) const;
    int Rank(real_t tol) const;
 
+   /** Compute the Null Space of the matrix, such that A x = 0,
+       where A = *this* and x is a column of ns */
+   void NullSpace(DenseMatrix &ns, real_t tol);
+
    /// Return the i-th singular value (decreasing order) of NxN matrix, N=1,2,3.
    real_t CalcSingularvalue(const int i) const;
 
@@ -318,7 +338,6 @@ public:
    void SetCol(int c, const real_t* col);
    void SetCol(int c, const Vector &col);
 
-
    /// Set all entries of a row to the specified value.
    void SetRow(int row, real_t value);
    /// Set all entries of a column to the specified value.
@@ -342,6 +361,7 @@ public:
    void Transpose(const DenseMatrix &A);
    /// (*this) = 1/2 ((*this) + (*this)^t)
    void Symmetrize();
+   bool IsSymmetric(real_t tol = 1e-10);
 
    void Lump();
 

linalg/hypre.cpp

@@ -410,19 +410,19 @@ HYPRE_Int HypreParVector::Randomize(HYPRE_Int seed)
    return hypre_ParVectorSetRandomValues(x,seed);
 }
 
-void HypreParVector::Print(const char *fname) const
+void HypreParVector::Print(const std::string &fname) const
 {
-   hypre_ParVectorPrint(x,fname);
+   hypre_ParVectorPrint(x, fname.c_str());
 }
 
-void HypreParVector::Read(MPI_Comm comm, const char *fname)
+void HypreParVector::Read(MPI_Comm comm, const std::string &fname)
 {
    if (own_ParVector)
    {
       hypre_ParVectorDestroy(x);
    }
    data.Delete();
-   x = hypre_ParVectorRead(comm, fname);
+   x = hypre_ParVectorRead(comm, fname.c_str());
    own_ParVector = true;
    _SetDataAndSize_();
 }
@@ -792,6 +792,44 @@ static void SyncBackBoolCSR(Table *bool_csr, MemoryIJData &mem_csr)
    }
 }
 
+/// @brief Return the size of the partitioning arrays, see @ref
+/// hypre_partitioning_descr.
+static int GetPartitioningArraySize(MPI_Comm comm)
+{
+   if (HYPRE_AssumedPartitionCheck())
+   {
+      return 2;
+   }
+   else
+   {
+      int comm_size;
+      MPI_Comm_size(comm, &comm_size);
+      return comm_size + 1;
+   }
+}
+
+/// @brief Returns true if the row and col arrays are equal (across all MPI
+/// ranks).
+///
+/// Both @a row and @a col are partitioning arrays, whose length is returned by
+/// GetPartitioningArraySize(), see @ref hypre_partitioning_descr.
+static bool RowAndColStartsAreEqual(MPI_Comm comm, HYPRE_BigInt *rows,
+                                    HYPRE_BigInt *cols)
+{
+   const int part_size = GetPartitioningArraySize(comm);
+   bool are_equal = true;
+   for (int i = 0; i < part_size; ++i)
+   {
+      if (rows[i] != cols[i])
+      {
+         are_equal = false;
+         break;
+      }
+   }
+   MPI_Allreduce(MPI_IN_PLACE, &are_equal, 1,  MPI_C_BOOL, MPI_LAND, comm);
+   return are_equal;
+}
+
 // static method
 signed char HypreParMatrix::HypreCsrToMem(hypre_CSRMatrix *h_mat,
                                           MemoryType h_mat_mt,
@@ -924,7 +962,7 @@ HypreParMatrix::HypreParMatrix(MPI_Comm comm,
    hypre_ParCSRMatrixSetNumNonzeros(A);
 
    /* Make sure that the first entry in each row is the diagonal one. */
-   if (row_starts == col_starts)
+   if (RowAndColStartsAreEqual(comm, row_starts, col_starts))
    {
       HypreReadWrite();
       hypre_CSRMatrixReorder(hypre_ParCSRMatrixDiag(A));
@@ -974,11 +1012,12 @@ HypreParMatrix::HypreParMatrix(MPI_Comm comm,
    hypre_ParCSRMatrixSetNumNonzeros(A);
 
    /* Make sure that the first entry in each row is the diagonal one. */
-   if (row_starts == col_starts)
+   if (RowAndColStartsAreEqual(comm, row_starts, col_starts))
    {
       HypreReadWrite();
       hypre_CSRMatrixReorder(hypre_ParCSRMatrixDiag(A));
-      SyncBackCSR(diag, mem_diag); // update diag, if needed
+      // update diag, if needed
+      if (!own_diag_offd) { SyncBackCSR(diag, mem_diag); }
    }
 
    hypre_MatvecCommPkgCreate(A);
@@ -1031,7 +1070,7 @@ HypreParMatrix::HypreParMatrix(
    hypre_ParCSRMatrixSetNumNonzeros(A);
 
    /* Make sure that the first entry in each row is the diagonal one. */
-   if (row_starts == col_starts)
+   if (RowAndColStartsAreEqual(comm, row_starts, col_starts))
    {
       hypre_CSRMatrixReorder(hypre_ParCSRMatrixDiag(A));
    }
@@ -1093,7 +1132,7 @@ HypreParMatrix::HypreParMatrix(MPI_Comm comm,
    hypre_CSRMatrixDestroy(csr_a);
 
    /* Make sure that the first entry in each row is the diagonal one. */
-   if (row_starts == col_starts)
+   if (RowAndColStartsAreEqual(comm, row_starts, col_starts))
    {
       hypre_CSRMatrixReorder(hypre_ParCSRMatrixDiag(new_A));
    }
@@ -1132,7 +1171,7 @@ HypreParMatrix::HypreParMatrix(MPI_Comm comm,
    hypre_ParCSRMatrixSetNumNonzeros(A);
 
    /* Make sure that the first entry in each row is the diagonal one. */
-   if (row_starts == col_starts)
+   if (RowAndColStartsAreEqual(comm, row_starts, col_starts))
    {
       HypreReadWrite();
       hypre_CSRMatrixReorder(hypre_ParCSRMatrixDiag(A));
@@ -1247,11 +1286,10 @@ HypreParMatrix::HypreParMatrix(MPI_Comm comm, int nrows,
    Init();
 
    // Determine partitioning size, and my column start and end
-   int part_size;
+   const int part_size = GetPartitioningArraySize(comm);
    HYPRE_BigInt my_col_start, my_col_end; // my range: [my_col_start, my_col_end)
    if (HYPRE_AssumedPartitionCheck())
    {
-      part_size = 2;
       my_col_start = cols[0];
       my_col_end = cols[1];
    }
@@ -1259,15 +1297,14 @@ HypreParMatrix::HypreParMatrix(MPI_Comm comm, int nrows,
    {
       int myid;
       MPI_Comm_rank(comm, &myid);
-      MPI_Comm_size(comm, &part_size);
-      part_size++;
       my_col_start = cols[myid];
       my_col_end = cols[myid+1];
    }
 
    // Copy in the row and column partitionings
+   const bool rows_eq_cols = RowAndColStartsAreEqual(comm, rows, cols);
    HYPRE_BigInt *row_starts, *col_starts;
-   if (rows == cols)
+   if (rows_eq_cols)
    {
       row_starts = col_starts = mfem_hypre_TAlloc_host(HYPRE_BigInt, part_size);
       for (int i = 0; i < part_size; i++)
@@ -1360,14 +1397,14 @@ HypreParMatrix::HypreParMatrix(MPI_Comm comm, int nrows,
    }
 
    hypre_ParCSRMatrixSetNumNonzeros(A);
-   /* Make sure that the first entry in each row is the diagonal one. */
-   if (row_starts == col_starts)
+   // Make sure that the first entry in each row is the diagonal one.
+   if (rows_eq_cols)
    {
       hypre_CSRMatrixReorder(hypre_ParCSRMatrixDiag(A));
    }
 #if MFEM_HYPRE_VERSION > 22200
    mfem_hypre_TFree_host(row_starts);
-   if (rows != cols)
+   if (!rows_eq_cols)
    {
       mfem_hypre_TFree_host(col_starts);
    }
@@ -1480,16 +1517,7 @@ void HypreParMatrix::CopyRowStarts()
       return;
    }
 
-   int row_starts_size;
-   if (HYPRE_AssumedPartitionCheck())
-   {
-      row_starts_size = 2;
-   }
-   else
-   {
-      MPI_Comm_size(hypre_ParCSRMatrixComm(A), &row_starts_size);
-      row_starts_size++; // num_proc + 1
-   }
+   const int row_starts_size = GetPartitioningArraySize(hypre_ParCSRMatrixComm(A));
 
    HYPRE_BigInt *old_row_starts = hypre_ParCSRMatrixRowStarts(A);
    HYPRE_BigInt *new_row_starts = mfem_hypre_CTAlloc_host(HYPRE_BigInt,
@@ -1520,16 +1548,7 @@ void HypreParMatrix::CopyColStarts()
       return;
    }
 
-   int col_starts_size;
-   if (HYPRE_AssumedPartitionCheck())
-   {
-      col_starts_size = 2;
-   }
-   else
-   {
-      MPI_Comm_size(hypre_ParCSRMatrixComm(A), &col_starts_size);
-      col_starts_size++; // num_proc + 1
-   }
+   const int col_starts_size = GetPartitioningArraySize(hypre_ParCSRMatrixComm(A));
 
    HYPRE_BigInt *old_col_starts = hypre_ParCSRMatrixColStarts(A);
    HYPRE_BigInt *new_col_starts = mfem_hypre_CTAlloc_host(HYPRE_BigInt,
@@ -2291,13 +2310,8 @@ void HypreParMatrix::Threshold(real_t threshold)
    A = parcsr_A_ptr;
 
    hypre_ParCSRMatrixSetNumNonzeros(A);
-   /* Make sure that the first entry in each row is the diagonal one. */
-#if MFEM_HYPRE_VERSION <= 22200
-   if (row_starts == col_starts)
-#else
-   if ((row_starts[0] == col_starts[0]) &&
-       (row_starts[1] == col_starts[1]))
-#endif
+   // Make sure that the first entry in each row is the diagonal one.
+   if (RowAndColStartsAreEqual(comm, row_starts, col_starts))
    {
       hypre_CSRMatrixReorder(hypre_ParCSRMatrixDiag(A));
    }
@@ -2623,48 +2637,38 @@ void HypreParMatrix::EliminateBC(const Array<int> &ess_dofs,
    mfem_hypre_TFree(eliminate_col);
 }
 
-void HypreParMatrix::Print(const char *fname, HYPRE_Int offi,
+void HypreParMatrix::Print(const std::string &fname, HYPRE_Int offi,
                            HYPRE_Int offj) const
 {
    HostRead();
-   hypre_ParCSRMatrixPrintIJ(A,offi,offj,fname);
+   hypre_ParCSRMatrixPrintIJ(A, offi, offj, fname.c_str());
    HypreRead();
 }
 
-void HypreParMatrix::Read(MPI_Comm comm, const char *fname)
+void HypreParMatrix::Read(MPI_Comm comm, const std::string &fname)
 {
-   Destroy();
-   Init();
-
+   HYPRE_ParCSRMatrix A_parcsr;
    HYPRE_Int base_i, base_j;
-   hypre_ParCSRMatrixReadIJ(comm, fname, &base_i, &base_j, &A);
+   hypre_ParCSRMatrixReadIJ(comm, fname.c_str(), &base_i, &base_j, &A_parcsr);
+
+   WrapHypreParCSRMatrix(A_parcsr, true);
+
    hypre_ParCSRMatrixSetNumNonzeros(A);
-
    if (!hypre_ParCSRMatrixCommPkg(A)) { hypre_MatvecCommPkgCreate(A); }
-
-   height = GetNumRows();
-   width = GetNumCols();
 }
 
-void HypreParMatrix::Read_IJMatrix(MPI_Comm comm, const char *fname)
+void HypreParMatrix::Read_IJMatrix(MPI_Comm comm, const std::string &fname)
 {
-   Destroy();
-   Init();
-
    HYPRE_IJMatrix A_ij;
-   HYPRE_IJMatrixRead(fname, comm, 5555, &A_ij); // HYPRE_PARCSR = 5555
+   HYPRE_IJMatrixRead(fname.c_str(), comm, 5555, &A_ij); // HYPRE_PARCSR = 5555
 
    HYPRE_ParCSRMatrix A_parcsr;
    HYPRE_IJMatrixGetObject(A_ij, (void**) &A_parcsr);
 
-   A = (hypre_ParCSRMatrix*)A_parcsr;
+   WrapHypreParCSRMatrix(A_parcsr, true);
 
    hypre_ParCSRMatrixSetNumNonzeros(A);
-
    if (!hypre_ParCSRMatrixCommPkg(A)) { hypre_MatvecCommPkgCreate(A); }
-
-   height = GetNumRows();
-   width = GetNumCols();
 }
 
 void HypreParMatrix::PrintCommPkg(std::ostream &os) const

linalg/hypre.hpp

@@ -364,10 +364,10 @@ public:
    HYPRE_Int Randomize(HYPRE_Int seed);
 
    /// Prints the locally owned rows in parallel
-   void Print(const char *fname) const;
+   void Print(const std::string &fname) const;
 
    /// Reads a HypreParVector from files saved with HypreParVector::Print
-   void Read(MPI_Comm comm, const char *fname);
+   void Read(MPI_Comm comm, const std::string &fname);
 
    /// Calls hypre's destroy function
    ~HypreParVector();
@@ -919,12 +919,14 @@ public:
    const Memory<HYPRE_Int> &GetDiagMemoryJ() const { return mem_diag.J; }
    const Memory<real_t> &GetDiagMemoryData() const { return mem_diag.data; }
 
-   /// Prints the locally owned rows in parallel
-   void Print(const char *fname, HYPRE_Int offi = 0, HYPRE_Int offj = 0) const;
+   /// @brief Prints the locally owned rows in parallel. The resulting files can
+   /// be read with Read_IJMatrix().
+   void Print(const std::string &fname, HYPRE_Int offi = 0,
+              HYPRE_Int offj = 0) const;
    /// Reads the matrix from a file
-   void Read(MPI_Comm comm, const char *fname);
+   void Read(MPI_Comm comm, const std::string &fname);
    /// Read a matrix saved as a HYPRE_IJMatrix
-   void Read_IJMatrix(MPI_Comm comm, const char *fname);
+   void Read_IJMatrix(MPI_Comm comm, const std::string &fname);
 
    /// Print information about the hypre_ParCSRCommPkg of the HypreParMatrix.
    void PrintCommPkg(std::ostream &out = mfem::out) const;

linalg/lapack.hpp

@@ -0,0 +1,141 @@
+// Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
+// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+// LICENSE and NOTICE for details. LLNL-CODE-806117.
+//
+// This file is part of the MFEM library. For more information and source code
+// availability visit https://mfem.org.
+//
+// MFEM is free software; you can redistribute it and/or modify it under the
+// terms of the BSD-3 license. We welcome feedback and contributions, see file
+// CONTRIBUTING.md for details.
+
+#ifndef MFEM_LAPACK_HPP
+#define MFEM_LAPACK_HPP
+
+#include "../config/config.hpp"
+
+#ifdef MFEM_USE_LAPACK
+
+#ifdef MFEM_USE_SINGLE
+#define MFEM_LAPACK_PREFIX(stub) s##stub
+#define MFEM_LAPACK_COMPLEX(stub) c##stub
+#elif defined(MFEM_USE_DOUBLE)
+#define MFEM_LAPACK_PREFIX(stub) d##stub
+#define MFEM_LAPACK_COMPLEX(stub) z##stub
+#endif
+
+namespace mfem
+{
+
+extern "C" void
+MFEM_LAPACK_PREFIX(gemm_)(char *, char *, int *, int *, int *, real_t *,
+                          real_t *, int *, real_t *, int *, real_t *, real_t *,
+                          int *);
+extern "C" void
+MFEM_LAPACK_PREFIX(getrf_)(int *, int *, real_t *, int *, int *, int *);
+extern "C" void
+MFEM_LAPACK_PREFIX(getrs_)(char *, int *, int *, real_t *, int *, int *,
+                           real_t *, int *, int *);
+extern "C" void
+MFEM_LAPACK_PREFIX(getri_)(int *N, real_t *A, int *LDA, int *IPIV, real_t *WORK,
+                           int *LWORK, int *INFO);
+extern "C" void
+MFEM_LAPACK_PREFIX(syevr_)(char *JOBZ, char *RANGE, char *UPLO, int *N,
+                           real_t *A, int *LDA, real_t *VL, real_t *VU, int *IL,
+                           int *IU, real_t *ABSTOL, int *M, real_t *W,
+                           real_t *Z, int *LDZ, int *ISUPPZ, real_t *WORK,
+                           int *LWORK, int *IWORK, int *LIWORK, int *INFO);
+extern "C" void
+MFEM_LAPACK_PREFIX(syev_)(char *JOBZ, char *UPLO, int *N, real_t *A, int *LDA,
+                          real_t *W, real_t *WORK, int *LWORK, int *INFO);
+extern "C" void
+MFEM_LAPACK_PREFIX(sygv_) (int *ITYPE, char *JOBZ, char *UPLO, int * N,
+                           real_t *A, int *LDA, real_t *B, int *LDB, real_t *W,
+                           real_t *WORK, int *LWORK, int *INFO);
+extern "C" void
+MFEM_LAPACK_PREFIX(sygvx_)(int *ITYPE, char *JOBZ, char *RANGE, char *UPLO,
+                           int *N, double *A, int *LDA, double *B, int *LDB,
+                           double *VL, double *VU, int *IL, int *IU,
+                           double *ABSTOL, int *M, double *W, double *Z,
+                           int *LDZ, double *WORK, int *LWORK,int *IWORK,
+                           int *IFAIL, int *INFO);
+extern "C" void
+MFEM_LAPACK_PREFIX(gesvd_)(char *JOBU, char *JOBVT, int *M, int *N, real_t *A,
+                           int *LDA, real_t *S, real_t *U, int *LDU, real_t *VT,
+                           int *LDVT, real_t *WORK, int *LWORK, int *INFO);
+extern "C" void
+MFEM_LAPACK_PREFIX(trsm_)(char *side, char *uplo, char *transa, char *diag,
+                          int *m, int *n, real_t *alpha, real_t *a, int *lda,
+                          real_t *b, int *ldb);
+extern "C" void
+MFEM_LAPACK_PREFIX(ggev_)(char *jobvl, char *jobvr, int *n, real_t *a, int *lda,
+                          real_t *B, int *ldb, real_t *alphar, real_t *alphai,
+                          real_t *beta, real_t *vl, int * ldvl, real_t * vr,
+                          int * ldvr, real_t * work, int * lwork, int* info);
+
+// Cholesky factorizations/solves
+extern "C" void
+MFEM_LAPACK_PREFIX(potrf_)(char *, int *, real_t *, int *, int *);
+// Solve
+extern "C" void
+MFEM_LAPACK_PREFIX(potrs_)(char *, int *, int *, real_t *, int *, real_t *,
+                           int *, int *);
+// Triangular Solves
+extern "C" void
+MFEM_LAPACK_PREFIX(trtrs_)(char *, char*, char *, int *, int *, real_t *, int *,
+                           real_t *, int *, int *);
+extern "C" void
+MFEM_LAPACK_PREFIX(potri_)(char *, int *, real_t *, int*, int *);
+
+// LAPACK routines for NNLSSolver
+extern "C" void
+MFEM_LAPACK_PREFIX(ormqr_)(char *, char *, int *, int *, int *, real_t *, int*,
+                           real_t *, real_t *, int *, real_t *, int*, int*);
+
+extern "C" void
+MFEM_LAPACK_PREFIX(geqrf_)(int *, int *, real_t *, int *, real_t *, real_t *,
+                           int *, int *);
+
+extern "C" void
+MFEM_LAPACK_PREFIX(gemv_)(char *, int *, int *, real_t *, real_t *, int *,
+                          real_t *, int *, real_t *, real_t *, int *);
+
+extern "C" void
+MFEM_LAPACK_PREFIX(trsm_)(char *side, char *uplo, char *transa, char *diag,
+                          int *m, int *n, real_t *alpha, real_t *a, int *lda,
+                          real_t *b, int *ldb);
+
+// Complex
+extern "C" void
+MFEM_LAPACK_COMPLEX(getrf_)(int *, int *, std::complex<real_t> *, int *, int *,
+                            int *);
+extern "C" void
+MFEM_LAPACK_COMPLEX(getrs_)(char *, int *, int *, std::complex<real_t> *, int *,
+                            int *, std::complex<real_t> *, int *, int *);
+extern "C" void
+MFEM_LAPACK_COMPLEX(getri_)(int *, std::complex<real_t> *, int *, int *,
+                            std::complex<real_t> *, int *, int *);
+extern "C" void
+MFEM_LAPACK_COMPLEX(trsm_)(char *, char *, char *, char *, int *, int *,
+                           std::complex<real_t> *, std::complex<real_t> *,
+                           int *, std::complex<real_t> *, int *);
+extern "C" void
+MFEM_LAPACK_COMPLEX(potrf_)(char *, int *, std::complex<real_t> *, int *,
+                            int *);
+
+extern "C" void
+MFEM_LAPACK_COMPLEX(trtrs_)(char *, char*, char *, int *, int *,
+                            std::complex<real_t> *, int *,
+                            std::complex<real_t> *, int *, int *);
+extern "C" void
+MFEM_LAPACK_COMPLEX(potri_)(char *, int *, std::complex<real_t> *, int*, int *);
+
+extern "C" void
+MFEM_LAPACK_COMPLEX(potrs_)(char *, int *, int *, std::complex<real_t> *, int *,
+                            std::complex<real_t> *, int *, int *);
+
+}
+
+#endif
+
+#endif

linalg/operator.hpp

@@ -309,18 +309,51 @@ public:
 
 
 /// Base abstract class for first order time dependent operators.
-/** Operator of the form: (x,t) -> f(x,t), where k = f(x,t) generally solves the
-    algebraic equation F(x,k,t) = G(x,t). The functions F and G represent the
-    _implicit_ and _explicit_ parts of the operator, respectively. For explicit
-    operators, F(x,k,t) = k, so f(x,t) = G(x,t). */
+/** Operator of the form: (u,t) -> k(u,t), where k generally solves the
+    algebraic equation F(u,k,t) = G(u,t). The functions F and G represent the
+    _implicit_ and _explicit_ parts of the operator, respectively.
+
+    A common use for this class is representing a differential algebraic
+    equation of the form $ F(y,\frac{dy}{dt},t) = G(y,t) $.
+
+    For example, consider an ordinary differential equation of the form
+    $ M \frac{dy}{dt} = g(y,t) $. There are various ways of expressing this ODE
+    as a TimeDependentOperator depending on the choices for F and G. Here are
+    some common choices:
+
+      1. F(u,k,t) = k and G(u,t) = inv(M) g(u,t),
+      2. F(u,k,t) = M k and G(u,t) = g(u,t),
+      3. F(u,k,t) = M k - g(u,t) and G(u,t) = 0.
+
+    Note that depending on the ODE solver, some of the above choices may be
+    preferable to the others.
+*/
 class TimeDependentOperator : public Operator
 {
 public:
+   /// Enum used to describe the form of the time-dependent operator.
+   /** The type should be set by classes derived from TimeDependentOperator to
+       describe the form, in terms of the functions F and G, used by the
+       specific derived class. This information can be queried by classes or
+       functions (like time stepping algorithms) to make choices about the
+       algorithm to use, or to ensure that the TimeDependentOperator uses the
+       form expected by the class/function.
+
+       For example, assume that a derived class is implementing the ODE
+       $M \frac{dy}{dt} = g(y,t)$ and chooses to define $F(u,k,t) = M k$ and
+       $G(u,t) = g(u,t)$. Then it cannot use type EXPLICIT, unless $M = I$, or
+       type HOMOGENEOUS, unless $g(u,t) = 0$. If, on the other hand, the derived
+       class chooses to define $F(u,k,t) = k$ and $G(u,t) = M^{-1} g(y,t)$, then
+       the natural choice is to set the type to EXPLICIT, even though setting it
+       to IMPLICIT is also not wrong -- doing so will simply fail to inform
+       methods that query this information that it uses a more specific
+       implementation, EXPLICIT, that may allow the use of algorithms that
+       support only the EXPLICIT type. */
    enum Type
    {
-      EXPLICIT,   ///< This type assumes F(x,k,t) = k, i.e. k = f(x,t) = G(x,t).
+      EXPLICIT,   ///< This type assumes F(u,k,t) = k.
       IMPLICIT,   ///< This is the most general type, no assumptions on F and G.
-      HOMOGENEOUS ///< This type assumes that G(x,t) = 0.
+      HOMOGENEOUS ///< This type assumes that G(u,t) = 0.
    };
 
    /// Evaluation mode. See SetEvalMode() for details.
@@ -328,29 +361,30 @@ public:
    {
       /** Normal evaluation. */
       NORMAL,
-      /** Assuming additive split, f(x,t) = f1(x,t) + f2(x,t), evaluate the
-          first term, f1. */
+      /** Assuming additive split, k(u,t) = k1(u,t) + k2(u,t), evaluate the
+          first term, k1. */
       ADDITIVE_TERM_1,
-      /** Assuming additive split, f(x,t) = f1(x,t) + f2(x,t), evaluate the
-          second term, f2. */
+      /** Assuming additive split, k(u,t) = k1(u,t) + k2(u,t), evaluate the
+          second term, k2. */
       ADDITIVE_TERM_2
    };
 
 protected:
    real_t t;  ///< Current time.
-   Type type; ///< Describes the form of the TimeDependentOperator.
+   Type type; /**< @brief Describes the form of the TimeDependentOperator, see
+                   the documentation of #Type. */
    EvalMode eval_mode; ///< Current evaluation mode.
 
 public:
-   /** @brief Construct a "square" TimeDependentOperator y = f(x,t), where x and
-       y have the same dimension @a n. */
+   /** @brief Construct a "square" TimeDependentOperator (u,t) -> k(u,t), where
+       u and k have the same dimension @a n. */
    explicit TimeDependentOperator(int n = 0, real_t t_ = 0.0,
                                   Type type_ = EXPLICIT)
       : Operator(n) { t = t_; type = type_; eval_mode = NORMAL; }
 
-   /** @brief Construct a TimeDependentOperator y = f(x,t), where x and y have
-       dimensions @a w and @a h, respectively. */
-   TimeDependentOperator(int h, int w, real_t t_ = 0.0, Type type_ = EXPLICIT)
+   /** @brief Construct a TimeDependentOperator (u,t) -> k(u,t), where u and k
+       have dimensions @a w and @a h, respectively. */
+   TimeDependentOperator(int h, int w, double t_ = 0.0, Type type_ = EXPLICIT)
       : Operator(h, w) { t = t_; type = type_; eval_mode = NORMAL; }
 
    /// Read the currently set time.
@@ -373,7 +407,7 @@ public:
    /** The evaluation mode is a switch that allows time-stepping methods to
        request evaluation of separate components/terms of the time-dependent
        operator. For example, IMEX methods typically assume additive split of
-       the operator: f(x,t) = f1(x,t) + f2(x,t) and they rely on the ability to
+       the operator: k(u,t) = k1(u,t) + k2(u,t) and they rely on the ability to
        evaluate the two terms separately.
 
        Generally, setting the evaluation mode should affect the behavior of all
@@ -384,62 +418,104 @@ public:
    { eval_mode = new_eval_mode; }
 
    /** @brief Perform the action of the explicit part of the operator, G:
-       @a y = G(@a x, t) where t is the current time.
+       @a v = G(@a u, t) where t is the current time.
 
        Presently, this method is used by some PETSc ODE solvers, for more
        details, see the PETSc Manual. */
-   virtual void ExplicitMult(const Vector &x, Vector &y) const;
+   virtual void ExplicitMult(const Vector &u, Vector &v) const;
 
    /** @brief Perform the action of the implicit part of the operator, F:
-       @a y = F(@a x, @a k, t) where t is the current time.
+       @a v = F(@a u, @a k, t) where t is the current time.
 
        Presently, this method is used by some PETSc ODE solvers, for more
        details, see the PETSc Manual.*/
-   virtual void ImplicitMult(const Vector &x, const Vector &k, Vector &y) const;
+   virtual void ImplicitMult(const Vector &u, const Vector &k, Vector &v) const;
 
-   /** @brief Perform the action of the operator: @a y = k = f(@a x, t), where
-       k solves the algebraic equation F(@a x, k, t) = G(@a x, t) and t is the
-       current time. */
-   virtual void Mult(const Vector &x, Vector &y) const;
+   /** @brief Perform the action of the operator (u,t) -> k(u,t) where t is the
+       current time set by SetTime() and @a k satisfies
+       F(@a u, @a k, t) = G(@a u, t).
 
-   /** @brief Solve the equation: @a k = f(@a x + @a dt @a k, t), for the
-       unknown @a k at the current time t.
+       For solving an ordinary differential equation of the form
+       $ M \frac{dy}{dt} = g(y,t) $, recall that F and G can be defined in
+       various ways, e.g.:
 
-       For general F and G, the equation for @a k becomes:
-       F(@a x + @a dt @a k, @a k, t) = G(@a x + @a dt @a k, t).
+         1. F(u,k,t) = k and G(u,t) = inv(M) g(u,t)
+         2. F(u,k,t) = M k and G(u,t) = g(u,t)
+         3. F(u,k,t) = M k - g(u,t) and G(u,t) = 0.
 
-       The input vector @a x corresponds to time index (or cycle) n, while the
-       currently set time, #t, and the result vector @a k correspond to time
-       index n+1. The time step @a dt corresponds to the time interval between
-       cycles n and n+1.
+       Regardless of the choice of F and G, this function should always compute
+       @a k = inv(M) g(@a u, t). */
+   virtual void Mult(const Vector &u, Vector &v) const override;
 
-       This method allows for the abstract implementation of some time
-       integration methods, including diagonal implicit Runge-Kutta (DIRK)
-       methods and the backward Euler method in particular.
+   /** @brief Solve for the unknown @a k, at the current time t, the following
+       equation:
+       F(@a u + @a gamma @a k, @a k, t) = G(@a u + @a gamma @a k, t).
+
+       For solving an ordinary differential equation of the form
+       $ M \frac{dy}{dt} = g(y,t) $, recall that F and G can be defined in
+       various ways, e.g.:
+
+         1. F(u,k,t) = k and G(u,t) = inv(M) g(u,t)
+         2. F(u,k,t) = M k and G(u,t) = g(u,t)
+         3. F(u,k,t) = M k - g(u,t) and G(u,t) = 0
+
+       Regardless of the choice of F and G, this function should solve for @a k
+       in M @a k = g(@a u + @a gamma @a k, t).
+
+       To see how @a k can be useful, consider the backward Euler method defined
+       by $ y(t + \Delta t) = y(t) + \Delta t k_0 $ where
+       $ M k_0 = g \big( y(t) + \Delta t k_0, t + \Delta t \big) $. A backward
+       Euler integrator can use @a k from this function for $k_0$, with the call
+       using @a u set to $ y(t) $, @a gamma set to $ \Delta t$, and time set to
+       $t + \Delta t$. See class BackwardEulerSolver.
+
+       Generalizing further, consider a diagonally implicit Runge-Kutta (DIRK)
+       method defined by
+       $ y(t + \Delta t) = y(t) + \Delta t \sum_{i=1}^s b_i k_i $ where
+       $ M k_i = g \big( y(t) + \Delta t \sum_{j=1}^i a_{ij} k_j,
+                         t + c_i \Delta t \big) $.
+       A DIRK integrator can use @a k from this function, with @a u set to
+       $ y(t) + \Delta t \sum_{j=1}^{i-1} a_{ij} k_j $ and @a gamma set to
+       $ a_{ii} \Delta t $, for $ k_i $. For example, see class SDIRK33Solver.
 
        If not re-implemented, this method simply generates an error. */
-   virtual void ImplicitSolve(const real_t dt, const Vector &x, Vector &k);
+   virtual void ImplicitSolve(const real_t gamma, const Vector &u, Vector &k);
 
-   /** @brief Return an Operator representing (dF/dk @a shift + dF/dx) at the
-       given @a x, @a k, and the currently set time.
+   /** @brief Return an Operator representing (dF/dk @a shift + dF/du) at the
+       given @a u, @a k, and the currently set time.
 
        Presently, this method is used by some PETSc ODE solvers, for more
        details, see the PETSc Manual. */
-   virtual Operator& GetImplicitGradient(const Vector &x, const Vector &k,
+   virtual Operator& GetImplicitGradient(const Vector &u, const Vector &k,
                                          real_t shift) const;
 
-   /** @brief Return an Operator representing dG/dx at the given point @a x and
+   /** @brief Return an Operator representing dG/du at the given point @a u and
        the currently set time.
 
        Presently, this method is used by some PETSc ODE solvers, for more
        details, see the PETSc Manual. */
-   virtual Operator& GetExplicitGradient(const Vector &x) const;
+   virtual Operator& GetExplicitGradient(const Vector &u) const;
 
-   /** @brief Setup the ODE linear system $ A(x,t) = (I - gamma J) $ or
-       $ A = (M - gamma J) $, where $ J(x,t) = \frac{df}{dt(x,t)} $.
+   /** @brief Setup a linear system as needed by some SUNDIALS ODE solvers.
 
-       @param[in]  x     The state at which $A(x,t)$ should be evaluated.
-       @param[in]  fx    The current value of the ODE rhs function, $f(x,t)$.
+       For solving an ordinary differential equation of the form
+       $ M \frac{dy}{dt} = g(y,t) $, recall that F and G can be defined as one
+       of the following:
+
+         1. F(u,k,t) = k and G(u,t) = inv(M) g(u,t)
+         2. F(u,k,t) = M k and G(u,t) = g(u,t)
+         3. F(u,k,t) = M k - g(u,t) and G(u,t) = 0
+
+       This function performs setup to solve $ A x = b $ where A is either
+
+         1. A(@a y,t) = I - @a gamma inv(M) J(@a y,t)
+         2. A(@a y,t) = M - @a gamma J(@a y,t)
+         3. A(@a y,t) = M - @a gamma J(@a y,t)
+
+       with J = dg/dy (or a reasonable approximation thereof).
+
+       @param[in]  y     The state at which A(@a y,t) should be evaluated.
+       @param[in]  v     The value of inv(M) g(y,t) for 1 or g(y,t) for 2 & 3.
        @param[in]  jok   Flag indicating if the Jacobian should be updated.
        @param[out] jcur  Flag to signal if the Jacobian was updated.
        @param[in]  gamma The scaled time step value.
@@ -448,10 +524,10 @@ public:
 
        Presently, this method is used by SUNDIALS ODE solvers, for more
        details, see the SUNDIALS User Guides. */
-   virtual int SUNImplicitSetup(const Vector &x, const Vector &fx,
+   virtual int SUNImplicitSetup(const Vector &y, const Vector &v,
                                 int jok, int *jcur, real_t gamma);
 
-   /** @brief Solve the ODE linear system $ A x = b $ as setup by
+   /** @brief Solve the ODE linear system A @a x = @a b, where A is defined by
        the method SUNImplicitSetup().
 
        @param[in]      b   The linear system right-hand side.
@@ -464,7 +540,8 @@ public:
        details, see the SUNDIALS User Guides. */
    virtual int SUNImplicitSolve(const Vector &b, Vector &x, real_t tol);
 
-   /** @brief Setup the mass matrix in the ODE system $ M y' = f(y,t) $ .
+   /** @brief Setup the mass matrix in the ODE system
+       $ M \frac{dy}{dt} = g(y,t) $ .
 
        If not re-implemented, this method simply generates an error.
 
@@ -472,8 +549,8 @@ public:
        details, see the ARKode User Guide. */
    virtual int SUNMassSetup();
 
-   /** @brief Solve the mass matrix linear system $ M x = b $
-       as setup by the method SUNMassSetup().
+   /** @brief Solve the mass matrix linear system  M @a x = @a b, where M is
+       defined by the method SUNMassSetup().
 
