update removeinternalboundariesnotadjacentto

added method to remove internal boundaries except for those adjacent to specified regions
added new method to remove mesh internal boundaries except for those in a array of boundary attributes to keep
2020-11-21 17:29:02 -05:00 · 2020-11-19 14:22:00 -05:00 · 2020-11-19 13:37:58 -05:00
111 changed files with 1349 additions and 12002 deletions
@@ -215,7 +215,6 @@ miniapps/meshing/polar-nc.mesh

 miniapps/navier/navier_mms
 miniapps/navier/navier_kovasznay
-miniapps/navier/navier_kovasznay_vs
 miniapps/navier/navier_tgv
 miniapps/navier/navier_shear
 miniapps/navier/navier_3dfoc
@@ -11,12 +11,6 @@
 Version 4.2.1 (development)
 ===========================

- Added high-order matrix-free auxiliary Maxwell solver for H(curl) problems,
-  as described in Barker and Kolev 2020 (https://doi.org/10.1002/nla.2348).
-
- Added matrix-free GPU-enabled implementations of GradientInterpolator and
-  IdentityInterpolator.
-
 - Added interface to MUMPS direct solver. Its usage is demonstrated in ex25p.
  See http://mumps.enseeiht.fr/ for more details. Supported versions >= 5.1.1.

@@ -26,41 +20,11 @@ Version 4.2.1 (development)
 - Added a "scaled Jacobian" visualization option in the Mesh Explorer miniapp to
  help identify elements with poor mesh quality.

- Added support for the "BR2" discontinuous Galerkin discretization for
-  diffusion via DGDiffusionBR2Integrator (see Example 14/14p).
-
- Generalized the Multigrid class to support non-geometric multigrid. The
-  previous functionality, based on FiniteElementSpaceHierarchy, is now available
-  in the derived class GeometricMultigrid.
-
 - Upgraded the Catch unit test framework from version 2.13.0 to version 2.13.2.

 - Implemented a filter method for the Navier miniapp to stabilize highly
  turbulent flows in direct numerical simulation.

- Added partial assembly and device support to Example 25/25p, with diagonal
-  preconditioning.
-
- Implemented a variable step-size IMEX (VSSIMEX) method for the Navier miniapp.
-
- Added new mesh quality metrics and improved the untangling capabilities of the
-  TMOP-based mesh optimization algorithms.
-
- Changed the interface for the error estimator.
-
- Implemented the parallel Kelly error indicator for scalar-valued problems.
-
- Added new classes DenseSymmetricMatrix and SymmetricMatrixCoefficient for
-  efficient evaluation of symmetric matrix coefficients. This replaces the now
-  deprecated EvalSymmetric in MatrixCoefficient. Added DiagonalMatrixCoefficient
-  for clarity, which is a typedef of VectorCoefficient.
-
- Implemented an adaptive linear solver tolerance option for NewtonSolver based
-  on the algorithm of Eisenstat and Walker.
-
- Extending support for L2 basis functions using MapTypes VALUE and INTEGRAL in
-  linear interpolators and GridFunction "GetValue" methods.
-

 Version 4.2, released on October 30, 2020
 =========================================
@@ -9,8 +9,7 @@
 # terms of the BSD-3 license. We welcome feedback and contributions, see file
 # CONTRIBUTING.md for details.

-# The variable CMAKE_CXX_STANDARD and related were introduced in CMake v3.1
-cmake_minimum_required(VERSION 3.1)
+cmake_minimum_required(VERSION 2.8.11)
 set(USER_CONFIG "${CMAKE_CURRENT_SOURCE_DIR}/config/user.cmake" CACHE PATH
  "Path to optional user configuration file.")

@@ -710,10 +710,10 @@ The specific libraries and their options are:
  Versions: libCEED >= 0.7.

 - RAJA (optional), used when MFEM_USE_RAJA = YES.
-  Beginning with MFEM v4.3, only RAJA v0.13.0+ is supported.
+  Beginning with MFEM v4.1, only RAJA v0.10.0+ is supported.
  URL: https://github.com/LLNL/RAJA
  Options: RAJA_DIR, RAJA_OPT, RAJA_LIB.
-  Versions: RAJA >= 0.13.0.
+  Versions: RAJA >= 0.10.0.

 - Umpire, used when MFEM_USE_UMPIRE = YES.
  URL: https://github.com/LLNL/Umpire
@@ -25,16 +25,16 @@ mfem_find_package(METIS METIS METIS_DIR "include;Lib" "metis.h"

 int main()
 {
-    idx_t n = 10;
-    idx_t nparts = 5;
-    idx_t edgecut;
-    idx_t* partitioning = new idx_t[10];
-    idx_t* I = partitioning,
+    int n = 10;
+    int nparts = 5;
+    int edgecut;
+    int* partitioning = new int[10];
+    int* I = partitioning,
       * J = partitioning;

-    idx_t ncon = 1;
+    int ncon = 1;
    int err;
-    idx_t options[40];
+    int options[40];

    METIS_SetDefaultOptions(options);
    options[10] = 1; // set METIS_OPTION_CONTIG
@@ -754,13 +754,7 @@ function(mfem_export_mk_files)
  set(MFEM_CXX ${CMAKE_CXX_COMPILER})
  set(MFEM_HOST_CXX ${MFEM_CXX})
  set(MFEM_CPPFLAGS "")
-  get_target_property(cxx_std mfem CXX_STANDARD)
-  # For now, we ignore the setting of the CXX_EXTENSIONS property. If this
-  # property is set, then we need to use a variable like:
-  #    CMAKE_CXX11_EXTENSION_COMPILE_OPTION
-  set(cxx_std_flag ${CMAKE_CXX${cxx_std}_STANDARD_COMPILE_OPTION})
-  string(STRIP
-         "${cxx_std_flag} ${CMAKE_CXX_FLAGS_${BUILD_TYPE}} ${CMAKE_CXX_FLAGS}"
+  string(STRIP "${CMAKE_CXX_FLAGS_${BUILD_TYPE}} ${CMAKE_CXX_FLAGS}"
         MFEM_CXXFLAGS)
  set(MFEM_TPLFLAGS "")
  foreach(dir ${MFEM_TPL_INCLUDE_DIRS})
@@ -161,7 +161,7 @@ endif
 ZLIB_DIR =
 ZLIB_OPT = $(if $(ZLIB_DIR),-I$(ZLIB_DIR)/include)
 ZLIB_LIB = $(if $(ZLIB_DIR),$(ZLIB_RPATH) -L$(ZLIB_DIR)/lib ,)-lz
-ZLIB_RPATH = $(XLINKER)-rpath,$(ZLIB_DIR)/lib
+ZLIB_RPATH = -Wl,-rpath,$(ZLIB_DIR)/lib

 LIBUNWIND_OPT = -g
 LIBUNWIND_LIB = $(if $(NOTMAC),-lunwind -ldl,)
@@ -231,21 +231,19 @@ MESQUITE_LIB = -L$(MESQUITE_DIR)/lib -lmesquite
 LIB_RT = $(if $(NOTMAC),-lrt,)
 SUITESPARSE_DIR = @MFEM_DIR@/../SuiteSparse
 SUITESPARSE_OPT = -I$(SUITESPARSE_DIR)/include
-SUITESPARSE_LIB = $(XLINKER)-rpath,$(SUITESPARSE_DIR)/lib\
- -L$(SUITESPARSE_DIR)/lib -lklu -lbtf -lumfpack -lcholmod -lcolamd -lamd -lcamd\
- -lccolamd -lsuitesparseconfig $(LIB_RT) $(METIS_LIB) $(LAPACK_LIB)
+SUITESPARSE_LIB = -Wl,-rpath,$(SUITESPARSE_DIR)/lib -L$(SUITESPARSE_DIR)/lib\
+ -lklu -lbtf -lumfpack -lcholmod -lcolamd -lamd -lcamd -lccolamd\
+ -lsuitesparseconfig $(LIB_RT) $(METIS_LIB) $(LAPACK_LIB)

 # SuperLU library configuration
 ifeq ($(MFEM_USE_SUPERLU5),YES)
   SUPERLU_DIR = @MFEM_DIR@/../SuperLU_DIST_5.1.0
   SUPERLU_OPT = -I$(SUPERLU_DIR)/include
-   SUPERLU_LIB = $(XLINKER)-rpath,$(SUPERLU_DIR)/lib -L$(SUPERLU_DIR)/lib\
-      -lsuperlu_dist_5.1.0
+   SUPERLU_LIB = -Wl,-rpath,$(SUPERLU_DIR)/lib -L$(SUPERLU_DIR)/lib -lsuperlu_dist_5.1.0
 else
   SUPERLU_DIR = @MFEM_DIR@/../SuperLU_DIST_6.3.1
   SUPERLU_OPT = -I$(SUPERLU_DIR)/include
-   SUPERLU_LIB = $(XLINKER)-rpath,$(SUPERLU_DIR)/lib64 -L$(SUPERLU_DIR)/lib64\
-      -lsuperlu_dist -lblas
+   SUPERLU_LIB = -Wl,-rpath,$(SUPERLU_DIR)/lib64 -L$(SUPERLU_DIR)/lib64 -lsuperlu_dist -lblas
 endif

 # SCOTCH library configuration (required by STRUMPACK <= v2.1.0, optional in
@@ -271,7 +269,7 @@ MPI_FORTRAN_LIB = -lmpifort
 # MUMPS library configuration
 MUMPS_DIR = @MFEM_DIR@/../MUMPS_5.2.0
 MUMPS_OPT = -I$(MUMPS_DIR)/include
-MUMPS_LIB = $(XLINKER)-rpath,$(MUMPS_DIR)/lib -L$(MUMPS_DIR)/lib -ldmumps\
+MUMPS_LIB = -Wl,-rpath,$(MUMPS_DIR)/lib -L$(MUMPS_DIR)/lib -ldmumps\
 -lmumps_common -lpord $(SCALAPACK_LIB) $(LAPACK_LIB) $(MPI_FORTRAN_LIB)