        @param[in]      b   The linear system right-hand side.
        @param[in,out]  x   On input, the initial guess. On output, the solution.
@@ -485,7 +562,8 @@ public:
        details, see the ARKode User Guide. */
    virtual int SUNMassSolve(const Vector &b, Vector &x, real_t tol);
 
-   /** @brief Compute the mass matrix-vector product $ v = M x $ .
+   /** @brief Compute the mass matrix-vector product @a v = M @a x, where M is
+       defined by the method SUNMassSetup().
 
        @param[in]   x The vector to multiply.
        @param[out]  v The result of the matrix-vector product.

linalg/solvers.cpp

@@ -10,6 +10,7 @@
 // CONTRIBUTING.md for details.
 
 #include "linalg.hpp"
+#include "lapack.hpp"
 #include "../general/annotation.hpp"
 #include "../general/forall.hpp"
 #include "../general/globals.hpp"
@@ -3543,38 +3544,6 @@ void AuxSpaceSmoother::Mult(const Vector &x, Vector &y, bool transpose) const
 #endif // MFEM_USE_MPI
 
 #ifdef MFEM_USE_LAPACK
-// LAPACK routines for NNLSSolver
-#ifdef MFEM_USE_SINGLE
-extern "C" void
-sormqr_(char *, char *, int *, int *, int *, float *, int*, float *,
-        float *, int *, float *, int*, int*);
-
-extern "C" void
-sgeqrf_(int *, int *, float *, int *, float *, float *, int *, int *);
-
-extern "C" void
-sgemv_(char *, int *, int *, float *, float *, int *, float *, int *,
-       float *, float *, int *);
-
-extern "C" void
-strsm_(char *side, char *uplo, char *transa, char *diag, int *m, int *n,
-       float *alpha, float *a, int *lda, float *b, int *ldb);
-#elif defined MFEM_USE_DOUBLE
-extern "C" void
-dormqr_(char *, char *, int *, int *, int *, double *, int*, double *,
-        double *, int *, double *, int*, int*);
-
-extern "C" void
-dgeqrf_(int *, int *, double *, int *, double *, double *, int *, int *);
-
-extern "C" void
-dgemv_(char *, int *, int *, double *, double *, int *, double *, int *,
-       double *, double *, int *);
-
-extern "C" void
-dtrsm_(char *side, char *uplo, char *transa, char *diag, int *m, int *n,
-       double *alpha, double *a, int *lda, double *b, int *ldb);
-#endif
 
 NNLSSolver::NNLSSolver()
    : Solver(0), mat(nullptr), const_tol_(1.0e-14), min_nnz_(0),
@@ -3938,25 +3907,19 @@ void NNLSSolver::Solve(const Vector& rhs_lb, const Vector& rhs_ub,
             lwork = -1;
             work.resize(10);
 
-#ifdef MFEM_USE_SINGLE
-            sormqr_(&lside, &trans, &m, &n_update, &i_qr_start,
-#elif defined MFEM_USE_DOUBLE
-            dormqr_(&lside, &trans, &m, &n_update, &i_qr_start,
-#endif
-                    mat_qr_data.GetData(), &m, tau.GetData(),
-                    mat_qr_data.GetData() + (i_qr_start * m), &m,
-                    work.data(), &lwork, &info);
+            MFEM_LAPACK_PREFIX(ormqr_)(&lside, &trans, &m, &n_update,
+                                       &i_qr_start, mat_qr_data.GetData(), &m,
+                                       tau.GetData(),
+                                       mat_qr_data.GetData() + (i_qr_start * m),
+                                       &m, work.data(), &lwork, &info);
             MFEM_VERIFY(info == 0, ""); // Q^T A update work calculation failed
             lwork = static_cast<int>(work[0]);
             work.resize(lwork);
-#ifdef MFEM_USE_SINGLE
-            sormqr_(&lside, &trans, &m, &n_update, &i_qr_start,
-#elif defined MFEM_USE_DOUBLE
-            dormqr_(&lside, &trans, &m, &n_update, &i_qr_start,
-#endif
-                    mat_qr_data.GetData(), &m, tau.GetData(),
-                    mat_qr_data.GetData() + (i_qr_start * m), &m,
-                    work.data(), &lwork, &info);
+            MFEM_LAPACK_PREFIX(ormqr_)(&lside, &trans, &m, &n_update,
+                                       &i_qr_start, mat_qr_data.GetData(), &m,
+                                       tau.GetData(),
+                                       mat_qr_data.GetData() + (i_qr_start * m),
+                                       &m, work.data(), &lwork, &info);
             MFEM_VERIFY(info == 0, ""); // Q^T A update failed
             // Compute QR factorization of the submatrix
             lwork = -1;
@@ -3977,24 +3940,16 @@ void NNLSSolver::Solve(const Vector& rhs_lb, const Vector& rhs_ub,
                sub_tau[j] = tau[i_qr_start + j];
             }
 
-#ifdef MFEM_USE_SINGLE
-            sgeqrf_(&m_update, &n_update,
-#elif defined MFEM_USE_DOUBLE
-            dgeqrf_(&m_update, &n_update,
-#endif
-                    submat_data.GetData(), &m_update, sub_tau.GetData(),
-                    work.data(), &lwork, &info);
+            MFEM_LAPACK_PREFIX(geqrf_)(&m_update, &n_update, submat_data.GetData(),
+                                       &m_update, sub_tau.GetData(), work.data(),
+                                       &lwork, &info);
             MFEM_VERIFY(info == 0, ""); // QR update factorization work calc
             lwork = static_cast<int>(work[0]);
             if (lwork == 0) { lwork = 1; }
             work.resize(lwork);
-#ifdef MFEM_USE_SINGLE
-            sgeqrf_(&m_update, &n_update,
-#elif defined MFEM_USE_DOUBLE
-            dgeqrf_(&m_update, &n_update,
-#endif
-                    submat_data.GetData(), &m_update, sub_tau.GetData(),
-                    work.data(), &lwork, &info);
+            MFEM_LAPACK_PREFIX(geqrf_)(&m_update, &n_update, submat_data.GetData(),
+                                       &m_update, sub_tau.GetData(), work.data(),
+                                       &lwork, &info);
             MFEM_VERIFY(info == 0, ""); // QR update factorization failed
 
             // Copy result back
@@ -4023,23 +3978,13 @@ void NNLSSolver::Solve(const Vector& rhs_lb, const Vector& rhs_ub,
             // perform qr)
             lwork = -1;
             work.resize(10);
-#ifdef MFEM_USE_SINGLE
-            sgeqrf_(&m, &n_glob,
-#elif defined MFEM_USE_DOUBLE
-            dgeqrf_(&m, &n_glob,
-#endif
-                    mat_qr_data.GetData(), &m, tau.GetData(),
-                    work.data(), &lwork, &info);
+            MFEM_LAPACK_PREFIX(geqrf_)(&m, &n_glob, mat_qr_data.GetData(), &m,
+                                       tau.GetData(), work.data(), &lwork, &info);
             MFEM_VERIFY(info == 0, ""); // QR factorization work calculation
             lwork = static_cast<int>(work[0]);
             work.resize(lwork);
-#ifdef MFEM_USE_SINGLE
-            sgeqrf_(&m, &n_glob,
-#elif defined MFEM_USE_DOUBLE
-            dgeqrf_(&m, &n_glob,
-#endif
-                    mat_qr_data.GetData(), &m, tau.GetData(),
-                    work.data(), &lwork, &info);
+            MFEM_LAPACK_PREFIX(geqrf_)(&m, &n_glob, mat_qr_data.GetData(), &m,
+                                       tau.GetData(), work.data(), &lwork, &info);
             MFEM_VERIFY(info == 0, ""); // QR factorization failed
          }
 
@@ -4067,25 +4012,17 @@ void NNLSSolver::Solve(const Vector& rhs_lb, const Vector& rhs_ub,
 
             sub_tau[0] = tau[i_qr_start];
 
-#ifdef MFEM_USE_SINGLE
-            sormqr_(&lside, &trans, &m_update, &ione, &ione,
-#elif defined MFEM_USE_DOUBLE
-            dormqr_(&lside, &trans, &m_update, &ione, &ione,
-#endif
-                    submat_data.GetData(), &m_update, sub_tau.GetData(),
-                    sub_qt.GetData(), &m_update,
-                    work.data(), &lwork, &info);
+            MFEM_LAPACK_PREFIX(ormqr_)(&lside, &trans, &m_update, &ione, &ione,
+                                       submat_data.GetData(), &m_update,
+                                       sub_tau.GetData(), sub_qt.GetData(),
+                                       &m_update, work.data(), &lwork, &info);
             MFEM_VERIFY(info == 0, ""); // H_last y work calculation failed
             lwork = static_cast<int>(work[0]);
             work.resize(lwork);
-#ifdef MFEM_USE_SINGLE
-            sormqr_(&lside, &trans, &m_update, &ione, &ione,
-#elif defined MFEM_USE_DOUBLE
-            dormqr_(&lside, &trans, &m_update, &ione, &ione,
-#endif
-                    submat_data.GetData(), &m_update, sub_tau.GetData(),
-                    sub_qt.GetData(), &m_update,
-                    work.data(), &lwork, &info);
+            MFEM_LAPACK_PREFIX(ormqr_)(&lside, &trans, &m_update, &ione, &ione,
+                                       submat_data.GetData(), &m_update,
+                                       sub_tau.GetData(), sub_qt.GetData(),
+                                       &m_update, work.data(), &lwork, &info);
             MFEM_VERIFY(info == 0, ""); // H_last y failed
             // Copy result back
             for (int i=0; i<m_update; ++i)
@@ -4099,25 +4036,17 @@ void NNLSSolver::Solve(const Vector& rhs_lb, const Vector& rhs_ub,
             qt_rhs_glob = rhs_avg_glob;
             lwork = -1;
             work.resize(10);
-#ifdef MFEM_USE_SINGLE
-            sormqr_(&lside, &trans, &m, &ione, &n_glob,
-#elif defined MFEM_USE_DOUBLE
-            dormqr_(&lside, &trans, &m, &ione, &n_glob,
-#endif
-                    mat_qr_data.GetData(), &m, tau.GetData(),
-                    qt_rhs_glob.GetData(), &m,
-                    work.data(), &lwork, &info);
+            MFEM_LAPACK_PREFIX(ormqr_)(&lside, &trans, &m, &ione, &n_glob,
+                                       mat_qr_data.GetData(), &m, tau.GetData(),
+                                       qt_rhs_glob.GetData(), &m,
+                                       work.data(), &lwork, &info);
             MFEM_VERIFY(info == 0, ""); // Q^T b work calculation failed
             lwork = static_cast<int>(work[0]);
             work.resize(lwork);
-#ifdef MFEM_USE_SINGLE
-            sormqr_(&lside, &trans, &m, &ione, &n_glob,
-#elif defined MFEM_USE_DOUBLE
-            dormqr_(&lside, &trans, &m, &ione, &n_glob,
-#endif
-                    mat_qr_data.GetData(), &m, tau.GetData(),
-                    qt_rhs_glob.GetData(), &m,
-                    work.data(), &lwork, &info);
+            MFEM_LAPACK_PREFIX(ormqr_)(&lside, &trans, &m, &ione, &n_glob,
+                                       mat_qr_data.GetData(), &m, tau.GetData(),
+                                       qt_rhs_glob.GetData(), &m,
+                                       work.data(), &lwork, &info);
             MFEM_VERIFY(info == 0, ""); // Q^T b failed
          }
 
@@ -4130,14 +4059,10 @@ void NNLSSolver::Solve(const Vector& rhs_lb, const Vector& rhs_ub,
          char upper = 'U';
          char nounit = 'N';
          vec1 = qt_rhs_glob;
-#ifdef MFEM_USE_SINGLE
-         strsm_(&lside, &upper, &notrans, &nounit,
-#elif defined MFEM_USE_DOUBLE
-         dtrsm_(&lside, &upper, &notrans, &nounit,
-#endif
-                &n_glob, &ione, &fone,
-                mat_qr_data.GetData(), &m,
-                vec1.GetData(), &n_glob);
+         MFEM_LAPACK_PREFIX(trsm_)(&lside, &upper, &notrans, &nounit,
+                                   &n_glob, &ione, &fone,
+                                   mat_qr_data.GetData(), &m,
+                                   vec1.GetData(), &n_glob);
 
          if (verbosity_ > 2)
          {
@@ -4360,14 +4285,10 @@ void NNLSSolver::Solve(const Vector& rhs_lb, const Vector& rhs_ub,
       {
          res_glob = rhs_avg_glob;
          real_t fmone = -1.0;
-#ifdef MFEM_USE_SINGLE
-         sgemv_(&notrans, &m, &n_glob, &fmone,
-#elif defined MFEM_USE_DOUBLE
-         dgemv_(&notrans, &m, &n_glob, &fmone,
-#endif
-                mat_0_data.GetData(), &m,
-                soln_nz_glob.GetData(), &ione, &fone,
-                res_glob.GetData(), &ione);
+         MFEM_LAPACK_PREFIX(gemv_)(&notrans, &m, &n_glob, &fmone,
+                                   mat_0_data.GetData(), &m,
+                                   soln_nz_glob.GetData(), &ione, &fone,
+                                   res_glob.GetData(), &ione);
       }
       else
       {
@@ -4381,24 +4302,18 @@ void NNLSSolver::Solve(const Vector& rhs_lb, const Vector& rhs_ub,
             qqt_rhs_glob(i) = qt_rhs_glob(i);
          }
 
-#ifdef MFEM_USE_SINGLE
-         sormqr_(&lside, &notrans, &m, &ione, &n_glob, mat_qr_data.GetData(), &m,
-#elif defined MFEM_USE_DOUBLE
-         dormqr_(&lside, &notrans, &m, &ione, &n_glob, mat_qr_data.GetData(), &m,
-#endif
-                 tau.GetData(), qqt_rhs_glob.GetData(), &m,
-                 work.data(), &lwork, &info);
+         MFEM_LAPACK_PREFIX(ormqr_)(&lside, &notrans, &m, &ione, &n_glob,
+                                    mat_qr_data.GetData(), &m,
+                                    tau.GetData(), qqt_rhs_glob.GetData(), &m,
+                                    work.data(), &lwork, &info);
 
          MFEM_VERIFY(info == 0, ""); // Q Q^T b work calculation failed.
          lwork = static_cast<int>(work[0]);
          work.resize(lwork);
-#ifdef MFEM_USE_SINGLE
-         sormqr_(&lside, &notrans, &m, &ione, &n_glob, mat_qr_data.GetData(), &m,
-#elif defined MFEM_USE_DOUBLE
-         dormqr_(&lside, &notrans, &m, &ione, &n_glob, mat_qr_data.GetData(), &m,
-#endif
-                 tau.GetData(), qqt_rhs_glob.GetData(), &m,
-                 work.data(), &lwork, &info);
+         MFEM_LAPACK_PREFIX(ormqr_)(&lside, &notrans, &m, &ione, &n_glob,
+                                    mat_qr_data.GetData(), &m,
+                                    tau.GetData(), qqt_rhs_glob.GetData(), &m,
+                                    work.data(), &lwork, &info);
          MFEM_VERIFY(info == 0, ""); // Q Q^T b calculation failed.
          res_glob = rhs_avg_glob;
          res_glob -= qqt_rhs_glob;

linalg/symmat.cpp

@@ -58,6 +58,20 @@ DenseSymmetricMatrix &DenseSymmetricMatrix::operator=(real_t c)
    return *this;
 }
 
+DenseSymmetricMatrix &DenseSymmetricMatrix::operator=(const DenseSymmetricMatrix
+                                                      &m)
+{
+   SetSize(m.height);
+
+   const int hw = m.GetStoredSize();
+   for (int i = 0; i < hw; i++)
+   {
+      data[i] = m.data[i];
+   }
+
+   return *this;
+}
+
 real_t &DenseSymmetricMatrix::Elem(int i, int j)
 {
    return (*this)(i,j);
@@ -89,11 +103,6 @@ MatrixInverse *DenseSymmetricMatrix::Inverse() const
    return nullptr;
 }
 
-void DenseSymmetricMatrix::Print (std::ostream & os, int width_) const
-{
-   mfem_error("DenseSymmetricMatrix::Print() not implemented!");
-}
-
 DenseSymmetricMatrix::~DenseSymmetricMatrix()
 {
    data.Delete();

linalg/symmat.hpp

@@ -102,6 +102,9 @@ public:
 
    DenseSymmetricMatrix &operator*=(real_t c);
 
+   /// Sets the matrix size and elements equal to those of m
+   DenseSymmetricMatrix &operator=(const DenseSymmetricMatrix &m);
+
    std::size_t MemoryUsage() const { return data.Capacity() * sizeof(real_t); }
 
    /// Shortcut for mfem::Read( GetMemory(), TotalSize(), on_dev).
@@ -134,9 +137,6 @@ public:
    /// Returns a pointer to (an approximation) of the matrix inverse.
    virtual MatrixInverse *Inverse() const;
 
-   /// Prints matrix to stream out.
-   virtual void Print (std::ostream & out = mfem::out, int width_ = 4) const;
-
    /// Destroys the symmetric matrix.
    virtual ~DenseSymmetricMatrix();
 };

mesh/mesh.cpp

@@ -466,7 +466,8 @@ void Mesh::GetBdrElementTransformation(int i,
          {
             for (int j = 0; j < n; j++)
             {
-               pm(k,j) = nodes(vdofs[n*k+j]);
+               int idx = vdofs[n*k+j];
+               pm(k,j) = nodes((idx<0)? -1-idx:idx);
             }
          }
          ElTr->SetFE(bdr_el);
@@ -7131,17 +7132,15 @@ Table *Mesh::GetEdgeVertexTable() const
 
 Table *Mesh::GetVertexToElementTable()
 {
-   int i, j, nv, *v;
-
    Table *vert_elem = new Table;
 
    vert_elem->MakeI(NumOfVertices);
 
-   for (i = 0; i < NumOfElements; i++)
+   for (int i = 0; i < NumOfElements; i++)
    {
-      nv = elements[i]->GetNVertices();
-      v  = elements[i]->GetVertices();
-      for (j = 0; j < nv; j++)
+      const int nv = elements[i]->GetNVertices();
+      const int *v = elements[i]->GetVertices();
+      for (int j = 0; j < nv; j++)
       {
          vert_elem->AddAColumnInRow(v[j]);
       }
@@ -7149,11 +7148,11 @@ Table *Mesh::GetVertexToElementTable()
 
    vert_elem->MakeJ();
 
-   for (i = 0; i < NumOfElements; i++)
+   for (int i = 0; i < NumOfElements; i++)
    {
-      nv = elements[i]->GetNVertices();
-      v  = elements[i]->GetVertices();
-      for (j = 0; j < nv; j++)
+      const int nv = elements[i]->GetNVertices();
+      const int *v = elements[i]->GetVertices();
+      for (int j = 0; j < nv; j++)
       {
          vert_elem->AddConnection(v[j], i);
       }
@@ -7164,6 +7163,39 @@ Table *Mesh::GetVertexToElementTable()
    return vert_elem;
 }
 
+Table *Mesh::GetVertexToBdrElementTable()
+{
+   Table *vert_bdr_elem = new Table;
+
+   vert_bdr_elem->MakeI(NumOfVertices);
+
+   for (int i = 0; i < NumOfBdrElements; i++)
+   {
+      const int nv = boundary[i]->GetNVertices();
+      const int *v = boundary[i]->GetVertices();
+      for (int j = 0; j < nv; j++)
+      {
+         vert_bdr_elem->AddAColumnInRow(v[j]);
+      }
+   }
+
+   vert_bdr_elem->MakeJ();
+
+   for (int i = 0; i < NumOfBdrElements; i++)
+   {
+      const int nv = boundary[i]->GetNVertices();
+      const int *v = boundary[i]->GetVertices();
+      for (int j = 0; j < nv; j++)
+      {
+         vert_bdr_elem->AddConnection(v[j], i);
+      }
+   }
+
+   vert_bdr_elem->ShiftUpI();
+
+   return vert_bdr_elem;
+}
+
 Table *Mesh::GetFaceToElementTable() const
 {
    Table *face_elem = new Table;

mesh/mesh.hpp

@@ -857,7 +857,9 @@ public:
        vectors using Mesh::CreatePeriodicVertexMapping.
        @note MFEM requires that each edge of the resulting mesh be uniquely
        identifiable by a pair of distinct vertices. As a consequence, periodic
-       boundaries must be connected by at least three edges. */
+       boundaries must be separated by at least two interior vertices.
+       @note The resulting mesh uses a discontinuous nodal function, see
+       SetCurvature() for further details. */
    static Mesh MakePeriodic(const Mesh &orig_mesh, const std::vector<int> &v2v);
 
    ///@}
@@ -1535,6 +1537,9 @@ public:
    /// @note The returned Table should be deleted by the caller
    Table *GetVertexToElementTable();
 
+   /// @note The returned Table should be deleted by the caller
+   Table *GetVertexToBdrElementTable();
+
    /// Return the "face"-element Table. Here "face" refers to face (3D),
    /// edge (2D), or vertex (1D).
    ///

mesh/mesh_readers.cpp

@@ -1151,15 +1151,24 @@ void Mesh::ReadXML_VTKMesh(std::istream &input, int &curved, int &read_gf,
    }
    if (cells_xml == NULL) { MFEM_ABORT(erstr); }
 
-   // Read the element attributes, which are stored as CellData named "material"
+   // Read the element attributes, which are stored as CellData named either
+   // "material" or "attribute". We prioritize "material" over "attribute" for
+   // backwards compatibility.
    Array<int> cell_attributes;
+   bool found_attributes = false;
    for (const XMLElement *cell_data_xml = piece->FirstChildElement();
         cell_data_xml != NULL;
         cell_data_xml = cell_data_xml->NextSiblingElement())
    {
-      if (StringCompare(cell_data_xml->Name(), "CellData")
-          && StringCompare(cell_data_xml->Attribute("Scalars"), "material"))
+      const bool is_cell_data =
+         StringCompare(cell_data_xml->Name(), "CellData");
+      const bool is_material =
+         StringCompare(cell_data_xml->Attribute("Scalars"), "material");
+      const bool is_attribute =
+         StringCompare(cell_data_xml->Attribute("Scalars"), "attribute");
+      if (is_cell_data && (is_material || (is_attribute && !found_attributes)))
       {
+         found_attributes = true;
          const XMLElement *data_xml = cell_data_xml->FirstChildElement();
          if (data_xml != NULL && StringCompare(data_xml->Name(), "DataArray"))
          {
@@ -1274,6 +1283,7 @@ void Mesh::ReadVTKMesh(std::istream &input, int &curved, int &read_gf,
    // Read the cell materials
    // bool found_material = false;
    Array<int> cell_attributes;
+   bool found_attributes = false;
    while ((input.good()))
    {
       getline(input, buff);
@@ -1281,8 +1291,10 @@ void Mesh::ReadVTKMesh(std::istream &input, int &curved, int &read_gf,
       {
          break; // We have entered the POINT_DATA block. Quit.
       }
-      else if (buff.rfind("SCALARS material") == 0)
+      else if (buff.rfind("SCALARS material") == 0 ||
+               (buff.rfind("SCALARS attribute") == 0 && !found_attributes))
       {
+         found_attributes = true;
          getline(input, buff); // LOOKUP_TABLE default
          if (buff.rfind("LOOKUP_TABLE default") != 0)
          {

mesh/nurbs.cpp

@@ -1857,7 +1857,7 @@ NURBSPatch *Revolve3D(NURBSPatch &patch, real_t n[], real_t ang, int times)
 {
    if (patch.Dim != 4)
    {
-      mfem_error("Revolve3D(NURBSPatch &, double [], double)");
+      mfem_error("Revolve3D(NURBSPatch &, real_t [], real_t)");
    }
 
    int size = 1, ns;
@@ -2008,23 +2008,23 @@ NURBSExtension::NURBSExtension(std::istream &input, bool spacing)
          input >> numSpacing;
          for (int j = 0; j < numSpacing; j++)
          {
-            int ki, spacingType, numIntParam, numDoubleParam;
-            input >> ki >> spacingType >> numIntParam >> numDoubleParam;
+            int ki, spacingType, numIntParam, numRealParam;
+            input >> ki >> spacingType >> numIntParam >> numRealParam;
 
             MFEM_VERIFY(0 <= ki && ki < NumOfKnotVectors,
                         "Invalid knotvector index");
-            MFEM_VERIFY(numIntParam >= 0 && numDoubleParam >= 0,
+            MFEM_VERIFY(numIntParam >= 0 && numRealParam >= 0,
                         "Invalid number of parameters in KnotVector");
 
             Array<int> ipar(numIntParam);
-            Vector dpar(numDoubleParam);
+            Vector dpar(numRealParam);
 
             for (int i=0; i<numIntParam; ++i)
             {
                input >> ipar[i];
             }
 
-            for (int i=0; i<numDoubleParam; ++i)
+            for (int i=0; i<numRealParam; ++i)
             {
                input >> dpar[i];
             }
@@ -2064,7 +2064,7 @@ NURBSExtension::NURBSExtension(std::istream &input, bool spacing)
                   new KnotVector(*patches[p]->GetKV(0));
             }
          }
-         if (Dimension() == 2)
+         else if (Dimension() == 2)
          {
             patchTopo->GetElementEdges(p, edges, oedge);
             if (knotVectors[KnotInd(edges[0])] == NULL)
@@ -2230,7 +2230,8 @@ NURBSExtension::NURBSExtension(NURBSExtension *parent, int newOrder)
 }
 
 NURBSExtension::NURBSExtension(NURBSExtension *parent,
-                               const Array<int> &newOrders)
+                               const Array<int> &newOrders, Mode mode)
+   : mode(mode)
 {
    newOrders.Copy(mOrders);
    SetOrderFromOrders();
@@ -3891,7 +3892,16 @@ void NURBSExtension::GenerateBdrElementDofTable()
    int ndof = bel_dof->Size_of_connections();
    for (int i = 0; i < ndof; i++)
    {
-      dof[i] = activeDof[dof[i]] - 1;
+      int idx = dof[i];
+      if (idx < 0)
+      {
+         dof[i] = -1 - (activeDof[-1-idx] - 1);
+         dof[i] = -activeDof[-1-idx];
+      }
+      else
+      {
+         dof[i] = activeDof[idx] - 1;
+      }
    }
 }
 
@@ -3943,6 +3953,22 @@ void NURBSExtension::Generate2DBdrElementDofTable()
       // Load dofs
       const int nks0 = kv[0]->GetNKS();
       const int ord0 = kv[0]->GetOrder();
+
+      bool add_dofs = true;
+      int  s = 1;
+
+      if (mode == Mode::H_DIV)
+      {
+         int fn = patchTopo->GetBdrElementFaceIndex(b);
+         if (ord0 == mOrders.Max()) { add_dofs = false; }
+         if (fn == 0) { s = -1; }
+         if (fn == 2) { s = -1; }
+      }
+      else if (mode == Mode::H_CURL)
+      {
+         if (ord0 == mOrders.Max()) { add_dofs = false; }
+      }
+
       for (int i = 0; i < nks0; i++)
       {
          if (kv[0]->isElement(i))
@@ -3950,10 +3976,14 @@ void NURBSExtension::Generate2DBdrElementDofTable()
             if (activeBdrElem[gbe])
             {
                Connection conn(lbe,0);
-               for (int ii = 0; ii <= ord0; ii++)
+               if (add_dofs)
                {
-                  conn.to = DofMap(p2g[(okv[0] >= 0) ? (i+ii) : (nx-i-ii)]);
-                  bel_dof_list.Append(conn);
+                  for (int ii = 0; ii <= ord0; ii++)
+                  {
+                     conn.to = DofMap(p2g[(okv[0] >= 0) ? (i+ii) : (nx-i-ii)]);
+                     if (s == -1) { conn.to = -1 -conn.to; }
+                     bel_dof_list.Append(conn);
+                  }
                }
                bel_to_patch[lbe] = b;
                bel_to_IJK(lbe,0) = (okv[0] >= 0) ? i : (-1-i);
@@ -3990,6 +4020,25 @@ void NURBSExtension::Generate3DBdrElementDofTable()
       const int ord0 = kv[0]->GetOrder();
       const int nks1 = kv[1]->GetNKS();
       const int ord1 = kv[1]->GetOrder();
+
+      // Check if dofs are actually defined on boundary
+      bool add_dofs = true;
+      int  s = 1;
+
+      if (mode == Mode::H_DIV)
+      {
+         int fn = patchTopo->GetBdrElementFaceIndex(b);
+         if (ord0 != ord1) { add_dofs = false; }
+         if (fn == 4) { s = -1; }
+         if (fn == 1) { s = -1; }
+         if (fn == 0) { s = -1; }
+      }
+      else if (mode == Mode::H_CURL)
+      {
+         if (ord0 == ord1) { add_dofs = false; }
+      }
+
+
       for (int j = 0; j < nks1; j++)
       {
          if (kv[1]->isElement(j))
@@ -4001,14 +4050,18 @@ void NURBSExtension::Generate3DBdrElementDofTable()
                   if (activeBdrElem[gbe])
                   {
                      Connection conn(lbe,0);
-                     for (int jj = 0; jj <= ord1; jj++)
+                     if (add_dofs)
                      {
-                        const int jj_ = (okv[1] >= 0) ? (j+jj) : (ny-j-jj);
-                        for (int ii = 0; ii <= ord0; ii++)
+                        for (int jj = 0; jj <= ord1; jj++)
                         {
-                           const int ii_ = (okv[0] >= 0) ? (i+ii) : (nx-i-ii);
-                           conn.to = DofMap(p2g(ii_, jj_));
-                           bel_dof_list.Append(conn);
+                           const int jj_ = (okv[1] >= 0) ? (j+jj) : (ny-j-jj);
+                           for (int ii = 0; ii <= ord0; ii++)
+                           {
+                              const int ii_ = (okv[0] >= 0) ? (i+ii) : (nx-i-ii);
+                              conn.to = DofMap(p2g(ii_, jj_));
+                              if (s == -1) { conn.to = -1 -conn.to; }
+                              bel_dof_list.Append(conn);
+                           }
                         }
                      }
                      bel_to_patch[lbe] = b;
@@ -4241,6 +4294,38 @@ void NURBSExtension::DegreeElevate(int rel_degree, int degree)
    }
 }
 
+NURBSExtension* NURBSExtension::GetDivExtension(int component)
+{
+   // Smarter routine
+   if (GetNP() > 1)
+   {
+      mfem_error("NURBSExtension::GetDivExtension currently "
+                 "only works for single patch NURBS meshes ");
+   }
+
+   Array<int> newOrders  = GetOrders();
+   newOrders[component] += 1;
+
+   return new NURBSExtension(this, newOrders, Mode::H_DIV);
+}
+
+NURBSExtension* NURBSExtension::GetCurlExtension(int component)
+{
+   // Smarter routine
+   if (GetNP() > 1)
+   {
+      mfem_error("NURBSExtension::GetCurlExtension currently "
+                 "only works for single patch NURBS meshes ");
+   }
+
+   Array<int> newOrders  = GetOrders();
+   for (int c = 0; c < newOrders.Size(); c++) { newOrders[c]++; }
+   newOrders[component] -= 1;
+
+   return new NURBSExtension(this, newOrders, Mode::H_CURL);
+}
+
+
 void NURBSExtension::UniformRefinement(Array<int> const& rf)
 {
    for (int p = 0; p < patches.Size(); p++)

mesh/nurbs.hpp

@@ -426,6 +426,16 @@ class NURBSExtension
    friend class NURBSPatchMap;
 
 protected:
+
+   /// Flag for indicating what type of NURBS fespace this extension is used for.
+   enum class Mode
+   {
+      H_1,    ///> Extension for a standard scalar-valued space
+      H_DIV,  ///> Extension for a divergence conforming vector-valued space
+      H_CURL, ///> Extension for a curl conforming vector-valued space
+   };
+   Mode mode = Mode::H_1;
+
    /// Order of KnotVectors, see GetOrder() for description.
    int mOrder;
 
@@ -655,8 +665,10 @@ public:
    /** @a note If a KnotVector in @a parent already has order greater than or
        equal to the corresponding entry in @a newOrder, it will be used
        unmodified. */
-   NURBSExtension(NURBSExtension *parent, const Array<int> &newOrders);
+   NURBSExtension(NURBSExtension *parent, const Array<int> &newOrders,
+                  Mode mode = Mode::H_1);
    /// Construct a NURBSExtension by merging a partitioned NURBS mesh.
+
    NURBSExtension(Mesh *mesh_array[], int num_pieces);
 
    /// Copy assignment not supported.
@@ -841,6 +853,16 @@ public:
    void KnotInsert(Array<KnotVector *> &kv);
    void KnotInsert(Array<Vector *> &kv);
 
+   /** Returns the NURBSExtension to be used for @a component of
+       an H(div) conforming NURBS space. Caller gets ownership of
+       the returned object, and is responsible for deletion.*/
+   NURBSExtension* GetDivExtension(int component);
+
+   /** Returns the NURBSExtension to be used for @a component of
+       an H(curl) conforming NURBS space. Caller gets ownership of
+       the returned object, and is responsible for deletion.*/
+   NURBSExtension* GetCurlExtension(int component);
+
    void KnotRemove(Array<Vector *> &kv, real_t tol = 1.0e-12);
 
    /** Calls GetCoarseningFactors for each patch and finds the minimum factor
@@ -848,6 +870,7 @@ public:
        non-nested spacing functions. */
    void GetCoarseningFactors(Array<int> & f) const;
 
+
    /// Returns the index of the patch containing element @a elem.
    int GetElementPatch(int elem) const { return el_to_patch[elem]; }
 

mesh/pumi.cpp

@@ -819,7 +819,7 @@ ParPumiMesh::ParPumiMesh(MPI_Comm comm, apf::Mesh2* apf_mesh,
          apf::Downward verts;
          apf_mesh->getDownward(ent,0,verts);
 
-         int *v, nv = 0;
+         int *v = nullptr, nv = 0;
          apf::Mesh::Type ftype = apf_mesh->getType(ent);
          if (ftype == apf::Mesh::TRIANGLE)
          {
@@ -890,9 +890,9 @@ GridFunctionPumi::GridFunctionPumi(Mesh* m, apf::Mesh2* PumiM,
 {
    int spDim = m->SpaceDimension();
    // Note: default BasisType for 'fec' is GaussLobatto.
-   fec = new H1_FECollection(mesh_order, m->Dimension());
+   fec_owned = new H1_FECollection(mesh_order, m->Dimension());
    int ordering = Ordering::byVDIM; // x1y1z1/x2y2z2/...
-   fes = new FiniteElementSpace(m, fec, spDim, ordering);
+   fes = new FiniteElementSpace(m, fec_owned, spDim, ordering);
    int data_size = fes->GetVSize();
 
    // Read PUMI mesh data

miniapps/CMakeLists.txt

@@ -37,3 +37,4 @@ add_subdirectory(tribol)
 add_subdirectory(hooke)
 add_subdirectory(dpg)
 add_subdirectory(hdiv-linear-solver)
+add_subdirectory(stabilized)

miniapps/dpg/acoustics.cpp

@@ -15,11 +15,11 @@
 //
 // Sample runs
 //
-// acoustics -ref 4 -o 1 -rnum 1.0
-// acoustics -m ../../data/inline-tri.mesh -ref 4 -o 2 -sc -rnum 3.0
-// acoustics -m ../../data/amr-quad.mesh -ref 3 -o 3 -sc -rnum 4.5 -prob 1
-// acoustics -m ../../data/inline-quad.mesh -ref 2 -o 4 -sc -rnum 11.5 -prob 1
-// acoustics -m ../../data/inline-hex.mesh -ref 1 -o 2 -sc -rnum 1.0
+//  acoustics -ref 4 -o 1 -rnum 1.0
+//  acoustics -m ../../data/inline-tri.mesh -ref 4 -o 2 -sc -rnum 3.0
+//  acoustics -m ../../data/amr-quad.mesh -ref 3 -o 3 -sc -rnum 4.5 -prob 1
+//  acoustics -m ../../data/inline-quad.mesh -ref 2 -o 4 -sc -rnum 11.5 -prob 1
+//  acoustics -m ../../data/inline-hex.mesh -ref 1 -o 2 -sc -rnum 1.0
 