 # STRUMPACK library configuration
@@ -301,8 +299,8 @@ GNUTLS_LIB = -lgnutls
 NETCDF_DIR = $(HOME)/local
 HDF5_DIR   = $(HOME)/local
 NETCDF_OPT = -I$(NETCDF_DIR)/include -I$(HDF5_DIR)/include $(ZLIB_OPT)
-NETCDF_LIB = $(XLINKER)-rpath,$(NETCDF_DIR)/lib -L$(NETCDF_DIR)/lib\
- $(XLINKER)-rpath,$(HDF5_DIR)/lib -L$(HDF5_DIR)/lib\
+NETCDF_LIB = -Wl,-rpath,$(NETCDF_DIR)/lib -L$(NETCDF_DIR)/lib\
+ -Wl,-rpath,$(HDF5_DIR)/lib -L$(HDF5_DIR)/lib\
 -lnetcdf -lhdf5_hl -lhdf5 $(ZLIB_LIB)

 # PETSc library configuration (version greater or equal to 3.8 or the dev branch)
@@ -314,10 +312,9 @@ PETSC_INC_VAR = PETSC_CC_INCLUDES
 PETSC_LIB_VAR = PETSC_EXTERNAL_LIB_BASIC
 ifeq ($(PETSC_FOUND),YES)
   PETSC_OPT := $(shell sed -n "s/$(PETSC_INC_VAR) = *//p" $(PETSC_VARS))
-   PETSC_DEP := $(shell sed -n "s/$(PETSC_LIB_VAR) = *//p" $(PETSC_VARS))
-   PETSC_LIB = $(XLINKER)-rpath,$(abspath $(PETSC_DIR))/lib\
-      -L$(abspath $(PETSC_DIR))/lib -lpetsc\
-      $(subst $(CXX_XLINKER),$(XLINKER),$(PETSC_DEP))
+   PETSC_LIB := $(shell sed -n "s/$(PETSC_LIB_VAR) = *//p" $(PETSC_VARS))
+   PETSC_LIB := -Wl,-rpath,$(abspath $(PETSC_DIR))/lib\
+      -L$(abspath $(PETSC_DIR))/lib -lpetsc $(PETSC_LIB)
 endif

 SLEPC_DIR := $(MFEM_DIR)/../slepc
@@ -329,10 +326,9 @@ ifeq ($(SLEPC_FOUND),YES)
   SLEPC_OPT := $(shell sed -n "s/$(SLEPC_INC_VAR) *= *//p" $(SLEPC_VARS))
   # Some additional external libraries might be defined in this file
   -include ${SLEPC_DIR}/${PETSC_ARCH}/lib/slepc/conf/slepcvariables
-   SLEPC_DEP := $(shell sed -n "s/$(SLEPC_LIB_VAR) *= *//p" $(SLEPC_VARS))
-   SLEPC_LIB = $(XLINKER)-rpath,$(abspath $(SLEPC_DIR))/$(PETSC_ARCH)/lib\
-      -L$(abspath $(SLEPC_DIR))/$(PETSC_ARCH)/lib -lslepc\
-      $(subst $(CXX_XLINKER),$(XLINKER),$(SLEPC_DEP))
+   SLEPC_LIB := $(shell sed -n "s/$(SLEPC_LIB_VAR) *= *//p" $(SLEPC_VARS))
+   SLEPC_LIB := -Wl,-rpath,$(abspath $(SLEPC_DIR))/$(PETSC_ARCH)/lib\
+      -L$(abspath $(SLEPC_DIR))/$(PETSC_ARCH)/lib -lslepc $(SLEPC_LIB)
 endif

 # MPFR library configuration
@@ -343,7 +339,7 @@ MPFR_LIB = -lmpfr
 CONDUIT_DIR = @MFEM_DIR@/../conduit
 CONDUIT_OPT = -I$(CONDUIT_DIR)/include/conduit
 CONDUIT_LIB = \
-   $(XLINKER)-rpath,$(CONDUIT_DIR)/lib -L$(CONDUIT_DIR)/lib \
+   -Wl,-rpath,$(CONDUIT_DIR)/lib -L$(CONDUIT_DIR)/lib \
   -lconduit -lconduit_relay -lconduit_blueprint  -ldl

 # Check if Conduit was built with hdf5 support, by looking
@@ -351,7 +347,7 @@ CONDUIT_LIB = \
 CONDUIT_HDF5_HEADER=$(CONDUIT_DIR)/include/conduit/conduit_relay_hdf5.hpp
 ifneq (,$(wildcard $(CONDUIT_HDF5_HEADER)))
   CONDUIT_OPT += -I$(HDF5_DIR)/include
-   CONDUIT_LIB += $(XLINKER)-rpath,$(HDF5_DIR)/lib -L$(HDF5_DIR)/lib \
+   CONDUIT_LIB += -Wl,-rpath,$(HDF5_DIR)/lib -L$(HDF5_DIR)/lib \
                  -lhdf5 $(ZLIB_LIB)
 endif

@@ -361,9 +357,9 @@ SIDRE_DIR = @MFEM_DIR@/../axom
 SIDRE_OPT = -I$(SIDRE_DIR)/include -I$(CONDUIT_DIR)/include/conduit\
 -I$(HDF5_DIR)/include
 SIDRE_LIB = \
-   $(XLINKER)-rpath,$(SIDRE_DIR)/lib -L$(SIDRE_DIR)/lib \
-   $(XLINKER)-rpath,$(CONDUIT_DIR)/lib -L$(CONDUIT_DIR)/lib \
-   $(XLINKER)-rpath,$(HDF5_DIR)/lib -L$(HDF5_DIR)/lib \
+   -Wl,-rpath,$(SIDRE_DIR)/lib -L$(SIDRE_DIR)/lib \
+   -Wl,-rpath,$(CONDUIT_DIR)/lib -L$(CONDUIT_DIR)/lib \
+   -Wl,-rpath,$(HDF5_DIR)/lib -L$(HDF5_DIR)/lib \
   -laxom -lconduit -lconduit_relay -lconduit_blueprint -lhdf5 $(ZLIB_LIB) -ldl

 # PUMI
@@ -419,9 +415,9 @@ MKL_CPARDISO_DIR ?=
 MKL_MPI_WRAPPER ?= mkl_blacs_mpich_lp64
 MKL_LIBRARY_SUBDIR ?= lib
 MKL_CPARDISO_OPT = -I$(MKL_CPARDISO_DIR)/include
-MKL_CPARDISO_LIB = $(XLINKER)-rpath,$(MKL_CPARDISO_DIR)/$(MKL_LIBRARY_SUBDIR)\
-   -L$(MKL_CPARDISO_DIR)/$(MKL_LIBRARY_SUBDIR) -l$(MKL_MPI_WRAPPER)\
-   -lmkl_intel_lp64 -lmkl_sequential -lmkl_core
+MKL_CPARDISO_LIB = -Wl,-rpath,$(MKL_CPARDISO_DIR)/$(MKL_LIBRARY_SUBDIR)\
+                   -L$(MKL_CPARDISO_DIR)/$(MKL_LIBRARY_SUBDIR) -l$(MKL_MPI_WRAPPER)\
+                   -lmkl_intel_lp64 -lmkl_sequential -lmkl_core

 # If YES, enable some informational messages
 VERBOSE = NO
@@ -31,14 +31,14 @@ POINTS 27 double
 7 0.5 1
 8 0.5 1
 CELLS 8 56
-6 0 18 9 1 19 10
-6 1 19 10 2 20 11
-6 2 20 11 3 21 12
-6 3 21 12 4 22 13
-6 4 22 13 5 23 14
-6 5 23 14 6 24 15
-6 6 24 15 7 25 16
-6 7 25 16 8 26 17
+6 0 9 18 1 10 19
+6 1 10 19 2 11 20
+6 2 11 20 3 12 21
+6 3 12 21 4 13 22
+6 4 13 22 5 14 23
+6 5 14 23 6 15 24
+6 6 15 24 7 16 25
+6 7 16 25 8 17 26
 CELL_TYPES 8
 13
 13
@@ -205,14 +205,10 @@ int main(int argc, char *argv[])

      if (amgx_solver)
      {
-         amgx.SetConvergenceCheck(true);
         amgx.Mult(B,X);
      }
      else
      {
-         // Omit convergence check at the AmgX level when using as a
-         // preconditioner.
-         amgx.SetConvergenceCheck(false);
         PCG(*A.As<SparseMatrix>(), amgx, B, X, 3, 40, 1e-12, 0.0);
      }
   }
@@ -264,7 +264,6 @@ int main(int argc, char *argv[])
      }

      amgx.SetOperator(*A.As<HypreParMatrix>());
-      amgx.SetConvergenceCheck(true);
      amgx.Mult(B, X);

      // Release MPI communicators and resources created by AmgX
@@ -30,7 +30,6 @@
 // Device sample runs:
 //               ex1 -pa -d cuda
 //               ex1 -pa -d raja-cuda
-//             * ex1 -pa -d raja-hip
 //               ex1 -pa -d occa-cuda
 //               ex1 -pa -d raja-omp
 //               ex1 -pa -d occa-omp
@@ -178,7 +178,6 @@ int main(int argc, char *argv[])
   double visc = 1e-2;
   double mu = 0.25;
   double K = 5.0;
-   bool adaptive_lin_rtol = true;
   bool visualization = true;
   int vis_steps = 1;