 // Description:
 // This example code demonstrates the use of MFEM to define and solve

miniapps/dpg/convection-diffusion.cpp

@@ -14,14 +14,14 @@
 // Compile with: make convection-diffusion
 //
 // sample runs
-// convection-diffusion -m ../../data/star.mesh -o 2 -ref 2 -theta 0.0 -eps 1e-1 -beta '2 3'
-// convection-diffusion -m ../../data/beam-hex.mesh -o 2 -ref 2 -theta 0.0 -eps 1e0 -beta '1 0 2'
-// convection-diffusion -m ../../data/inline-tri.mesh -o 3 -ref 2 -theta 0.0 -eps 1e-2 -beta '4 2' -sc
+//  convection-diffusion -m ../../data/star.mesh -o 2 -ref 2 -theta 0.0 -eps 1e-1 -beta '2 3'
+//  convection-diffusion -m ../../data/beam-hex.mesh -o 2 -ref 2 -theta 0.0 -eps 1e0 -beta '1 0 2'
+//  convection-diffusion -m ../../data/inline-tri.mesh -o 3 -ref 2 -theta 0.0 -eps 1e-2 -beta '4 2' -sc
 
 // AMR runs
-// convection-diffusion  -o 3 -ref 5 -prob 1 -eps 1e-1 -theta 0.75
-// convection-diffusion  -o 2 -ref 9 -prob 1 -eps 1e-2 -theta 0.75
-// convection-diffusion  -o 3 -ref 9 -prob 1 -eps 1e-3 -theta 0.75 -sc
+//  convection-diffusion -o 3 -ref 5 -prob 1 -eps 1e-1 -theta 0.75
+//  convection-diffusion -o 2 -ref 9 -prob 1 -eps 1e-2 -theta 0.75
+//  convection-diffusion -o 3 -ref 9 -prob 1 -eps 1e-3 -theta 0.75 -sc
 
 // Description:
 // This example code demonstrates the use of MFEM to define and solve

miniapps/dpg/maxwell.cpp

@@ -14,10 +14,10 @@
 // Compile with: make maxwell
 //
 // Sample runs
-// maxwell -m ../../data/inline-tri.mesh -ref 4 -o 1 -rnum 1.0
-// maxwell -m ../../data/amr-quad.mesh -ref 3 -o 2 -rnum 1.6 -sc
-// maxwell -m ../../data/inline-quad.mesh -ref 2 -o 3 -rnum 4.2 -sc
-// maxwell -m ../../data/inline-hex.mesh -ref 1 -o 2 -sc -rnum 1.0
+//  maxwell -m ../../data/inline-tri.mesh -ref 4 -o 1 -rnum 1.0
+//  maxwell -m ../../data/amr-quad.mesh -ref 3 -o 2 -rnum 1.6 -sc
+//  maxwell -m ../../data/inline-quad.mesh -ref 2 -o 3 -rnum 4.2 -sc
+//  maxwell -m ../../data/inline-hex.mesh -ref 1 -o 2 -sc -rnum 1.0
 
 // Description:
 // This example code demonstrates the use of MFEM to define and solve

miniapps/dpg/pacoustics.cpp

@@ -15,19 +15,19 @@
 //
 // sample runs
 
-// mpirun -np 4 pacoustics -o 3 -m ../../data/star.mesh -sref 1 -pref 2 -rnum 1.9 -sc -prob 0
-// mpirun -np 4 pacoustics -o 3 -m ../../data/inline-quad.mesh -sref 1 -pref 2  -rnum 5.2 -sc -prob 1
-// mpirun -np 4 pacoustics -o 4 -m ../../data/inline-tri.mesh -sref 1 -pref 2  -rnum 7.1 -sc -prob 1
-// mpirun -np 4 pacoustics -o 2 -m ../../data/inline-hex.mesh -sref 0 -pref 1 -rnum 1.9 -sc -prob 0
-// mpirun -np 4 pacoustics -o 3 -m ../../data/inline-quad.mesh -sref 2 -pref 1 -rnum 7.1 -sc -prob 2
-// mpirun -np 4 pacoustics -o 2 -m ../../data/inline-hex.mesh -sref 0 -pref 1  -rnum 4.1 -sc -prob 2
-// mpirun -np 4 pacoustics -o 3 -m meshes/scatter.mesh -sref 1 -pref 1  -rnum 7.1 -sc -prob 3
-// mpirun -np 4 pacoustics -o 4 -m meshes/scatter.mesh -sref 1 -pref 1  -rnum 10.1 -sc -prob 4
-// mpirun -np 4 pacoustics -o 4 -m meshes/scatter.mesh -sref 1 -pref 1  -rnum 12.1 -sc -prob 5
+//  mpirun -np 4 pacoustics -o 3 -m ../../data/star.mesh -sref 1 -pref 2 -rnum 1.9 -sc -prob 0
+//  mpirun -np 4 pacoustics -o 3 -m ../../data/inline-quad.mesh -sref 1 -pref 2  -rnum 5.2 -sc -prob 1
+//  mpirun -np 4 pacoustics -o 4 -m ../../data/inline-tri.mesh -sref 1 -pref 2  -rnum 7.1 -sc -prob 1
+//  mpirun -np 4 pacoustics -o 2 -m ../../data/inline-hex.mesh -sref 0 -pref 1 -rnum 1.9 -sc -prob 0
+//  mpirun -np 4 pacoustics -o 3 -m ../../data/inline-quad.mesh -sref 2 -pref 1 -rnum 7.1 -sc -prob 2
+//  mpirun -np 4 pacoustics -o 2 -m ../../data/inline-hex.mesh -sref 0 -pref 1  -rnum 4.1 -sc -prob 2
+//  mpirun -np 4 pacoustics -o 3 -m meshes/scatter.mesh -sref 1 -pref 1  -rnum 7.1 -sc -prob 3
+//  mpirun -np 4 pacoustics -o 4 -m meshes/scatter.mesh -sref 1 -pref 1  -rnum 10.1 -sc -prob 4
+//  mpirun -np 4 pacoustics -o 4 -m meshes/scatter.mesh -sref 1 -pref 1  -rnum 12.1 -sc -prob 5
 
 // AMR runs
-// mpirun -np 4 pacoustics -o 3 -m meshes/scatter.mesh -sref 0 -pref 7 -theta 0.75 -rnum 10.1 -sc -prob 3
-// mpirun -np 4 pacoustics -o 3 -m meshes/scatter.mesh -sref 0 -pref 12 -theta 0.75 -rnum 20.1 -sc -prob 3
+//  mpirun -np 4 pacoustics -o 3 -m meshes/scatter.mesh -sref 0 -pref 7 -theta 0.75 -rnum 10.1 -sc -prob 3
+//  mpirun -np 4 pacoustics -o 3 -m meshes/scatter.mesh -sref 0 -pref 12 -theta 0.75 -rnum 20.1 -sc -prob 3
 
 // Description:
 // This example code demonstrates the use of MFEM to define and solve

miniapps/dpg/pconvection-diffusion.cpp

@@ -11,17 +11,17 @@
 //
 //             MFEM Ultraweak DPG parallel example for convection-diffusion
 //
-// Compile with: make  pconvection-diffusion
+// Compile with: make pconvection-diffusion
 //
 // sample runs
-// mpirun -np 4 pconvection-diffusion  -o 2 -ref 3 -prob 0 -eps 1e-1 -beta '4 2' -theta 0.0
-// mpirun -np 4 pconvection-diffusion  -o 3 -ref 3 -prob 0 -eps 1e-2 -beta '2 3' -theta 0.0
-// mpirun -np 4 pconvection-diffusion -m ../../data/inline-hex.mesh -o 2 -ref 1 -prob 0 -sc -eps 1e-1 -theta 0.0
+//  mpirun -np 4 pconvection-diffusion -o 2 -ref 3 -prob 0 -eps 1e-1 -beta '4 2' -theta 0.0
+//  mpirun -np 4 pconvection-diffusion -o 3 -ref 3 -prob 0 -eps 1e-2 -beta '2 3' -theta 0.0
+//  mpirun -np 4 pconvection-diffusion -m ../../data/inline-hex.mesh -o 2 -ref 1 -prob 0 -sc -eps 1e-1 -theta 0.0
 
 // AMR runs
-// mpirun -np 4 pconvection-diffusion  -o 3 -ref 10 -prob 1 -eps 1e-3 -beta '1 0' -theta 0.7 -sc
-// mpirun -np 4 pconvection-diffusion  -o 3 -ref 15 -prob 2 -eps 5e-3 -theta 0.7 -sc
-// mpirun -np 4 pconvection-diffusion  -o 2 -ref 12 -prob 3 -eps 1e-2 -beta '1 2' -theta 0.7 -sc
+//  mpirun -np 4 pconvection-diffusion -o 3 -ref 10 -prob 1 -eps 1e-3 -beta '1 0' -theta 0.7 -sc
+//  mpirun -np 4 pconvection-diffusion -o 3 -ref 15 -prob 2 -eps 5e-3 -theta 0.7 -sc
+//  mpirun -np 4 pconvection-diffusion -o 2 -ref 12 -prob 3 -eps 1e-2 -beta '1 2' -theta 0.7 -sc
 
 // Description:
 // This example code demonstrates the use of MFEM to define and solve a parallel

miniapps/dpg/pdiffusion.cpp

@@ -14,18 +14,18 @@
 // Compile with: make pdiffusion
 //
 // Sample runs
-// mpirun -np 4 pdiffusion -m ../../data/inline-quad.mesh -o 3 -sref 1 -pref 2 -theta 0.0 -prob 0
-// mpirun -np 4 pdiffusion -m ../../data/inline-hex.mesh -o 2 -sref 0 -pref 1 -theta 0.0 -prob 0 -sc
-// mpirun -np 4 pdiffusion -m ../../data/beam-tet.mesh -o 3 -sref 0 -pref 2 -theta 0.0 -prob 0 -sc
+//  mpirun -np 4 pdiffusion -m ../../data/inline-quad.mesh -o 3 -sref 1 -pref 2 -theta 0.0 -prob 0
+//  mpirun -np 4 pdiffusion -m ../../data/inline-hex.mesh -o 2 -sref 0 -pref 1 -theta 0.0 -prob 0 -sc
+//  mpirun -np 4 pdiffusion -m ../../data/beam-tet.mesh -o 3 -sref 0 -pref 2 -theta 0.0 -prob 0 -sc
 
 // L-shape runs
 // Note: uniform ref are expected to give sub-optimal rate for the L-shape problem (rate = 2/3)
-// mpirun -np 4 pdiffusion -o 2 -sref 1 -pref 5 -theta 0.0 -prob 1
+//  mpirun -np 4 pdiffusion -o 2 -sref 1 -pref 5 -theta 0.0 -prob 1
 
 // L-shape AMR runs
-// mpirun -np 4 pdiffusion -o 1 -sref 1 -pref 10 -theta 0.8 -prob 1
-// mpirun -np 4 pdiffusion -o 2 -sref 1 -pref 8 -theta 0.75 -prob 1 -sc
-// mpirun -np 4 pdiffusion -o 3 -sref 1 -pref 6 -theta 0.75 -prob 1 -sc -do 2
+//  mpirun -np 4 pdiffusion -o 1 -sref 1 -pref 10 -theta 0.8 -prob 1
+//  mpirun -np 4 pdiffusion -o 2 -sref 1 -pref 8 -theta 0.75 -prob 1 -sc
+//  mpirun -np 4 pdiffusion -o 3 -sref 1 -pref 6 -theta 0.75 -prob 1 -sc -do 2
 
 // Description:
 // This example code demonstrates the use of MFEM to define and solve

miniapps/dpg/pmaxwell.cpp

@@ -14,16 +14,16 @@
 // Compile with: make pmaxwell
 //
 // sample run
-// mpirun -np 4 pmaxwell -m ../../data/star.mesh -o 2 -sref 0 -pref 3 -rnum 0.5 -prob 0
-// mpirun -np 4 pmaxwell -m ../../data/inline-quad.mesh -o 3 -sref 0 -pref 3 -rnum 4.8 -sc -prob 0
-// mpirun -np 4 pmaxwell -m ../../data/inline-hex.mesh -o 2 -sref 0 -pref 1 -rnum 0.8 -sc -prob 0
-// mpirun -np 4 pmaxwell -m ../../data/inline-quad.mesh -o 3 -sref 1 -pref 3 -rnum 4.8 -sc -prob 2
-// mpirun -np 4 pmaxwell -o 3 -sref 1 -pref 2 -rnum 11.8 -sc -prob 3
-// mpirun -np 4 pmaxwell -o 3 -sref 1 -pref 2 -rnum 9.8 -sc -prob 4
+//  mpirun -np 4 pmaxwell -m ../../data/star.mesh -o 2 -sref 0 -pref 3 -rnum 0.5 -prob 0
+//  mpirun -np 4 pmaxwell -m ../../data/inline-quad.mesh -o 3 -sref 0 -pref 3 -rnum 4.8 -sc -prob 0
+//  mpirun -np 4 pmaxwell -m ../../data/inline-hex.mesh -o 2 -sref 0 -pref 1 -rnum 0.8 -sc -prob 0
+//  mpirun -np 4 pmaxwell -m ../../data/inline-quad.mesh -o 3 -sref 1 -pref 3 -rnum 4.8 -sc -prob 2
+//  mpirun -np 4 pmaxwell -o 3 -sref 1 -pref 2 -rnum 11.8 -sc -prob 3
+//  mpirun -np 4 pmaxwell -o 3 -sref 1 -pref 2 -rnum 9.8 -sc -prob 4
 
 // AMR run. Note that this is a computationally intensive sample run.
 // We recommend trying it on a large machine with more mpi ranks
-// mpirun -np 4 pmaxwell -o 3 -sref 0 -pref 15 -prob 1 -theta 0.7 -sc
+//  mpirun -np 4 pmaxwell -o 3 -sref 0 -pref 15 -prob 1 -theta 0.7 -sc
 
 // Description:
 // This example code demonstrates the use of MFEM to define and solve

miniapps/meshing/fit-node-position.cpp

@@ -206,7 +206,7 @@ int main (int argc, char *argv[])
    solver.SetMaxIter(200);
    solver.SetRelTol(1e-10);
    solver.SetAbsTol(0.0);
-   solver.EnableAdaptiveSurfaceFitting();
+   solver.SetAdaptiveSurfaceFittingScalingFactor(10);
    solver.SetTerminationWithMaxSurfaceFittingError(1e-3);
 
    // Solve.

miniapps/meshing/mesh-fitting.hpp

@@ -34,6 +34,22 @@ real_t circle_level_set(const Vector &x)
    }
 }
 
+real_t squircle_level_set(const Vector &x)
+{
+   const int dim = x.Size();
+   if (dim == 2)
+   {
+      const real_t xc = x(0) - 0.5, yc = x(1) - 0.5;
+      return std::pow(xc, 4.0) + std::pow(yc, 4.0) - std::pow(0.24, 4.0);
+   }
+   else
+   {
+      const real_t xc = x(0) - 0.5, yc = x(1) - 0.5, zc = x(2) - 0.5;
+      return std::pow(xc, 4.0) + std::pow(yc, 4.0) +
+             std::pow(zc, 4.0) - std::pow(0.24, 4.0);
+   }
+}
+
 real_t in_circle(const Vector &x, const Vector &x_center, real_t radius)
 {
    Vector x_current = x;

miniapps/meshing/pmesh-fitting.cpp

@@ -33,13 +33,15 @@
 // Compile with: make pmesh-fitting
 //
 // Sample runs:
-//  Interface fitting:
-//    mpirun -np 4 pmesh-fitting -o 3 -mid 58 -tid 1 -ni 200 -vl 1 -sfc 5e4 -rtol 1e-5
-//    mpirun -np 4 pmesh-fitting -m square01-tri.mesh -o 3 -rs 0 -mid 58 -tid 1 -ni 200 -vl 1 -sfc 1e4 -rtol 1e-5
+//  Surface fitting:
+//    mpirun -np 4 pmesh-fitting -o 3 -mid 58 -tid 1 -vl 1 -sfc 5e4 -rtol 1e-5
+//    mpirun -np 4 pmesh-fitting -m square01-tri.mesh -o 3 -rs 0 -mid 58 -tid 1 -vl 1 -sfc 1e4 -rtol 1e-5
 //  Surface fitting with weight adaptation and termination based on fitting error:
-//    mpirun -np 4 pmesh-fitting -o 2 -mid 2 -tid 1 -ni 100 -vl 2 -sfc 10 -rtol 1e-20 -st 0 -sfa 10.0 -sft 1e-5
-//  Fitting to Fischer-Tropsch reactor like domain (requires GSLIB):
-//  * mpirun -np 6 pmesh-fitting -m ../../data/inline-tri.mesh -o 2 -rs 4 -mid 2 -tid 1 -vl 2 -sfc 100 -rtol 1e-12 -ni 100 -li 40 -ae 1 -bnd -sbgmesh -slstype 2 -smtype 0 -sfa 10.0 -sft 1e-4 -amriter 5 -dist -mod-bndr-attr
+//    mpirun -np 4 pmesh-fitting -o 2 -mid 2 -tid 1 -vl 2 -sfc 10 -rtol 1e-20 -sfa 10.0 -sft 1e-5 -no-resid
+//  Surface fitting with weight adaptation, limit on max weight, and convergence based on residual.
+//  * mpirun -np 4 pmesh-fitting -m ../../data/inline-tri.mesh -o 2 -mid 2 -tid 4 -vl 2 -sfc 10 -rtol 1e-10 -sfa 10.0 -sft 1e-5 -bgamriter 3 -sbgmesh -ae 1 -marking -slstype 3 -resid -sfcmax 10000 -mod-bndr-attr
+//  Surface fitting to Fischer-Tropsch reactor like domain (requires GSLIB):
+//  * mpirun -np 6 pmesh-fitting -m ../../data/inline-tri.mesh -o 2 -rs 4 -mid 2 -tid 1 -vl 2 -sfc 100 -rtol 1e-12 -li 20 -ae 1 -bnd -sbgmesh -slstype 2 -smtype 0 -sfa 10.0 -sft 1e-4 -no-resid -bgamriter 5 -dist -mod-bndr-attr
 
 #include "mesh-fitting.hpp"
 
@@ -48,12 +50,16 @@ using namespace std;
 
 int main (int argc, char *argv[])
 {
-   // 0. Initialize MPI and HYPRE.
+#ifdef HYPRE_USING_GPU
+   cout << "\nThis miniapp is NOT supported with the GPU version of hypre.\n\n";
+   return MFEM_SKIP_RETURN_VALUE;
+#endif
+
    Mpi::Init(argc, argv);
    int myid = Mpi::WorldRank();
    Hypre::Init();
 
-   // 1. Set the method's default parameters.
+   // Set the method's default parameters.
    const char *mesh_file = "square01.mesh";
    int mesh_poly_deg     = 1;
    int rs_levels         = 1;
@@ -61,7 +67,6 @@ int main (int argc, char *argv[])
    int metric_id         = 2;
    int target_id         = 1;
    real_t surface_fit_const = 100.0;
-   int quad_type         = 1;
    int quad_order        = 8;
    int solver_type       = 0;
    int solver_iter       = 20;
@@ -70,27 +75,28 @@ int main (int argc, char *argv[])
 #else
    real_t solver_rtol    = 1e-10;
 #endif
-   int solver_art_type   = 0;
    int lin_solver        = 2;
    int max_lin_iter      = 100;
    bool move_bnd         = true;
-   bool visualization    = true;
+   bool visualization    = false;
    int verbosity_level   = 0;
    int adapt_eval        = 0;
    const char *devopt    = "cpu";
    real_t surface_fit_adapt = 0.0;
    real_t surface_fit_threshold = -10;
+   real_t surf_fit_const_max    = 1e20;
    bool adapt_marking     = false;
-   bool surf_bg_mesh     = false;
-   bool comp_dist     = false;
-   int surf_ls_type      = 1;
-   int marking_type      = 0;
-   bool mod_bndr_attr    = false;
-   bool material         = false;
+   bool surf_bg_mesh      = false;
+   bool comp_dist         = false;
+   int surf_ls_type       = 1;
+   int marking_type       = 0;
+   bool mod_bndr_attr     = false;
+   bool material          = false;
    int mesh_node_ordering = 0;
-   int amr_iters         = 0;
+   int bg_amr_iters       = 0;
+   bool conv_residual     = true;
 
-   // 2. Parse command-line options.
+   // Parse command-line options.
    OptionsParser args(argc, argv);
    args.AddOption(&mesh_file, "-m", "--mesh",
                   "Mesh file to use.");
@@ -111,11 +117,6 @@ int main (int argc, char *argv[])
                   "5: Ideal shape, given size (in physical space)");
    args.AddOption(&surface_fit_const, "-sfc", "--surface-fit-const",
                   "Surface preservation constant.");
-   args.AddOption(&quad_type, "-qt", "--quad-type",
-                  "Quadrature rule type:\n\t"
-                  "1: Gauss-Lobatto\n\t"
-                  "2: Gauss-Legendre\n\t"
-                  "3: Closed uniform points");
    args.AddOption(&quad_order, "-qo", "--quad_order",
                   "Order of the quadrature rule.");
    args.AddOption(&solver_type, "-st", "--solver-type",
@@ -124,11 +125,6 @@ int main (int argc, char *argv[])
                   "Maximum number of Newton iterations.");
    args.AddOption(&solver_rtol, "-rtol", "--newton-rel-tolerance",
                   "Relative tolerance for the Newton solver.");
-   args.AddOption(&solver_art_type, "-art", "--adaptive-rel-tol",
-                  "Type of adaptive relative linear solver tolerance:\n\t"
-                  "0: None (default)\n\t"
-                  "1: Eisenstat-Walker type 1\n\t"
-                  "2: Eisenstat-Walker type 2");
    args.AddOption(&lin_solver, "-ls", "--lin-solver",
                   "Linear solver:\n\t"
                   "0: l1-Jacobi\n\t"
@@ -151,10 +147,12 @@ int main (int argc, char *argv[])
    args.AddOption(&devopt, "-d", "--device",
                   "Device configuration string, see Device::Configure().");
    args.AddOption(&surface_fit_adapt, "-sfa", "--adaptive-surface-fit",
-                  "Enable or disable adaptive surface fitting.");
+                  "Scaling factor for surface fitting weight.");
    args.AddOption(&surface_fit_threshold, "-sft", "--surf-fit-threshold",
                   "Set threshold for surface fitting. TMOP solver will"
                   "terminate when max surface fitting error is below this limit");
+   args.AddOption(&surf_fit_const_max, "-sfcmax", "--surf-fit-const-max",
+                  "Max surface fitting weight allowed");
    args.AddOption(&adapt_marking, "-marking", "--adaptive-marking", "-no-amarking",
                   "--no-adaptive-marking",
                   "Enable or disable adaptive marking surface fitting.");
@@ -165,9 +163,9 @@ int main (int argc, char *argv[])
                   "-no-dist","--no-comp-dist",
                   "Compute distance from 0 level set or not.");
    args.AddOption(&surf_ls_type, "-slstype", "--surf-ls-type",
-                  "1 - Circle (DEFAULT), 2 - Squircle, 3 - Butterfly.");
+                  "1 - Circle (DEFAULT), 2 - reactor level-set, 3 - squircle.");
    args.AddOption(&marking_type, "-smtype", "--surf-marking-type",
-                  "1 - Interface (DEFAULT), 2 - Boundary attribute.");
+                  "0 - Interface (DEFAULT), otherwise Boundary attribute.");
    args.AddOption(&mod_bndr_attr, "-mod-bndr-attr", "--modify-boundary-attribute",
                   "-fix-bndr-attr", "--fix-boundary-attribute",
                   "Change boundary attribute based on alignment with Cartesian axes.");
@@ -176,8 +174,11 @@ int main (int argc, char *argv[])
    args.AddOption(&mesh_node_ordering, "-mno", "--mesh_node_ordering",
                   "Ordering of mesh nodes."
                   "0 (default): byNodes, 1: byVDIM");
-   args.AddOption(&amr_iters, "-amriter", "--amr-iter",
+   args.AddOption(&bg_amr_iters, "-bgamriter", "--amr-iter",
                   "Number of amr iterations on background mesh");
+   args.AddOption(&conv_residual, "-resid", "--resid", "-no-resid",
+                  "--no-resid",
+                  "Enable residual based convergence.");
    args.Parse();
    if (!args.Good())
    {
@@ -189,7 +190,11 @@ int main (int argc, char *argv[])
    Device device(devopt);
    if (myid == 0) { device.Print();}
 
-   // 3. Initialize and refine the starting mesh.
+   MFEM_VERIFY(surface_fit_const > 0.0,
+               "This miniapp is for surface fitting only. See (p)mesh-optimizer"
+               "miniapps for general high-order mesh optimization.");
+
+   // Initialize and refine the starting mesh.
    Mesh *mesh = new Mesh(mesh_file, 1, 1, false);
    for (int lev = 0; lev < rs_levels; lev++)
    {
@@ -207,6 +212,10 @@ int main (int argc, char *argv[])
    {
       ls_coeff = new FunctionCoefficient(reactor);
    }
+   else if (surf_ls_type == 3) //squircle
+   {
+      ls_coeff = new FunctionCoefficient(squircle_level_set);
+   }
    else if (surf_ls_type == 6) // 3D shape
    {
       ls_coeff = new FunctionCoefficient(csg_cubecylsph);
@@ -220,7 +229,7 @@ int main (int argc, char *argv[])
    delete mesh;
    for (int lev = 0; lev < rp_levels; lev++) { pmesh->UniformRefinement(); }
 
-   // 4. Setup background mesh for surface fitting
+   // Setup background mesh for surface fitting
    ParMesh *pmesh_surf_fit_bg = NULL;
    if (surf_bg_mesh)
    {
@@ -240,10 +249,10 @@ int main (int argc, char *argv[])
       delete mesh_surf_fit_bg;
    }
 
-   // 5. Define a finite element space on the mesh. Here we use vector finite
-   //    elements which are tensor products of quadratic finite elements. The
-   //    number of components in the vector finite element space is specified by
-   //    the last parameter of the FiniteElementSpace constructor.
+   // Define a finite element space on the mesh. Here we use vector finite
+   // elements which are tensor products of quadratic finite elements. The
+   // number of components in the vector finite element space is specified by
+   // the last parameter of the FiniteElementSpace constructor.
    FiniteElementCollection *fec;
    if (mesh_poly_deg <= 0)
    {
@@ -254,21 +263,21 @@ int main (int argc, char *argv[])
    ParFiniteElementSpace *pfespace =
       new ParFiniteElementSpace(pmesh, fec, dim, mesh_node_ordering);
 
-   // 6. Make the mesh curved based on the above finite element space. This
-   //    means that we define the mesh elements through a fespace-based
-   //    transformation of the reference element.
+   // Make the mesh curved based on the above finite element space. This
+   // means that we define the mesh elements through a fespace-based
+   // transformation of the reference element.
    pmesh->SetNodalFESpace(pfespace);
 
-   // 7. Get the mesh nodes (vertices and other degrees of freedom in the finite
-   //    element space) as a finite element grid function in fespace. Note that
-   //    changing x automatically changes the shapes of the mesh elements.
+   // Get the mesh nodes (vertices and other degrees of freedom in the finite
+   // element space) as a finite element grid function in fespace. Note that
+   // changing x automatically changes the shapes of the mesh elements.
    ParGridFunction x(pfespace);
    pmesh->SetNodalGridFunction(&x);
    x.SetTrueVector();
 
-   // 10. Save the starting (prior to the optimization) mesh to a file. This
-   //     output can be viewed later using GLVis: "glvis -m perturbed -np
-   //     num_mpi_tasks".
+   // Save the starting (prior to the optimization) mesh to a file. This
+   // output can be viewed later using GLVis: "glvis -m perturbed -np
+   // num_mpi_tasks".
    {
       ostringstream mesh_name;
       mesh_name << "perturbed.mesh";
@@ -326,16 +335,7 @@ int main (int argc, char *argv[])
    TMOP_Integrator *tmop_integ = new TMOP_Integrator(metric, target_c);
 
    // Setup the quadrature rules for the TMOP integrator.
-   IntegrationRules *irules = NULL;
-   switch (quad_type)
-   {
-      case 1: irules = &IntRulesLo; break;
-      case 2: irules = &IntRules; break;
-      case 3: irules = &IntRulesCU; break;
-      default:
-         if (myid == 0) { cout << "Unknown quad_type: " << quad_type << endl; }
-         return 3;
-   }
+   IntegrationRules *irules = &IntRulesLo;
    tmop_integ->SetIntegrationRules(*irules, quad_order);
    if (myid == 0 && dim == 2)
    {
@@ -423,7 +423,7 @@ int main (int argc, char *argv[])
       if (surf_bg_mesh)
       {
          OptimizeMeshWithAMRAroundZeroLevelSet(*pmesh_surf_fit_bg, *ls_coeff,
-                                               amr_iters, *surf_fit_bg_gf0);
+                                               bg_amr_iters, *surf_fit_bg_gf0);
          pmesh_surf_fit_bg->Rebalance();
          surf_fit_bg_fes->Update();
          surf_fit_bg_gf0->Update();
@@ -505,6 +505,7 @@ int main (int argc, char *argv[])
          ModifyAttributeForMarkingDOFS(pmesh, mat, 0);
          ModifyAttributeForMarkingDOFS(pmesh, mat, 1);
       }
+      pmesh->SetAttributes();
 
       GridFunctionCoefficient coeff_mat(&mat);
       surf_fit_mat_gf.ProjectDiscCoefficient(coeff_mat,
@@ -579,6 +580,22 @@ int main (int argc, char *argv[])
          }
       }
 
+      // Unify marker across processor boundary
+      surf_fit_mat_gf.ExchangeFaceNbrData();
+      {
+         GroupCommunicator &gcomm = surf_fit_mat_gf.ParFESpace()->GroupComm();
+         Array<real_t> gf_array(surf_fit_mat_gf.GetData(),
+                                surf_fit_mat_gf.Size());
+         gcomm.Reduce<real_t>(gf_array, GroupCommunicator::Max);
+         gcomm.Bcast(gf_array);
+      }
+      surf_fit_mat_gf.ExchangeFaceNbrData();
+
+      for (int i = 0; i < surf_fit_mat_gf.Size(); i++)
+      {
+         surf_fit_marker[i] = surf_fit_mat_gf(i) == 1.0;
+      }
+
       // Set AdaptivityEvaluators for transferring information from initial
       // mesh to current mesh as it moves during adaptivity.
       if (adapt_eval == 0)
@@ -590,11 +607,8 @@ int main (int argc, char *argv[])
       {
 #ifdef MFEM_USE_GSLIB
          adapt_surface = new InterpolatorFP;
-         if (surf_bg_mesh)
-         {
-            adapt_grad_surface = new InterpolatorFP;
-            adapt_hess_surface = new InterpolatorFP;
-         }
+         adapt_grad_surface = new InterpolatorFP;
+         adapt_hess_surface = new InterpolatorFP;
 #else
          MFEM_ABORT("MFEM is not built with GSLIB support!");
 #endif
@@ -604,7 +618,9 @@ int main (int argc, char *argv[])
       if (!surf_bg_mesh)
       {
          tmop_integ->EnableSurfaceFitting(surf_fit_gf0, surf_fit_marker,
-                                          surf_fit_coeff, *adapt_surface);
+                                          surf_fit_coeff, *adapt_surface,
+                                          adapt_grad_surface,
+                                          adapt_hess_surface);
       }
       else
       {
@@ -632,16 +648,9 @@ int main (int argc, char *argv[])
          }
       }
    }
-   pmesh->SetAttributes();
 
-   // 13. Setup the final NonlinearForm (which defines the integral of interest,
-   //     its first and second derivatives). Here we can use a combination of
-   //     metrics, i.e., optimize the sum of two integrals, where both are
-   //     scaled by used-defined space-dependent weights.  Note that there are
-   //     no command-line options for the weights and the type of the second
-   //     metric; one should update those in the code.
+   // Setup the final NonlinearForm.
    ParNonlinearForm a(pfespace);
-   ConstantCoefficient *metric_coeff1 = NULL;
    a.AddDomainIntegrator(tmop_integ);
 
    // Compute the minimum det(J) of the starting mesh.
@@ -674,10 +683,10 @@ int main (int argc, char *argv[])
       surf_fit_coeff.constant  = surface_fit_const;
    }
 
-   // 14. Fix all boundary nodes, or fix only a given component depending on the
-   //     boundary attributes of the given mesh.  Attributes 1/2/3 correspond to
-   //     fixed x/y/z components of the node.  Attribute dim+1 corresponds to
-   //     an entirely fixed node.
+   // Fix all boundary nodes, or fix only a given component depending on the
+   // boundary attributes of the given mesh.  Attributes 1/2/3 correspond to
+   // fixed x/y/z components of the node.  Attribute dim+1 corresponds to
+   // an entirely fixed node.
    if (move_bnd == false)
    {
       Array<int> ess_bdr(pmesh->bdr_attributes.Max());
@@ -733,8 +742,7 @@ int main (int argc, char *argv[])
       a.SetEssentialVDofs(ess_vdofs);
    }
 
-   // 15. As we use the Newton method to solve the resulting nonlinear system,
-   //     here we setup the linear solver for the system's Jacobian.
+   // Setup the linear solver for the system's Jacobian.
    Solver *S = NULL, *S_prec = NULL;
 #ifdef MFEM_USE_SINGLE
    const real_t linsol_rtol = 1e-5;
@@ -784,8 +792,14 @@ int main (int argc, char *argv[])
    }
    if (surface_fit_threshold > 0)
    {
-      solver.SetTerminationWithMaxSurfaceFittingError(surface_fit_threshold);
+      solver.SetSurfaceFittingMaxErrorLimit(surface_fit_threshold);
    }
+   solver.SetSurfaceFittingConvergenceBasedOnError(!conv_residual);
+   if (conv_residual)
+   {
+      solver.SetSurfaceFittingWeightLimit(surf_fit_const_max);
+   }
+
    // Provide all integration rules in case of a mixed mesh.
    solver.SetIntegrationRules(*irules, quad_order);
    if (solver_type == 0)
@@ -797,18 +811,14 @@ int main (int argc, char *argv[])
    solver.SetRelTol(solver_rtol);
    solver.SetAbsTol(0.0);
    solver.SetMinimumDeterminantThreshold(0.001*min_detJ);
-   if (solver_art_type > 0)
-   {
-      solver.SetAdaptiveLinRtol(solver_art_type, 0.5, 0.9);
-   }
    solver.SetPrintLevel(verbosity_level >= 1 ? 1 : -1);
    solver.SetOperator(a);
    Vector b(0);
    solver.Mult(b, x.GetTrueVector());
    x.SetFromTrueVector();
 
-   // 16. Save the optimized mesh to a file. This output can be viewed later
-   //     using GLVis: "glvis -m optimized -np num_mpi_tasks".
+   // Save the optimized mesh to a file. This output can be viewed later
+   // using GLVis: "glvis -m optimized -np num_mpi_tasks".
    {
       ostringstream mesh_name;
       mesh_name << "optimized.mesh";
@@ -842,9 +852,13 @@ int main (int argc, char *argv[])
 
    if (surface_fit_const > 0.0)
    {
+      adapt_surface->ComputeAtNewPosition(x, surf_fit_gf0,
+                                          x.FESpace()->GetOrdering());
       if (visualization)
       {
-         socketstream vis2, vis3;
+         socketstream vis1, vis2, vis3;
+         common::VisualizeField(vis1, "localhost", 19916, surf_fit_gf0,
+                                "Level Set", 000, 400, 300, 300);
          common::VisualizeField(vis2, "localhost", 19916, mat,
                                 "Materials", 300, 400, 300, 300);
          common::VisualizeField(vis3, "localhost", 19916, surf_fit_mat_gf,
@@ -859,7 +873,7 @@ int main (int argc, char *argv[])
       }
    }
 