@@ -207,9 +206,6 @@ int main(int argc, char *argv[])
                  "Shear modulus in the Neo-Hookean hyperelastic model.");
   args.AddOption(&K, "-K", "--bulk-modulus",
                  "Bulk modulus in the Neo-Hookean hyperelastic model.");
-   args.AddOption(&adaptive_lin_rtol, "-alrtol", "--adaptive-lin-rtol",
-                  "-no-alrtol", "--no-adaptive-lin-rtol",
-                  "Enable or disable adaptive linear solver rtol.");
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
@@ -578,7 +574,6 @@ HyperelasticOperator::HyperelasticOperator(ParFiniteElementSpace &f,
   newton_solver.SetPrintLevel(1); // print Newton iterations
   newton_solver.SetRelTol(rel_tol);
   newton_solver.SetAbsTol(0.0);
-   newton_solver.SetAdaptiveLinRtol(2, 0.5, 0.9);
   newton_solver.SetMaxIter(10);
 }

@@ -5,7 +5,6 @@
 // Sample runs:  ex14 -m ../data/inline-quad.mesh -o 0
 //               ex14 -m ../data/star.mesh -r 4 -o 2
 //               ex14 -m ../data/star-mixed.mesh -r 4 -o 2
-//               ex14 -m ../data/star-mixed.mesh -r 2 -o 2 -k 0 -e 1
 //               ex14 -m ../data/escher.mesh -s 1
 //               ex14 -m ../data/fichera.mesh -s 1 -k 1
 //               ex14 -m ../data/fichera-mixed.mesh -s 1 -k 1
@@ -45,7 +44,6 @@ int main(int argc, char *argv[])
   int order = 1;
   double sigma = -1.0;
   double kappa = -1.0;
-   double eta = 0.0;
   bool visualization = 1;

   OptionsParser args(argc, argv);
@@ -56,12 +54,11 @@ int main(int argc, char *argv[])
   args.AddOption(&order, "-o", "--order",
                  "Finite element order (polynomial degree) >= 0.");
   args.AddOption(&sigma, "-s", "--sigma",
-                  "One of the three DG penalty parameters, typically +1/-1."
+                  "One of the two DG penalty parameters, typically +1/-1."
                  " See the documentation of class DGDiffusionIntegrator.");
   args.AddOption(&kappa, "-k", "--kappa",
-                  "One of the three DG penalty parameters, should be positive."
+                  "One of the two DG penalty parameters, should be positive."
                  " Negative values are replaced with (order+1)^2.");
-   args.AddOption(&eta, "-e", "--eta", "BR2 penalty parameter.");
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
@@ -133,11 +130,6 @@ int main(int argc, char *argv[])
   a->AddDomainIntegrator(new DiffusionIntegrator(one));
   a->AddInteriorFaceIntegrator(new DGDiffusionIntegrator(one, sigma, kappa));
   a->AddBdrFaceIntegrator(new DGDiffusionIntegrator(one, sigma, kappa));
-   if (eta > 0)
-   {
-      a->AddInteriorFaceIntegrator(new DGDiffusionBR2Integrator(fespace, eta));
-      a->AddBdrFaceIntegrator(new DGDiffusionBR2Integrator(fespace, eta));
-   }
   a->Assemble();
   a->Finalize();
   const SparseMatrix &A = a->SpMat();
@@ -5,7 +5,6 @@
 // Sample runs:  mpirun -np 4 ex14p -m ../data/inline-quad.mesh -o 0
 //               mpirun -np 4 ex14p -m ../data/star.mesh -o 2
 //               mpirun -np 4 ex14p -m ../data/star-mixed.mesh -o 2
-//               mpirun -np 4 ex14p -m ../data/star-mixed.mesh -o 2 -k 0 -e 1
 //               mpirun -np 4 ex14p -m ../data/escher.mesh -s 1
 //               mpirun -np 4 ex14p -m ../data/fichera.mesh -s 1 -k 1
 //               mpirun -np 4 ex14p -m ../data/fichera-mixed.mesh -s 1 -k 1
@@ -83,7 +82,6 @@ int main(int argc, char *argv[])
   int order = 1;
   double sigma = -1.0;
   double kappa = -1.0;
-   double eta = 0.0;
   bool visualization = 1;

   OptionsParser args(argc, argv);
@@ -97,12 +95,11 @@ int main(int argc, char *argv[])
   args.AddOption(&order, "-o", "--order",
                  "Finite element order (polynomial degree) >= 0.");
   args.AddOption(&sigma, "-s", "--sigma",
-                  "One of the three DG penalty parameters, typically +1/-1."
+                  "One of the two DG penalty parameters, typically +1/-1."
                  " See the documentation of class DGDiffusionIntegrator.");
   args.AddOption(&kappa, "-k", "--kappa",
-                  "One of the three DG penalty parameters, should be positive."
+                  "One of the two DG penalty parameters, should be positive."
                  " Negative values are replaced with (order+1)^2.");
-   args.AddOption(&eta, "-e", "--eta", "BR2 penalty parameter.");
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
@@ -197,11 +194,6 @@ int main(int argc, char *argv[])
   a->AddDomainIntegrator(new DiffusionIntegrator(one));
   a->AddInteriorFaceIntegrator(new DGDiffusionIntegrator(one, sigma, kappa));
   a->AddBdrFaceIntegrator(new DGDiffusionIntegrator(one, sigma, kappa));
-   if (eta > 0)
-   {
-      a->AddInteriorFaceIntegrator(new DGDiffusionBR2Integrator(fespace, eta));
-      a->AddBdrFaceIntegrator(new DGDiffusionBR2Integrator(fespace, eta));
-   }
   a->Assemble();
   a->Finalize();

@@ -19,13 +19,6 @@
 //               mpirun -np 4 ex15p -m ../data/square-disc.mesh
 //               mpirun -np 4 ex15p -m ../data/escher.mesh -r 2 -tf 0.3
 //
-//               Different estimators:
-//
-//               mpirun -np 4 ex15p -est 0 -e 1e-4
-//               mpirun -np 4 ex15p -est 1 -e 1e-6
-//               mpirun -np 4 ex15p -est 1 -o 3 -tf 0.3
-//               mpirun -np 4 ex15p -est 2 -o 2
-//
 // Description:  Building on Example 6, this example demonstrates dynamic AMR.
 //               The mesh is adapted to a time-dependent solution by refinement
 //               as well as by derefinement. For simplicity, the solution is
@@ -35,11 +28,8 @@
 //               At each outer iteration the right hand side function is changed
 //               to mimic a time dependent problem.  Within each inner iteration
 //               the problem is solved on a sequence of meshes which are locally
-//               refined according to a chosen error estimator. Currently there
-//               are three error estimators supported: A L2 formulation of the
-//               Zienkiewicz-Zhu error estimator (0), a Kelly error indicator (1)
-//               and a traditional Zienkiewicz-Zhu error estimator (2). At the
-//               end of the inner iteration the error estimates are also used to
+//               refined according to a simple ZZ error estimator.  At the end
+//               of the inner iteration the error estimates are also used to
 //               identify any elements which may be over-refined and a single
 //               derefinement step is performed.  After each refinement or
 //               derefinement step a rebalance operation is performed to keep
@@ -97,7 +87,6 @@ int main(int argc, char *argv[])
   int nc_limit = 3;         // maximum level of hanging nodes
   bool visualization = true;
   bool visit = false;
-   int which_estimator = 0;

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -118,9 +107,6 @@ int main(int argc, char *argv[])
                  "Maximum level of hanging nodes.");
   args.AddOption(&t_final, "-tf", "--t-final",
                  "Final time; start time is 0.");
-   args.AddOption(&which_estimator, "-est", "--estimator",
-                  "Which estimator to use: "
-                  "0 = L2ZZ, 1 = Kelly, 2 = ZZ. Defaults to L2ZZ.");
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
@@ -228,43 +214,17 @@ int main(int argc, char *argv[])
   //     provide the method ComputeElementFlux. We supply an L2 space for the
   //     discontinuous flux and an H(div) space for the smoothed flux.
   L2_FECollection flux_fec(order, dim);
+   ParFiniteElementSpace flux_fes(&pmesh, &flux_fec, sdim);
   RT_FECollection smooth_flux_fec(order-1, dim);
-   ErrorEstimator* estimator;
-   switch (which_estimator)
-   {
-      case 1:
-      {
-         auto flux_fes = new ParFiniteElementSpace(&pmesh, &flux_fec, sdim);
-         estimator = new KellyErrorEstimator(*integ, x, flux_fes);
-         break;
-      }
-      case 2:
-      {
-         auto flux_fes = new ParFiniteElementSpace(&pmesh, &fec, sdim);
-         estimator = new ZienkiewiczZhuEstimator(*integ, x, flux_fes);
-         break;
-      }
-
-      default:
-         if (myid == 0)
-         {
-            std::cout << "Unkown estimator. Falling back to L2ZZ." << std::endl;
-         }
-      case 0:
-      {
-         auto flux_fes = new ParFiniteElementSpace(&pmesh, &flux_fec, sdim);
-         auto smooth_flux_fes = new ParFiniteElementSpace(&pmesh, &smooth_flux_fec);
-         estimator = new L2ZienkiewiczZhuEstimator(*integ, x, flux_fes, smooth_flux_fes);
-         break;
-      }
-   }
+   ParFiniteElementSpace smooth_flux_fes(&pmesh, &smooth_flux_fec);
+   L2ZienkiewiczZhuEstimator estimator(*integ, x, flux_fes, smooth_flux_fes);