-   // 18. Visualize the mesh displacement.
+   // Visualize the mesh displacement.
    if (visualization)
    {
       x0 -= x;
@@ -870,7 +884,6 @@ int main (int argc, char *argv[])
 
    delete S;
    delete S_prec;
-   delete metric_coeff1;
    delete adapt_surface;
    delete adapt_grad_surface;
    delete adapt_hess_surface;

miniapps/multidomain/makefile

@@ -55,6 +55,9 @@ MESH_FILES = $(notdir $(wildcard $(SRC)*.mesh))
 $(MESH_FILES): %: $(SRC)%
 	ln -sf $(<) .
 multidomain: | $(MESH_FILES)
+# The target 'copy-data' is used by the makefile in ../../tests/unit
+.PHONY: copy-data
+copy-data: | $(MESH_FILES)
 endif
 
 MFEM_TESTS = MINIAPPS

miniapps/nurbs/CMakeLists.txt

@@ -13,6 +13,18 @@ add_mfem_miniapp(nurbs_ex1
   MAIN nurbs_ex1.cpp
   LIBRARIES mfem)
 
+add_mfem_miniapp(nurbs_ex3
+  MAIN nurbs_ex3.cpp
+  LIBRARIES mfem)
+
+add_mfem_miniapp(nurbs_ex5
+  MAIN nurbs_ex5.cpp
+  LIBRARIES mfem)
+
+add_mfem_miniapp(nurbs_ex24
+  MAIN nurbs_ex24.cpp
+  LIBRARIES mfem)
+
 add_mfem_miniapp(nurbs_curveint
   MAIN nurbs_curveint.cpp
   LIBRARIES mfem)
@@ -29,6 +41,14 @@ add_mfem_miniapp(nurbs_patch_ex1
   MAIN nurbs_patch_ex1.cpp
   LIBRARIES mfem)
 
+add_mfem_miniapp(nurbs_solenoidal
+  MAIN nurbs_solenoidal.cpp
+  LIBRARIES mfem)
+
+add_mfem_miniapp(nurbs_biharm
+  MAIN nurbs_biharm.cpp
+  LIBRARIES mfem)
+
 if (MFEM_ENABLE_TESTING)
   add_test(NAME nurbs_ex1_1d_r1_o2_ser
     COMMAND $<TARGET_FILE:nurbs_ex1> -no-vis
@@ -64,6 +84,10 @@ if (MFEM_ENABLE_TESTING)
     COMMAND $<TARGET_FILE:nurbs_ex1> -no-vis
     -m ${PROJECT_SOURCE_DIR}/data/pipe-nurbs-2d.mesh -o 2 --weak-bc -r 2)
 
+  add_test(NAME nurbs_ex1_neu_r2_ser
+    COMMAND $<TARGET_FILE:nurbs_ex1> -no-vis
+    -m ${PROJECT_SOURCE_DIR}/data/pipe-nurbs-2d.mesh -o 2 -r 2 --neu "3")
+
   add_test(NAME nurbs_ex1_weak_mp_ser
     COMMAND $<TARGET_FILE:nurbs_ex1> -no-vis
     -m ${PROJECT_SOURCE_DIR}/data/ball-nurbs.mesh -o 2 --weak-bc -r 0)
@@ -125,9 +149,64 @@ if (MFEM_ENABLE_TESTING)
     COMMAND $<TARGET_FILE:nurbs_ex1> -no-vis
     -m ${PROJECT_SOURCE_DIR}/miniapps/nurbs/meshes/two-cubes-nurbs-autoedge.mesh -o 1 -r 3 -rf ${PROJECT_SOURCE_DIR}/miniapps/nurbs/meshes/two-cubes.ref)
 
+  add_test(NAME nurbs_ex1_periodic_2d
+    COMMAND $<TARGET_FILE:nurbs_ex1> -no-vis
+    -m ${PROJECT_SOURCE_DIR}/data/pipe-nurbs-2d.mesh -o 2 -r 2 --master "3" --slave "4")
+
   add_test(NAME nurbs_ex1_periodic_3d
     COMMAND $<TARGET_FILE:nurbs_ex1> -no-vis
-    -m ${PROJECT_SOURCE_DIR}/miniapps/nurbs/meshes/cube-nurbs.mesh -pm "1" -ps "2" -rf ${PROJECT_SOURCE_DIR}/miniapps/nurbs/meshes/cube.ref)
+    -m ${PROJECT_SOURCE_DIR}/miniapps/nurbs/meshes/cube-nurbs.mesh
+    -rf ${PROJECT_SOURCE_DIR}/miniapps/nurbs/meshes/cube.ref
+    --master "1" --slave "2")
+
+  add_test(NAME nurbs_ex3_2d_r1_o2_ser
+    COMMAND $<TARGET_FILE:nurbs_ex3> -no-vis
+        -m ${PROJECT_SOURCE_DIR}/data/square-nurbs.mesh -r 1 -o 2)
+
+  add_test(NAME nurbs_ex3_3d_r1_o2_ser
+    COMMAND $<TARGET_FILE:nurbs_ex3> -no-vis
+        -m ${PROJECT_SOURCE_DIR}/data/cube-nurbs.mesh -r 1 -o 2)
+
+  add_test(NAME nurbs_ex5_2d_r1_o2_ser
+    COMMAND $<TARGET_FILE:nurbs_ex5> -no-vis
+        -m ${PROJECT_SOURCE_DIR}/data/square-nurbs.mesh -r 1 -o 2)
+
+  add_test(NAME nurbs_ex5_3d_r1_o2_ser
+    COMMAND $<TARGET_FILE:nurbs_ex5> -no-vis
+        -m ${PROJECT_SOURCE_DIR}/data/cube-nurbs.mesh -r 1 -o 2)
+
+  add_test(NAME nurbs_ex24_2d_r1_o2_p0_ser
+    COMMAND $<TARGET_FILE:nurbs_ex24> -no-vis
+        -m ${PROJECT_SOURCE_DIR}/data/pipe-nurbs-2d.mesh -r 1 -o 2 -p 0)
+
+  add_test(NAME nurbs_ex24_2d_r1_o2_p2_ser
+    COMMAND $<TARGET_FILE:nurbs_ex24> -no-vis
+        -m ${PROJECT_SOURCE_DIR}/data/pipe-nurbs-2d.mesh -r 1 -o 2 -p 2)
+
+  add_test(NAME nurbs_ex24_3d_r1_o2_p0_ser
+    COMMAND $<TARGET_FILE:nurbs_ex24> -no-vis
+        -m ${PROJECT_SOURCE_DIR}/data/cube-nurbs.mesh -r 1 -o 2 -p 0)
+
+  add_test(NAME nurbs_ex24_3d_r1_o2_p1_ser
+    COMMAND $<TARGET_FILE:nurbs_ex24> -no-vis
+        -m ${PROJECT_SOURCE_DIR}/data/cube-nurbs.mesh -r 1 -o 2 -p 1)
+
+  add_test(NAME nurbs_ex24_3d_r1_o2_p2_ser
+    COMMAND $<TARGET_FILE:nurbs_ex24> -no-vis
+        -m ${PROJECT_SOURCE_DIR}/data/cube-nurbs.mesh -r 1 -o 2 -p 2)
+
+  add_test(NAME nurbs_solenoidal_2d_r1_o2_ser
+    COMMAND $<TARGET_FILE:nurbs_solenoidal> -no-vis
+        -m ${PROJECT_SOURCE_DIR}/data/pipe-nurbs-2d.mesh -r 1 -o 2)
+
+  add_test(NAME nurbs_solenoidal_3d_r1_o2_ser
+    COMMAND $<TARGET_FILE:nurbs_solenoidal> -no-vis
+        -m ${PROJECT_SOURCE_DIR}/data/cube-nurbs.mesh -r 1 -o 2)
+
+  add_test(NAME nurbs_biharm_ser
+    COMMAND $<TARGET_FILE:nurbs_biharm> -no-vis
+        -m ${PROJECT_SOURCE_DIR}/data/square-nurbs.mesh -r 1 -o 2)
+
 endif()
 
 if (MFEM_USE_MPI)
@@ -135,6 +214,10 @@ if (MFEM_USE_MPI)
     MAIN nurbs_ex1p.cpp
     LIBRARIES mfem)
 
+  add_mfem_miniapp(nurbs_ex11p
+    MAIN nurbs_ex11p.cpp
+    LIBRARIES mfem)
+
   if (MFEM_ENABLE_TESTING)
     add_test(NAME nurbs_ex1p_np=4
       COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
@@ -169,13 +252,7 @@ if (MFEM_USE_MPI)
         COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
         ${MPIEXEC_PREFLAGS}  $<TARGET_FILE:nurbs_ex1p> -no-vis
         -m ${PROJECT_SOURCE_DIR}/data/square-disc-nurbs-patch.mesh -o 2 --weak-bc -r 1)
-  endif()
 
-  add_mfem_miniapp(nurbs_ex11p
-    MAIN nurbs_ex11p.cpp
-    LIBRARIES mfem)
-
-  if (MFEM_ENABLE_TESTING)
     add_test(NAME nurbs_ex11p_np=4
       COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
       ${MPIEXEC_PREFLAGS} $<TARGET_FILE:nurbs_ex11p> -no-vis

miniapps/nurbs/makefile

@@ -21,8 +21,7 @@ CONFIG_MK = $(MFEM_BUILD_DIR)/config/config.mk
 MFEM_LIB_FILE = mfem_is_not_built
 -include $(CONFIG_MK)
 
-SEQ_MINIAPPS = nurbs_ex1 nurbs_patch_ex1 nurbs_curveint nurbs_printfunc nurbs_naca_cmesh
-
+SEQ_MINIAPPS = nurbs_ex1 nurbs_patch_ex1 nurbs_ex3 nurbs_ex5 nurbs_ex24 nurbs_curveint nurbs_printfunc nurbs_solenoidal nurbs_naca_cmesh
 PAR_MINIAPPS = nurbs_ex1p nurbs_ex11p
 ifeq ($(MFEM_USE_MPI),NO)
    MINIAPPS = $(SEQ_MINIAPPS)
@@ -103,6 +102,36 @@ endif
 	@$(call mfem-test,$<,, NURBS miniapp,$(EX1PATCH_ARGS_2))
 	@$(call mfem-test,$<,, NURBS miniapp,$(EX1PATCH_ARGS_3))
 
+EX3_ARGS_1 := -m $(MFEM_DIR)/data/square-nurbs.mesh -r 1 -o 2
+EX3_ARGS_2 := -m $(MFEM_DIR)/data/cube-nurbs.mesh -r 1 -o 2
+nurbs_ex3-test-seq: nurbs_ex3
+	@$(call mfem-test,$<,, NURBS miniapp,$(EX3_ARGS_1))
+	@$(call mfem-test,$<,, NURBS miniapp,$(EX3_ARGS_2))
+
+EX5_ARGS_1 := -m $(MFEM_DIR)/data/square-nurbs.mesh -r 1 -o 2
+EX5_ARGS_2 := -m $(MFEM_DIR)/data/cube-nurbs.mesh -r 1 -o 2
+nurbs_ex5-test-seq: nurbs_ex5
+	@$(call mfem-test,$<,, NURBS miniapp,$(EX5_ARGS_1))
+	@$(call mfem-test,$<,, NURBS miniapp,$(EX5_ARGS_2))
+
+EX24_ARGS_1 := -m $(MFEM_DIR)/data/pipe-nurbs-2d.mesh -r 1 -o 2 -p 0
+EX24_ARGS_2 := -m $(MFEM_DIR)/data/pipe-nurbs-2d.mesh -r 1 -o 2 -p 2
+EX24_ARGS_3 := -m $(MFEM_DIR)/data/cube-nurbs.mesh -r 1 -o 2 -p 0
+EX24_ARGS_4 := -m $(MFEM_DIR)/data/cube-nurbs.mesh -r 1 -o 2 -p 1
+EX24_ARGS_5 := -m $(MFEM_DIR)/data/cube-nurbs.mesh -r 1 -o 2 -p 2
+nurbs_ex24-test-seq: nurbs_ex24
+	@$(call mfem-test,$<,, NURBS miniapp,$(EX24_ARGS_1))
+	@$(call mfem-test,$<,, NURBS miniapp,$(EX24_ARGS_2))
+	@$(call mfem-test,$<,, NURBS miniapp,$(EX24_ARGS_3))
+	@$(call mfem-test,$<,, NURBS miniapp,$(EX24_ARGS_4))
+	@$(call mfem-test,$<,, NURBS miniapp,$(EX24_ARGS_5))
+
+SOL_ARGS_1 := -m $(MFEM_DIR)/data/pipe-nurbs-2d.mesh -r 1 -o 2
+SOL_ARGS_1 := -m $(MFEM_DIR)/data/cube-nurbs.mesh -r 1 -o 2
+nurbs_sol-test-seq: nurbs_solenoidal
+	@$(call mfem-test,$<,, NURBS miniapp,$(SOL_ARGS_1))
+	@$(call mfem-test,$<,, NURBS miniapp,$(SOl_ARGS_2))
+
 CI_ARGS_1 := -uw -n 9 -no-visit
 CI_ARGS_2 := -nw -n 9 -no-visit
 
@@ -151,6 +180,6 @@ clean-build:
 	rm -rf *.dSYM *.TVD.*breakpoints
 
 clean-exec:
-	@rm -f refined.mesh sin-fit.mesh mesh.* sol.* mode_* naca-cmesh.mesh
-	@rm -rf Example1*
+	@rm -f refined.mesh sin-fit.mesh ex5.mesh exsol.mesh mesh.* sol.* mode_* naca-cmesh.mesh sol_?.gf
+	@rm -rf Example1* Example3* Example5* Solenoidal_* ParaView
 	@rm -rf CurveInt Naca_cmesh glvis_naca-cmesh.mesh

miniapps/nurbs/nurbs_biharm.cpp

@@ -0,0 +1,408 @@
+// Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
+// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+// LICENSE and NOTICE for details. LLNL-CODE-806117.
+//
+// This file is part of the MFEM library. For more information and source code
+// availability visit https://mfem.org.
+//
+// MFEM is free software; you can redistribute it and/or modify it under the
+// terms of the BSD-3 license. We welcome feedback and contributions, see file
+// CONTRIBUTING.md for details.
+//
+// Stabilized Convection-Diffusion
+
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+#include <list>
+
+
+using namespace std;
+using namespace mfem;
+
+real_t kappa_param = 1.0;
+
+real_t dif_fun(const Vector & x)
+{
+   return kappa_param;
+}
+
+real_t force_fun(const Vector & x)
+{
+   int d = x.Size();
+
+   real_t kappa = dif_fun(x);
+
+   real_t pi  = (real_t)(M_PI);
+
+   real_t sx = sin(pi*x[0]);
+   real_t sy = 1.0;
+   real_t sz = 1.0;
+
+   if (d >= 2)
+   {
+      sy = sin(pi*x[1]);
+   }
+   if (d >= 3)
+   {
+      sz = sin(pi*x[2]);
+   }
+
+   return d*d*kappa*pi*pi*pi*pi*sx*sy*sz;
+}
+
+real_t sol_fun(const Vector & x)
+{
+   real_t pi  = (real_t)(M_PI);
+
+   real_t sx = sin(pi*x[0]);
+   real_t sy = 1.0;
+   real_t sz = 1.0;
+
+   int d = x.Size();
+   if (d >= 2)
+   {
+      sy = sin(pi*x[1]);
+   }
+   if (d >= 3)
+   {
+      sz = sin(pi*x[2]);
+   }
+
+   return sx*sy*sz;
+}
+
+void grad_fun(const Vector & x, Vector & a)
+{
+   real_t pi  = (real_t)(M_PI);
+
+   real_t sx = sin(pi*x[0]);
+   real_t cx = cos(pi*x[0]);
+   real_t sy = 1.0;
+   real_t cy = 1.0;
+   real_t sz = 1.0;
+   real_t cz = 1.0;
+
+   int d = x.Size();
+   if (d >= 2)
+   {
+      sy = sin(pi*x[1]);
+      cy = cos(pi*x[1]);
+   }
+   if (d >= 3)
+   {
+      sz = sin(pi*x[2]);
+      cz = cos(pi*x[2]);
+   }
+
+   a[0] = pi*cx*sy;
+   a[1] = pi*sx*cy;
+}
+
+//----------------------------------------------------------
+real_t lap_fun(const Vector & x)
+{
+   real_t pi  = (real_t)(M_PI);
+
+   real_t sx = sin(pi*x[0]);
+   real_t cx = cos(pi*x[0]);
+   real_t sy = 1.0;
+   real_t cy = 1.0;
+   real_t sz = 1.0;
+   real_t cz = 1.0;
+
+   int d = x.Size();
+   if (d >= 2)
+   {
+      sy = sin(pi*x[1]);
+      cy = cos(pi*x[1]);
+   }
+   if (d >= 3)
+   {
+      sz = sin(pi*x[2]);
+      cz = cos(pi*x[2]);
+   }
+
+   return -d*pi*pi*sx*sy*sz;
+}
+
+
+int main(int argc, char *argv[])
+{
+   // 1. Parse command-line options.
+   const char *mesh_file = "../../data/square-nurbs.mesh";
+   const char *per_file  = "none";
+   const char *ref_file  = "";
+   int ref_levels = 0;
+   Array<int> master(0);
+   Array<int> slave(0);
+   bool static_cond = false;
+   bool visualization = false;
+   real_t penalty = -1;
+   Array<int> order(1);
+   order[0] = 2;
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&ref_levels, "-r", "--refine",
+                  "Number of times to refine the mesh uniformly, -1 for auto.");
+   args.AddOption(&per_file, "-p", "--per",
+                  "Periodic BCS file.");
+   args.AddOption(&ref_file, "-rf", "--ref-file",
+                  "File with refinement data");
+   args.AddOption(&master, "-pm", "--master",
+                  "Master boundaries for periodic BCs");
+   args.AddOption(&slave, "-ps", "--slave",
+                  "Slave boundaries for periodic BCs");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree) or -1 for"
+                  " isoparametric space.");
+   args.AddOption(&kappa_param, "-k", "--kappa",
+                  "Sets the diffusion parameters, should be positive."
+                  " Negative values are replaced with function defined in source.");
+   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
+                  "--no-static-condensation", "Enable static condensation.");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.Parse();
+   if (!args.Good())
+   {
+      args.PrintUsage(mfem::out);
+      return 1;
+   }
+
+   args.PrintOptions(mfem::out);
+
+   if (order.Min()< 2)
+   {
+      mfem_error("Wrong order.");
+   }
+   // 2. Read the mesh from the given mesh file. We can handle triangular,
+   //    quadrilateral, tetrahedral, hexahedral, surface and volume meshes with
+   //    the same code.
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
+   int dim = mesh->Dimension();
+
+   // 3. Refine the mesh to increase the resolution. In this example we do
+   //    'ref_levels' of uniform refinement and knot insertion of knots defined
+   //    in a refinement file. We choose 'ref_levels' to be the largest number
+   //    that gives a final mesh with no more than 50,000 elements.
+   {
+      // Mesh refinement as defined in refinement file
+      if (mesh->NURBSext && (strlen(ref_file) != 0))
+      {
+         mesh->RefineNURBSFromFile(ref_file);
+      }
+
+      for (int l = 0; l < ref_levels; l++)
+      {
+         mesh->UniformRefinement();
+      }
+      mesh->PrintInfo();
+   }
+
+   // 4. Define a finite element space on the mesh. Here we use continuous
+   //    Lagrange finite elements of the specified order. If order < 1, we
+   //    instead use an isoparametric/isogeometric space.
+   FiniteElementCollection *fec;
+   NURBSExtension *NURBSext = NULL;
+   int own_fec = 0;
+
+   if (mesh->NURBSext)
+   {
+      fec = new NURBSFECollection(order[0]);
+      own_fec = 1;
+
+      int nkv = mesh->NURBSext->GetNKV();
+      if (order.Size() == 1)
+      {
+         int tmp = order[0];
+         order.SetSize(nkv);
+         order = tmp;
+      }
+
+      if (order.Size() != nkv ) { mfem_error("Wrong number of orders set."); }
+      NURBSext = new NURBSExtension(mesh->NURBSext, order);
+
+      // Read periodic BCs from file
+      std::ifstream in;
+      in.open(per_file, std::ifstream::in);
+      if (in.is_open())
+      {
+         int psize;
+         in >> psize;
+         master.SetSize(psize);
+         slave.SetSize(psize);
+         master.Load(in, psize);
+         slave.Load(in, psize);
+         in.close();
+      }
+      master.Print();
+      slave.Print();
+      NURBSext->ConnectBoundaries(master,slave);
+   }
+   else if (order[0] == -1) // Isoparametric
+   {
+      if (mesh->GetNodes())
+      {
+         fec = mesh->GetNodes()->OwnFEC();
+         own_fec = 0;
+         mfem::out << "Using isoparametric FEs: " << fec->Name() << endl;
+      }
+      else
+      {
+         mfem::out <<"Mesh does not have FEs --> Assume order 1.\n";
+         fec = new H1_FECollection(1, dim);
+         own_fec = 1;
+      }
+   }
+   else
+   {
+      if (order.Size() > 1) { cout <<"Wrong number of orders set, needs one.\n"; }
+      fec = new H1_FECollection(abs(order[0]), dim);
+      own_fec = 1;
+   }
+
+   FiniteElementSpace *fespace = new FiniteElementSpace(mesh, NURBSext, fec);
+   mfem::out << "Number of finite element unknowns: "
+             << fespace->GetTrueVSize() << endl;
+
+   // 5. Determine the list of true (i.e. conforming) essential boundary dofs.
+   //    In this example, the boundary conditions are defined by marking all
+   //    the boundary attributes from the mesh as essential (Dirichlet) and
+   //    converting them to a list of true dofs.
+   Array<int> ess_tdof_list;
+   if (mesh->bdr_attributes.Size())
+   {
+      Array<int> ess_bdr(mesh->bdr_attributes.Max());
+      ess_bdr = 1;
+
+      // Remove periodic BCs
+      for (int i = 0; i < master.Size(); i++)
+      {
+         ess_bdr[master[i]-1] = 0;
+         ess_bdr[slave[i]-1] = 0;
+      }
+      fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+   }
+
+   // 6. Set up the linear form b(.) which corresponds to the right-hand side of
+   //    the FEM linear system, which in this case is (1,phi_i) where phi_i are
+   //    the basis functions in the finite element fespace.
+   ConstantCoefficient u_dir(0.0);
+
+   Coefficient *kappa_tmp;
+   if (kappa_param < 0.0)
+   {
+      kappa_tmp = new FunctionCoefficient(dif_fun);
+   }
+   else
+   {
+      kappa_tmp = new ConstantCoefficient(kappa_param);
+   }
+
+   Coefficient& kappa = *kappa_tmp;
+   FunctionCoefficient force(force_fun);
+
+   LinearForm *b = new LinearForm(fespace);
+   b->AddDomainIntegrator(new DomainLFIntegrator(force));
+   b->Assemble();
+
+   // 7. Define the solution vector x as a finite element grid function
+   //    corresponding to fespace. Initialize x with initial guess of zero,
+   //    which satisfies the boundary conditions.
+   GridFunction x(fespace);
+   x = 0.0;
+
+   // 8. Set up the bilinear form a(.,.) on the finite element space
+   //    corresponding to the Laplacian operator -Delta, by adding the Diffusion
+   //    domain integrator.
+   BilinearForm *a = new BilinearForm(fespace);
+   a->AddDomainIntegrator(new LaplaceLaplaceIntegrator(kappa));
+
+   // 9. Assemble the bilinear form and the corresponding linear system,
+   //    applying any necessary transformations such as: eliminating boundary
+   //    conditions, applying conforming constraints for non-conforming AMR,
+   //    static condensation, etc.
+   if (static_cond) { a->EnableStaticCondensation(); }
+   a->Assemble();
+
+   SparseMatrix A;
+   Vector B, X;
+   a->FormLinearSystem(ess_tdof_list, x, *b, A, X, B);
+
+   mfem::out << "Size of linear system: " << A.Height() << endl;
+
+#ifndef MFEM_USE_SUITESPARSE
+   // 10. Define a simple Jacobi preconditioner and use it to
+   //     solve the system A X = B with PCG.
+   GSSmoother M(A);
+   GMRES(A, M, B, X, 1, 2000, 2000, 1e-16, 0.0);
+#else
+   // 10. If MFEM was compiled with SuiteSparse, use UMFPACK to solve the system.
+   UMFPackSolver umf_solver;
+   umf_solver.Control[UMFPACK_ORDERING] = UMFPACK_ORDERING_METIS;
+   umf_solver.SetOperator(A);
+   umf_solver.Mult(B, X);
+#endif
+
+   // 11. Recover the solution as a finite element grid function.
+   a->RecoverFEMSolution(X, *b, x);
+
+   // 12. Save the refined mesh and the solution. This output can be viewed later
+   //     using GLVis: "glvis -m refined.mesh -g sol.gf".
+   {
+      ofstream mesh_ofs("refined.mesh");
+      mesh_ofs.precision(8);
+      mesh->Print(mesh_ofs);
+      ofstream sol_ofs("sol.gf");
+      sol_ofs.precision(8);
+      x.Save(sol_ofs);
+      sol_ofs.close();
+   }
+
+   // 13. Send the solution by socket to a GLVis server.
+   if (visualization)
+   {
+      char vishost[] = "localhost";
+      int  visport   = 19916;
+      socketstream sol_sock(vishost, visport);
+      sol_sock.precision(8);
+      sol_sock << "solution\n" << *mesh << x << flush;
+   }
+
+   // 14. Error computation
+   Vector norm(3);
+   int order_quad = 3*order.Max() + 4;
+   const IntegrationRule *irs[Geometry::NumGeom];
+   for (int i=0; i < Geometry::NumGeom; ++i)
+   {
+      irs[i] = &(IntRules.Get(i, order_quad));
+   }
+   FunctionCoefficient sol_cf(sol_fun);
+   VectorFunctionCoefficient grad_cf(mesh->Dimension(), grad_fun);
+   FunctionCoefficient lap_cf(lap_fun);
+
+   norm[0]= x.ComputeL2Error(sol_cf,irs);
+   norm[1]= x.ComputeGradError(&grad_cf, irs);
+   norm[2] = x.ComputeLaplaceError(&lap_cf, irs);
+
+   mfem::out << "|| x_h - x_ex ||            = " << norm[0]  << "\n";
+   mfem::out << "|| grad x_h - grad x_ex ||  = " << norm[1] << "\n";
+   mfem::out << "|| lap x_h - lap x_ex ||    = " << norm[2] << "\n";
+
+   // 15. Save data in the VisIt format
+   VisItDataCollection visit_dc("Biharm", mesh);
+   visit_dc.RegisterField("solution", &x);
+   visit_dc.Save();
+
+   // 16. Free the used memory.
+   delete fespace;
+   if (own_fec) { delete fec; }
+   delete mesh;
+
+   return 0;
+}
+

miniapps/nurbs/nurbs_ex1.cpp

@@ -6,6 +6,7 @@
 //               nurbs_ex1 -m ../../data/square-nurbs.mesh -o 2 --weak-bc
 //               nurbs_ex1 -m ../../data/cube-nurbs.mesh -o 2 -no-ibp
 //               nurbs_ex1 -m ../../data/pipe-nurbs-2d.mesh -o 2 -no-ibp
+//               nurbs_ex1 -m ../../data/pipe-nurbs-2d.mesh -o 2 -r 2 --neu "3"
 //               nurbs_ex1 -m ../../data/square-disc-nurbs.mesh -o -1
 //               nurbs_ex1 -m ../../data/disc-nurbs.mesh -o -1
 //               nurbs_ex1 -m ../../data/pipe-nurbs.mesh -o -1
@@ -52,101 +53,17 @@ public:
 inline bool operator==(const Data& d1,const Data& d2) { return (d1.x == d2.x); }
 inline bool operator <(const Data& d1,const Data& d2) { return (d1.x  < d2.x); }
 
-/** Class for integrating the bilinear form a(u,v) := (Q Laplace u, v) where Q
-    can be a scalar coefficient. */
-class Diffusion2Integrator: public BilinearFormIntegrator
-{
-private:
-#ifndef MFEM_THREAD_SAFE
-   Vector shape,laplace;
-#endif
-   Coefficient *Q;
-
-public:
-   /// Construct a diffusion integrator with coefficient Q = 1
-   Diffusion2Integrator() { Q = NULL; }
-
-   /// Construct a diffusion integrator with a scalar coefficient q
-   Diffusion2Integrator (Coefficient &q) : Q(&q) { }
-
-   /** Given a particular Finite Element
-       computes the element stiffness matrix elmat. */
-   virtual void AssembleElementMatrix(const FiniteElement &el,
-                                      ElementTransformation &Trans,
-                                      DenseMatrix &elmat)
-   {
-      int nd = el.GetDof();
-      int dim = el.GetDim();
-      real_t w;
-
-#ifdef MFEM_THREAD_SAFE
-      Vector shape(nd);
-      Vector laplace(nd);
-#else
-      shape.SetSize(nd);
-      laplace.SetSize(nd);
-#endif
-      elmat.SetSize(nd);
-
-      const IntegrationRule *ir = IntRule;
-      if (ir == NULL)
-      {
-         int order;
-         if (el.Space() == FunctionSpace::Pk)
-         {
-            order = 2*el.GetOrder() - 2;
-         }
-         else
-         {
-            order = 2*el.GetOrder() + dim - 1;
-         }
-
-         if (el.Space() == FunctionSpace::rQk)
-         {
-            ir = &RefinedIntRules.Get(el.GetGeomType(),order);
-         }
-         else
-         {
-            ir = &IntRules.Get(el.GetGeomType(),order);
-         }
-      }
-
-      elmat = 0.0;
-      for (int i = 0; i < ir->GetNPoints(); i++)
-      {
-         const IntegrationPoint &ip = ir->IntPoint(i);
-         Trans.SetIntPoint(&ip);
-         w = -ip.weight * Trans.Weight();
-
-         el.CalcShape(ip, shape);
-         el.CalcPhysLaplacian(Trans, laplace);
-
-         if (Q)
-         {
-            w *= Q->Eval(Trans, ip);
-         }
-
-         for (int jj = 0; jj < nd; jj++)
-         {
-            for (int ii = 0; ii < nd; ii++)
-            {
-               elmat(ii, jj) += w*shape(ii)*laplace(jj);
-            }
-         }
-      }
-   }
-
-};
 
 int main(int argc, char *argv[])
 {
    // 1. Parse command-line options.
-   const char *mesh_file = "../../data/star.mesh";
+   const char *mesh_file = "../../data/square-nurbs.mesh";
    const char *per_file  = "none";
    const char *ref_file  = "";
    int ref_levels = -1;
    Array<int> master(0);
    Array<int> slave(0);
+   Array<int> neu(0);
    bool static_cond = false;
    bool visualization = 1;
    int lod = 0;
@@ -169,6 +86,8 @@ int main(int argc, char *argv[])
                   "Master boundaries for periodic BCs");
    args.AddOption(&slave, "-ps", "--slave",
                   "Slave boundaries for periodic BCs");
+   args.AddOption(&neu, "-n", "--neu",
+                  "Boundaries with Neumann BCs");
    args.AddOption(&order, "-o", "--order",
                   "Finite element order (polynomial degree) or -1 for"
                   " isoparametric space.");
@@ -266,8 +185,6 @@ int main(int argc, char *argv[])
          slave.Load(in, psize);
          in.close();
       }
-      master.Print();
-      slave.Print();
       NURBSext->ConnectBoundaries(master,slave);
    }
    else if (order[0] == -1) // Isoparametric
@@ -323,39 +240,84 @@ int main(int argc, char *argv[])
    //    In this example, the boundary conditions are defined by marking all
    //    the boundary attributes from the mesh as essential (Dirichlet) and
    //    converting them to a list of true dofs.
-   Array<int> ess_tdof_list;
+   Array<int> ess_bdr(0);
+   Array<int> neu_bdr(0);
+   Array<int> per_bdr(0);
    if (mesh->bdr_attributes.Size())
    {
-      Array<int> ess_bdr(mesh->bdr_attributes.Max());
-      if (strongBC)
+      ess_bdr.SetSize(mesh->bdr_attributes.Max());
+      neu_bdr.SetSize(mesh->bdr_attributes.Max());
+      per_bdr.SetSize(mesh->bdr_attributes.Max());
+
+      ess_bdr = 1;
+      neu_bdr = 0;
+      per_bdr = 0;
+
+      // Apply Neumann BCs
+      for (int i = 0; i < neu.Size(); i++)
       {
-         ess_bdr = 1;
-      }
-      else
-      {
-         ess_bdr = 0;
+         if ( neu[i]-1 >= 0 &&
+              neu[i]-1 < mesh->bdr_attributes.Max())
+         {
+            ess_bdr[neu[i]-1] = 0;
+            neu_bdr[neu[i]-1] = 1;
+         }
+         else
+         {
+            cout <<"Neumann boundary "<<neu[i]<<" out of range -- discarded"<< endl;
+         }
       }
 
-      // Remove periodic BCs
+      // Correct for periodic BCs
       for (int i = 0; i < master.Size(); i++)
       {
-         ess_bdr[master[i]-1] = 0;
-         ess_bdr[slave[i]-1] = 0;
+         if ( master[i]-1 >= 0 &&
+              master[i]-1 < mesh->bdr_attributes.Max())
+         {
+            ess_bdr[master[i]-1] = 0;
+            neu_bdr[master[i]-1] = 0;
+            per_bdr[master[i]-1] = 1;
+         }
+         else
+         {
+            cout <<"Master boundary "<<master[i]<<" out of range -- discarded"<< endl;
+         }
+      }
+      for (int i = 0; i < slave.Size(); i++)
+      {
+         if ( slave[i]-1 >= 0 &&
+              slave[i]-1 < mesh->bdr_attributes.Max())
+         {
+            ess_bdr[slave[i]-1] = 0;
+            neu_bdr[slave[i]-1] = 0;
+            per_bdr[slave[i]-1] = 1;
+         }
+         else
+         {
+            cout <<"Slave boundary "<<slave[i]<<" out of range -- discarded"<< endl;
+         }
       }
-      fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
    }
+   cout <<"Boundary conditions:"<< endl;
+   cout <<" - Periodic  : "; per_bdr.Print();
+   cout <<" - Essential : "; ess_bdr.Print();
+   cout <<" - Neumann   : "; neu_bdr.Print();
+
 
    // 6. Set up the linear form b(.) which corresponds to the right-hand side of
    //    the FEM linear system, which in this case is (1,phi_i) where phi_i are
    //    the basis functions in the finite element fespace.
    ConstantCoefficient one(1.0);
+   ConstantCoefficient mone(-1.0);
    ConstantCoefficient zero(0.0);
 
    LinearForm *b = new LinearForm(fespace);
    b->AddDomainIntegrator(new DomainLFIntegrator(one));
+   b->AddBoundaryIntegrator( new BoundaryLFIntegrator(one),neu_bdr);
    if (!strongBC)
       b->AddBdrFaceIntegrator(
-         new DGDirichletLFIntegrator(zero, one, -1.0, kappa));
+         new DGDirichletLFIntegrator(zero, one, -1.0, kappa), ess_bdr);
+
    b->Assemble();
 
    // 7. Define the solution vector x as a finite element grid function
@@ -374,12 +336,13 @@ int main(int argc, char *argv[])
    }
    else
    {
-      a->AddDomainIntegrator(new Diffusion2Integrator(one));
+      a->AddDomainIntegrator(new LaplaceIntegrator(one, -1.0));
+      a->AddBdrFaceIntegrator(new DGDiffusionIntegrator(mone, 0.0, 0.0), neu_bdr);
    }
 
    if (!strongBC)
    {
-      a->AddBdrFaceIntegrator(new DGDiffusionIntegrator(one, -1.0, kappa));
+      a->AddBdrFaceIntegrator(new DGDiffusionIntegrator(one, -1.0, kappa), ess_bdr);
    }
 
    // 9. Assemble the bilinear form and the corresponding linear system,
@@ -391,6 +354,11 @@ int main(int argc, char *argv[])
 
    SparseMatrix A;
    Vector B, X;
+   Array<int> ess_tdof_list(0);
+   if (strongBC)
+   {
+      fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+   }
    a->FormLinearSystem(ess_tdof_list, x, *b, A, X, B);
 
    cout << "Size of linear system: " << A.Height() << endl;

miniapps/nurbs/nurbs_ex24.cpp

@@ -0,0 +1,459 @@
+//                                MFEM Example 24 -- modified for NURBS FE
+//
+// Compile with: make nurbs_ex24
+//
+// Sample runs:  nurbs_ex24 -m ../../data/pipe-nurbs-2d.mesh -o 2
+//               nurbs_ex24 -m ../../data/pipe-nurbs-2d.mesh -p 2
+//               nurbs_ex24 -m ../../data/cube-nurbs.mesh -o 2
+//               nurbs_ex24 -m ../../data/cube-nurbs.mesh -o 2 -p 1
+//               nurbs_ex24 -m ../../data/cube-nurbs.mesh -o 2 -p 2
+//               nurbs_ex24 -m ../../data/escher.mesh
+//               nurbs_ex24 -m ../../data/escher.mesh -o 2
+//               nurbs_ex24 -m ../../data/fichera.mesh
+//               nurbs_ex24 -m ../../data/fichera-q2.vtk
+//               nurbs_ex24 -m ../../data/fichera-q3.mesh
+//               nurbs_ex24 -m ../../data/amr-quad.mesh -o 2
+//               nurbs_ex24 -m ../../data/amr-hex.mesh
+//
+// Device sample runs -- do not work for NURBS:
+//               nurbs_ex24 -m ../../data/escher.mesh -pa -d cuda
+//               nurbs_ex24 -m ../../data/escher.mesh -pa -d raja-cuda
+//               nurbs_ex24 -m ../../data/escher.mesh -pa -d raja-omp
+//
+// Description:  This example code illustrates usage of mixed finite element
+//               spaces, with three variants:
+//
+//               0) (grad p, u) for p in H^1 tested against u in H(curl)
+//               1) (curl v, u) for v in H(curl) tested against u in H(div), 3D
+//               2) (div v, q) for v in H(div) tested against q in L_2
+//
+//               Using different approaches, we project the gradient, curl, or
+//               divergence to the appropriate space.
+//
+//               NURBS-based H(curl) and H(div) spaces only implemented
+//               for meshes consisting of a single patch.
+//
+//               We recommend viewing examples 1, 3, and 5 before viewing this
+//               example.
+
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+
+using namespace std;
+using namespace mfem;
+
+real_t p_exact(const Vector &x);
+void gradp_exact(const Vector &, Vector &);
+real_t div_gradp_exact(const Vector &x);
+void v_exact(const Vector &x, Vector &v);
+void curlv_exact(const Vector &x, Vector &cv);
+
+int dim;
+real_t freq = 1.0, kappa;
+
+int main(int argc, char *argv[])
+{
+   // 1. Parse command-line options.
+   const char *mesh_file = "../../data/cube-nurbs.mesh";
+   int ref_levels = -1;
+   int order = 1;
+   bool NURBS = true;
+   int prob = 0;
+   bool static_cond = false;
+   bool pa = false;
+   const char *device_config = "cpu";
+   bool visualization = 1;
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&ref_levels, "-r", "--refine",
+                  "Number of times to refine the mesh uniformly, -1 for auto.");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree).");
+   args.AddOption(&NURBS, "-n", "--nurbs", "-nn","--no-nurbs",
+                  "NURBS.");
+   args.AddOption(&prob, "-p", "--problem-type",
+                  "Choose between 0: grad, 1: curl, 2: div");
+   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
+                  "--no-static-condensation", "Enable static condensation.");
+   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
+                  "--no-partial-assembly", "Enable Partial Assembly.");
+   args.AddOption(&device_config, "-d", "--device",
+                  "Device configuration string, see Device::Configure().");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+
+   args.Parse();
+   if (!args.Good())
+   {
+      args.PrintUsage(cout);
+      return 1;
+   }
+   args.PrintOptions(cout);
+   kappa = freq * M_PI;
+
+   // 2. Enable hardware devices such as GPUs, and programming models such as
+   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
+   Device device(device_config);
+   device.Print();
+
+   // 3. Read the mesh from the given mesh file. We can handle triangular,
+   //    quadrilateral, tetrahedral, hexahedral, surface and volume meshes with
+   //    the same code.
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
+   dim = mesh->Dimension();
+   if ((prob == 1) &&(dim != 3))
+   {
+      MFEM_ABORT("The curl problem is only defined in 3D.");
+   }
+   int sdim = mesh->SpaceDimension();
+
+   // 4. Refine the mesh to increase the resolution. In this example we do
+   //    'ref_levels' of uniform refinement. We choose 'ref_levels' to be the
+   //    largest number that gives a final mesh with no more than 50,000
+   //    elements.
+   {
+      if (ref_levels < 0)
+      {
+         ref_levels = (int)floor(log(50000./mesh->GetNE())/log(2.)/dim);
+      }
+      for (int l = 0; l < ref_levels; l++)
+      {
+         mesh->UniformRefinement();
+      }
+   }
+
+   // 5. Define a finite element space on the mesh. Here we use Nedelec or
+   //    Raviart-Thomas finite elements of the specified order.
+   FiniteElementCollection *trial_fec = nullptr;
+   FiniteElementCollection *test_fec = nullptr;
+   NURBSExtension *NURBSext = nullptr;
+   if (mesh->NURBSext && NURBS)
+   {
+      NURBSext  = new NURBSExtension(mesh->NURBSext, order);
+      if (prob == 0)
+      {
+         trial_fec  = new NURBSFECollection(order);
+         test_fec = new NURBS_HCurlFECollection(order, dim);
+      }
+      else if (prob == 1)
+      {
+         trial_fec = new NURBS_HCurlFECollection(order, dim);
+         test_fec = new NURBS_HDivFECollection(order, dim);
+      }
+      else
+      {
+         trial_fec = new NURBS_HDivFECollection(order, dim);
+         test_fec  = new NURBSFECollection(order);
+      }
+      mfem::out<<"Create NURBS fec and ext"<<std::endl;
+   }
+   else
+   {
+      if (prob == 0)
+      {
+         trial_fec = new H1_FECollection(order, dim);
+         test_fec = new ND_FECollection(order, dim);
+      }
+      else if (prob == 1)
+      {
+         trial_fec = new ND_FECollection(order, dim);
+         test_fec = new RT_FECollection(order-1, dim);
+      }
+      else
+      {
+         trial_fec = new RT_FECollection(order-1, dim);
+         test_fec = new L2_FECollection(order-1, dim);
+      }
+   }
+
+   FiniteElementSpace trial_fes(mesh, NURBSext, trial_fec);
+   FiniteElementSpace test_fes(mesh,trial_fes.StealNURBSext(), test_fec);
+
+   int trial_size = trial_fes.GetTrueVSize();
+   int test_size = test_fes.GetTrueVSize();
+
+   if (prob == 0)
+   {
+      cout << "Number of Nedelec finite element unknowns: " << test_size << endl;
+      cout << "Number of H1 finite element unknowns: " << trial_size << endl;
+   }
+   else if (prob == 1)
+   {
+      cout << "Number of Nedelec finite element unknowns: " << trial_size << endl;
+      cout << "Number of Raviart-Thomas finite element unknowns: " << test_size <<
+           endl;
+   }
+   else
+   {
+      cout << "Number of Raviart-Thomas finite element unknowns: "
+           << trial_size << endl;
+      cout << "Number of L2 finite element unknowns: " << test_size << endl;
+   }
+
+   // 6. Define the solution vector as a finite element grid function
+   //    corresponding to the trial fespace.
+   GridFunction gftest(&test_fes);
+   GridFunction gftrial(&trial_fes);
+   GridFunction x(&test_fes);
+   FunctionCoefficient p_coef(p_exact);
+   VectorFunctionCoefficient gradp_coef(sdim, gradp_exact);
+   VectorFunctionCoefficient v_coef(sdim, v_exact);
+   VectorFunctionCoefficient curlv_coef(sdim, curlv_exact);
+   FunctionCoefficient divgradp_coef(div_gradp_exact);
+
+   if (prob == 0)
+   {
+      gftrial.ProjectCoefficient(p_coef);
+   }
+   else if (prob == 1)
+   {
+      gftrial.ProjectCoefficient(v_coef);
+   }
+   else
+   {
+      gftrial.ProjectCoefficient(gradp_coef);
+   }
+
+   gftrial.SetTrueVector();
+   gftrial.SetFromTrueVector();
+
+   // 7. Set up the bilinear forms for L2 projection.
+   ConstantCoefficient one(1.0);
+   BilinearForm a(&test_fes);
+   MixedBilinearForm a_mixed(&trial_fes, &test_fes);
+   if (pa)
+   {
+      a.SetAssemblyLevel(AssemblyLevel::PARTIAL);
+      a_mixed.SetAssemblyLevel(AssemblyLevel::PARTIAL);
+   }
+
+   if (prob == 0)
+   {
+      a.AddDomainIntegrator(new VectorFEMassIntegrator(one));
+      a_mixed.AddDomainIntegrator(new MixedVectorGradientIntegrator(one));
+   }
+   else if (prob == 1)
+   {
+      a.AddDomainIntegrator(new VectorFEMassIntegrator(one));
+      a_mixed.AddDomainIntegrator(new MixedVectorCurlIntegrator(one));
+   }
+   else
+   {
+      a.AddDomainIntegrator(new MassIntegrator(one));
+      a_mixed.AddDomainIntegrator(new VectorFEDivergenceIntegrator(one));
+   }
+
+   // 8. Assemble the bilinear form and the corresponding linear system,
+   //    applying any necessary transformations such as: eliminating boundary
+   //    conditions, applying conforming constraints for non-conforming AMR,
+   //    static condensation, etc.
+   if (static_cond) { a.EnableStaticCondensation(); }
+
+   a.Assemble();
+   if (!pa) { a.Finalize(); }
+
+   a_mixed.Assemble();
+   if (!pa) { a_mixed.Finalize(); }
+
+   if (pa)
+   {
+      a_mixed.Mult(gftrial, x);
+   }
+   else
+   {
+      SparseMatrix& mixed = a_mixed.SpMat();
+      mixed.Mult(gftrial, x);
+   }
+
+   // 9. Define and apply a PCG solver for Ax = b with Jacobi preconditioner.
+   {
+      GridFunction rhs(&test_fes);
+      rhs = x;
+      x = 0.0;
+
+      CGSolver cg;
+      cg.SetRelTol(1e-12);
+      cg.SetMaxIter(1000);
+      cg.SetPrintLevel(1);
+      if (pa)
+      {
+         Array<int> ess_tdof_list; // empty
+         OperatorJacobiSmoother Jacobi(a, ess_tdof_list);
+
+         cg.SetOperator(a);
+         cg.SetPreconditioner(Jacobi);
+         cg.Mult(rhs, x);
+      }
+      else
+      {
+         SparseMatrix& Amat = a.SpMat();
+         DSmoother Jacobi(Amat);
+
+         cg.SetOperator(Amat);
+         cg.SetPreconditioner(Jacobi);
+         cg.Mult(rhs, x);
+      }
+   }
+
+   // 10. Compute the projection of the exact field.
+   GridFunction exact_proj(&test_fes);
+   if (prob == 0)
+   {
+      exact_proj.ProjectCoefficient(gradp_coef);
+   }
+   else if (prob == 1)
+   {
+      exact_proj.ProjectCoefficient(curlv_coef);
+   }
+   else
+   {
+      exact_proj.ProjectCoefficient(divgradp_coef);
+   }
+
+   exact_proj.SetTrueVector();
+   exact_proj.SetFromTrueVector();
+
+   // 11. Compute and print the L_2 norm of the error.
+   if (prob == 0)
+   {
+      real_t errSol = x.ComputeL2Error(gradp_coef);
+      real_t errProj = exact_proj.ComputeL2Error(gradp_coef);
+
+      cout << "\n Solution of (E_h,v) = (grad p_h,v) for E_h and v in H(curl): "
+           "|| E_h - grad p ||_{L_2} = " << errSol << '\n' << endl;
+      cout << " Projection E_h of exact grad p in H(curl): || E_h - grad p "
+           "||_{L_2} = " << errProj << '\n' << endl;
+   }
+   else if (prob == 1)
+   {
+      real_t errSol = x.ComputeL2Error(curlv_coef);
+      real_t errProj = exact_proj.ComputeL2Error(curlv_coef);
+
+      cout << "\n Solution of (E_h,w) = (curl v_h,w) for E_h and w in H(div): "
+           "|| E_h - curl v ||_{L_2} = " << errSol << '\n' << endl;
+      cout << " Projection E_h of exact curl v in H(div): || E_h - curl v "
+           "||_{L_2} = " << errProj << '\n' << endl;
+   }
+   else
+   {
+      int order_quad = max(3, 2*order+1);
+      const IntegrationRule *irs[Geometry::NumGeom];
+      for (int i=0; i < Geometry::NumGeom; ++i)
+      {
+         irs[i] = &(IntRules.Get(i, order_quad));
+      }
+
+      real_t errSol = x.ComputeL2Error(divgradp_coef, irs);
+      real_t errProj = exact_proj.ComputeL2Error(divgradp_coef, irs);
+
+      cout << "\n Solution of (f_h,q) = (div v_h,q) for f_h and q in L_2: "
+           "|| f_h - div v ||_{L_2} = " << errSol << '\n' << endl;
+
+      cout << " Projection f_h of exact div v in L_2: || f_h - div v "
+           "||_{L_2} = " << errProj << '\n' << endl;
+   }
+
+   // 12. Save the refined mesh and the solution. This output can be viewed
+   //     later using GLVis: "glvis -m refined.mesh -g sol.gf".
+   ofstream mesh_ofs("refined.mesh");
+   mesh_ofs.precision(8);
+   mesh->Print(mesh_ofs);
+   ofstream sol_ofs("sol.gf");
+   sol_ofs.precision(8);
+   x.Save(sol_ofs);
+
+   // 13. Send the solution by socket to a GLVis server.
+   if (visualization)
+   {
+      char vishost[] = "localhost";
+      int  visport   = 19916;
+      socketstream sol_sock(vishost, visport);
+      sol_sock.precision(8);
+      sol_sock << "solution\n" << *mesh << x << flush;
+   }
+
+   // 14. Free the used memory.
+   delete trial_fec;
+   delete test_fec;
+   delete mesh;
+
+   return 0;
+}
+
+real_t p_exact(const Vector &x)
+{
+   if (dim == 3)
+   {
+      return sin(x(0)) * sin(x(1)) * sin(x(2));
+   }
+   else if (dim == 2)
+   {
+      return sin(x(0)) * sin(x(1));
+   }
+
+   return 0.0;
+}
+
+void gradp_exact(const Vector &x, Vector &f)
+{
+   if (dim == 3)
+   {
+      f(0) = cos(x(0)) * sin(x(1)) * sin(x(2));
+      f(1) = sin(x(0)) * cos(x(1)) * sin(x(2));
+      f(2) = sin(x(0)) * sin(x(1)) * cos(x(2));
+   }
+   else
+   {
+      f(0) = cos(x(0)) * sin(x(1));
+      f(1) = sin(x(0)) * cos(x(1));
+      if (x.Size() == 3) { f(2) = 0.0; }
+   }
+}
+
+real_t div_gradp_exact(const Vector &x)
+{
+   if (dim == 3)
+   {
+      return -3.0 * sin(x(0)) * sin(x(1)) * sin(x(2));
+   }
+   else if (dim == 2)
+   {
+      return -2.0 * sin(x(0)) * sin(x(1));
+   }
+
+   return 0.0;
+}
+
+void v_exact(const Vector &x, Vector &v)
+{
+   if (dim == 3)
+   {
+      v(0) = sin(kappa * x(1));
+      v(1) = sin(kappa * x(2));
+      v(2) = sin(kappa * x(0));
+   }
+   else
+   {
+      v(0) = sin(kappa * x(1));
+      v(1) = sin(kappa * x(0));
+      if (x.Size() == 3) { v(2) = 0.0; }
+   }
+}
+
+void curlv_exact(const Vector &x, Vector &cv)
+{
+   if (dim == 3)
+   {
+      cv(0) = -kappa * cos(kappa * x(2));
+      cv(1) = -kappa * cos(kappa * x(0));
+      cv(2) = -kappa * cos(kappa * x(1));
+   }
+   else
+   {
+      cv = 0.0;
+   }
+}

miniapps/nurbs/nurbs_ex3.cpp

@@ -0,0 +1,284 @@
+//                                MFEM Example 3 -- modified for NURBS FE
+//
+// Compile with: make nurbs_ex3
+//
+// Sample runs:  nurbs_ex3 -m ../../data/square-nurbs.mesh
+//               nurbs_ex3 -m ../../data/square-nurbs.mesh -o 2
+//               nurbs_ex3 -m ../../data/cube-nurbs.mesh
+//
+// Description:  This example code solves a simple electromagnetic diffusion
+//               problem corresponding to the second order definite Maxwell
+//               equation curl curl E + E = f with boundary condition
+//               E x n = <given tangential field>. Here, we use a given exact
+//               solution E and compute the corresponding r.h.s. f.
+//               We discretize with Nedelec finite elements in 2D or 3D.
+//
+//               The example demonstrates the use of H(curl) finite element
+//               spaces with the curl-curl and the (vector finite element) mass
+//               bilinear form, as well as the computation of discretization
+//               error when the exact solution is known. Static condensation is
+//               also illustrated.
+//
+//               NURBS-based H(curl) spaces only implemented for meshes
+//               consisting of a single patch.
+//
+//               We recommend viewing examples 1-2 before viewing this example.
+
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+
+using namespace std;
+using namespace mfem;
+
+// Exact solution, E, and r.h.s., f. See below for implementation.
+void E_exact(const Vector &, Vector &);
+void f_exact(const Vector &, Vector &);
+real_t freq = 1.0, kappa;
+int dim;
+
+int main(int argc, char *argv[])
+{
+   // 1. Parse command-line options.
+   const char *mesh_file = "../../data/square-nurbs.mesh";
+   int ref_levels = -1;
+   bool NURBS = true;
+   int order = 1;
+   bool static_cond = false;
+   bool pa = false;
+   const char *device_config = "cpu";
+   bool visualization = 1;
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&ref_levels, "-r", "--refine",
+                  "Number of times to refine the mesh uniformly, -1 for auto.");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree).");
+   args.AddOption(&NURBS, "-n", "--nurbs", "-nn","--no-nurbs",
+                  "NURBS.");
+   args.AddOption(&freq, "-f", "--frequency", "Set the frequency for the exact"
+                  " solution.");
+   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
+                  "--no-static-condensation", "Enable static condensation.");
+   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
+                  "--no-partial-assembly", "Enable Partial Assembly.");
+   args.AddOption(&device_config, "-d", "--device",
+                  "Device configuration string, see Device::Configure().");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.Parse();
+   if (!args.Good())
+   {
+      args.PrintUsage(cout);
+      return 1;
+   }
+   args.PrintOptions(cout);
+   kappa = freq * M_PI;
+
+   // 2. Enable hardware devices such as GPUs, and programming models such as
+   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
+   Device device(device_config);
+   device.Print();
+
+   // 3. Read the mesh from the given mesh file. We can handle triangular,
+   //    quadrilateral, tetrahedral, hexahedral, surface and volume meshes with
+   //    the same code.
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
+   dim = mesh->Dimension();
+   int sdim = mesh->SpaceDimension();
+
+   // 4. Refine the mesh to increase the resolution. In this example we do
+   //    'ref_levels' of uniform refinement. We choose 'ref_levels' to be the
+   //    largest number that gives a final mesh with no more than 50,000
+   //    elements.
+   {
+      if (ref_levels < 0)
+      {
+         ref_levels =
+            (int)floor(log(50000./mesh->GetNE())/log(2.)/dim);
+      }
+      for (int l = 0; l < ref_levels; l++)
+      {
+         mesh->UniformRefinement();
+      }
+   }
+
+   // 5. Define a finite element space on the mesh. Here we use the
+   //    Raviart-Thomas finite elements of the specified order.
+   FiniteElementCollection *fec = nullptr;
+   NURBSExtension *NURBSext = nullptr;
+
+   if (mesh->NURBSext && NURBS)
+   {
+      fec = new NURBS_HCurlFECollection(order,dim);
+      NURBSext  = new NURBSExtension(mesh->NURBSext, order);
+      mfem::out<<"Create NURBS fec and ext"<<std::endl;
+   }
+   else
+   {
+      NURBS = false;
+      fec = new ND_FECollection(order, dim);
+      mfem::out<<"Create Normal fec"<<std::endl;
+   }
+
+   FiniteElementSpace *fespace = new FiniteElementSpace(mesh, NURBSext, fec);
+   cout << "Number of finite element unknowns: "
+        << fespace->GetTrueVSize() << endl;
+
+   // 6. Determine the list of true (i.e. conforming) essential boundary dofs.
+   //    In this example, the boundary conditions are defined by marking all
+   //    the boundary attributes from the mesh as essential (Dirichlet) and
+   //    converting them to a list of true dofs.
+   Array<int> ess_tdof_list;
+   if (mesh->bdr_attributes.Size())
+   {
+      Array<int> ess_bdr(mesh->bdr_attributes.Max());
+      ess_bdr = 1;
+      fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+   }
+   cout << "Number of knowns in essential BCs: "
+        << ess_tdof_list.Size() << endl;
+
+   // 7. Set up the linear form b(.) which corresponds to the right-hand side
+   //    of the FEM linear system, which in this case is (f,phi_i) where f is
+   //    given by the function f_exact and phi_i are the basis functions in the
+   //    finite element fespace.
+   VectorFunctionCoefficient f(sdim, f_exact);
+   LinearForm *b = new LinearForm(fespace);
+   b->AddDomainIntegrator(new VectorFEDomainLFIntegrator(f));
+   b->Assemble();
+
+   // 8. Define the solution vector x as a finite element grid function
+   //    corresponding to fespace. Initialize x by projecting the exact
+   //    solution. Note that only values from the boundary edges will be used
+   //    when eliminating the non-homogeneous boundary condition to modify the
+   //    r.h.s. vector b.
+   GridFunction x(fespace);
+   VectorFunctionCoefficient E(sdim, E_exact);
+   x.ProjectCoefficient(E);
+
+   // 9. Set up the bilinear form corresponding to the EM diffusion operator
+   //    curl muinv curl + sigma I, by adding the curl-curl and the mass domain
+   //    integrators.
+   Coefficient *muinv = new ConstantCoefficient(1.0);
+   Coefficient *sigma = new ConstantCoefficient(1.0);
+   BilinearForm *a = new BilinearForm(fespace);
+   if (pa) { a->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
+   a->AddDomainIntegrator(new CurlCurlIntegrator(*muinv));
+   a->AddDomainIntegrator(new VectorFEMassIntegrator(*sigma));
+
+   // 10. Assemble the bilinear form and the corresponding linear system,
+   //     applying any necessary transformations such as: eliminating boundary
+   //     conditions, applying conforming constraints for non-conforming AMR,
+   //     static condensation, etc.
+   if (static_cond) { a->EnableStaticCondensation(); }
+   a->Assemble();
+
+   OperatorPtr A;
+   Vector B, X;
+   a->FormLinearSystem(ess_tdof_list, x, *b, A, X, B);
+
+   cout << "Size of linear system: " << A->Height() << endl;
+
+   // 11. Solve the linear system A X = B.
+   if (pa) // Jacobi preconditioning in partial assembly mode
+   {
+      OperatorJacobiSmoother M(*a, ess_tdof_list);
+      PCG(*A, M, B, X, 1, 1000, 1e-12, 0.0);
+   }
+   else
+   {
+#ifndef MFEM_USE_SUITESPARSE
+      // 11. Define a simple symmetric Gauss-Seidel preconditioner and use it to
+      //     solve the system Ax=b with PCG.
+      GSSmoother M((SparseMatrix&)(*A));
+      PCG(*A, M, B, X, 1, 500, 1e-12, 0.0);
+#else
+      // 11. If MFEM was compiled with SuiteSparse, use UMFPACK to solve the
+      //     system.
+      UMFPackSolver umf_solver;
+      umf_solver.Control[UMFPACK_ORDERING] = UMFPACK_ORDERING_METIS;
+      umf_solver.SetOperator(*A);
+      umf_solver.Mult(B, X);
+#endif
+   }
+
+   // 12. Recover the solution as a finite element grid function.
+   a->RecoverFEMSolution(X, *b, x);
+
+   // 13. Compute and print the L^2 norm of the error.
+   cout << "\n|| E_h - E ||_{L^2} = " << x.ComputeL2Error(E) << '\n' << endl;
+
+   // 14. Save the refined mesh and the solution. This output can be viewed
+   //     later using GLVis: "glvis -m refined.mesh -g sol.gf".
+   {
+      ofstream mesh_ofs("refined.mesh");
+      mesh_ofs.precision(8);
+      mesh->Print(mesh_ofs);
+      ofstream sol_ofs("sol.gf");
+      sol_ofs.precision(8);
+      x.Save(sol_ofs);
+   }
+
+   // 15. Send the solution by socket to a GLVis server.
+   if (visualization)
+   {
+      char vishost[] = "localhost";
+      int  visport   = 19916;
+      socketstream sol_sock(vishost, visport);
+      sol_sock.precision(8);
+      sol_sock << "solution\n" << *mesh << x << flush;
+   }
+
+   // 16. Create output in visit format
+   VisItDataCollection visit_dc("Example3", mesh);
+   visit_dc.RegisterField("x", &x);
+   visit_dc.Save();
+
+   // 17. Free the used memory.
+   delete a;
+   delete sigma;
+   delete muinv;
+   delete b;
+   delete fespace;
+   delete fec;
+   delete mesh;
+
+   return 0;
+}
+
+
+void E_exact(const Vector &x, Vector &E)
+{
+   if (dim == 3)
+   {
+      E(0) = sin(kappa * x(1));
+      E(1) = sin(kappa * x(2));
+      E(2) = sin(kappa * x(0));
+   }
+   else
+   {
+      E(0) = sin(kappa * x(1));
+      E(1) = sin(kappa * x(0));
+      if (x.Size() == 3) { E(2) = 0.0; }
+   }
+}
+
+void f_exact(const Vector &x, Vector &f)
+{
+   if (dim == 3)
+   {
+      f(0) = (1. + kappa * kappa) * sin(kappa * x(1));
+      f(1) = (1. + kappa * kappa) * sin(kappa * x(2));
+      f(2) = (1. + kappa * kappa) * sin(kappa * x(0));
+   }
+   else
+   {
+      f(0) = (1. + kappa * kappa) * sin(kappa * x(1));
+      f(1) = (1. + kappa * kappa) * sin(kappa * x(0));
+      if (x.Size() == 3) { f(2) = 0.0; }
+   }
+}