   // 11. As in Example 6p, we also need a refiner. This time the refinement
   //     strategy is based on a fixed threshold that is applied locally to each
   //     element. The global threshold is turned off by setting the total error
   //     fraction to zero. We also enforce a maximum refinement ratio between
   //     adjacent elements.
-   ThresholdRefiner refiner(*estimator);
+   ThresholdRefiner refiner(estimator);
   refiner.SetTotalErrorFraction(0.0); // use purely local threshold
   refiner.SetLocalErrorGoal(max_elem_error);
   refiner.PreferConformingRefinement();
@@ -273,7 +233,7 @@ int main(int argc, char *argv[])
   // 12. A derefiner selects groups of elements that can be coarsened to form
   //     a larger element. A conservative enough threshold needs to be set to
   //     prevent derefining elements that would immediately be refined again.
-   ThresholdDerefiner derefiner(*estimator);
+   ThresholdDerefiner derefiner(estimator);
   derefiner.SetThreshold(hysteresis * max_elem_error);
   derefiner.SetNCLimit(nc_limit);

@@ -356,7 +316,7 @@ int main(int argc, char *argv[])
         refiner.Apply(pmesh);
         if (myid == 0)
         {
-            cout << ", total error: " << estimator->GetTotalError() << endl;
+            cout << ", total error: " << estimator.GetTotalError() << endl;
         }

         // 21. Quit the AMR loop if the termination criterion has been met
@@ -386,8 +346,6 @@ int main(int argc, char *argv[])
      }
   }

-   delete estimator;
-
   // 25. Exit
   MPI_Finalize();
   return 0;
@@ -1,435 +0,0 @@
-//                       MFEM Example 3 - Parallel Version
-//
-// Compile with: make ex3p_complex
-//
-
-#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.
-double E_exact(const Vector &);
-void gradE_exact(const Vector &, Vector &);
-double f_exact(const Vector &);
-double freq = 1.0, kappa;
-int dim;
-
-
-#define COMPLEX_VERSION
-#define NEUMANN
-
-const double omega = 1.4;
-const double eps = 1.0e-8;
-
-
-int main(int argc, char *argv[])
-{
-   // 1. Initialize MPI.
-   int num_procs, myid;
-   MPI_Init(&argc, &argv);
-   MPI_Comm_size(MPI_COMM_WORLD, &num_procs);
-   MPI_Comm_rank(MPI_COMM_WORLD, &myid);
-
-   // 2. Parse command-line options.
-   const char *mesh_file = "../data/beam-tet.mesh";
-   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(&order, "-o", "--order",
-                  "Finite element order (polynomial degree).");
-   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())
-   {
-      if (myid == 0)
-      {
-         args.PrintUsage(cout);
-      }
-      MPI_Finalize();
-      return 1;
-   }
-   if (myid == 0)
-   {
-      args.PrintOptions(cout);
-   }
-   kappa = freq * M_PI;
-
-   // 3. 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);
-   if (myid == 0) { device.Print(); }
-
-   // 4. Read the (serial) mesh from the given mesh file on all processors.  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();
-
-   // 5. Refine the serial mesh on all processors 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 1,000 elements.
-   {
-      int ref_levels = (int)floor(log(1000./mesh->GetNE())/log(2.)/dim);
-      for (int l = 0; l < ref_levels; l++)
-      {
-         mesh->UniformRefinement();
-      }
-   }
-
-   // 6. Define a parallel mesh by a partitioning of the serial mesh. Refine
-   //    this mesh further in parallel to increase the resolution. Once the
-   //    parallel mesh is defined, the serial mesh can be deleted. Tetrahedral
-   //    meshes need to be reoriented before we can define high-order Nedelec
-   //    spaces on them.
-   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
-   delete mesh;
-   {
-      int par_ref_levels = 0;
-      for (int l = 0; l < par_ref_levels; l++)
-      {
-         pmesh->UniformRefinement();
-      }
-   }
-   pmesh->ReorientTetMesh();
-
-   // 7. Define a parallel finite element space on the parallel mesh. Here we
-   //    use the Nedelec finite elements of the specified order.
-   FiniteElementCollection *fec = new H1_FECollection(order, dim);
-   ParFiniteElementSpace *fespace = new ParFiniteElementSpace(pmesh, fec);
-   HYPRE_Int size = fespace->GlobalTrueVSize();
-   if (myid == 0)
-   {
-      cout << "Number of finite element unknowns: " << size << endl;
-   }
-
-   // 8. Determine the list of true (i.e. parallel 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;
-
-#ifndef NEUMANN
-   if (pmesh->bdr_attributes.Size())
-   {
-      Array<int> ess_bdr(pmesh->bdr_attributes.Max());
-      ess_bdr = 1;
-      fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
-   }
-#endif
-
-   //const double imscale = 0.0;
-   const double imscale = -omega;
-
-   Coefficient *im = new ConstantCoefficient(imscale);  // im part
-   //Coefficient *im = new ConstantCoefficient(0.0);  // im part
-
-   FunctionCoefficient E_coef(E_exact);
-   VectorFunctionCoefficient grad_E(sdim, gradE_exact);
-   ProductCoefficient omegaE(imscale, E_coef);
-
-   // 9. Set up the parallel 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.
-   FunctionCoefficient f(f_exact);
-#ifdef COMPLEX_VERSION
-   ParComplexLinearForm *b = new ParComplexLinearForm(fespace);
-   b->AddDomainIntegrator(new DomainLFIntegrator(f), NULL);
-#ifdef NEUMANN
-   b->AddBoundaryIntegrator(NULL, new BoundaryNormalLFIntegrator(grad_E));
-#endif
-   b->AddBoundaryIntegrator(NULL, new BoundaryLFIntegrator(omegaE)); // im part
-#endif
-
-   b->Assemble();
-
-   // 10. Define the solution vector x as a parallel 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.
-   /*
-   ParGridFunction x(fespace);
-   VectorFunctionCoefficient E(sdim, E_exact);
-   x.ProjectCoefficient(E);
-   */
-
-#ifdef COMPLEX_VERSION
-   // Complex version
-   ParComplexGridFunction x(fespace);
-   x = 0.0;
-   ConstantCoefficient E_im(0.0);
-   //x.ProjectBdrCoefficientTangent(E_Re, E_Im, ess_bdr);
-   x.ProjectCoefficient(E_coef, E_im);
-#endif
-
-   // 11. Set up the parallel 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 *epscoef = new ConstantCoefficient(eps);
-   Coefficient *imabs = new ConstantCoefficient(fabs(imscale));  // im part
-
-#ifdef COMPLEX_VERSION
-   // Complex version
-   ParSesquilinearForm *a = new ParSesquilinearForm(fespace);
-   if (pa) { a->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
-   a->AddDomainIntegrator(new DiffusionIntegrator(*muinv), NULL);
-   a->AddDomainIntegrator(new MassIntegrator(*epscoef), NULL);
-   a->AddBoundaryIntegrator(NULL, new MassIntegrator(*im));  // im part
-#endif
-
-   // 12. Assemble the parallel bilinear form and the corresponding linear
-   //     system, applying any necessary transformations such as: parallel
-   //     assembly, 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);
-
-   ParBilinearForm a_Re(fespace);
-   a_Re.AddDomainIntegrator(new DiffusionIntegrator(*muinv));
-   a_Re.AddDomainIntegrator(new MassIntegrator(*epscoef));
-
-   if (pa) { a_Re.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
-   a_Re.Assemble();
-
-   OperatorPtr A_Re;
-   a_Re.FormSystemMatrix(ess_tdof_list, A_Re);
-
-   ParBilinearForm a_Im(fespace);
-   a_Im.AddBoundaryIntegrator(new MassIntegrator(*imabs));
-   a_Im.Assemble();
-
-   OperatorPtr A_Im;
-   a_Im.FormSystemMatrix(ess_tdof_list, A_Im);
-
-   // 13. Solve the system AX=B using PCG with the AMS preconditioner from hypre
-   //     (in the full assembly case) or CG with Jacobi preconditioner (in the
-   //     partial assembly case).
-
-   Array<int> offsets(3);
-   offsets[0] = 0;
-   offsets[1] = fespace->GetTrueVSize();
-   offsets[2] = fespace->GetTrueVSize();
-   offsets.PartialSum();
-
-   //OperatorJacobiSmoother massJacobi(a_Im, ess_tdof_list);
-
-   StopWatch sw;
-   sw.Clear();
-   sw.Start();
-
-   if (pa) // Jacobi preconditioning in partial assembly mode
-   {
-      MFEM_VERIFY(false, "TODO");
-      //OperatorJacobiSmoother Jacobi(*a, ess_tdof_list);
-
-      CGSolver cg(MPI_COMM_WORLD);
-      cg.SetRelTol(1e-12);
-      cg.SetMaxIter(1000);
-      cg.SetPrintLevel(1);
-      cg.SetOperator(*A);
-      //cg.SetPreconditioner(Jacobi);
-      cg.Mult(B, X);
-   }
-   else
-   {
-      if (myid == 0)
-      {
-         cout << "Size of linear system: "
-              << A.As<HypreParMatrix>()->GetGlobalNumRows() << endl;
-      }
-
-      HypreBoomerAMG amg(*A_Re.As<HypreParMatrix>());
-
-#ifdef COMPLEX_VERSION
-      BlockDiagonalPreconditioner BlockDP(offsets);
-      BlockDP.SetDiagonalBlock(0, &amg);
-      BlockDP.SetDiagonalBlock(1, &amg);
-
-      Complex_PMHSS PMHSS(A_Re.Ptr(), A_Im.Ptr(), &BlockDP, NULL, 1.0);
-
-      ComplexOperator AspdComplex(A_Re.Ptr(), A_Im.Ptr(), false, false);
-
-      GMRESSolver PMHSSgmres(MPI_COMM_WORLD);
-      PMHSSgmres.SetPrintLevel(1);
-      PMHSSgmres.SetKDim(100);
-      PMHSSgmres.SetMaxIter(100);
-      PMHSSgmres.SetRelTol(1e-6);
-      PMHSSgmres.SetAbsTol(0.0);
-      PMHSSgmres.SetOperator(AspdComplex);
-      PMHSSgmres.SetPreconditioner(PMHSS);
-
-      GMRESSolver gmres(MPI_COMM_WORLD);
-      gmres.SetPrintLevel(1);
-      gmres.SetKDim(1000);
-      gmres.SetMaxIter(100);
-      gmres.SetRelTol(1e-8);
-      gmres.SetAbsTol(0.0);
-      gmres.SetOperator(*A);
-      //gmres.SetPreconditioner(BlockDP);
-      gmres.SetPreconditioner(PMHSS);
-      //gmres.SetPreconditioner(PMHSSgmres);
-#else
-      GMRESSolver gmres(MPI_COMM_WORLD);
-      gmres.SetPrintLevel(1);
-      gmres.SetKDim(1000);
-      gmres.SetMaxIter(100);
-      gmres.SetRelTol(1e-8);
-      gmres.SetAbsTol(0.0);
-      gmres.SetOperator(*A);
-      gmres.SetPreconditioner(ams);
-#endif
-
-      gmres.Mult(B, X);
-   }
-
-   sw.Stop();
-   mfem::out << "Total solve time " <<sw.RealTime() << endl;
-
-   // 14. Recover the parallel grid function corresponding to X. This is the
-   //     local finite element solution on each processor.
-   a->RecoverFEMSolution(X, *b, x);
-
-   // 15. Compute and print the L^2 norm of the error.
-   {
-#ifdef COMPLEX_VERSION
-      double err = x.real().ComputeL2Error(E_coef);
-#else
-      double err = x.ComputeL2Error(E_coef);
-#endif
-      if (myid == 0)
-      {
-         cout << "\n|| E_h - E ||_{L^2} = " << err << '\n' << endl;
-      }
-   }
-
-   // 16. Save the refined mesh and the solution in parallel. This output can
-   //     be viewed later using GLVis: "glvis -np <np> -m mesh -g sol".
-   {
-      ostringstream mesh_name, sol_name;
-      mesh_name << "mesh." << setfill('0') << setw(6) << myid;
-      sol_name << "sol." << setfill('0') << setw(6) << myid;
-
-      ofstream mesh_ofs(mesh_name.str().c_str());
-      mesh_ofs.precision(8);
-      pmesh->Print(mesh_ofs);
-
-      ofstream sol_ofs(sol_name.str().c_str());
-      sol_ofs.precision(8);
-#ifdef COMPLEX_VERSION
-      x.real().Save(sol_ofs);
-#else
-      x.Save(sol_ofs);
-#endif
-   }
-
-   // 17. Send the solution by socket to a GLVis server.
-   if (visualization)
-   {
-      char vishost[] = "localhost";
-      int  visport   = 19916;
-      socketstream sol_sock(vishost, visport);
-      sol_sock << "parallel " << num_procs << " " << myid << "\n";
-      sol_sock.precision(8);
-#ifdef COMPLEX_VERSION
-      sol_sock << "solution\n" << *pmesh << x.real() << flush;
-#else
-      sol_sock << "solution\n" << *pmesh << x << flush;
-#endif
-   }
-
-   // 18. Free the used memory.
-   delete a;
-   delete muinv;
-   delete b;
-   delete fespace;
-   delete fec;
-   delete pmesh;
-
-   MPI_Finalize();
-
-   return 0;
-}
-
-#define VERSION_COS
-
-double E_exact(const Vector &x)
-{
-   if (dim == 3)
-   {
-#ifdef VERSION_COS
-      return cos(kappa * x(0)) * cos(kappa * x(1)) * cos(kappa * x(2));
-#else
-      return sin(kappa * x(0)) * sin(kappa * x(1)) * sin(kappa * x(2));
-#endif
-   }
-   else
-   {
-      return 0.0;
-   }
-}
-
-void gradE_exact(const Vector &x, Vector &grad)
-{
-   if (dim == 3)
-   {
-#ifdef VERSION_COS
-      grad(0) = -kappa * sin(kappa * x(0)) * cos(kappa * x(1)) * cos(kappa * x(2));
-      grad(1) = -kappa * sin(kappa * x(1)) * cos(kappa * x(0)) * cos(kappa * x(2));
-      grad(2) = -kappa * sin(kappa * x(2)) * cos(kappa * x(0)) * cos(kappa * x(1));
-#else
-      grad(0) = kappa * cos(kappa * x(0)) * sin(kappa * x(1)) * sin(kappa * x(2));
-      grad(1) = kappa * cos(kappa * x(1)) * sin(kappa * x(0)) * sin(kappa * x(2));
-      grad(2) = kappa * cos(kappa * x(2)) * sin(kappa * x(0)) * sin(kappa * x(1));
-#endif
-   }
-   else
-   {
-      MFEM_VERIFY(false, "");
-   }
-}
-
-// (grad u, grad v) + eps (u, v) = <grad u . n, v> - (div grad u, v) + eps (u, v)
-double f_exact(const Vector &x)
-{
-   if (dim == 3)
-   {
-      const double c = 3.0 * kappa * kappa;
-#ifdef VERSION_COS
-      return (eps + c) * cos(kappa * x(0)) * cos(kappa * x(1)) * cos(kappa * x(2));
-#else
-      return (eps + c) * sin(kappa * x(0)) * sin(kappa * x(1)) * sin(kappa * x(2));
-#endif
-   }
-   else
-   {
-      return 0.0;
-   }
-}
@@ -10,10 +10,6 @@
 //               ex25 -o 2 -f 8.0 -ref 3 -prob 4 -m ../data/inline-quad.mesh
 //               ex25 -o 2 -f 2.0 -ref 1 -prob 4 -m ../data/inline-hex.mesh
 //
-// Device sample runs:
-//               ex25 -o 2 -f 8.0 -ref 3 -prob 4 -m ../data/inline-quad.mesh -pa -d cuda
-//               ex25 -o 2 -f 2.0 -ref 1 -prob 4 -m ../data/inline-hex.mesh -pa -d cuda
-//
 // Description:  This example code solves a simple electromagnetic wave
 //               propagation problem corresponding to the second order
 //               indefinite Maxwell equation
@@ -161,10 +157,7 @@ int main(int argc, char *argv[])
   int iprob = 4;
   double freq = 5.0;
   bool herm_conv = true;
-   bool umf_solver = false;
   bool visualization = 1;
-   bool pa = false;
-   const char *device_config = "cpu";

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -183,28 +176,15 @@ int main(int argc, char *argv[])
                  "Frequency (in Hz).");
   args.AddOption(&herm_conv, "-herm", "--hermitian", "-no-herm",
                  "--no-hermitian", "Use convention for Hermitian operators.");
-#ifdef MFEM_USE_SUITESPARSE
-   args.AddOption(&umf_solver, "-umf", "--umfpack", "-no-umf",
-                  "--no-umfpack", "Use the UMFPack Solver.");
-#endif
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
-   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.Parse();

   if (iprob > 4) { iprob = 4; }
   prob = (prob_type)iprob;

-   // 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. Setup the mesh
+   // 2. Setup the mesh
   if (!mesh_file)
   {
      exact_known = true;
@@ -245,7 +225,7 @@ int main(int argc, char *argv[])
   // Setup PML length
   Array2D<double> length(dim, 2); length = 0.0;

-   // 4. Setup the Cartesian PML region.
+   // 3. Setup the Cartesian PML region.
   switch (prob)
   {
      case disc:
@@ -271,19 +251,19 @@ int main(int argc, char *argv[])
   comp_domain_bdr = pml->GetCompDomainBdr();
   domain_bdr = pml->GetDomainBdr();

-   // 5. Refine the mesh to increase the resolution.
+   // 4. Refine the mesh to increase the resolution.
   for (int l = 0; l < ref_levels; l++)
   {
      mesh->UniformRefinement();
   }

-   // 6. Reorient mesh in case of a tet mesh
+   // 5. Reorient mesh in case of a tet mesh
   mesh->ReorientTetMesh();

   // Set element attributes in order to distinguish elements in the PML region
   pml->SetAttributes(mesh);

-   // 7. Define a finite element space on the mesh. Here we use the Nedelec
+   // 6. Define a finite element space on the mesh. Here we use the Nedelec
   //    finite elements of the specified order.
   FiniteElementCollection *fec = new ND_FECollection(order, dim);
   FiniteElementSpace *fespace = new FiniteElementSpace(mesh, fec);
@@ -291,7 +271,7 @@ int main(int argc, char *argv[])

   cout << "Number of finite element unknowns: " << size << endl;

-   // 8. Determine the list of true essential boundary dofs. In this example,
+   // 7. Determine the list of true essential boundary dofs. In this example,
   //    the boundary conditions are defined based on the specific mesh and the
   //    problem type.
   Array<int> ess_tdof_list;
@@ -333,12 +313,12 @@ int main(int argc, char *argv[])
   }
   fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);

-   // 9. Setup Complex Operator convention
+   // 8. Setup Complex Operator convention
   ComplexOperator::Convention conv =
      herm_conv ? ComplexOperator::HERMITIAN : ComplexOperator::BLOCK_SYMMETRIC;

-   // 10. Set up the linear form b(.) which corresponds to the right-hand side of
-   //     the FEM linear system.
+   // 9. Set up the linear form b(.) which corresponds to the right-hand side of
+   //    the FEM linear system.
   VectorFunctionCoefficient f(dim, source);
   ComplexLinearForm b(fespace, conv);
   if (prob == load_src)
@@ -348,7 +328,7 @@ int main(int argc, char *argv[])
   b.Vector::operator=(0.0);
   b.Assemble();