miniapps/nurbs/nurbs_ex5.cpp

@@ -0,0 +1,498 @@
+//                                MFEM Example 5 -- modified for NURBS FE
+//
+// Compile with: make nurbs_ex5
+//
+// Sample runs:  nurbs_ex5 -m ../../data/square-nurbs.mesh -o 3
+//               nurbs_ex5 -m ../../data/cube-nurbs.mesh -r 3
+//               nurbs_ex5 -m ../../data/pipe-nurbs-2d.mesh
+//               nurbs_ex5 -m ../../data/beam-tet.mesh
+//               nurbs_ex5 -m ../../data/beam-hex.mesh
+//               nurbs_ex5 -m ../../data/escher.mesh
+//               nurbs_ex5 -m ../../data/fichera.mesh
+//
+// Device sample runs -- do not work for NURBS:
+//               nurbs_ex5 -m ../../data/escher.mesh -pa -d cuda
+//               nurbs_ex5 -m ../../data/escher.mesh -pa -d raja-cuda
+//               nurbs_ex5 -m ../../data/escher.mesh -pa -d raja-omp
+//
+// Description:  This example code solves a simple 2D/3D mixed Darcy problem
+//               corresponding to the saddle point system
+//
+//                                 k*u + grad p = f
+//                                 - div u      = g
+//
+//               with natural boundary condition -p = <given pressure>.
+//               Here, we use a given exact solution (u,p) and compute the
+//               corresponding r.h.s. (f,g).  We discretize with Raviart-Thomas
+//               finite elements (velocity u) and piecewise discontinuous
+//               polynomials (pressure p).
+//
+//               NURBS-based H(div) spaces only implemented for meshes
+//               consisting of a single patch.
+//
+//               The example demonstrates the use of the BlockOperator class, as
+//               well as the collective saving of several grid functions in
+//               VisIt (visit.llnl.gov) and ParaView (paraview.org) formats.
+//
+//               We recommend viewing examples 1-4 before viewing this example.
+
+// Sample runs:  nurbs_ex3 -m ../../data/square-nurbs.mesh
+//               nurbs_ex3 -m ../../data/square-nurbs.mesh -o 2
+//               nurbs_ex3 -m ../../data/cube-nurbs.mesh
+
+
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+#include <algorithm>
+
+using namespace std;
+using namespace mfem;
+
+// Define the analytical solution and forcing terms / boundary conditions
+void uFun_ex(const Vector & x, Vector & u);
+real_t pFun_ex(const Vector & x);
+void fFun(const Vector & x, Vector & f);
+real_t gFun(const Vector & x);
+real_t f_natural(const Vector & x);
+
+int main(int argc, char *argv[])
+{
+   StopWatch chrono;
+
+   // 1. Parse command-line options.
+   const char *mesh_file = "../../data/square-nurbs.mesh";
+   int ref_levels = -1;
+   int order = 1;
+   bool pa = false;
+   const char *device_config = "cpu";
+   bool visualization = 1;
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&ref_levels, "-r", "--refine",
+                  "Number of times to refine the mesh uniformly, -1 for auto.");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree).");
+   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
+                  "--no-partial-assembly", "Enable Partial Assembly.");
+   args.AddOption(&device_config, "-d", "--device",
+                  "Device configuration string, see Device::Configure().");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.Parse();
+   if (!args.Good())
+   {
+      args.PrintUsage(cout);
+      return 1;
+   }
+   args.PrintOptions(cout);
+
+   // 2. Enable hardware devices such as GPUs, and programming models such as
+   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
+   Device device(device_config);
+   device.Print();
+
+   // 3. Read the mesh from the given mesh file. We can handle triangular,
+   //    quadrilateral, tetrahedral, hexahedral, surface and volume meshes with
+   //    the same code.
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
+   int dim = mesh->Dimension();
+
+   // 4. Refine the mesh to increase the resolution. In this example we do
+   //    'ref_levels' of uniform refinement. We choose 'ref_levels' to be the
+   //    largest number that gives a final mesh with no more than 10,000
+   //    elements.
+   {
+      if (ref_levels < 0)
+      {
+         ref_levels =
+            (int)floor(log(10000./mesh->GetNE())/log(2.)/dim);
+      }
+      for (int l = 0; l < ref_levels; l++)
+      {
+         mesh->UniformRefinement();
+      }
+   }
+
+   // 5. Define a finite element space on the mesh. Here we use the
+   //    Raviart-Thomas finite elements of the specified order.
+   FiniteElementCollection *hdiv_coll = nullptr;
+   FiniteElementCollection *l2_coll = nullptr;
+   NURBSExtension *NURBSext = nullptr;
+
+   if (mesh->NURBSext && !pa)
+   {
+      hdiv_coll = new NURBS_HDivFECollection(order,dim);
+      l2_coll   = new NURBSFECollection(order);
+      NURBSext  = new NURBSExtension(mesh->NURBSext, order);
+      mfem::out<<"Create NURBS fec and ext"<<std::endl;
+   }
+   else
+   {
+      hdiv_coll = new RT_FECollection(order, dim);
+      l2_coll   = new L2_FECollection(order, dim);
+      mfem::out<<"Create Normal fec"<<std::endl;
+   }
+   pa = false;
+   FiniteElementSpace *W_space = new FiniteElementSpace(mesh, NURBSext, l2_coll);
+   FiniteElementSpace *R_space = new FiniteElementSpace(mesh,
+                                                        W_space->StealNURBSext(),
+                                                        hdiv_coll);
+
+   // 6. Define the BlockStructure of the problem, i.e. define the array of
+   //    offsets for each variable. The last component of the Array is the sum
+   //    of the dimensions of each block.
+   Array<int> block_offsets(3); // number of variables + 1
+   block_offsets[0] = 0;
+   block_offsets[1] = R_space->GetVSize();
+   block_offsets[2] = W_space->GetVSize();
+   block_offsets.PartialSum();
+
+   std::cout << "***********************************************************\n";
+   std::cout << "dim(R) = " << block_offsets[1] - block_offsets[0] << "\n";
+   std::cout << "dim(W) = " << block_offsets[2] - block_offsets[1] << "\n";
+   std::cout << "dim(R+W) = " << block_offsets.Last() << "\n";
+   std::cout << "***********************************************************\n";
+   {
+      Array<int> ess_tdof_list;
+      if (mesh->bdr_attributes.Size())
+      {
+         Array<int> ess_bdr(mesh->bdr_attributes.Max());
+         ess_bdr = 1;
+         R_space->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+      }
+      cout << "Number boundary dofs in H(div): "
+           << ess_tdof_list.Size() << endl;
+   }
+
+   {
+      Array<int> ess_tdof_list;
+      if (mesh->bdr_attributes.Size())
+      {
+         Array<int> ess_bdr(mesh->bdr_attributes.Max());
+         ess_bdr = 1;
+         W_space->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+      }
+      cout << "Number boundary dofs in H1: "
+           << ess_tdof_list.Size() << endl;
+   }
+
+   // 7. Define the coefficients, analytical solution, and rhs of the PDE.
+   ConstantCoefficient k(1.0);
+
+   VectorFunctionCoefficient fcoeff(dim, fFun);
+   FunctionCoefficient fnatcoeff(f_natural);
+   FunctionCoefficient gcoeff(gFun);
+
+   VectorFunctionCoefficient ucoeff(dim, uFun_ex);
+   FunctionCoefficient pcoeff(pFun_ex);
+
+   // 8. Allocate memory (x, rhs) for the analytical solution and the right hand
+   //    side.  Define the GridFunction u,p for the finite element solution and
+   //    linear forms fform and gform for the right hand side.  The data
+   //    allocated by x and rhs are passed as a reference to the grid functions
+   //    (u,p) and the linear forms (fform, gform).
+   MemoryType mt = device.GetMemoryType();
+   BlockVector x(block_offsets, mt), rhs(block_offsets, mt);
+
+   LinearForm *fform(new LinearForm);
+   fform->Update(R_space, rhs.GetBlock(0), 0);
+   fform->AddDomainIntegrator(new VectorFEDomainLFIntegrator(fcoeff));
+   fform->AddBoundaryIntegrator(new VectorFEBoundaryFluxLFIntegrator(fnatcoeff));
+   fform->Assemble();
+   fform->SyncAliasMemory(rhs);
+
+   LinearForm *gform(new LinearForm);
+   gform->Update(W_space, rhs.GetBlock(1), 0);
+   gform->AddDomainIntegrator(new DomainLFIntegrator(gcoeff));
+   gform->Assemble();
+   gform->SyncAliasMemory(rhs);
+
+   // 9. Assemble the finite element matrices for the Darcy operator
+   //
+   //                            D = [ M  B^T ]
+   //                                [ B   0  ]
+   //     where:
+   //
+   //     M = \int_\Omega k u_h \cdot v_h d\Omega   u_h, v_h \in R_h
+   //     B   = -\int_\Omega \div u_h q_h d\Omega   u_h \in R_h, q_h \in W_h
+   BilinearForm *mVarf(new BilinearForm(R_space));
+   MixedBilinearForm *bVarf(new MixedBilinearForm(R_space, W_space));
+
+   if (pa) { mVarf->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
+   mVarf->AddDomainIntegrator(new VectorFEMassIntegrator(k));
+   mVarf->Assemble();
+   if (!pa) { mVarf->Finalize(); }
+
+   if (pa) { bVarf->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
+   bVarf->AddDomainIntegrator(new VectorFEDivergenceIntegrator);
+   bVarf->Assemble();
+   if (!pa) { bVarf->Finalize(); }
+
+   BlockOperator darcyOp(block_offsets);
+
+   TransposeOperator *Bt = NULL;
+
+   if (pa)
+   {
+      Bt = new TransposeOperator(bVarf);
+
+      darcyOp.SetBlock(0,0, mVarf);
+      darcyOp.SetBlock(0,1, Bt, -1.0);
+      darcyOp.SetBlock(1,0, bVarf, -1.0);
+   }
+   else
+   {
+      SparseMatrix &M(mVarf->SpMat());
+      SparseMatrix &B(bVarf->SpMat());
+      B *= -1.;
+      Bt = new TransposeOperator(&B);
+
+      darcyOp.SetBlock(0,0, &M);
+      darcyOp.SetBlock(0,1, Bt);
+      darcyOp.SetBlock(1,0, &B);
+   }
+
+   // 10. Construct the operators for preconditioner
+   //
+   //                 P = [ diag(M)         0         ]
+   //                     [  0       B diag(M)^-1 B^T ]
+   //
+   //     Here we use Symmetric Gauss-Seidel to approximate the inverse of the
+   //     pressure Schur Complement
+   SparseMatrix *MinvBt = NULL;
+   Vector Md(mVarf->Height());
+
+   BlockDiagonalPreconditioner darcyPrec(block_offsets);
+   Solver *invM, *invS;
+   SparseMatrix *S = NULL;
+
+   if (pa)
+   {
+      mVarf->AssembleDiagonal(Md);
+      auto Md_host = Md.HostRead();
+      Vector invMd(mVarf->Height());
+      for (int i=0; i<mVarf->Height(); ++i)
+      {
+         invMd(i) = 1.0 / Md_host[i];
+      }
+
+      Vector BMBt_diag(bVarf->Height());
+      bVarf->AssembleDiagonal_ADAt(invMd, BMBt_diag);
+
+      Array<int> ess_tdof_list;  // empty
+
+      invM = new OperatorJacobiSmoother(Md, ess_tdof_list);
+      invS = new OperatorJacobiSmoother(BMBt_diag, ess_tdof_list);
+   }
+   else
+   {
+      SparseMatrix &M(mVarf->SpMat());
+      M.GetDiag(Md);
+      Md.HostReadWrite();
+
+      SparseMatrix &B(bVarf->SpMat());
+      MinvBt = Transpose(B);
+
+      for (int i = 0; i < Md.Size(); i++)
+      {
+         MinvBt->ScaleRow(i, 1./Md(i));
+      }
+
+      S = Mult(B, *MinvBt);
+
+      invM = new DSmoother(M);
+
+#ifndef MFEM_USE_SUITESPARSE
+      invS = new GSSmoother(*S);
+#else
+      invS = new UMFPackSolver(*S);
+#endif
+   }
+
+   invM->iterative_mode = false;
+   invS->iterative_mode = false;
+
+   darcyPrec.SetDiagonalBlock(0, invM);
+   darcyPrec.SetDiagonalBlock(1, invS);
+
+   // 11. Solve the linear system with MINRES.
+   //     Check the norm of the unpreconditioned residual.
+   int maxIter(10000);
+   real_t rtol(1.e-10);
+   real_t atol(1.e-10);
+
+   chrono.Clear();
+   chrono.Start();
+   MINRESSolver solver;
+   solver.SetAbsTol(atol);
+   solver.SetRelTol(rtol);
+   solver.SetMaxIter(maxIter);
+   solver.SetOperator(darcyOp);
+   solver.SetPreconditioner(darcyPrec);
+   solver.SetPrintLevel(1);
+   x = 0.0;
+   solver.Mult(rhs, x);
+   if (device.IsEnabled()) { x.HostRead(); }
+   chrono.Stop();
+
+   if (solver.GetConverged())
+   {
+      std::cout << "MINRES converged in " << solver.GetNumIterations()
+                << " iterations with a residual norm of "
+                << solver.GetFinalNorm() << ".\n";
+   }
+   else
+   {
+      std::cout << "MINRES did not converge in " << solver.GetNumIterations()
+                << " iterations. Residual norm is " << solver.GetFinalNorm()
+                << ".\n";
+   }
+   std::cout << "MINRES solver took " << chrono.RealTime() << "s.\n";
+
+   // 12. Create the grid functions u and p. Compute the L2 error norms.
+   GridFunction u, p;
+   u.MakeRef(R_space, x.GetBlock(0), 0);
+   p.MakeRef(W_space, x.GetBlock(1), 0);
+
+   int order_quad = max(2, 2*order+1);
+   const IntegrationRule *irs[Geometry::NumGeom];
+   for (int i=0; i < Geometry::NumGeom; ++i)
+   {
+      irs[i] = &(IntRules.Get(i, order_quad));
+   }
+
+   real_t err_u  = u.ComputeL2Error(ucoeff, irs);
+   real_t norm_u = ComputeLpNorm(2., ucoeff, *mesh, irs);
+   real_t err_p  = p.ComputeL2Error(pcoeff, irs);
+   real_t norm_p = ComputeLpNorm(2., pcoeff, *mesh, irs);
+
+   std::cout << "|| u_h - u_ex || / || u_ex || = " << err_u / norm_u << "\n";
+   std::cout << "|| p_h - p_ex || / || p_ex || = " << err_p / norm_p << "\n";
+
+   // 13. Save the mesh and the solution. This output can be viewed later using
+   //     GLVis: "glvis -m ex5.mesh -g sol_u.gf" or "glvis -m ex5.mesh -g
+   //     sol_p.gf".
+   {
+      ofstream mesh_ofs("ex5.mesh");
+      mesh_ofs.precision(8);
+      mesh->Print(mesh_ofs);
+
+      ofstream u_ofs("sol_u.gf");
+      u_ofs.precision(8);
+      u.Save(u_ofs);
+
+      ofstream p_ofs("sol_p.gf");
+      p_ofs.precision(8);
+      p.Save(p_ofs);
+   }
+
+   // 14. Save data in the VisIt format
+   VisItDataCollection visit_dc("Example5", mesh);
+   visit_dc.RegisterField("velocity", &u);
+   visit_dc.RegisterField("pressure", &p);
+   visit_dc.Save();
+
+   // 15. Save data in the ParaView format
+   ParaViewDataCollection paraview_dc("Example5", mesh);
+   paraview_dc.SetPrefixPath("ParaView");
+   paraview_dc.SetLevelsOfDetail(order);
+   paraview_dc.SetCycle(0);
+   paraview_dc.SetDataFormat(VTKFormat::BINARY);
+   paraview_dc.SetHighOrderOutput(true);
+   paraview_dc.SetTime(0.0); // set the time
+   paraview_dc.RegisterField("velocity",&u);
+   paraview_dc.RegisterField("pressure",&p);
+   paraview_dc.Save();
+
+   // 16. Send the solution by socket to a GLVis server.
+   if (visualization)
+   {
+      char vishost[] = "localhost";
+      int  visport   = 19916;
+      socketstream u_sock(vishost, visport);
+      u_sock.precision(8);
+      u_sock << "solution\n" << *mesh << u << "window_title 'Velocity'" << endl;
+      socketstream p_sock(vishost, visport);
+      p_sock.precision(8);
+      p_sock << "solution\n" << *mesh << p << "window_title 'Pressure'" << endl;
+   }
+
+   // 17. Free the used memory.
+   delete fform;
+   delete gform;
+   delete invM;
+   delete invS;
+   delete S;
+   delete Bt;
+   delete MinvBt;
+   delete mVarf;
+   delete bVarf;
+   delete W_space;
+   delete R_space;
+   delete l2_coll;
+   delete hdiv_coll;
+   delete mesh;
+
+   return 0;
+}
+
+
+void uFun_ex(const Vector & x, Vector & u)
+{
+   real_t xi(x(0));
+   real_t yi(x(1));
+   real_t zi(0.0);
+   if (x.Size() == 3)
+   {
+      zi = x(2);
+   }
+
+   u(0) = - exp(xi)*sin(yi)*cos(zi);
+   u(1) = - exp(xi)*cos(yi)*cos(zi);
+
+   if (x.Size() == 3)
+   {
+      u(2) = exp(xi)*sin(yi)*sin(zi);
+   }
+}
+
+// Change if needed
+real_t pFun_ex(const Vector & x)
+{
+   real_t xi(x(0));
+   real_t yi(x(1));
+   real_t zi(0.0);
+
+   if (x.Size() == 3)
+   {
+      zi = x(2);
+   }
+
+   return exp(xi)*sin(yi)*cos(zi);
+}
+
+void fFun(const Vector & x, Vector & f)
+{
+   f = 0.0;
+}
+
+real_t gFun(const Vector & x)
+{
+   if (x.Size() == 3)
+   {
+      return -pFun_ex(x);
+   }
+   else
+   {
+      return 0;
+   }
+}
+
+real_t f_natural(const Vector & x)
+{
+   return (-pFun_ex(x));
+}

miniapps/nurbs/nurbs_patch_ex1.cpp

@@ -153,7 +153,8 @@ int main(int argc, char *argv[])
    if (patchAssembly && reducedIntegration && !pa)
    {
 #ifdef MFEM_USE_SINGLE
-      MFEM_ABORT("Reduced integration is not supported in single precision.");
+      cout << "Reduced integration is not supported in single precision.\n";
+      return MFEM_SKIP_RETURN_VALUE;
 #endif
 
       di->SetIntegrationMode(NonlinearFormIntegrator::Mode::PATCHWISE_REDUCED);

miniapps/nurbs/nurbs_solenoidal.cpp

@@ -0,0 +1,401 @@
+// Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
+// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+// LICENSE and NOTICE for details. LLNL-CODE-806117.
+//
+// This file is part of the MFEM library. For more information and source code
+// availability visit https://mfem.org.
+//
+// MFEM is free software; you can redistribute it and/or modify it under the
+// terms of the BSD-3 license. We welcome feedback and contributions, see file
+// CONTRIBUTING.md for details.
+//
+//        ------------------------------------------------------------
+//        NURBS Solenoidal Miniapp: Project solenoidal velocity
+//        ------------------------------------------------------------
+//
+//
+// Compile with: make nurbs_solenoidal
+//
+// Sample runs:  nurbs_solenoidal -m ../../data/square-nurbs.mesh -o 2
+//               nurbs_solenoidal -m ../../data/cube-nurbs.mesh -o 2
+//
+// Description:  This code projects a velocity field, and forces this field
+//               to be solenoidal, viz. the divergence is zero. If the correct
+//               discrete spaces are chosen the divergence is pointwise zero.
+//
+//               This is achieved by solving a simple 2D/3D mixed Darcy problem
+//               corresponding to the saddle point system   (similar to ex5)
+//
+//                                 u + grad p = u_ex
+//                                    - div u = 0
+//
+//               NURBS-based H(div) spaces only implemented for meshes
+//               consisting of a single patch.
+//
+//               Here, u_ex is the specified velocity field. If u_ex is
+//               divergence free, we expect the pressure to converge to zero.
+//               We discretize with H(div) and L2/H1 conforming elements.
+
+
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+#include <algorithm>
+
+using namespace std;
+using namespace mfem;
+
+
+void u_2d(const Vector & x, Vector & u)
+{
+   real_t xi(x(0));
+   real_t yi(x(1));
+
+   int p = 4;
+
+   u(0) =  pow(xi,p + 1)*pow(yi,p    );
+   u(1) = -pow(xi,p    )*pow(yi,p + 1);
+}
+
+void u_3d(const Vector & x, Vector & u)
+{
+   real_t xi(x(0));
+   real_t yi(x(1));
+   real_t zi(x(2));
+
+   int p = 4;
+
+   real_t cx = 3.0/4.0;
+   real_t cy = 2.0/3.0;
+   real_t cz = -cx - cy;
+
+   u(0) = cx*pow(xi,p + 1)*pow(yi,p    )*pow(zi,p    );
+   u(1) = cy*pow(xi,p    )*pow(yi,p + 1)*pow(zi,p    );
+   u(2) = cz*pow(xi,p    )*pow(yi,p    )*pow(zi,p + 1);
+}
+
+// Define the analytical solution and forcing terms / boundary conditions
+void u_ex(const Vector & x, Vector & u)
+{
+   if (x.Size() == 2)
+   {
+      u_2d(x, u);
+   }
+   else if (x.Size() == 3)
+   {
+      u_3d(x, u);
+   }
+}
+
+int main(int argc, char *argv[])
+{
+   StopWatch chrono;
+
+   // 1. Parse command-line options.
+   const char *mesh_file = "../../data/square-nurbs.mesh";
+   int ref_levels = -1;
+   int order = 1;
+   const char *device_config = "cpu";
+   bool visualization = 1;
+   bool NURBS = true;
+   bool div_free = true;
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&ref_levels, "-r", "--refine",
+                  "Number of times to refine the mesh uniformly, -1 for auto.");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree).");
+   args.AddOption(&div_free, "-df", "--div-free", "-p","--proj",
+                  "Div-free or standard projection.");
+   args.AddOption(&NURBS, "-n", "--nurbs", "-nn","--no-nurbs",
+                  "NURBS.");
+   args.AddOption(&device_config, "-d", "--device",
+                  "Device configuration string, see Device::Configure().");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.Parse();
+   if (!args.Good())
+   {
+      args.PrintUsage(mfem::out);
+      return 1;
+   }
+   args.PrintOptions(mfem::out);
+
+   // 2. Enable hardware devices such as GPUs, and programming models such as
+   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
+   Device device(device_config);
+   device.Print();
+
+   // 3. Read the mesh from the given mesh file. We can handle triangular,
+   //    quadrilateral, tetrahedral, hexahedral, surface and volume meshes with
+   //    the same code.
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
+   int dim = mesh->Dimension();
+
+   // 4. Refine the mesh to increase the resolution. In this example we do
+   //    'ref_levels' of uniform refinement. We choose 'ref_levels' to be the
+   //    largest number that gives a final mesh with no more than 10,000
+   //    elements.
+   {
+      if (ref_levels < 0)
+      {
+         ref_levels =
+            (int)floor(log(5000./mesh->GetNE())/log(2.)/dim);
+      }
+      for (int l = 0; l < ref_levels; l++)
+      {
+         mesh->UniformRefinement();
+      }
+   }
+
+   // 5. Define a finite element space on the mesh. Here we use the
+   //    Raviart-Thomas finite elements of the specified order.
+   FiniteElementCollection *hdiv_coll = nullptr;
+   FiniteElementCollection *l2_coll = nullptr;
+   NURBSExtension *NURBSext = nullptr;
+
+   if (mesh->NURBSext&& NURBS)
+   {
+      hdiv_coll = new NURBS_HDivFECollection(order, dim);
+      l2_coll   = new NURBSFECollection(order);
+      NURBSext  = new NURBSExtension(mesh->NURBSext, order);
+      mfem::out<<"Create NURBS fec and ext"<<std::endl;
+   }
+   else
+   {
+      NURBS = false;
+      hdiv_coll = new RT_FECollection(order, dim);
+      l2_coll   = new L2_FECollection(order, dim);
+      mfem::out<<"Create Normal fec"<<std::endl;
+   }
+
+   FiniteElementSpace *W_space = new FiniteElementSpace(mesh, NURBSext, l2_coll);
+   FiniteElementSpace *R_space = new FiniteElementSpace(mesh,
+                                                        W_space->StealNURBSext(),
+                                                        hdiv_coll);
+
+   // 6. Define the BlockStructure of the problem, i.e. define the array of
+   //    offsets for each variable. The last component of the Array is the sum
+   //    of the dimensions of each block.
+   Array<int> block_offsets(3); // number of variables + 1
+   block_offsets[0] = 0;
+   block_offsets[1] = R_space->GetVSize();
+   block_offsets[2] = W_space->GetVSize();
+   block_offsets.PartialSum();
+
+   mfem::out << "***********************************************************\n";
+   mfem::out << "dim(R) = " << block_offsets[1] - block_offsets[0] << "\n";
+   mfem::out << "dim(W) = " << block_offsets[2] - block_offsets[1] << "\n";
+   mfem::out << "dim(R+W) = " << block_offsets.Last() << "\n";
+   mfem::out << "***********************************************************\n";
+
+   // 7. Define the coefficients, analytical solution, and rhs of the PDE.
+   ConstantCoefficient one(1.0);
+   ConstantCoefficient zero(0.0);
+   VectorFunctionCoefficient ucoeff(dim, u_ex);
+
+   // 8. Allocate memory (x, rhs) for the analytical solution and the right hand
+   //    side.  Define the GridFunction u,p for the finite element solution and
+   //    linear forms fform and gform for the right hand side.  The data
+   //    allocated by x and rhs are passed as a reference to the grid functions
+   //    (u,p) and the linear forms (fform, gform).
+   MemoryType mt = device.GetMemoryType();
+   BlockVector x(block_offsets, mt), rhs(block_offsets, mt);
+   rhs = 0.0;
+
+   LinearForm *fform(new LinearForm);
+   fform->Update(R_space, rhs.GetBlock(0), 0);
+   fform->AddDomainIntegrator(new VectorFEDomainLFIntegrator(ucoeff));
+   fform->Assemble();
+   fform->SyncAliasMemory(rhs);
+
+   // 9. Assemble the finite element matrices for the Darcy operator
+   //
+   //                            D = [ M  B^T ]
+   //                                [ B   0  ]
+   //     where:
+   //
+   //     M = \int_\Omega k u_h \cdot v_h d\Omega   u_h, v_h \in R_h
+   //     B   = -\int_\Omega \div u_h q_h d\Omega   u_h \in R_h, q_h \in W_h
+   BilinearForm *mVarf(new BilinearForm(R_space));
+   MixedBilinearForm *bVarf(new MixedBilinearForm(R_space, W_space));
+
+   mVarf->AddDomainIntegrator(new VectorFEMassIntegrator(one));
+   mVarf->Assemble();
+   mVarf->Finalize();
+
+   bVarf->AddDomainIntegrator(new VectorFEDivergenceIntegrator);
+   bVarf->Assemble();
+   bVarf->Finalize();
+
+   SparseMatrix &M(mVarf->SpMat());
+   SparseMatrix &B(bVarf->SpMat());
+   B *= -1.;
+   TransposeOperator *Bt = new TransposeOperator(&B);
+
+   BlockOperator darcyOp(block_offsets);
+   darcyOp.SetBlock(0,0, &M);
+   if (div_free) { darcyOp.SetBlock(0,1, Bt); }
+   if (div_free) { darcyOp.SetBlock(1,0, &B); }
+
+   // 10. Construct the operators for preconditioner
+   //
+   //                 P = [ diag(M)         0         ]
+   //                     [  0       B diag(M)^-1 B^T ]
+   //
+   //     Here we use Symmetric Gauss-Seidel to approximate the inverse of the
+   //     pressure Schur Complement
+   Vector Md(mVarf->Height());
+   M.GetDiag(Md);
+   Md.HostReadWrite();
+   SparseMatrix *MinvBt = Transpose(B);
+   for (int i = 0; i < Md.Size(); i++)
+   {
+      MinvBt->ScaleRow(i, 1./Md(i));
+   }
+   SparseMatrix *S = Mult(B, *MinvBt);
+   Solver *invS;
+#ifndef MFEM_USE_SUITESPARSE
+   invS = new GSSmoother(*S);
+#else
+   invS = new UMFPackSolver(*S);
+#endif
+   invS->iterative_mode = false;
+
+   Solver *invM = new GSSmoother(M);
+   invM->iterative_mode = false;
+
+   BlockDiagonalPreconditioner darcyPrec(block_offsets);
+   darcyPrec.SetDiagonalBlock(0, invM);
+   darcyPrec.SetDiagonalBlock(1, invS);
+
+   // 11. Solve the linear system with MINRES.
+   //     Check the norm of the unpreconditioned residual.
+   int maxIter(10000);
+   real_t rtol(10*std::numeric_limits<real_t>::epsilon());
+   real_t atol(10*std::numeric_limits<real_t>::epsilon());
+
+   chrono.Clear();
+   chrono.Start();
+   MINRESSolver solver;
+   solver.SetAbsTol(atol);
+   solver.SetRelTol(rtol);
+   solver.SetMaxIter(maxIter);
+   solver.SetOperator(darcyOp);
+   solver.SetPreconditioner(darcyPrec);
+   solver.SetPrintLevel(2);
+   x = 0.0;
+   solver.Mult(rhs, x);
+   if (device.IsEnabled()) { x.HostRead(); }
+   chrono.Stop();
+
+   if (solver.GetConverged())
+   {
+      mfem::out << "MINRES converged in " << solver.GetNumIterations()
+                << " iterations with a residual norm of "
+                << solver.GetFinalNorm() << ".\n";
+   }
+   else
+   {
+      mfem::out << "MINRES did not converge in " << solver.GetNumIterations()
+                << " iterations. Residual norm is " << solver.GetFinalNorm()
+                << ".\n";
+   }
+   mfem::out << "MINRES solver took " << chrono.RealTime() << "s.\n";
+
+   // 12. Create the grid functions u and p
+   GridFunction u, p, uu, vv, ww;
+   u.MakeRef(R_space, x.GetBlock(0), 0);
+   p.MakeRef(W_space, x.GetBlock(1), 0);
+
+   // 13. Save the mesh and the solution. This output can be viewed later using
+   //     GLVis: "glvis -m exsol.mesh -g sol_u.gf" or "glvis -m exsol.mesh -g
+   //     sol_p.gf".
+   {
+      ofstream mesh_ofs("exsol.mesh");
+      mesh_ofs.precision(8);
+      mesh->Print(mesh_ofs);
+
+      ofstream u_ofs("sol_u.gf");
+      u_ofs.precision(8);
+      u.Save(u_ofs);
+
+      ofstream p_ofs("sol_p.gf");
+      p_ofs.precision(8);
+      p.Save(p_ofs);
+   }
+
+   // 14. Save data in the VisIt format
+   VisItDataCollection visit_dc("Solenoidal", mesh);
+   visit_dc.RegisterField("velocity", &u);
+   visit_dc.RegisterField("pressure", &p);
+   visit_dc.Save();
+
+   // 15. Save data in the ParaView format
+   ParaViewDataCollection paraview_dc("Solenoidal", mesh);
+   paraview_dc.SetPrefixPath("ParaView");
+   paraview_dc.SetLevelsOfDetail(order);
+   paraview_dc.SetCycle(0);
+   paraview_dc.SetDataFormat(VTKFormat::BINARY);
+   paraview_dc.SetHighOrderOutput(true);
+   paraview_dc.SetTime(0.0); // set the time
+   paraview_dc.RegisterField("velocity",&u);
+   paraview_dc.RegisterField("pressure",&p);
+   paraview_dc.Save();
+
+   // 16. Send the solution by socket to a GLVis server.
+   if (visualization)
+   {
+      char vishost[] = "localhost";
+      int  visport   = 19916;
+      socketstream u_sock(vishost, visport);
+      u_sock.precision(8);
+      u_sock << "solution\n" << *mesh << u << "window_title 'Velocity'" << endl;
+      socketstream p_sock(vishost, visport);
+      p_sock.precision(8);
+      p_sock << "solution\n" << *mesh << p << "window_title 'Pressure'" << endl;
+   }
+
+   // 17. Compute errors
+   int order_quad = 2*order+2;
+   const IntegrationRule *irs[Geometry::NumGeom];
+   for (int i=0; i < Geometry::NumGeom; ++i)
+   {
+      irs[i] = &(IntRules.Get(i, order_quad));
+   }
+
+   real_t err_u  = u.ComputeL2Error(ucoeff, irs);
+   real_t err_p  = p.ComputeL2Error(zero, irs);
+   real_t err_div  = u.ComputeDivError(&zero, irs);
+
+   mfem::out << "|| u_h - u_ex ||  = " << err_u  << "\n";
+   mfem::out << "|| div u_h - div u_ex ||  = " << err_div  << "\n";
+   mfem::out << "|| p_h - p_ex ||  = " << err_p  << "\n";
+
+   // 18. Free the used memory.
+   delete fform;
+   delete invM;
+   delete invS;
+   delete S;
+   delete Bt;
+   delete MinvBt;
+   delete mVarf;
+   delete bVarf;
+   delete W_space;
+   delete R_space;
+   delete l2_coll;
+   delete hdiv_coll;
+   delete mesh;
+
+   if (err_div > 1e4*std::numeric_limits<real_t>::epsilon() )
+   {
+      mfem::out << "std::numeric_limits<real_t>::epsilon() = "
+                << std::numeric_limits<real_t>::epsilon()  << "\n";
+      mfem_error("Divergence error larger than expected");
+   }
+
+   return 0;
+}

miniapps/stabilized/CMakeLists.txt

@@ -0,0 +1,51 @@
+# Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
+# at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+# LICENSE and NOTICE for details. LLNL-CODE-806117.
+#
+# This file is part of the MFEM library. For more information and source code
+# availability visit https://mfem.org.
+#
+# MFEM is free software; you can redistribute it and/or modify it under the
+# terms of the BSD-3 license. We welcome feedback and contributions, see file
+# CONTRIBUTING.md for details.
+
+
+list(APPEND STAB_COMMON_SOURCES
+     stab_tau.cpp stab_condif.cpp stab_navsto.cpp)
+list(APPEND STAB_COMMON_HEADERS
+     stab_tau.hpp stab_condif.hpp stab_navsto.hpp manu.hpp skew.hpp)
+
+set(STAB_COMMON_FILES
+    EXTRA_SOURCES ${STAB_COMMON_SOURCES}
+    EXTRA_HEADERS ${STAB_COMMON_HEADERS})
+
+add_mfem_miniapp(condif
+    MAIN condif.cpp
+    ${STAB_COMMON_FILES}
+    LIBRARIES mfem)
+
+add_mfem_miniapp(navsto
+    MAIN navsto.cpp
+    ${STAB_COMMON_FILES}
+    LIBRARIES mfem)
+
+if (MFEM_ENABLE_TESTING)
+  add_test(NAME ex_condif
+    COMMAND $<TARGET_FILE:condif> -no-vis
+    -m ${PROJECT_SOURCE_DIR}/data/square-nurbs.mesh -r 1 -o 2)
+endif()
+
+if (MFEM_USE_MPI)
+  add_mfem_miniapp(navsto_p
+    MAIN navsto_p.cpp
+    ${STAB_COMMON_FILES}
+    LIBRARIES mfem)
+
+  if (MFEM_ENABLE_TESTING)
+    add_test(NAME navsto_np=4
+      COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
+      ${MPIEXEC_PREFLAGS} $<TARGET_FILE:navsto> -no-vis
+      ${MPIEXEC_POSTFLAGS}
+      -m ${PROJECT_SOURCE_DIR}/data/square-nurbs.mesh -r 1 -o 2)
+  endif()
+endif()

miniapps/stabilized/condif.cpp

@@ -0,0 +1,405 @@
+// Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
+// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+// LICENSE and NOTICE for details. LLNL-CODE-806117.
+//
+// This file is part of the MFEM library. For more information and source code
+// availability visit https://mfem.org.
+//
+// MFEM is free software; you can redistribute it and/or modify it under the
+// terms of the BSD-3 license. We welcome feedback and contributions, see file
+// CONTRIBUTING.md for details.
+//
+// Stabilized Convection-Diffusion
+
+#include "stab_condif.hpp"
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+#include <list>
+
+using namespace std;
+using namespace mfem;
+
+real_t att_param = 1.0;
+real_t kappa_param = 1.0;
+real_t pi  = (real_t)(M_PI);
+
+using VectorFun = std::function<void(const Vector & x, Vector & a)>;
+using ScalarFun = std::function<real_t(const Vector & x)>;
+
+#include "skew.hpp"
+#include "manu.hpp"
+
+void evaluate1D(Vector &x, Vector &f, GridFunction *gf, int lod)
+{
+   // Get Mesh and Nodes gridfunction
+   Mesh *mesh = gf->FESpace()->GetMesh();
+   GridFunction *nodes = mesh->GetNodes();
+   if (!nodes)
+   {
+      nodes = new GridFunction(gf->FESpace());
+      mesh->GetNodes(*nodes);
+   }
+
+   // Evaluate
+   std::list<pair<real_t,real_t>> sol;
+   Vector vals,coords;
+   for (int i = 0; i <  mesh->GetNE(); i++)
+   {
+      int geom       = mesh->GetElementBaseGeometry(i);
+      RefinedGeometry *refined_geo = GlobGeometryRefiner.Refine(( Geometry::Type)geom, 1, lod);
+
+      gf->GetValues(i, refined_geo->RefPts, vals);
+      nodes->GetValues(i, refined_geo->RefPts, coords);
+
+      for (int j = 0; j < vals.Size(); j++)
+      {
+         sol.push_back(std::make_pair(coords[j],vals[j]));
+      }
+   }
+
+   // Sort and make unique
+   sol.sort();
+   sol.unique();
+
+   // Convert to Vectors
+   x.SetSize(sol.size());
+   f.SetSize(sol.size());
+   int i = 0;
+   for (std::list<pair<real_t,real_t>>::iterator d = sol.begin() ; d != sol.end(); ++d, i++)
+   {
+      x[i] = d->first;
+      f[i] = d->second;
+   }
+}
+
+StabType GetStabilisationType(int stype)
+{
+   switch (stype)
+   {
+      case GALERKIN:
+         mfem::out<<"Galerkin formulation"<<std::endl;
+         break;
+      case SUPG:
+         mfem::out<<"SUPG formulation"<<std::endl;
+         break;
+      case GLS:
+         mfem::out<<"GLS formulation"<<std::endl;
+         break;
+      case VMS:
+         mfem::out<<"VMS formulation"<<std::endl;
+         break;
+      default:
+         mfem::out<<"GAL"<<"\t"<<"SUPG"<<"\t"<<"GLS"<<"\t"<<"VMS"<<std::endl;
+         mfem::out<<GALERKIN<<"\t"<<SUPG<<"\t"<<GLS<<"\t"<<VMS<<std::endl;
+         mfem_error("Wrong formulation");
+   }
+   return (StabType) stype;
+}
+
+int main(int argc, char *argv[])
+{
+   // 1. Parse command-line options.
+   const char *mesh_file = "../../data/inline-quad.mesh";
+   const char *ref_file  = "";
+   int problem = 0;
+   int sstype = -2;
+   bool static_cond = false;
+   bool visualization = false;
+   int lod = 0;
+   real_t penalty = -1;
+   Array<int> order(1);
+   order[0] = 2;
+   int ref_levels = 0;
+
+   bool mono = true;
+
+   OptionsParser args(argc, argv);
+
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&ref_file, "-rf", "--ref-file",
+                  "File with refinement data");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree) or -1 for"
+                  " isoparametric space.");
+   args.AddOption(&ref_levels, "-r", "--refine",
+                  "Number of times to refine the mesh.");
+   args.AddOption(&kappa_param , "-k", "--kappa",
+                  "Sets the diffusion parameters, should be positive.");
+   args.AddOption(&att_param , "-a", "--att",
+                  "Sets the velocity direction");
+   args.AddOption(&problem, "-p", "--problem",
+                  "Select the problem to solve:\n\t"
+                  "  0 = convection skew-to-the mesh\n\t"
+                  "  1 = manufactured solution\n");
+   args.AddOption(&sstype, "-s", "--stab", " Stabilization type:\n\t"
+                  " -2 = Galerkin\n\t"
+                  " -1 = GLS\n\t"
+                  "  0 = SUPG\n\t"
+                  "  1 = VMS\n");
+   args.AddOption(&mono, "-mo", "--mono", "-co",
+                  "--comp",
+                  "Use a monolithic integrator or a composed one.");
+   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
+                  "--no-static-condensation", "Enable static condensation.");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.AddOption(&lod, "-lod", "--level-of-detail",
+                  "Refinement level for 1D solution output (0 means no output).");
+   args.Parse();
+   if (!args.Good())
+   {
+      args.PrintUsage(cout);
+
+   }
+   args.PrintOptions(mfem::out);
+
+
+
+   // 2. Read the mesh from the given mesh file. We can handle triangular,
+   //    quadrilateral, tetrahedral, hexahedral, surface and volume meshes with
+   //    the same code.
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
+   int dim = mesh->Dimension();
+
+   // 3. Refine the mesh to increase the resolution. In this example we do
+   //    'ref_levels' of uniform refinement and knot insertion of knots defined
+   //    in a refinement file. We choose 'ref_levels' to be the largest number
+   //    that gives a final mesh with no more than 50,000 elements.
+   {
+      // Mesh refinement as defined in refinement file
+      if (mesh->NURBSext && (strlen(ref_file) != 0))
+      {
+         mesh->RefineNURBSFromFile(ref_file);
+      }
+
+      for (int l = 0; l < ref_levels; l++)
+      {
+         mesh->UniformRefinement();
+      }
+      mesh->PrintInfo();
+   }
+
+   // 4. Define a finite element space on the mesh. Here we use continuous
+   //    Lagrange finite elements of the specified order. If order < 1, we
+   //    instead use an isoparametric/isogeometric space.
+   FiniteElementCollection *fec;
+   NURBSExtension *NURBSext = nullptr;
+   int own_fec = 1;
+
+   if (mesh->NURBSext)
+   {
+      fec = new NURBSFECollection(order[0]);
+
+      int nkv = mesh->NURBSext->GetNKV();
+      if (order.Size() == 1)
+      {
+         int tmp = order[0];
+         order.SetSize(nkv);
+         order = tmp;
+      }
+
+      if (order.Size() != nkv ) { mfem_error("Wrong number of orders set."); }
+      NURBSext = new NURBSExtension(mesh->NURBSext, order);
+   }
+   else if (order[0] == -1) // Isoparametric
+   {
+      if (mesh->GetNodes())
+      {
+         fec = mesh->GetNodes()->OwnFEC();
+         own_fec = 0;
+         mfem::out << "Using isoparametric FEs: " << fec->Name() << endl;
+      }
+      else
+      {
+         mfem::out <<"Mesh does not have FEs --> Assume order 1.\n";
+         fec = new H1_FECollection(1, dim);
+      }
+   }
+   else
+   {
+      if (order.Size() > 1) { cout <<"Wrong number of orders set, needs one.\n"; }
+      fec = new H1_FECollection(abs(order[0]), dim);
+   }
+
+   FiniteElementSpace *fespace = new FiniteElementSpace(mesh, NURBSext, fec);
+   mfem::out << "Number of finite element unknowns: "
+             << fespace->GetTrueVSize() << endl;
+
+   // 5. Determine the list of true (i.e. conforming) essential boundary dofs.
+   //    In this example, the boundary conditions are defined by marking all
+   //    the boundary attributes from the mesh as essential (Dirichlet) and
+   //    converting them to a list of true dofs.
+   Array<int> ess_tdof_list;
+   if (mesh->bdr_attributes.Size())
+   {
+      Array<int> ess_bdr(mesh->bdr_attributes.Max());
+      ess_bdr = 1;
+      fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+   }
+
+   // 6. Set up the linear form b(.) which corresponds to the right-hand side of
+   //    the FEM linear system, which in this case is (1,phi_i) where phi_i are
+   //    the basis functions in the finite element fespace.
+   VectorFunctionCoefficient *adv, *grad;
+   FunctionCoefficient *kappa,*force, *sol, *lap;
+
+   if (problem == 0)
+   {
+      if (mesh->Dimension() != 2) mfem_error("Advection skew to the mesh needs a 2D mesh!");
+      adv = new VectorFunctionCoefficient(mesh->Dimension(), skew::adv);
+      kappa= new FunctionCoefficient(skew::kappa);
+
+      force = new FunctionCoefficient(skew::force);
+      sol = new FunctionCoefficient(skew::sol);
+      grad = new VectorFunctionCoefficient(mesh->Dimension(), skew::grad);
+      lap = new FunctionCoefficient(skew::laplace);
+
+   }
+   else if (problem == 1)
+   {
+      adv = new VectorFunctionCoefficient(mesh->Dimension(), manufactured::adv);
+      kappa= new FunctionCoefficient(manufactured::kappa);
+
+      force = new FunctionCoefficient(manufactured::force);
+      sol = new FunctionCoefficient(manufactured::sol);
+      grad = new VectorFunctionCoefficient(mesh->Dimension(), manufactured::grad);
+      lap = new FunctionCoefficient(manufactured::laplace);
+   }
+   else 
+   {
+      mfem_error("Incorrect problem!");
+   }
+
+   // 7. Define the solution vector x as a finite element grid function
+   //    corresponding to fespace. Initialize x with initial guess of zero,
+   //    which satisfies the boundary conditions.
+   GridFunction x(fespace);
+   x.ProjectCoefficient(*sol);
+
+   if (problem == 1)
+   {
+      Vector norm(3);
+      norm[0] = x.ComputeL2Error(*sol);
+      norm[1] = x.ComputeGradError(grad);
+      norm[2] = x.ComputeLaplaceError(lap);
+
+      mfem::out << "|| x_h - x_ex || = " << norm[0]  << "\n";
+      mfem::out << "|| grad x_h - grad x_ex ||    = " << norm[1] << "\n";
+      mfem::out << "|| lap x_h - lap x_ex ||      = " << norm[2] << "\n";
+   }
+
+   // 9. Assemble the bilinear form and the corresponding linear system,
+   //    applying any necessary transformations such as: eliminating boundary
+   //    conditions, applying conforming constraints for non-conforming AMR,
+   //    static condensation, etc.
+   StabType stype = GetStabilisationType(sstype);
+   FFH92Tau tau (adv, kappa, fespace);
+   StabConDifComposition stab_condif_comp(adv, kappa, force, &tau);
+
+   BilinearForm a(fespace);
+   LinearForm b(fespace);
+
+   if (mono)
+   {
+      a.AddDomainIntegrator(new StabConDifIntegrator(adv, kappa, force, &tau, stype));
+      b.AddDomainIntegrator(new StabConDifIntegrator(adv, kappa, force, &tau, stype));
+   }
+   else
+   {
+      stab_condif_comp.SetBilinearIntegrators(&a, stype);
+      stab_condif_comp.SetLinearIntegrators(&b, stype);
+   }
+
+   a.Assemble();
+   b.Assemble();
+
+   if (static_cond) { a.EnableStaticCondensation(); }
+   SparseMatrix A;
+   Vector B, X;
+
+   a.FormLinearSystem(ess_tdof_list, x, b, A, X, B);
+
+   mfem::out << "Size of linear system: " << A.Height() << endl;
+
+#ifndef MFEM_USE_SUITESPARSE
+   // 10. Define a simple Jacobi preconditioner and use it to
+   //     solve the system A X = B with PCG.
+   GSSmoother M(A);
+   GMRES(A, M, B, X, 1, 2000, 2000, 1e-16, 0.0);
+#else
+   // 10. If MFEM was compiled with SuiteSparse, use UMFPACK to solve the system.
+   UMFPackSolver umf_solver;
+   umf_solver.Control[UMFPACK_ORDERING] = UMFPACK_ORDERING_METIS;
+   umf_solver.SetOperator(A);
+   umf_solver.Mult(B, X);
+#endif
+
+   // 11. Recover the solution as a finite element grid function.
+   a.RecoverFEMSolution(X, b, x);
+
+   // 12. Save the refined mesh and the solution. This output can be viewed later
+   //     using GLVis: "glvis -m refined.mesh -g sol.gf".
+   {
+      ofstream mesh_ofs("refined.mesh");
+      mesh_ofs.precision(8);
+      mesh->Print(mesh_ofs);
+      ofstream sol_ofs("sol.gf");
+      sol_ofs.precision(8);
+      x.Save(sol_ofs);
+      sol_ofs.close();
+   }
+
+   // 13. Send the solution by socket to a GLVis server.
+   if (visualization)
+   {
+      char vishost[] = "localhost";
+      int  visport   = 19916;
+      socketstream sol_sock(vishost, visport);
+      sol_sock.precision(8);
+      sol_sock << "solution\n" << *mesh << x << flush;
+   }
+
+   if (mesh->Dimension() == 1 && lod > 0)
+   {
+      Vector coord, val;
+      evaluate1D(coord, val, &x, lod);
+
+      ofstream sol_ofs("solution.dat");
+      for (int i = 0; i < x.Size();i++)
+      {
+         sol_ofs<<coord[i] <<"\t"<<val[i]<<endl;
+      }
+      sol_ofs.close();
+   }
+
+   // 14. Error computation
+   if (problem == 1)
+   {
+      Vector norm(3);
+      norm[0] = x.ComputeL2Error(*sol);
+      norm[1] = x.ComputeGradError(grad);
+      norm[2] = x.ComputeLaplaceError(lap);
+
+      mfem::out << "|| x_h - x_ex || = " << norm[0]  << "\n";
+      mfem::out << "|| grad x_h - grad x_ex ||    = " << norm[1] << "\n";
+      mfem::out << "|| lap x_h - lap x_ex ||      = " << norm[2] << "\n";
+   }
+
+   // 15. Save data in the VisIt format
+   VisItDataCollection visit_dc("condif", mesh);
+   visit_dc.RegisterField("solution", &x);
+   visit_dc.Save();
+
+   // 16. Free the used memory.
+   delete fespace;
+   if (own_fec) { delete fec; }
+   delete mesh;
+   delete adv, grad;
+   delete kappa, force, sol, lap;
+
+   return 0;
+}
+

miniapps/stabilized/manu.hpp

@@ -0,0 +1,114 @@
+namespace manufactured
+{
+
+//----------------------------------------------------------
+void adv(const Vector & x, Vector & a)
+{
+   a[1] = 1.0/(1.0 + att_param*att_param);
+   a[0] = sqrt(1.0 - a[1]*a[1]);
+}
+
+//----------------------------------------------------------
+real_t kappa(const Vector & x)
+{
+   return kappa_param;
+}
+
+//----------------------------------------------------------
+real_t force(const Vector & x)
+{
+   int d = x.Size();
+
+   Vector a(d);
+   adv(x, a);
+   real_t ax = a[0];
+   real_t ay = 0.0;
+   real_t az = 0.0;
+   real_t k = kappa(x);
+
+   real_t sx = sin(pi*x[0]);
+   real_t cx = cos(pi*x[0]);
+   real_t sy = 1.0;
+   real_t cy = 1.0;
+   real_t sz = 1.0;
+   real_t cz = 1.0;
+
+   if (d >= 2)
+   {
+      sy = sin(pi*x[1]);
+      cy = cos(pi*x[1]);
+      ay = a[1];
+   }
+   if (d >= 3)
+   {
+      sz = sin(pi*x[2]);
+      cz = cos(pi*x[2]);
+      az = a[2];
+   }
+
+   return ax*pi*cx*sy*sz
+        + ay*pi*sx*cy*sz
+        + az*pi*sx*sy*cz + d*k*pi*pi*sx*sy*sz;
+}
+
+//----------------------------------------------------------
+real_t sol(const Vector & x)
+{
+   real_t sx = sin(pi*x[0]);
+   real_t sy = 1.0;
+   real_t sz = 1.0;
+
+   int d = x.Size();
+   if (d >= 2)
+   {
+      sy = sin(pi*x[1]);
+   }
+   if (d >= 3)
+   {
+      sz = sin(pi*x[2]);
+   }
+
+   return sx*sy*sz;
+}
+
+
+//----------------------------------------------------------
+void grad(const Vector & x, Vector &grad)
+{
+   real_t sx = sin(pi*x[0]);
+   real_t sy = 1.0;
+   real_t sz = 1.0;
+
+   real_t gx = pi*cos(pi*x[0]);
+   real_t gy = 0.0;
+   real_t gz = 0.0;
+
+   grad[0] = gx;
+
+   int d = x.Size();
+   if (d >= 2)
+   {
+      sy = sin(pi*x[1]);
+      gy = pi*cos(pi*x[1]);
+
+      grad[0] = gx*sy;
+      grad[1] = sx*gy;
+   }
+   if (d >= 3)
+   {
+      sz = sin(pi*x[2]);
+      gz = pi*cos(pi*x[2]);
+
+      grad[0] = gx*sy*sz;
+      grad[1] = sx*gy*sz;
+      grad[2] = sx*sy*gz;
+   }
+}
+
+//----------------------------------------------------------
+real_t laplace(const Vector & x)
+{
+   return -x.Size()*pi*pi*sol(x);
+}
+
+}

miniapps/stabilized/navsto.cpp

@@ -0,0 +1,282 @@
+// Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
+// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+// LICENSE and NOTICE for details. LLNL-CODE-806117.
+//
+// This file is part of the MFEM library. For more information and source code
+// availability visit https://mfem.org.
+//
+// MFEM is free software; you can redistribute it and/or modify it under the
+// terms of the BSD-3 license. We welcome feedback and contributions, see file
+// CONTRIBUTING.md for details.
+//
+// Stabilized  Navier-Stokes
+
+#include "stab_navsto.hpp"
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+#include <list>
+
+using namespace std;
+using namespace mfem;
+
+real_t kappa_param = 1.0;
+real_t pi  = (real_t)(M_PI);
+
+using VectorFun = std::function<void(const Vector & x, Vector & a)>;
+using ScalarFun = std::function<real_t(const Vector & x)>;
+
+void sol_fun(const Vector & x, Vector &sol)
+{
+   sol = 0.0;
+   if ((x[1] - 0.99 > 0.0) &&
+       (fabs(x[0] - 0.5) < 0.49) )
+   {
+      sol[0] = 1.0;
+   }
+}
+
+real_t kappa_fun(const Vector & x)
+{
+   return kappa_param;
+}
+
+void force_fun(const Vector & x, Vector &f)
+{
+   f = 0.0;
+  // f[0] = x[1]*(1.0-x[1])*x[0]*(1.0-x[0]);
+}
+
+StabType GetStabilisationType(int stype)
+{
+   switch (stype)
+   {
+      case GALERKIN:
+         mfem::out<<"Galerkin formulation"<<std::endl;
+         break;
+      case SUPG:
+         mfem::out<<"SUPG formulation"<<std::endl;
+         break;
+      case GLS:
+         mfem::out<<"GLS formulation"<<std::endl;
+         break;
+      case VMS:
+         mfem::out<<"VMS formulation"<<std::endl;
+         break;
+      default:
+         mfem::out<<"GAL"<<"\t"<<"SUPG"<<"\t"<<"GLS"<<"\t"<<"VMS"<<std::endl;
+         mfem::out<<GALERKIN<<"\t"<<SUPG<<"\t"<<GLS<<"\t"<<VMS<<std::endl;
+         mfem_error("Wrong formulation");
+   }
+   return (StabType) stype;
+}
+
+int main(int argc, char *argv[])
+{
+   // Parse command-line options.
+   const char *mesh_file = "../../data/inline-quad.mesh";
+   const char *ref_file  = "";
+   int problem = 0;
+   int sstype = -2;
+   bool static_cond = false;
+   bool visualization = false;
+
+   real_t penalty = -1;
+   int order = 1;
+   int ref_levels = 0;
+
+   bool mono = true;
+
+   OptionsParser args(argc, argv);
+
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&ref_file, "-rf", "--ref-file",
+                  "File with refinement data");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order isoparametric space.");
+   args.AddOption(&ref_levels, "-r", "--refine",
+                  "Number of times to refine the mesh.");
+   args.AddOption(&kappa_param , "-k", "--kappa",
+                  "Sets the diffusion parameters, should be positive.");
+   args.AddOption(&problem, "-p", "--problem",
+                  "Select the problem to solve:\n\t"
+                  "  0 = convection skew-to-the mesh\n\t"
+                  "  1 = manufactured solution\n");
+   args.AddOption(&sstype, "-s", "--stab", " Stabilization type:\n\t"
+                  " -2 = Galerkin\n\t"
+                  " -1 = GLS\n\t"
+                  "  0 = SUPG\n\t"
+                  "  1 = VMS\n");
+   args.AddOption(&mono, "-mo", "--mono", "-co",
+                  "--comp",
+                  "Use a monolithic integrator or a composed one.");
+   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
+                  "--no-static-condensation", "Enable static condensation.");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.Parse();
+   if (!args.Good())
+   {
+      args.PrintUsage(cout);
+
+   }
+   args.PrintOptions(mfem::out);
+
+   // Read the mesh from the given mesh file. We can handle triangular,
+   // quadrilateral, tetrahedral, hexahedral, surface and volume meshes with
+   // the same code.
+   Mesh mesh(mesh_file, 1, 1);
+   int dim = mesh.Dimension();
+
+   // Refine the mesh to increase the resolution. In this example we do
+   // 'ref_levels' of uniform refinement and knot insertion of knots defined
+   // in a refinement file. We choose 'ref_levels' to be the largest number
+   // that gives a final mesh with no more than 50,000 elements.
+   {
+      // Mesh refinement as defined in refinement file
+      if (mesh.NURBSext && (strlen(ref_file) != 0))
+      {
+         mesh.RefineNURBSFromFile(ref_file);
+      }
+
+      for (int l = 0; l < ref_levels; l++)
+      {
+         mesh.UniformRefinement();
+      }
+      mesh.PrintInfo();
+   }
+
+   // Define a finite element space on the mesh. Here we use continuous
+   // Lagrange finite elements of the specified order. If order < 1, we
+   // instead use an isoparametric/isogeometric space.
+   Array<FiniteElementCollection *> fecs(2);
+   fecs[0] = new H1_FECollection(order, dim);
+   fecs[1] = new H1_FECollection(order, dim);
+
+   Array<FiniteElementSpace *> spaces(2);
+   spaces[0] = new FiniteElementSpace(&mesh, fecs[0], dim);
+   spaces[1] = new FiniteElementSpace(&mesh, fecs[1]);
+
+   mfem::out << "Number of finite element unknowns:\n"
+             << "\tVelocity = "<<spaces[0]->GetTrueVSize() << endl
+             << "\tPressure = "<<spaces[1]->GetTrueVSize() << endl;
+   // Determine the list of true (i.e. conforming) essential boundary dofs.
+   // In this example, the boundary conditions are defined by marking all
+   // the boundary attributes from the mesh as essential (Dirichlet) and
+   // converting them to a list of true dofs.
+   Array<Array<int> *> ess_bdr(2);
+   Array<int> ess_tdof_list;
+
+   Array<int> ess_bdr_u(spaces[0]->GetMesh()->bdr_attributes.Max());
+   Array<int> ess_bdr_p(spaces[1]->GetMesh()->bdr_attributes.Max());
+
+   ess_bdr_p = 0;
+   ess_bdr_u = 1;
+
+   ess_bdr[0] = &ess_bdr_u;
+   ess_bdr[1] = &ess_bdr_p;
+
+   // Set up the linear form b(.) which corresponds to the right-hand side of
+   // the FEM linear system, which in this case is (1,phi_i) where phi_i are
+   // the basis functions in the finite element fespace.
+
+   // Define the solution vector xp as a finite element grid function
+   Array<int> bOffsets(3);
+   bOffsets[0] = 0;
+   bOffsets[1] = spaces[0]->GetTrueVSize();
+   bOffsets[2] = spaces[1]->GetTrueVSize();
+   bOffsets.PartialSum();
+
+   BlockVector xp(bOffsets);
+
+   GridFunction x_u(spaces[0]);
+   GridFunction x_p(spaces[1]);
+
+   x_u.MakeTRef(spaces[0], xp.GetBlock(0), 0);
+   x_p.MakeTRef(spaces[1], xp.GetBlock(1), 0);
+
+   VectorFunctionCoefficient sol(dim, sol_fun);
+
+   x_u.ProjectCoefficient(sol);
+   x_p = 0.0;
+
+   x_u.SetTrueVector();
+   x_p.SetTrueVector();
+
+   // Define the output
+   VisItDataCollection visit_dc("navsto", &mesh);
+   visit_dc.RegisterField("u", &x_u);
+   visit_dc.RegisterField("p", &x_p);
+   visit_dc.SetCycle(0);
+   visit_dc.Save();
+
+   // Define the problem parameters
+   FunctionCoefficient kappa(kappa_fun);
+   VectorFunctionCoefficient force(dim, force_fun);
+
+   // Define the stabilisation parameters
+   VectorGridFunctionCoefficient adv(&x_u);
+   ElasticInverseEstimateCoefficient invEst(spaces[0]);
+   FFH92Tau tau(&adv, &kappa, &invEst, 4.0);
+   FF91Delta delta(&adv, &kappa, &invEst);
+
+   tau.print = delta.print = true;
+
+   // Define the block nonlinear form
+   BlockNonlinearForm Hform(spaces);
+   Hform.AddDomainIntegrator(new StabInNavStoIntegrator(kappa, force, tau, delta));
+   Array<Vector *> rhs(2);
+   rhs = nullptr; // Set all entries in the array
+   Hform.SetEssentialBC(ess_bdr, rhs);
+
+   // Set up the preconditioner
+   JacobianPreconditioner jac_prec(bOffsets,
+                                   Array<Solver *>({new GSSmoother(0,5),
+                                                    new GSSmoother(0,5)}));
+
+   // Set up the Jacobian solver
+   GeneralResidualMonitor j_monitor("\t\t\t\tFGMRES", 25);
+   FGMRESSolver j_gmres;
+   j_gmres.iterative_mode = false;
+   j_gmres.SetRelTol(1e-2);
+   j_gmres.SetAbsTol(1e-12);
+   j_gmres.SetMaxIter(300);
+   j_gmres.SetPrintLevel(-1);
+   j_gmres.SetMonitor(j_monitor);
+   j_gmres.SetPreconditioner(jac_prec);
+
+   // Set up the newton solver
+   SystemResidualMonitor newton_monitor("Newton", 1, bOffsets, &visit_dc);
+   NewtonSolver newton_solver;
+   newton_solver.iterative_mode = true;
+   newton_solver.SetPrintLevel(-1);
+   newton_solver.SetMonitor(newton_monitor);
+   newton_solver.SetRelTol(1e-4);
+   newton_solver.SetAbsTol(1e-8);
+   newton_solver.SetMaxIter(25);
+   newton_solver.SetSolver(j_gmres);
+   newton_solver.SetOperator(Hform);
+
+   // Solve the Newton system
+   Vector zero;
+   newton_solver.Mult(zero, xp);
+
+   // Save data in the VisIt format
+   visit_dc.SetCycle(999999);
+   visit_dc.Save();
+
+   // Free the used memory.
+   for (int i = 0; i < fecs.Size(); ++i)
+   {
+      delete fecs[i];
+   }
+   for (int i = 0; i < spaces.Size(); ++i)
+   {
+      delete spaces[i];
+   }
+
+   return 0;
+}
+

miniapps/stabilized/navsto_p.cpp

@@ -0,0 +1,297 @@
+// Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
+// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+// LICENSE and NOTICE for details. LLNL-CODE-806117.
+//
+// This file is part of the MFEM library. For more information and source code
+// availability visit https://mfem.org.
+//
+// MFEM is free software; you can redistribute it and/or modify it under the
+// terms of the BSD-3 license. We welcome feedback and contributions, see file
+// CONTRIBUTING.md for details.
+//
+// Stabilized  Navier-Stokes
+
+#include "stab_navsto.hpp"
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+#include <list>
+
+using namespace std;
+using namespace mfem;
+
+real_t kappa_param = 1.0;
+real_t pi  = (real_t)(M_PI);
+
+using VectorFun = std::function<void(const Vector & x, Vector & a)>;
+using ScalarFun = std::function<real_t(const Vector & x)>;
+
+void sol_fun(const Vector & x, Vector &sol)
+{
+   sol = 0.0;
+   if ((x[1] - 0.99 > 0.0) &&
+       (fabs(x[0] - 0.5) < 0.49) )
+   {
+      sol[0] = 1.0;
+   }
+}
+
+real_t kappa_fun(const Vector & x)
+{
+   return kappa_param;
+}
+
+void force_fun(const Vector & x, Vector &f)
+{
+   f = 0.0;
+  // f[0] = x[1]*(1.0-x[1])*x[0]*(1.0-x[0]);
+}
+
+StabType GetStabilisationType(int stype)
+{
+   switch (stype)
+   {
+      case GALERKIN:
+         mfem::out<<"Galerkin formulation"<<std::endl;
+         break;
+      case SUPG:
+         mfem::out<<"SUPG formulation"<<std::endl;
+         break;
+      case GLS:
+         mfem::out<<"GLS formulation"<<std::endl;
+         break;
+      case VMS:
+         mfem::out<<"VMS formulation"<<std::endl;
+         break;
+      default:
+         mfem::out<<"GAL"<<"\t"<<"SUPG"<<"\t"<<"GLS"<<"\t"<<"VMS"<<std::endl;
+         mfem::out<<GALERKIN<<"\t"<<SUPG<<"\t"<<GLS<<"\t"<<VMS<<std::endl;
+         mfem_error("Wrong formulation");
+   }
+   return (StabType) stype;
+}
+
+int main(int argc, char *argv[])
+{
+   // Initialize MPI and HYPRE.
+   Mpi::Init(argc, argv);
+   int num_procs = Mpi::WorldSize();
+   int myid = Mpi::WorldRank();
+   Hypre::Init();
+
+   // Parse command-line options.
+   const char *mesh_file = "../../data/inline-quad.mesh";
+   const char *ref_file  = "";
+   int problem = 0;
+   int sstype = -2;
+   bool static_cond = false;
+   bool visualization = false;
+
+   real_t penalty = -1;
+   int order = 1;
+   int ref_levels = 0;
+
+   bool mono = true;
+
+   OptionsParser args(argc, argv);
+
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&ref_file, "-rf", "--ref-file",
+                  "File with refinement data");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order isoparametric space.");
+   args.AddOption(&ref_levels, "-r", "--refine",
+                  "Number of times to refine the mesh.");
+   args.AddOption(&kappa_param , "-k", "--kappa",
+                  "Sets the diffusion parameters, should be positive.");
+   args.AddOption(&problem, "-p", "--problem",
+                  "Select the problem to solve:\n\t"
+                  "  0 = convection skew-to-the mesh\n\t"
+                  "  1 = manufactured solution\n");
+   args.AddOption(&sstype, "-s", "--stab", " Stabilization type:\n\t"
+                  " -2 = Galerkin\n\t"
+                  " -1 = GLS\n\t"
+                  "  0 = SUPG\n\t"
+                  "  1 = VMS\n");
+   args.AddOption(&mono, "-mo", "--mono", "-co",
+                  "--comp",
+                  "Use a monolithic integrator or a composed one.");
+   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
+                  "--no-static-condensation", "Enable static condensation.");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.Parse();
+   if (!args.Good())
+   {
+      if (myid == 0) args.PrintUsage(cout);
+      return 1;
+   }
+   if (myid == 0) args.PrintOptions(cout);
+
+   // Read the mesh from the given mesh file. We can handle triangular,
+   // quadrilateral, tetrahedral, hexahedral, surface and volume meshes with
+   // the same code.
+   Mesh mesh(mesh_file, 1, 1);
+   int dim = mesh.Dimension();
+
+   // Refine the mesh to increase the resolution. In this example we do
+   // 'ref_levels' of uniform refinement and knot insertion of knots defined
+   // in a refinement file. We choose 'ref_levels' to be the largest number
+   // that gives a final mesh with no more than 50,000 elements.
+   {
+      // Mesh refinement as defined in refinement file
+      if (mesh.NURBSext && (strlen(ref_file) != 0))
+      {
+         mesh.RefineNURBSFromFile(ref_file);
+      }
+
+      for (int l = 0; l < ref_levels; l++)
+      {
+         mesh.UniformRefinement();
+      }
+      if (myid == 0) mesh.PrintInfo();
+   }
+   ParMesh pmesh(MPI_COMM_WORLD, mesh);
+   mesh.Clear();
+
+   // Define a finite element space on the mesh. Here we use continuous
+   // Lagrange finite elements of the specified order. If order < 1, we
+   // instead use an isoparametric/isogeometric space.
+   Array<FiniteElementCollection *> fecs(2);
+   fecs[0] = new H1_FECollection(order, dim);
+   fecs[1] = new H1_FECollection(order, dim);
+
+   Array<ParFiniteElementSpace *> spaces(2);
+   spaces[0] = new ParFiniteElementSpace(&pmesh, fecs[0], dim);//, Ordering::byVDIM);
+   spaces[1] = new ParFiniteElementSpace(&pmesh, fecs[1]);
+
+   Array<int> tdof(num_procs),udof(num_procs),pdof(num_procs);
+   tdof = 0;
+   tdof[myid] = spaces[0]->TrueVSize();
+   MPI_Reduce(tdof.GetData(), udof.GetData(), num_procs, MPI_INT, MPI_MAX, 0, MPI_COMM_WORLD);
+
+   tdof = 0;
+   tdof[myid] = spaces[1]->TrueVSize();
+   MPI_Reduce(tdof.GetData(), pdof.GetData(), num_procs, MPI_INT, MPI_MAX, 0, MPI_COMM_WORLD);
+
+   if (myid == 0)
+   {
+      mfem::out << "Number of finite element unknowns:\n";
+      mfem::out << "\tVelocity = "<<spaces[0]->GlobalTrueVSize() << endl;
+      mfem::out << "\tPressure = "<<spaces[1]->GlobalTrueVSize() << endl;
+      mfem::out << "Number of finite element unknowns per partition:\n";
+      mfem::out <<  "\tVelocity = ";udof.Print(mfem::out, num_procs);
+      mfem::out <<  "\tPressure = ";pdof.Print(mfem::out, num_procs);
+   }
+
+   // Mark all velocity boundary dofs as essential
+   Array<Array<int> *> ess_bdr(2);
+  // Array<int> ess_tdof_list;
+
+   Array<int> ess_bdr_u(spaces[0]->GetMesh()->bdr_attributes.Max());
+   Array<int> ess_bdr_p(spaces[1]->GetMesh()->bdr_attributes.Max());
+
+   ess_bdr_p = 0;
+   ess_bdr_u = 1;
+
+   ess_bdr[0] = &ess_bdr_u;
+   ess_bdr[1] = &ess_bdr_p;
+
+   // Define the solution vector xp as a finite element grid function
+   Array<int> bOffsets(3);
+   bOffsets[0] = 0;
+   bOffsets[1] = spaces[0]->TrueVSize();
+   bOffsets[2] = spaces[1]->TrueVSize();
+   bOffsets.PartialSum();
+
+   BlockVector xp(bOffsets);
+
+   ParGridFunction x_u(spaces[0]);
+   ParGridFunction x_p(spaces[1]);
+
+   VectorFunctionCoefficient sol(dim, sol_fun);
+   x_u.ProjectCoefficient(sol);
+   x_p = 0.0;
+
+   x_u.GetTrueDofs(xp.GetBlock(0));
+   x_p.GetTrueDofs(xp.GetBlock(1));
+
+   VisItDataCollection visit_dc("navsto", &pmesh);
+   visit_dc.RegisterField("u", &x_u);
+   visit_dc.RegisterField("p", &x_p);
+   visit_dc.SetCycle(0);
+   visit_dc.Save();
+
+   // Define the problem parameters
+   FunctionCoefficient kappa(kappa_fun);
+   VectorFunctionCoefficient force(dim, force_fun);
+
+   // Define the stabilisation parameters
+   VectorGridFunctionCoefficient adv(&x_u);
+   ElasticInverseEstimateCoefficient invEst(spaces[0]);
+   FFH92Tau tau(&adv, &kappa, &invEst, 4.0);
+   FF91Delta delta(&adv, &kappa, &invEst);
+
+   tau.print = delta.print = (myid == 0);
+
+   // Define the block nonlinear form
+   ParBlockNonlinearForm Hform(spaces);
+   Hform.AddDomainIntegrator(new StabInNavStoIntegrator(kappa, force, tau, delta));
+   Array<Vector *> rhs(2);
+   rhs = nullptr; // Set all entries in the array
+   Hform.SetEssentialBC(ess_bdr, rhs);
+
+   // Set up the preconditioner
+   JacobianPreconditioner jac_prec(bOffsets,
+                                   Array<Solver *>({new HypreSmoother(),
+                                                    new HypreSmoother()}));
+
+   // Set up the Jacobian solver
+   GeneralResidualMonitor j_monitor(MPI_COMM_WORLD,"\t\t\t\tFGMRES", 25);
+   FGMRESSolver j_gmres(MPI_COMM_WORLD);
+   j_gmres.iterative_mode = false;
+   j_gmres.SetRelTol(1e-2);
+   j_gmres.SetAbsTol(1e-12);
+   j_gmres.SetMaxIter(300);
+   j_gmres.SetPrintLevel(-1);
+   j_gmres.SetMonitor(j_monitor);
+   j_gmres.SetPreconditioner(jac_prec);
+
+   // Set up the newton solver
+   SystemResidualMonitor newton_monitor(MPI_COMM_WORLD,"Newton", 1, bOffsets, &visit_dc, &xp,
+                                        Array<ParGridFunction *>({&x_u, &x_p}));
+   NewtonSolver newton_solver(MPI_COMM_WORLD);
+   newton_solver.iterative_mode = true;
+   newton_solver.SetPrintLevel(-1);
+   newton_solver.SetMonitor(newton_monitor);
+   newton_solver.SetRelTol(1e-4);
+   newton_solver.SetAbsTol(1e-8);
+   newton_solver.SetMaxIter(25);
+   newton_solver.SetSolver(j_gmres);
+   newton_solver.SetOperator(Hform);
+
+   // Solve the Newton system
+   Vector zero;
+   newton_solver.Mult(zero, xp);
+
+   // Save data in the VisIt format
+   // Define the output
+   // Save data in the VisIt format
+   visit_dc.SetCycle(999999);
+   visit_dc.Save();
+
+   // Free the used memory.
+   for (int i = 0; i < fecs.Size(); ++i)
+   {
+      delete fecs[i];
+   }
+   for (int i = 0; i < spaces.Size(); ++i)
+   {
+      delete spaces[i];
+   }
+
+   return 0;
+}
+

miniapps/stabilized/skew.hpp

@@ -0,0 +1,46 @@
+namespace skew
+{
+
+//----------------------------------------------------------
+void adv(const Vector & x, Vector & a)
+{
+   a[1] = 1.0/(1.0 + att_param*att_param);
+   a[0] = sqrt(1.0 - a[1]*a[1]);
+}
+
+//----------------------------------------------------------
+real_t kappa(const Vector & x)
+{
+   return kappa_param;
+}
+
+//----------------------------------------------------------
+real_t force(const Vector & x)
+{
+   return 0.0;
+}
+
+//----------------------------------------------------------
+real_t sol(const Vector & x)
+{
+   if ((x[1] - x[0] - 0.2 < 0.0)
+      &(x[0] + x[1] -0.99 < 0.0))
+   {
+      return 1.0;
+   }
+   return 0.0;
+}
+
+//----------------------------------------------------------
+void grad(const Vector & x, Vector &grad)
+{
+   grad = 0.0;
+}
+
+//----------------------------------------------------------
+real_t laplace(const Vector & x)
+{
+   return 0.0;
+}
+
+}

miniapps/stabilized/stab_condif.cpp

@@ -0,0 +1,234 @@
+// Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
+// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+// LICENSE and NOTICE for details. LLNL-CODE-806117.
+//
+// This file is part of the MFEM library. For more information and source code
+// availability visit https://mfem.org.
+//
+// MFEM is free software; you can redistribute it and/or modify it under the
+// terms of the BSD-3 license. We welcome feedback and contributions, see file
+// CONTRIBUTING.md for details.
+
+#include "stab_condif.hpp"
+
+using namespace mfem;
+
+StabConDifIntegrator::StabConDifIntegrator(VectorCoefficient *a,
+                                           Coefficient *k,
+                                           Coefficient *f,
+                                           Tau *t, StabType s)
+ : adv(a), kappa(k), force(f), tau(t), stab(s), own_tau(false)
+{
+   if (tau == nullptr)
+   {
+      tau = new FFH92Tau(adv, kappa, 12.0);
+      own_tau = true;
+   }
+   else
+   {
+      tau->SetConvection(adv);
+      tau->SetDiffusion(kappa);
+   }
+}
+
+StabConDifIntegrator::~StabConDifIntegrator()
+{
+   if (own_tau) { delete tau; }
+}
+
+const IntegrationRule &StabConDifIntegrator::GetRule(
+   const FiniteElement &trial_fe,
+   const FiniteElement &test_fe,
+   ElementTransformation &Trans)
+{
+   int order = trial_fe.GetOrder() + test_fe.GetOrder();
+   return IntRules.Get(trial_fe.GetGeomType(), order);
+}
+
+void StabConDifIntegrator::AssembleElementMatrix(const FiniteElement &el,
+                                                 ElementTransformation &Trans,
+                                                 DenseMatrix &elmat )
+{
+   int nd = el.GetDof();
+   int dim = el.GetDim();
+   real_t w,k,t = 0;
+   Vector a(dim);
+
+   elmat.SetSize(nd);
+   shape.SetSize(nd);
+   dshape.SetSize(nd,dim);
+   adshape.SetSize(nd);
+   laplace.SetSize(nd);
+   trail.SetSize(nd);
+   test.SetSize(nd);
+
+   const IntegrationRule *ir = NonlinearFormIntegrator::IntRule ? NonlinearFormIntegrator::IntRule : &GetRule(el, el, Trans);
+
+   elmat = 0.0;
+   for (int i = 0; i < ir->GetNPoints(); i++)
+   {
+      const IntegrationPoint &ip = ir->IntPoint(i);
+      Trans.SetIntPoint (&ip);
+      w = Trans.Weight() * ip.weight;
+
+      // Calculate shapes
+      el.CalcPhysShape(Trans, shape);
+      el.CalcPhysDShape(Trans, dshape);
+
+      // Evaluate coefficients
+      k = kappa->Eval(Trans, ip);
+      adv->Eval(a, Trans, ip);
+
+      // Galerkin convection term
+      dshape.Mult(a, adshape);
+      AddMult_a_VWt(w, shape, adshape, elmat);
+
+      // Galerkin diffusion term
+      AddMult_a_AAt(w*k, dshape, elmat);
+
+      if (stab != GALERKIN)
+      {
+         // Calculate shapes
+         el.CalcPhysLaplacian(Trans, laplace);
+
+         // Evaluate coefficients
+         t = tau->Eval(Trans, ip);
+
+         // Stablization term
+         // - GLS:  stab = -1
+         // - SUPG: stab =  0
+         // - VMS:  stab = +1
+         add(adshape, stab*k, laplace, test);
+         add(adshape, -k, laplace, trail);
+         AddMult_a_VWt(w*t, test, trail, elmat);
+      }
+   }
+}
+
+void StabConDifIntegrator::AssembleRHSElementVect(const FiniteElement &el,
+                                      ElementTransformation &Trans,
+                                      Vector &elvect)
+{
+   int nd = el.GetDof();
+   int dim = el.GetDim();
+   real_t w,k,t,f;
+   Vector a(dim);
+
+   elvect.SetSize(nd);
+   shape.SetSize(nd);
+   dshape.SetSize(nd,dim);
+   adshape.SetSize(nd);
+   laplace.SetSize(nd);
+   test.SetSize(nd);
+
+   const IntegrationRule *ir = LinearFormIntegrator::IntRule ? LinearFormIntegrator::IntRule : &GetRule(el, el, Trans);
+
+   elvect = 0.0;
+   for (int i = 0; i < ir->GetNPoints(); i++)
+   {
+      const IntegrationPoint &ip = ir->IntPoint(i);
+      Trans.SetIntPoint (&ip);
+      w = Trans.Weight() * ip.weight;
+
+      // Calculate shapes
+      el.CalcPhysShape(Trans, shape);
+
+      // Evaluate coefficients
+      f = force->Eval(Trans, ip);
+
+      // Galerkin term
+      elvect.Add(w*f, shape);
+
+      if (stab != GALERKIN)
+      {
+         // Calculate shapes
+         el.CalcPhysDShape(Trans, dshape);
+         el.CalcPhysLaplacian(Trans, laplace);
+
+         // Evaluate coefficients
+         k = kappa->Eval(Trans, ip);
+         adv->Eval(a, Trans, ip);
+         t = tau->Eval(Trans, ip);
+
+         // Advective derivative
+         dshape.Mult(a, adshape);
+
+         // Stablization term
+         // - GLS:  stab = -1
+         // - SUPG: stab =  0
+         // - VMS:  stab = +1
+         add(adshape, stab*k, laplace, test);
+         elvect.Add(w*f*t, test);
+      }
+   }
+}
+
+
+StabConDifComposition::StabConDifComposition(VectorCoefficient *a,
+                                             Coefficient *k,
+                                             Coefficient *f,
+                                             Tau *t)
+  :  adv(a), kappa(k), force(f), tau(t), own_tau(false)
+{
+   if (tau == nullptr)
+   {
+      tau = new FFH92Tau(adv, kappa, 12.0);
+      own_tau = true;
+   }
+   else
+   {
+      tau->SetConvection(adv);
+      tau->SetDiffusion(kappa);
+   }
+
+   // SUPG coefficients
+   adv_tau = new ScalarVectorProductCoefficient(*tau, *adv);
+   adv_tau_force = new ScalarVectorProductCoefficient(*force, *adv_tau);
+   adv_tau_kappa = new ScalarVectorProductCoefficient(*kappa, *adv_tau);
+   adv_tau_adv = new OuterProductCoefficient(*adv_tau, *adv);
+
+   // GLS/VMS coefficients
+   kappa_tau = new ProductCoefficient(*kappa, *tau);
+   kappa_tau_kappa = new ProductCoefficient(*kappa_tau, *kappa);
+   kappa_tau_force = new ProductCoefficient(*kappa_tau, *force);
+}
+
+StabConDifComposition::~StabConDifComposition()
+{
+   if (own_tau) { delete tau; }
+   delete adv_tau, adv_tau_kappa, adv_tau_adv, adv_tau_force,
+          kappa_tau, kappa_tau_kappa,kappa_tau_force;
+}
+
+void StabConDifComposition::SetBilinearIntegrators(BilinearForm *a, StabType stype)
+{
+   a->AddDomainIntegrator(new ConservativeConvectionIntegrator(*adv));
+   a->AddDomainIntegrator(new DiffusionIntegrator(*kappa));
+   if (stype == GALERKIN) return;
+
+   // Add SUPG terms
+   a->AddDomainIntegrator(new DiffusionIntegrator(*adv_tau_adv));
+   a->AddDomainIntegrator(new GradLaplaceIntegrator(*adv_tau_kappa, 1.0));
+   if (stype == SUPG) return;
+
+   // Add VMS/GLS terms
+   real_t  s = (stype == GLS)? -1.0: 1.0;
+   a->AddDomainIntegrator(new LaplaceGradIntegrator(*adv_tau_kappa,-s));
+   a->AddDomainIntegrator(new LaplaceLaplaceIntegrator(*kappa_tau_kappa,-s));
+}
+
+void StabConDifComposition::SetLinearIntegrators(LinearForm *b, StabType stype)
+{
+   b->AddDomainIntegrator(new DomainLFIntegrator(*force));
+   if (stype == GALERKIN) return;
+
+   // Add SUPG terms
+   b->AddDomainIntegrator(new DomainLFGradIntegrator(*adv_tau_force));
+   if (stype == SUPG) return;
+
+   // Add VMS/GLS terms
+   real_t  s = (stype == GLS)? -1.0: 1.0;
+   b->AddDomainIntegrator(new DomainLFLaplaceIntegrator(*kappa_tau_force,-s));
+}
+
+