-   // 11. Define the solution vector x as a complex finite element grid function
+   // 10. Define the solution vector x as a complex finite element grid function
   //     corresponding to fespace.
   ComplexGridFunction x(fespace);
   x = 0.0;
@@ -356,7 +336,7 @@ int main(int argc, char *argv[])
   VectorFunctionCoefficient E_Im(dim, E_bdr_data_Im);
   x.ProjectBdrCoefficientTangent(E_Re, E_Im, ess_bdr);

-   // 12. Set up the sesquilinear form a(.,.)
+   // 11. Set up the sesquilinear form a(.,.)
   //
   //     In Comp
   //     Domain:   1/mu (Curl E, Curl F) - omega^2 * epsilon (E,F)
@@ -410,35 +390,32 @@ int main(int argc, char *argv[])
   a.AddDomainIntegrator(new VectorFEMassIntegrator(restr_c2_Re),
                         new VectorFEMassIntegrator(restr_c2_Im));

-   // 13. Assemble the bilinear form and the corresponding linear system,
+   // 12. Assemble the bilinear form and the corresponding linear system,
   //     applying any necessary transformations such as: assembly, eliminating
   //     boundary conditions, applying conforming constraints for
   //     non-conforming AMR, etc.
-   if (pa) { a.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
   a.Assemble(0);

   OperatorPtr A;
   Vector B, X;
   a.FormLinearSystem(ess_tdof_list, x, b, A, X, B);

-   // 14. Solve using a direct or an iterative solver
+   // 13. Solve using a direct or an iterative solver
 #ifdef MFEM_USE_SUITESPARSE
-   if (!pa && umf_solver)
   {
      ComplexUMFPackSolver csolver(*A.As<ComplexSparseMatrix>());
      csolver.Control[UMFPACK_ORDERING] = UMFPACK_ORDERING_METIS;
      csolver.SetPrintLevel(1);
      csolver.Mult(B, X);
   }
-#endif
-   // 14a. Set up the Bilinear form a(.,.) for the preconditioner
+#else
+   // 13a. Set up the Bilinear form a(.,.) for the preconditioner
   //
   //    In Comp
   //    Domain:   1/mu (Curl E, Curl F) + omega^2 * epsilon (E,F)
   //
   //    In PML:   1/mu (abs(1/det(J) J^T J) Curl E, Curl F)
   //              + omega^2 * epsilon (abs(det(J) * (J^T J)^-1) * E, F)
-   if (pa || !umf_solver)
   {
      ConstantCoefficient absomeg(pow(omega, 2) * epsilon);
      RestrictedCoefficient restr_absomeg(absomeg,attr);
@@ -458,57 +435,39 @@ int main(int argc, char *argv[])
      prec.AddDomainIntegrator(new CurlCurlIntegrator(restr_c1_abs));
      prec.AddDomainIntegrator(new VectorFEMassIntegrator(restr_c2_abs));

-      if (pa) { prec.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
      prec.Assemble();

-      // 14b. Define and apply a GMRES solver for AU=B with a block diagonal
-      //      preconditioner based on the Gauss-Seidel or Jacobi sparse smoother.
+      OperatorPtr PCOpAh;
+      prec.FormSystemMatrix(ess_tdof_list, PCOpAh);
+
+      // 13b. Define and apply a GMRES solver for AU=B with a block diagonal
+      //      preconditioner based on the Gauss-Seidel sparse smoother.
      Array<int> offsets(3);
      offsets[0] = 0;
      offsets[1] = fespace->GetTrueVSize();
      offsets[2] = fespace->GetTrueVSize();
      offsets.PartialSum();

-      Operator *pc_r = nullptr;
-      Operator *pc_i = nullptr;
-      int s = (conv == ComplexOperator::HERMITIAN) ? -1.0 : 1.0;
-      if (pa)
-      {
-         // Jacobi Smoother
-         OperatorJacobiSmoother *d00 = new OperatorJacobiSmoother(prec, ess_tdof_list);
-         ScaledOperator *d11 = new ScaledOperator(d00, s);
-         pc_r = d00;
-         pc_i = d11;
-      }
-      else
-      {
-         OperatorPtr PCOpAh;
-         prec.SetDiagonalPolicy(mfem::Operator::DIAG_ONE);
-         prec.FormSystemMatrix(ess_tdof_list, PCOpAh);
-
-         // Gauss-Seidel Smoother
-         GSSmoother *gs00 = new GSSmoother(*PCOpAh.As<SparseMatrix>());
-         ScaledOperator *gs11 = new ScaledOperator(gs00, s);
-         pc_r = gs00;
-         pc_i = gs11;
-      }
-
-      BlockDiagonalPreconditioner BlockDP(offsets);
-      BlockDP.SetDiagonalBlock(0, pc_r);
-      BlockDP.SetDiagonalBlock(1, pc_i);
+      GSSmoother gs00(*PCOpAh.As<SparseMatrix>());
+      BlockDiagonalPreconditioner BlockGS(offsets);
+      ScaledOperator gs11(&gs00,
+                          (conv == ComplexOperator::HERMITIAN) ? -1.0 : 1.0);
+      BlockGS.SetDiagonalBlock(0,&gs00);
+      BlockGS.SetDiagonalBlock(1,&gs11);

      GMRESSolver gmres;
      gmres.SetPrintLevel(1);
      gmres.SetKDim(200);
-      gmres.SetMaxIter(pa ? 5000 : 2000);
+      gmres.SetMaxIter(2000);
      gmres.SetRelTol(1e-5);
      gmres.SetAbsTol(0.0);
      gmres.SetOperator(*A);
-      gmres.SetPreconditioner(BlockDP);
+      gmres.SetPreconditioner(BlockGS);
      gmres.Mult(B, X);
   }
+#endif

-   // 15. Recover the solution as a finite element grid function and compute the
+   // 14. Recover the solution as a finite element grid function and compute the
   //     errors if the exact solution is known.
   a.RecoverFEMSolution(X, b, x);

@@ -545,7 +504,7 @@ int main(int argc, char *argv[])
           << sqrt(L2Error_Re*L2Error_Re + L2Error_Im*L2Error_Im) << "\n\n";
   }

-   // 16. Save the refined mesh and the solution. This output can be viewed
+   // 15. Save the refined mesh and the solution. This output can be viewed
   //     later using GLVis: "glvis -m mesh -g sol".
   {
      ofstream mesh_ofs("ex25.mesh");
@@ -560,7 +519,7 @@ int main(int argc, char *argv[])
      x.imag().Save(sol_i_ofs);
   }

-   // 17. Send the solution by socket to a GLVis server.
+   // 16. Send the solution by socket to a GLVis server.
   if (visualization)
   {
      // Define visualization keys for GLVis (see GLVis documentation)
@@ -611,7 +570,7 @@ int main(int argc, char *argv[])
      }
   }

-   // 18. Free the used memory.
+   // 17. Free the used memory.
   delete pml;
   delete fespace;
   delete fec;
@@ -957,14 +916,7 @@ void CartesianPML::SetBoundaries()

 void CartesianPML::SetAttributes(Mesh *mesh_)
 {
-   // Initialize bdr attributes
-   for (int i = 0; i < mesh_->GetNBE(); ++i)
-   {
-      mesh_->GetBdrElement(i)->SetAttribute(i+1);
-   }
-
   int nrelem = mesh_->GetNE();
-
   elems.SetSize(nrelem);

   // Loop through the elements and identify which of them are in the PML
@@ -10,10 +10,6 @@
 //               mpirun -np 4 ex25p -o 2 -f 8.0 -rs 2 -rp 2 -prob 4 -m ../data/inline-quad.mesh
 //               mpirun -np 4 ex25p -o 2 -f 2.0 -rs 1 -rp 1 -prob 4 -m ../data/inline-hex.mesh
 //
-// Device sample runs:
-//               mpirun -np 4 ex25p -o 1 -f 3.0 -rs 3 -rp 1 -prob 2 -pa -d cuda
-//               mpirun -np 4 ex25p -o 2 -f 1.0 -rs 1 -rp 1 -prob 3 -pa -d cuda
-//
 // Description:  This example code solves a simple electromagnetic wave
 //               propagation problem corresponding to the second order
 //               indefinite Maxwell equation
@@ -171,8 +167,6 @@ int main(int argc, char *argv[])
   bool slu_solver  = false;
   bool mumps_solver = false;
   bool visualization = 1;
-   bool pa = false;
-   const char *device_config = "cpu";

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -204,10 +198,6 @@ int main(int argc, char *argv[])
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
-   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.Parse();
   if (slu_solver && mumps_solver)
   {
@@ -221,12 +211,7 @@ int main(int argc, char *argv[])
   if (iprob > 4) { iprob = 4; }
   prob = (prob_type)iprob;

-   // 3. 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);
-   if (myid == 0) { device.Print(); }
-
-   // 4. Setup the (serial) mesh on all processors.
+   // 3. Setup the (serial) mesh on all processors.
   if (!mesh_file)
   {
      exact_known = true;
@@ -274,7 +259,7 @@ int main(int argc, char *argv[])
   // Setup PML length
   Array2D<double> length(dim, 2); length = 0.0;

-   // 5. Setup the Cartesian PML region.
+   // 4. Setup the Cartesian PML region.
   switch (prob)
   {
      case disc:
@@ -300,13 +285,13 @@ int main(int argc, char *argv[])
   comp_domain_bdr = pml->GetCompDomainBdr();
   domain_bdr = pml->GetDomainBdr();

-   // 6. Refine the serial mesh on all processors to increase the resolution.
+   // 5. Refine the serial mesh on all processors to increase the resolution.
   for (int l = 0; l < ref_levels; l++)
   {
      mesh->UniformRefinement();
   }

-   // 7. Define a parallel mesh by a partitioning of the serial mesh.
+   // 6. Define a parallel mesh by a partitioning of the serial mesh.
   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
   delete mesh;
   {
@@ -316,13 +301,13 @@ int main(int argc, char *argv[])
      }
   }

-   // 7a. Reorient mesh in case of a tet mesh
+   // 6a. Reorient mesh in case of a tet mesh
   pmesh->ReorientTetMesh();

-   // 8. Set element attributes in order to distinguish elements in the PML
+   // 7. Set element attributes in order to distinguish elements in the PML
   pml->SetAttributes(pmesh);

-   // 9. Define a parallel finite element space on the parallel mesh. Here we
+   // 8. Define a parallel finite element space on the parallel mesh. Here we
   //    use the Nedelec finite elements of the specified order.
   FiniteElementCollection *fec = new ND_FECollection(order, dim);
   ParFiniteElementSpace *fespace = new ParFiniteElementSpace(pmesh, fec);
@@ -332,9 +317,9 @@ int main(int argc, char *argv[])
      cout << "Number of finite element unknowns: " << size << endl;
   }

-   // 10. Determine the list of true (i.e. parallel conforming) essential
-   //     boundary dofs. In this example, the boundary conditions are defined
-   //     based on the specific mesh and the problem type.
+   // 9. Determine the list of true (i.e. parallel conforming) essential
+   //    boundary dofs. In this example, the boundary conditions are defined
+   //    based on the specific mesh and the problem type.
   Array<int> ess_tdof_list;
   Array<int> ess_bdr;
   if (pmesh->bdr_attributes.Size())
@@ -374,11 +359,11 @@ int main(int argc, char *argv[])
   }
   fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);

-   // 11. Setup Complex Operator convention
+   // 10. Setup Complex Operator convention
   ComplexOperator::Convention conv =
      herm_conv ? ComplexOperator::HERMITIAN : ComplexOperator::BLOCK_SYMMETRIC;

-   // 12. Set up the parallel linear form b(.) which corresponds to the
+   // 11. Set up the parallel linear form b(.) which corresponds to the
   //     right-hand side of the FEM linear system.
   VectorFunctionCoefficient f(dim, source);
   ParComplexLinearForm b(fespace, conv);
@@ -389,7 +374,7 @@ int main(int argc, char *argv[])
   b.Vector::operator=(0.0);
   b.Assemble();

-   // 13. Define the solution vector x as a parallel complex finite element grid
+   // 12. Define the solution vector x as a parallel complex finite element grid
   //     function corresponding to fespace.
   ParComplexGridFunction x(fespace);
   x = 0.0;
@@ -397,7 +382,7 @@ int main(int argc, char *argv[])
   VectorFunctionCoefficient E_Im(dim, E_bdr_data_Im);
   x.ProjectBdrCoefficientTangent(E_Re, E_Im, ess_bdr);

-   // 14. Set up the parallel sesquilinear form a(.,.)
+   // 13. Set up the parallel sesquilinear form a(.,.)
   //
   //     In Comp
   //     Domain:   1/mu (Curl E, Curl F) - omega^2 * epsilon (E,F)
@@ -451,20 +436,19 @@ int main(int argc, char *argv[])
   a.AddDomainIntegrator(new VectorFEMassIntegrator(restr_c2_Re),
                         new VectorFEMassIntegrator(restr_c2_Im));

-   // 15. Assemble the parallel bilinear form and the corresponding linear
+   // 14. Assemble the parallel bilinear form and the corresponding linear
   //     system, applying any necessary transformations such as: parallel
   //     assembly, eliminating boundary conditions, applying conforming
   //     constraints for non-conforming AMR, etc.
-   if (pa) { a.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
   a.Assemble();

   OperatorPtr Ah;
   Vector B, X;
   a.FormLinearSystem(ess_tdof_list, x, b, Ah, X, B);

-   // 16. Solve using a direct or an iterative solver
+   // 15. Solve using a direct or an iterative solver
 #ifdef MFEM_USE_SUPERLU
-   if (!pa && slu_solver)
+   if (slu_solver)
   {
      // Transform to monolithic HypreParMatrix
      HypreParMatrix *A = Ah.As<ComplexHypreParMatrix>()->GetSystemMatrix();
@@ -479,7 +463,7 @@ int main(int argc, char *argv[])
   }
 #endif
 #ifdef MFEM_USE_MUMPS
-   if (!pa && mumps_solver)
+   if (mumps_solver)
   {
      HypreParMatrix *A = Ah.As<ComplexHypreParMatrix>()->GetSystemMatrix();
      MUMPSSolver mumps;
@@ -497,7 +481,7 @@ int main(int argc, char *argv[])
   //
   //    In PML:   1/mu (abs(1/det(J) J^T J) Curl E, Curl F)
   //              + omega^2 * epsilon (abs(det(J) * (J^T J)^-1) * E, F)
-   if (pa || (!slu_solver && !mumps_solver))
+   if (!slu_solver && !mumps_solver)
   {
      ConstantCoefficient absomeg(pow(omega, 2) * epsilon);
      RestrictedCoefficient restr_absomeg(absomeg,attr);
@@ -517,9 +501,11 @@ int main(int argc, char *argv[])
      prec.AddDomainIntegrator(new CurlCurlIntegrator(restr_c1_abs));
      prec.AddDomainIntegrator(new VectorFEMassIntegrator(restr_c2_abs));

-      if (pa) { prec.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
      prec.Assemble();

+      OperatorPtr PCOpAh;
+      prec.FormSystemMatrix(ess_tdof_list, PCOpAh);
+
      // 16b. Define and apply a parallel GMRES solver for AU=B with a block
      //      diagonal preconditioner based on hypre's AMS preconditioner.
      Array<int> offsets(3);
@@ -528,41 +514,21 @@ int main(int argc, char *argv[])
      offsets[2] = fespace->GetTrueVSize();
      offsets.PartialSum();

-      Operator *pc_r = nullptr;
-      Operator *pc_i = nullptr;
-      int s = (conv == ComplexOperator::HERMITIAN) ? -1.0 : 1.0;
-      if (pa)
-      {
-         // Jacobi Smoother
-         OperatorJacobiSmoother *d00 = new OperatorJacobiSmoother(prec, ess_tdof_list);
-         ScaledOperator *d11 = new ScaledOperator(d00, s);
-         pc_r = d00;
-         pc_i = d11;
-      }
-      else
-      {
-         OperatorPtr PCOpAh;
-         prec.FormSystemMatrix(ess_tdof_list, PCOpAh);
-
-         // Hypre AMS
-         HypreAMS *ams00 = new HypreAMS(*PCOpAh.As<HypreParMatrix>(), fespace);
-         ScaledOperator *ams11 = new ScaledOperator(ams00, s);
-         pc_r = ams00;
-         pc_i = ams11;
-      }
-
-      BlockDiagonalPreconditioner BlockDP(offsets);
-      BlockDP.SetDiagonalBlock(0, pc_r);
-      BlockDP.SetDiagonalBlock(1, pc_i);
+      HypreAMS ams00(*PCOpAh.As<HypreParMatrix>(),fespace);
+      BlockDiagonalPreconditioner BlockAMS(offsets);
+      ScaledOperator ams11(&ams00,
+                           (conv == ComplexOperator::HERMITIAN) ? -1.0 : 1.0);
+      BlockAMS.SetDiagonalBlock(0,&ams00);
+      BlockAMS.SetDiagonalBlock(1,&ams11);

      GMRESSolver gmres(MPI_COMM_WORLD);
      gmres.SetPrintLevel(1);
      gmres.SetKDim(200);
-      gmres.SetMaxIter(pa ? 5000 : 2000);
+      gmres.SetMaxIter(2000);
      gmres.SetRelTol(1e-5);
      gmres.SetAbsTol(0.0);
      gmres.SetOperator(*Ah);
-      gmres.SetPreconditioner(BlockDP);
+      gmres.SetPreconditioner(BlockAMS);
      gmres.Mult(B, X);
   }

@@ -1037,12 +1003,8 @@ void CartesianPML::SetBoundaries()

 void CartesianPML::SetAttributes(ParMesh *pmesh)
 {
-   // Initialize bdr attributes
-   for (int i = 0; i < pmesh->GetNBE(); ++i)
-   {
-      pmesh->GetBdrElement(i)->SetAttribute(i+1);
-   }
-
+   int myid;
+   MPI_Comm_rank(MPI_COMM_WORLD,&myid);
   int nrelem = pmesh->GetNE();

   // Initialize list with 1
@@ -40,7 +40,7 @@ using namespace mfem;
 // in the FiniteElementSpaceHierarchy. The preconditioner uses a CG solver on
 // the coarsest level and second order Chebyshev accelerated smoothers on the
 // other levels.
-class DiffusionMultigrid : public GeometricMultigrid
+class DiffusionMultigrid : public Multigrid
 {
 private:
   ConstantCoefficient one;
@@ -49,7 +49,7 @@ public:
   // Constructs a diffusion multigrid for the given FiniteElementSpaceHierarchy
   // and the array of essential boundaries
   DiffusionMultigrid(FiniteElementSpaceHierarchy& fespaces, Array<int>& ess_bdr)
-      : GeometricMultigrid(fespaces), one(1.0)
+      : Multigrid(fespaces), one(1.0)
   {
      ConstructCoarseOperatorAndSolver(fespaces.GetFESpaceAtLevel(0), ess_bdr);