miniapps/stabilized/stab_condif.hpp

@@ -0,0 +1,130 @@
+// Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
+// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+// LICENSE and NOTICE for details. LLNL-CODE-806117.
+//
+// This file is part of the MFEM library. For more information and source code
+// availability visit https://mfem.org.
+//
+// MFEM is free software; you can redistribute it and/or modify it under the
+// terms of the BSD-3 license. We welcome feedback and contributions, see file
+// CONTRIBUTING.md for details.
+
+#ifndef MFEM_STAB_CONDIF_HPP
+#define MFEM_STAB_CONDIF_HPP
+
+#include "mfem.hpp"
+#include "stab_tau.hpp"
+
+namespace mfem
+{
+
+/** This Class defines a monolithic integrator for stabilized multi-dimensional
+    convection-diffusion.
+
+     $(a \cdot \nabla u, v) + (\kappa \nabla u, \nabla v) 
+    + \sum (a \cdot \nabla u - \kappa \Delta u, \tau (a \cdot \nabla v + s \kappa \Delta v))_e$
+
+     $(f, \nabla v) 
+    + \sum (f, \tau (a \cdot \nabla v + s \kappa \Delta v))_e$
+*/
+class StabConDifIntegrator : public BilinearFormIntegrator,
+                             public LinearFormIntegrator
+{
+protected:
+   /// The advection field
+   VectorCoefficient *adv;
+   /// The diffusion parameter and force fields
+   Coefficient *kappa, *force;
+
+   /// The stabilization parameter
+   Tau *tau;
+   bool own_tau;
+
+   StabType stab;
+
+private:
+   Vector laplace, shape, adshape, trail, test;
+   DenseMatrix dshape;
+
+public:
+   StabConDifIntegrator(VectorCoefficient *a,
+                        Coefficient *k,
+                        Coefficient *f,
+                        Tau *t = nullptr, StabType s = GALERKIN);
+
+   ~StabConDifIntegrator();
+
+   virtual void AssembleElementMatrix(const FiniteElement &el,
+                                      ElementTransformation &Tr,
+                                      DenseMatrix &elmat);
+
+   static const IntegrationRule &GetRule(const FiniteElement &trial_fe,
+                                         const FiniteElement &test_fe,
+                                         ElementTransformation &Trans);
+
+   virtual void AssembleRHSElementVect(const FiniteElement &el,
+                                       ElementTransformation &Tr,
+                                       Vector &elvect);
+
+   using LinearFormIntegrator::AssembleRHSElementVect;
+}; 
+
+/** This Class composes standard integrators to obtain a stabilized formulation for 
+    multi-dimensional convection-diffusion.
+
+     $(a \cdot \nabla u, v) + (\kappa \nabla u, \nabla v) 
+    + \sum (a \cdot \nabla u - \kappa \Delta u, \tau (a \cdot \nabla v + s \kappa \Delta v))_e$
+
+     $(f, \nabla v) 
+    + \sum (f, \tau (a \cdot \nabla v + s \kappa \Delta v))_e$
+*/
+class StabConDifComposition
+{
+
+private:
+   /// The advection field
+   VectorCoefficient *adv;
+   /// The diffusion parameter and force fields
+   Coefficient *kappa, *force;
+
+   /// The stabilization parameter
+   Tau *tau;
+   bool own_tau;
+
+   //// Helper coefficients for defining the weak forms
+   VectorCoefficient *adv_tau;
+   Coefficient *kappa_tau;
+
+   /// SUPG coefficients
+   VectorCoefficient *adv_tau_force;
+   VectorCoefficient *adv_tau_kappa;
+   MatrixCoefficient *adv_tau_adv;
+
+   /// GLS/VMS coefficients
+   Coefficient *kappa_tau_kappa;
+   Coefficient *kappa_tau_force;
+
+public:
+
+   /** Constructor
+       @a a: is the advection velocity field.
+       @a k: is the diffusion param field.
+       @a f: is the force field. */
+   StabConDifComposition(VectorCoefficient *a,
+                         Coefficient *k,
+                         Coefficient *f,
+                         Tau *t = nullptr);
+
+   /// Destructor
+   ~StabConDifComposition();
+
+   /// This method sets the integrators for the bilinearform
+   void SetBilinearIntegrators(BilinearForm *a, StabType s);
+
+   /// This method sets the integrators for the linearform
+   void SetLinearIntegrators(LinearForm *b, StabType s);
+};
+
+} // namespace mfem
+
+#endif

miniapps/stabilized/stab_navsto.cpp

@@ -0,0 +1,394 @@
+// Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
+// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+// LICENSE and NOTICE for details. LLNL-CODE-806117.
+//
+// This file is part of the MFEM library. For more information and source code
+// availability visit https://mfem.org.
+//
+// MFEM is free software; you can redistribute it and/or modify it under the
+// terms of the BSD-3 license. We welcome feedback and contributions, see file
+// CONTRIBUTING.md for details.
+
+#include "stab_navsto.hpp"
+
+using namespace mfem;
+
+StabInNavStoIntegrator::StabInNavStoIntegrator(Coefficient &mu_,
+                                               VectorCoefficient &force_,
+                                               Tau &t, Tau &d, StabType s)
+ : c_mu(&mu_), c_force(&force_), tau(&t), delta(&d), stab(s)
+{ }
+
+void StabInNavStoIntegrator::SetDim(int dim_)
+{
+  if (dim_ != dim)
+  {
+     dim = dim_;
+     u.SetSize(dim);
+     f.SetSize(dim);
+     res.SetSize(dim);
+     up.SetSize(dim);
+     grad_u.SetSize(dim);
+     hess_u.SetSize(dim, (dim*(dim+1))/2);
+     grad_p.SetSize(dim);
+     hmap.SetSize(dim,dim);
+
+     if (dim == 2)
+     {
+        hmap(0,0) = 0;
+        hmap(0,1) =  hmap(1,0) =  1;
+        hmap(1,1) = 2; 
+     }
+     else if (dim == 2)
+     {
+        hmap(0,0) = 0;
+        hmap(0,1) = hmap(1,0) = 1;
+        hmap(0,2) = hmap(2,0) = 2;
+        hmap(1,1) = 3;
+        hmap(1,2) = hmap(2,1) = 4;
+        hmap(2,2) = 5;
+     }
+     else
+     {
+        mfem_error("Only implemented for 2D and 3D");
+     }
+  }
+}
+
+real_t StabInNavStoIntegrator::GetElementEnergy(
+   const Array<const FiniteElement *>&el,
+   ElementTransformation &Tr,
+   const Array<const Vector *>&elfun)
+{
+   if (el.Size() != 2)
+   {
+      mfem_error("StabInNavStoIntegrator::GetElementEnergy"
+                 " has incorrect block finite element space size!");
+   }
+   SetDim(el[0]->GetDim());
+   int dof_u = el[0]->GetDof();
+
+   sh_u.SetSize(dof_u);
+   elf_u.UseExternalData(elfun[0]->GetData(), dof_u, dim);
+
+   int intorder = 2*el[0]->GetOrder();
+   const IntegrationRule &ir = IntRules.Get(el[0]->GetGeomType(), intorder);
+
+   real_t energy = 0.0;
+
+   for (int i = 0; i < ir.GetNPoints(); ++i)
+   {
+      const IntegrationPoint &ip = ir.IntPoint(i);
+      Tr.SetIntPoint(&ip);
+
+      real_t w = ip.weight * Tr.Weight();
+
+      el[0]->CalcPhysShape(Tr, sh_u);
+      elf_u.MultTranspose(sh_u, u);
+
+      energy += w*(u*u)/2;
+   }
+
+   return energy;
+}
+
+void StabInNavStoIntegrator::AssembleElementVector(
+   const Array<const FiniteElement *> &el,
+   ElementTransformation &Tr,
+   const Array<const Vector *> &elfun,
+   const Array<Vector *> &elvec)
+{
+   if (el.Size() != 2)
+   {
+      mfem_error("StabInNavStoIntegrator::AssembleElementVector"
+                 " has finite element space of incorrect block number");
+   }
+
+   int dof_u = el[0]->GetDof();
+   int dof_p = el[1]->GetDof();
+
+   SetDim(el[0]->GetDim());
+   int spaceDim = Tr.GetSpaceDim();
+   bool hess = (el[0]->GetDerivType() == (int) FiniteElement::HESS);
+   if (dim != spaceDim)
+   {
+      mfem_error("StabInNavStoIntegrator::AssembleElementVector"
+                 " is not defined on manifold meshes");
+   }
+   elvec[0]->SetSize(dof_u*dim);
+   elvec[1]->SetSize(dof_p);
+
+   *elvec[0] = 0.0;
+   *elvec[1] = 0.0;
+
+   elf_u.UseExternalData(elfun[0]->GetData(), dof_u, dim);
+   elv_u.UseExternalData(elvec[0]->GetData(), dof_u, dim);
+
+   sh_u.SetSize(dof_u);
+   shg_u.SetSize(dof_u, dim);
+   ushg_u.SetSize(dof_u);
+   shh_u.SetSize(dof_u, (dim*(dim+1))/2);
+   sh_p.SetSize(dof_p);
+   shg_p.SetSize(dof_p, dim);
+
+   int intorder = 2*el[0]->GetOrder();
+   const IntegrationRule &ir = IntRules.Get(el[0]->GetGeomType(), intorder);
+
+   for (int i = 0; i < ir.GetNPoints(); ++i)
+   {
+      const IntegrationPoint &ip = ir.IntPoint(i);
+      Tr.SetIntPoint(&ip);
+      real_t w = ip.weight * Tr.Weight();
+      real_t mu = c_mu->Eval(Tr, ip);
+      c_force->Eval(f, Tr, ip);
+
+      // Compute shape and interpolate
+      el[0]->CalcPhysShape(Tr, sh_u);
+      elf_u.MultTranspose(sh_u, u);
+
+      el[0]->CalcPhysDShape(Tr, shg_u);
+      shg_u.Mult(u, ushg_u);
+      MultAtB(elf_u, shg_u, grad_u);
+
+      if (hess)
+      {
+         el[0]->CalcPhysHessian(Tr,shh_u);
+         MultAtB(elf_u, shh_u, hess_u);
+      }
+      else
+      {
+         shh_u = 0.0;
+         hess_u = 0.0;
+      }
+
+      el[1]->CalcPhysShape(Tr, sh_p);
+      real_t p = sh_p*(*elfun[1]);
+
+      el[1]->CalcPhysDShape(Tr, shg_p);
+      shg_p.MultTranspose(*elfun[1], grad_p);
+
+      // Compute strong residual
+      grad_u.Mult(u,res);   // Add convection
+      res += grad_p;        // Add pressure
+      res -= f;             // Subtract force
+      for (int i = 0; i < dim; ++i)
+      {
+         for (int j = 0; j < dim; ++j)
+         {
+            res[j] -= mu*(hess_u(j,hmap(i,i)) + 
+                          hess_u(i,hmap(j,i))); // Add diffusion
+         }
+      }
+
+      // Compute stability params
+      real_t t = tau->Eval(Tr, ip);
+      real_t d = delta->Eval(Tr, ip);
+
+      // Compute momentum weak residual
+      flux.Diag(-p + d*grad_u.Trace(),dim);  // Add pressure & LSIC to flux
+      grad_u.Symmetrize();                   // Grad to strain
+      flux.Add(2*mu,grad_u);                 // Add stress to flux
+      AddMult_a_VVt(-1.0, u, flux);          // Add convection to flux
+      AddMult_a_VWt(t, res, u, flux);        // Add SUPG to flux --> check order u and res
+      AddMult_a_ABt(w, shg_u, flux, elv_u);  // Add flux term to rhs
+      AddMult_a_VWt(-w, sh_u, f,    elv_u);  // Add force term to rhs
+
+      // Compute momentum weak residual
+      elvec[1]->Add(w*grad_u.Trace(), sh_p); // Add Galerkin term
+      shg_p.Mult(res, sh_p);                 // PSPG help term
+      elvec[1]->Add(w*t, sh_p);              // Add PSPG term  - sign looks worng?
+   }
+}
+
+void StabInNavStoIntegrator::AssembleElementGrad(
+   const Array<const FiniteElement*> &el,
+   ElementTransformation &Tr,
+   const Array<const Vector *> &elfun,
+   const Array2D<DenseMatrix *> &elmats)
+{
+   int dof_u = el[0]->GetDof();
+   int dof_p = el[1]->GetDof();
+
+   SetDim(el[0]->GetDim());
+   bool hess = (el[0]->GetDerivType() == (int) FiniteElement::HESS);
+
+   elf_u.UseExternalData(elfun[0]->GetData(), dof_u, dim);
+
+   elmats(0,0)->SetSize(dof_u*dim, dof_u*dim);
+   elmats(0,1)->SetSize(dof_u*dim, dof_p);
+   elmats(1,0)->SetSize(dof_p, dof_u*dim);
+   elmats(1,1)->SetSize(dof_p, dof_p);
+
+   *elmats(0,0) = 0.0;
+   *elmats(0,1) = 0.0;
+   *elmats(1,0) = 0.0;
+   *elmats(1,1) = 0.0;
+
+   sh_u.SetSize(dof_u);
+   shg_u.SetSize(dof_u, dim);
+   ushg_u.SetSize(dof_u);
+   sh_p.SetSize(dof_p);
+   shg_p.SetSize(dof_p, dim);
+
+   int intorder = 2*el[0]->GetOrder();
+   const IntegrationRule &ir = IntRules.Get(el[0]->GetGeomType(), intorder);
+
+   for (int i = 0; i < ir.GetNPoints(); ++i)
+   {
+      const IntegrationPoint &ip = ir.IntPoint(i);
+      Tr.SetIntPoint(&ip);
+      real_t w = ip.weight * Tr.Weight();
+      real_t mu = c_mu->Eval(Tr, ip);
+      real_t t = tau->Eval(Tr, ip);
+      real_t d = delta->Eval(Tr, ip);
+
+      el[0]->CalcPhysShape(Tr, sh_u);
+      elf_u.MultTranspose(sh_u, u);
+
+      el[0]->CalcPhysDShape(Tr, shg_u);
+      MultAtB(elf_u, shg_u, grad_u);
+
+      shg_u.Mult(u, ushg_u);
+
+      el[1]->CalcPhysShape(Tr, sh_p);
+      real_t p = sh_p*(*elfun[1]);
+
+      el[1]->CalcPhysDShape(Tr, shg_p);
+      shg_p.MultTranspose(*elfun[1], grad_p);
+
+      // u,u block
+      for (int i_u = 0; i_u < dof_u; ++i_u)
+      {
+         for (int j_u = 0; j_u < dof_u; ++j_u)
+         {
+            // Diffusion
+            real_t mat = 0.0;
+            for (int dim_u = 0; dim_u < dim; ++dim_u)
+            {
+              mat += shg_u(i_u,dim_u)*shg_u(j_u,dim_u);
+            }
+            mat *= mu;
+
+            // Convection
+            mat -= ushg_u(i_u)*sh_u(j_u);      // Galerkin
+            mat += t*ushg_u(i_u)*ushg_u(j_u);  // SUPG
+
+            mat *= w;
+            for (int dim_u = 0; dim_u < dim; ++dim_u)
+            {
+               (*elmats(0,0))(i_u + dim_u*dof_u, j_u + dim_u*dof_u) += mat;
+            }
+
+            for (int i_dim = 0; i_dim < dim; ++i_dim)
+            {
+               for (int j_dim = 0; j_dim < dim; ++j_dim)
+               {
+                  (*elmats(0,0))(i_u + i_dim*dof_u, j_u + j_dim*dof_u) +=
+                     (mu + d)*shg_u(i_u,j_dim)*shg_u(j_u,i_dim)*w;
+               }
+            }
+         }
+      }
+
+      // u,p and p,u blocks
+      for (int i_p = 0; i_p < dof_p; ++i_p)
+      {
+         for (int j_u = 0; j_u < dof_u; ++j_u)
+         {
+            for (int dim_u = 0; dim_u < dim; ++dim_u)
+            {
+               (*elmats(0,1))(j_u + dof_u * dim_u, i_p) += (shg_p(i_p, dim_u)*t*ushg_u(j_u)
+                                                           -shg_u(j_u,dim_u)*sh_p(i_p))*w;
+               (*elmats(1,0))(i_p, j_u + dof_u * dim_u) +=  shg_u(j_u,dim_u)*sh_p(i_p)*w;
+            }
+         }
+      }
+
+      // p,p block
+      AddMult_a_AAt(w*t, shg_p, *elmats(1,1));
+   }
+}
+
+
+void GeneralResidualMonitor::MonitorResidual(int it, real_t norm,
+                                             const Vector &r, bool final)
+{
+   if (it == 0)
+   {
+      norm0 = norm;
+   }
+
+   if ((print_level > 0 &&  it%print_level == 0) || final)
+   {
+      mfem::out << prefix << " iteration " << std::setw(2) << it
+                << " : ||r|| = " << norm
+                << ",  ||r||/||r_0|| = " << 100*norm/norm0<<" % \n";
+   }
+}
+
+void SystemResidualMonitor::MonitorResidual(int it, real_t norm,
+                                            const Vector &r, bool final)
+{
+   if (dc && (it > 0))
+   {
+      if (rank > 1)
+      {
+         for (int i = 0; i < nvar; ++i)
+         {
+            pgf[i]->Distribute(xp->GetBlock(i));
+         }
+      }
+      dc->SetCycle(it);
+      dc->Save();
+   }
+
+   Vector vnorm(nvar);
+
+   for (int i = 0; i < nvar; ++i)
+   {
+       Vector r_i(r.GetData() + bOffsets[i], bOffsets[i+1] - bOffsets[i]);
+       if ( rank == 1 )
+       {
+          vnorm[i] = r_i.Norml2();
+       }
+       else
+       {
+          vnorm[i] = sqrt(InnerProduct(MPI_COMM_WORLD, r_i, r_i));
+       }
+       if (it == 0) norm0[i] = vnorm[i];
+   }
+
+   bool print = (print_level > 0 &&  it%print_level == 0) || final;
+   if (print)
+   {
+      mfem::out << prefix << " iteration " << std::setw(3) << it <<"\n"
+                << " ||r||  \t"<< "||r||/||r_0||  \n";
+      for (int i = 0; i < nvar; ++i)
+      {
+         mfem::out <<vnorm[i]<<"\t"<< 100*vnorm[i]/norm0[i]<<" % \n";
+      }
+   }
+}
+
+
+void JacobianPreconditioner::SetOperator(const Operator &op)
+{
+   BlockOperator *jacobian = (BlockOperator *) &op;
+
+   for (int i = 0; i < prec.Size(); ++i)
+   {
+      prec[i]->SetOperator(jacobian->GetBlock(i,i));
+      SetDiagonalBlock(i, prec[i]);
+   }
+
+   SetBlock(1,0, const_cast<Operator*>(&jacobian->GetBlock(1,0)));
+}
+
+JacobianPreconditioner::~JacobianPreconditioner()
+{
+   for (int i = 0; i < prec.Size(); ++i)
+   {
+      delete prec[i];
+   }
+}
+

miniapps/stabilized/stab_navsto.hpp

@@ -0,0 +1,242 @@
+// Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
+// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+// LICENSE and NOTICE for details. LLNL-CODE-806117.
+//
+// This file is part of the MFEM library. For more information and source code
+// availability visit https://mfem.org.
+//
+// MFEM is free software; you can redistribute it and/or modify it under the
+// terms of the BSD-3 license. We welcome feedback and contributions, see file
+// CONTRIBUTING.md for details.
+
+#ifndef MFEM_STAB_NAVSTO_HPP
+#define MFEM_STAB_NAVSTO_HPP
+
+#include "mfem.hpp"
+#include "stab_tau.hpp"
+
+namespace mfem
+{
+
+/** Stabilized incompressible Navier-Stokes integrator
+    Start with Galerkin for stokes - done
+    Add convection - done
+    Modify diffusion - done
+    Add difussion to residual - done
+    CHECK NUMBERING HESSIAN -> NURBS = WRONG?? 2D = ok --> 3D???   --> DONE NEEDS CHECKING???
+    Inverse estimate check order -> done
+    Add force                    -> done
+    Parallel -->  done
+
+
+    Add supg   -  rhs done, jac conv + press --> ignore diffusion for now
+    Add pspg   -  rhs done, jac conv + press --> ignore diffusion for now
+    Add lsic    -  rhs done, jac conv + press --> ignore diffusion for now
+
+    Add correct inverse estimate -> done?? number does not coincide with H&C
+
+    Add VMS/GLS
+    Add selection option of different stab modes
+
+    Add Hessian check to inverse estimate
+
+
+    Check
+    - Hessian numbering in 3D
+    - Power method --> Laplack / null-space
+    - Elastic Inverse estimate
+
+    Leopoldo P. Franca, Sérgio L. Frey
+    Stabilized finite element methods:
+    II. The incompressible Navier-Stokes equations.
+    Computer Methods in Applied Mechanics and Engineering, 99(2-3), 209-233.
+
+    https://doi.org/10.1016/0045-7825(92)90041-H
+    https://www.sciencedirect.com/science/article/pii/004578259290041H
+
+*/
+class StabInNavStoIntegrator : public BlockNonlinearFormIntegrator
+{
+private:
+   Coefficient *c_mu;
+   VectorCoefficient *c_force;
+   Vector u, f, grad_p;
+   DenseMatrix flux;
+
+   DenseMatrix elf_u, elv_u;
+//   Vector elf_u, elv_u;//
+   DenseMatrix elf_p, elv_p;
+   Vector sh_u, ushg_u, sh_p;
+   DenseMatrix shg_u, shh_u, shg_p, grad_u, hess_u;
+   Array2D<int> hmap;
+
+   /// The stabilization parameters
+   StabType stab;
+   Tau *tau = nullptr;
+   Tau *delta = nullptr;
+   Vector res, up;
+
+   /// The advection field
+   VectorCoefficient *adv = nullptr; // tbd???
+
+   int dim = -1;
+   void SetDim(int dim);
+
+public:
+   StabInNavStoIntegrator(Coefficient &mu_,
+                          VectorCoefficient &force_,
+                          Tau &t, Tau &d,
+                          StabType s = GALERKIN);
+
+   virtual real_t GetElementEnergy(const Array<const FiniteElement *>&el,
+                                   ElementTransformation &Tr,
+                                   const Array<const Vector *> &elfun);
+
+   /// Perform the local action of the NonlinearFormIntegrator
+   virtual void AssembleElementVector(const Array<const FiniteElement *> &el,
+                                      ElementTransformation &Tr,
+                                      const Array<const Vector *> &elfun,
+                                      const Array<Vector *> &elvec);
+
+   /// Assemble the local gradient matrix
+   virtual void AssembleElementGrad(const Array<const FiniteElement*> &el,
+                                    ElementTransformation &Tr,
+                                    const Array<const Vector *> &elfun,
+                                    const Array2D<DenseMatrix *> &elmats);
+};
+
+class GeneralResidualMonitor : public IterativeSolverMonitor
+{
+public:
+   GeneralResidualMonitor(const std::string& prefix_, int print_lvl)
+      : prefix(prefix_)
+   {
+      print_level = print_lvl;
+      rank = 1;
+   }
+
+   GeneralResidualMonitor(MPI_Comm comm,
+                          const std::string& prefix_, int print_lvl)
+      : prefix(prefix_)
+   {
+#ifndef MFEM_USE_MPI
+      print_level = print_lvl;
+#else
+      MPI_Comm_rank(comm, &rank);
+      if (rank == 0)
+      {
+         print_level = print_lvl;
+      }
+      else
+      {
+         print_level = -1;
+      }
+#endif
+   }
+
+   virtual void MonitorResidual(int it, real_t norm, const Vector &r, bool final);
+
+private:
+   const std::string prefix;
+   int rank, print_level;
+   mutable real_t norm0;
+};
+
+class SystemResidualMonitor : public IterativeSolverMonitor
+{
+public:
+   SystemResidualMonitor(const std::string& prefix_,
+                         int print_lvl,
+                         Array<int> &offsets,
+                         DataCollection *dc_ = nullptr)
+      : prefix(prefix_), bOffsets(offsets), dc(dc_)
+   {
+      print_level = print_lvl;
+      nvar = bOffsets.Size()-1;
+      norm0.SetSize(nvar);
+      rank = 1;
+   }
+
+   SystemResidualMonitor(MPI_Comm comm,
+                         const std::string& prefix_,
+                         int print_lvl,
+                         Array<int> &offsets)
+      : prefix(prefix_), bOffsets(offsets), dc(nullptr), xp(nullptr)
+   {
+#ifndef MFEM_USE_MPI
+      print_level = print_lvl;
+      rank = 1;
+#else
+      MPI_Comm_rank(comm, &rank);
+      if (rank == 0)
+      {
+         print_level = print_lvl;
+      }
+      else
+      {
+         print_level = -1;
+      }
+#endif
+      nvar = bOffsets.Size()-1;
+      norm0.SetSize(nvar);
+   }
+   SystemResidualMonitor(MPI_Comm comm,
+                         const std::string& prefix_,
+                         int print_lvl,
+                         Array<int> &offsets,
+                         DataCollection *dc_,
+                         BlockVector *x,
+                         Array<ParGridFunction *> pgf_)
+      : prefix(prefix_), bOffsets(offsets), dc(dc_), xp(x), pgf(pgf_)
+   {
+#ifndef MFEM_USE_MPI
+      print_level = print_lvl;
+      rank = 1;
+#else
+      MPI_Comm_rank(comm, &rank);
+      if (rank == 0)
+      {
+         print_level = print_lvl;
+      }
+      else
+      {
+         print_level = -1;
+      }
+#endif
+      nvar = bOffsets.Size()-1;
+      norm0.SetSize(nvar);
+   }
+
+
+   virtual void MonitorResidual(int it, real_t norm, const Vector &r, bool final);
+
+private:
+   const std::string prefix;
+   int print_level, nvar, rank;
+   mutable Vector norm0;
+  // Offsets for extracting block vector segments
+   Array<int> &bOffsets;
+   DataCollection *dc;
+   BlockVector *xp;
+   Array<ParGridFunction *> pgf;
+};
+
+// Custom block preconditioner for the Jacobian
+class JacobianPreconditioner : public BlockLowerTriangularPreconditioner //BlockDiagonalPreconditioner
+{
+protected:
+   Array<Solver *> prec;
+public:
+   JacobianPreconditioner(Array<int> &offsets, Array<Solver *> p)
+     : BlockLowerTriangularPreconditioner (offsets), prec(p)
+   { MFEM_VERIFY(offsets.Size()-1 == p.Size(), ""); };
+
+   virtual void SetOperator(const Operator &op);
+
+   virtual ~JacobianPreconditioner();
+};
+
+
+} // namespace mfem
+
+#endif

miniapps/stabilized/stab_tau.cpp

@@ -0,0 +1,124 @@
+// Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
+// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+// LICENSE and NOTICE for details. LLNL-CODE-806117.
+//
+// This file is part of the MFEM library. For more information and source code
+// availability visit https://mfem.org.
+//
+// MFEM is free software; you can redistribute it and/or modify it under the
+// terms of the BSD-3 license. We welcome feedback and contributions, see file
+// CONTRIBUTING.md for details.
+
+#include "stab_tau.hpp"
+
+using namespace mfem;
+
+real_t FFH92Tau::GetElementSize(ElementTransformation &T)
+{
+   const DenseMatrix &dxdxi = T.Jacobian();
+   row.SetSize(dim);
+   h.SetSize(dim);
+
+   for (int i = 0; i < dim; i++)
+   {
+      dxdxi.GetRow(i, row);
+      h[i] = row.Norml2();
+   }
+   switch (dim)
+   {
+      case 1:
+         return h[0];
+      case 2:
+         return h[0]*h[1]*sqrt(2.0/(h[0]*h[0] + h[1]*h[1]));
+      case 3:
+         return h[0]*h[1]*h[2]*(3.0/(h[0]*h[0] + h[1]*h[1] + h[2]*h[2]));
+   }
+   mfem_error("Wrong dim!");
+   return -1.0;
+}
+
+real_t FFH92Tau::GetInverseEstimate(ElementTransformation &T,
+                                    const IntegrationPoint &ip, real_t scale)
+{
+   if (Ci>0.0)
+   {
+      return 1.0/Ci;
+   }
+   else
+   {
+      return 1.0/(invEst_cf->Eval(T,ip)*scale);
+   }
+}
+
+real_t FFH92Tau::Eval(ElementTransformation &T,
+                      const IntegrationPoint &ip)
+{
+   real_t k = kappa->Eval(T, ip);
+   adv->Eval(a, T, ip);
+   real_t hk = GetElementSize(T);
+   real_t ci = GetInverseEstimate(T, ip, hk*hk);
+   real_t mk = std::min(1.0/3.0, 2*ci);
+   real_t ap = a.Normlp(p);
+   // Prevent division by zero
+   ap = std::max(ap,std::numeric_limits<real_t>::min());
+   real_t pe = mk*ap*hk/(k_fac*k); // k_fac = 2 for CD and k_fac = 4 for NS
+   real_t xi = std::min(pe,1.0);
+   real_t tau = hk*xi/(2*ap);
+
+   if (print)
+   {
+      std::cout<<"\n==========================\n";
+      std::cout<<"  kappa = "<<k  <<std::endl;
+      std::cout<<"  adv   = "; a.Print(std::cout);
+      std::cout<<"  h     = "<<hk <<std::endl;
+      std::cout<<"  Ci    = "<<ci<<" "
+               <<( (Ci<0) ? "(Computed)" :"(Specified)")<<std::endl;
+      std::cout<<"  1/Ci  = "<<1.0/ci<<std::endl;
+      std::cout<<"  mk    = "<<mk <<std::endl;
+      std::cout<<"  |a|_p = "<<ap <<std::endl;
+      std::cout<<"  Pe    = "<<pe <<std::endl;
+      std::cout<<"  xi    = "<<xi <<std::endl;
+      std::cout<<"  tau   = "<<tau<<std::endl;
+      std::cout<<"  tau   = "<<mk*hk*hk/(8*k)<<std::endl;
+      std::cout<<"==========================\n\n";
+      print = false;
+   }
+
+   return tau;
+}
+
+real_t FF91Delta::Eval(ElementTransformation &T,
+                      const IntegrationPoint &ip)
+{
+   real_t k = kappa->Eval(T, ip);
+   adv->Eval(a, T, ip);
+   real_t hk = GetElementSize(T);
+   real_t ci = GetInverseEstimate(T, ip, hk*hk);
+   real_t mk = std::min(1.0/3.0, 2*ci);
+   real_t ap = a.Normlp(p); // Preventing division by zero not necessary
+   real_t pe = mk*ap*hk/(k_fac*k);
+   real_t xi = std::min(pe,1.0);
+   real_t delta = lambda*ap*hk*xi;
+
+   if (print)
+   {
+      std::cout<<"\n==========================\n";
+      std::cout<<"  kappa  = "<<k  <<std::endl;
+      std::cout<<"  adv    = "; a.Print(std::cout);
+      std::cout<<"  h      = "<<hk <<std::endl;
+      std::cout<<"  Ci     = "<<ci<<" "
+               <<( (Ci<0) ? "(Computed)" :"(Specified)")<<std::endl;
+      std::cout<<"  1/Ci   = "<<1.0/ci<<std::endl;
+      std::cout<<"  mk     = "<<mk <<std::endl;
+      std::cout<<"  |a|_p  = "<<ap <<std::endl;
+      std::cout<<"  Pe     = "<<pe <<std::endl;
+      std::cout<<"  xi     = "<<xi <<std::endl;
+      std::cout<<"  lambda = "<<lambda <<std::endl;
+      std::cout<<"  delta  = "<<delta<<std::endl;
+      std::cout<<"==========================\n\n";
+      print = false;
+   }
+
+   return delta;
+}
+