@@ -37,7 +37,7 @@ using namespace mfem;
 // all spaces except the coarsest one in the ParFiniteElementSpaceHierarchy.
 // The multigrid uses a PCG solver preconditioned with AMG on the coarsest level
 // and second order Chebyshev accelerated smoothers on the other levels.
-class DiffusionMultigrid : public GeometricMultigrid
+class DiffusionMultigrid : public Multigrid
 {
 private:
   ConstantCoefficient one;
@@ -48,7 +48,7 @@ public:
   // and the array of essential boundaries
   DiffusionMultigrid(ParFiniteElementSpaceHierarchy& fespaces,
                      Array<int>& ess_bdr)
-      : GeometricMultigrid(fespaces), one(1.0)
+      : Multigrid(fespaces), one(1.0)
   {
      ConstructCoarseOperatorAndSolver(fespaces.GetFESpaceAtLevel(0), ess_bdr);

@@ -22,7 +22,6 @@
 //
 // Device sample runs:
 //               mpirun -np 4 ex3p -m ../data/star.mesh -pa -d cuda
-//               mpirun -np 4 ex3p -m ../data/star.mesh -no-pa -d cuda
 //               mpirun -np 4 ex3p -m ../data/star.mesh -pa -d raja-cuda
 //               mpirun -np 4 ex3p -m ../data/star.mesh -pa -d raja-omp
 //               mpirun -np 4 ex3p -m ../data/beam-hex.mesh -pa -d cuda
@@ -69,10 +68,7 @@ int main(int argc, char *argv[])
   bool static_cond = false;
   bool pa = false;
   const char *device_config = "cpu";
-   bool visualization = true;
-#ifdef MFEM_USE_AMGX
-   bool useAmgX = false;
-#endif
+   bool visualization = 1;

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -90,11 +86,6 @@ int main(int argc, char *argv[])
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
-#ifdef MFEM_USE_AMGX
-   args.AddOption(&useAmgX, "-amgx", "--useAmgX", "-no-amgx",
-                  "--no-useAmgX",
-                  "Enable or disable AmgX in MatrixFreeAMS.");
-#endif

   args.Parse();
   if (!args.Good())
@@ -167,10 +158,9 @@ int main(int argc, char *argv[])
   //    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;
   if (pmesh->bdr_attributes.Size())
   {
-      ess_bdr.SetSize(pmesh->bdr_attributes.Max());
+      Array<int> ess_bdr(pmesh->bdr_attributes.Max());
      ess_bdr = 1;
      fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
   }
@@ -214,20 +204,20 @@ int main(int argc, char *argv[])
   Vector B, X;
   a->FormLinearSystem(ess_tdof_list, x, *b, A, X, B);

-   // 13. Solve the system AX=B using PCG with an AMS preconditioner.
-   if (pa)
+   // 13. Solve the system AX=B using PCG with the AMS preconditioner from hypre
+   //     (in the full assembly case) or CG with Jacobi preconditioner (in the
+   //     partial assembly case).
+
+   if (pa) // Jacobi preconditioning in partial assembly mode
   {
-#ifdef MFEM_USE_AMGX
-      MatrixFreeAMS ams(*a, *A, *fespace, muinv, sigma, NULL, ess_bdr, useAmgX);
-#else
-      MatrixFreeAMS ams(*a, *A, *fespace, muinv, sigma, NULL, ess_bdr);
-#endif
+      OperatorJacobiSmoother Jacobi(*a, ess_tdof_list);
+
      CGSolver cg(MPI_COMM_WORLD);
      cg.SetRelTol(1e-12);
      cg.SetMaxIter(1000);
      cg.SetPrintLevel(1);
      cg.SetOperator(*A);
-      cg.SetPreconditioner(ams);
+      cg.SetPreconditioner(Jacobi);
      cg.Mult(B, X);
   }
   else
@@ -1,517 +0,0 @@
-//                       MFEM Example 3 - Parallel Version
-//
-// Compile with: make ex3p_complex
-//
-
-#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 curlE_exact(const Vector &, Vector &);
-void f_exact(const Vector &, Vector &);
-double freq = 1.0, kappa;
-int dim;
-
-
-#define COMPLEX_VERSION
-#define NEUMANN
-#define INDEFINITE
-
-const double omega = 1.4;
-
-
-int main(int argc, char *argv[])
-{
-   // 1. Initialize MPI.
-   int num_procs, myid;
-   MPI_Init(&argc, &argv);
-   MPI_Comm_size(MPI_COMM_WORLD, &num_procs);
-   MPI_Comm_rank(MPI_COMM_WORLD, &myid);
-
-   // 2. Parse command-line options.
-   const char *mesh_file = "../data/beam-tet.mesh";
-   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(&order, "-o", "--order",
-                  "Finite element order (polynomial degree).");
-   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())
-   {
-      if (myid == 0)
-      {
-         args.PrintUsage(cout);
-      }
-      MPI_Finalize();
-      return 1;
-   }
-   if (myid == 0)
-   {
-      args.PrintOptions(cout);
-   }
-   kappa = freq * M_PI;
-
-   // 3. 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);
-   if (myid == 0) { device.Print(); }
-
-   // 4. Read the (serial) mesh from the given mesh file on all processors.  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();
-
-   // 5. Refine the serial mesh on all processors 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 1,000 elements.
-   {
-      int ref_levels = (int)floor(log(1000./mesh->GetNE())/log(2.)/dim);
-      for (int l = 0; l < ref_levels; l++)
-      {
-         mesh->UniformRefinement();
-      }
-   }
-
-   // 6. Define a parallel mesh by a partitioning of the serial mesh. Refine
-   //    this mesh further in parallel to increase the resolution. Once the
-   //    parallel mesh is defined, the serial mesh can be deleted. Tetrahedral
-   //    meshes need to be reoriented before we can define high-order Nedelec
-   //    spaces on them.
-   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
-   delete mesh;
-   {
-      int par_ref_levels = 0;
-      for (int l = 0; l < par_ref_levels; l++)
-      {
-         pmesh->UniformRefinement();
-      }
-   }
-   pmesh->ReorientTetMesh();
-
-   // 7. Define a parallel finite element space on the parallel mesh. Here we
-   //    use the Nedelec finite elements of the specified order.
-   FiniteElementCollection *fec = new ND_FECollection(order, dim);
-   ParFiniteElementSpace *fespace = new ParFiniteElementSpace(pmesh, fec);
-   HYPRE_Int size = fespace->GlobalTrueVSize();
-   if (myid == 0)
-   {
-      cout << "Number of finite element unknowns: " << size << endl;
-   }
-
-   // 8. Determine the list of true (i.e. parallel 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;
-   Array<int> ess_bdr;
-   ess_bdr.SetSize(pmesh->bdr_attributes.Max());
-   ess_bdr = 0;
-
-#ifndef NEUMANN
-   if (pmesh->bdr_attributes.Size())
-   {
-      ess_bdr = 1;
-      fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
-   }
-#endif
-
-   //const double imscale = 0.0;
-   const double imscale = omega;
-
-   Coefficient *im = new ConstantCoefficient(imscale);  // im part
-   //Coefficient *im = new ConstantCoefficient(0.0);  // im part
-
-   VectorFunctionCoefficient E_Re(sdim, E_exact);
-   VectorFunctionCoefficient curlE_Re(sdim, curlE_exact);
-
-   ScalarVectorProductCoefficient omegaE(imscale, E_Re);  // im part
-   //ScalarVectorProductCoefficient omegaE(0.0, E_Re);  // im part
-
-   // 9. Set up the parallel 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);
-#ifdef COMPLEX_VERSION
-   ParComplexLinearForm *b = new ParComplexLinearForm(fespace);
-   b->AddDomainIntegrator(new VectorFEDomainLFIntegrator(f), NULL);
-   b->AddBoundaryIntegrator(NULL,
-                            new VectorFEDomainLFIntegrator(omegaE));  // im part
-#else
-   // Real version
-   ParLinearForm *b = new ParLinearForm(fespace);
-   b->AddDomainIntegrator(new VectorFEDomainLFIntegrator(f));
-#endif
-
-#ifdef NEUMANN
-   b->AddBoundaryIntegrator(new VectorFEBoundaryTangentLFIntegrator(curlE_Re),
-                            NULL);
-#endif
-
-   b->Assemble();
-
-   // 10. Define the solution vector x as a parallel 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.
-   /*
-   ParGridFunction x(fespace);
-   VectorFunctionCoefficient E(sdim, E_exact);
-   x.ProjectCoefficient(E);
-   */
-
-#ifdef COMPLEX_VERSION
-   // Complex version
-   ParComplexGridFunction x(fespace);
-   x = 0.0;
-   Vector zero(sdim);
-   zero = 0.0;
-   VectorConstantCoefficient E_Im(zero);
-   //x.ProjectBdrCoefficientTangent(E_Re, E_Im, ess_bdr);
-   x.ProjectCoefficient(E_Re, E_Im);
-#else
-   ParGridFunction x(fespace);
-   x = 0.0;
-   x.ProjectCoefficient(E_Re);
-#endif
-
-   // 11. Set up the parallel 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);
-#ifdef INDEFINITE
-   Coefficient *sigma = new ConstantCoefficient(
-      -omega*omega); // indefinite -, definite +
-#else
-   Coefficient *sigma = new ConstantCoefficient(
-      omega*omega); // indefinite -, definite +
-#endif
-   Coefficient *abssigma = new ConstantCoefficient(omega*omega);
-   Coefficient *imabs = new ConstantCoefficient(imscale);  // im part
-   //Coefficient *imabs = new ConstantCoefficient(0.0);  // im part
-   //Coefficient *im = new ConstantCoefficient(0.0);
-
-#ifdef COMPLEX_VERSION
-   // Complex version
-   ParSesquilinearForm *a = new ParSesquilinearForm(fespace);
-   if (pa) { a->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
-   a->AddDomainIntegrator(new CurlCurlIntegrator(*muinv), NULL);
-   //a->AddDomainIntegrator(new VectorFEMassIntegrator(*sigma), new VectorFEMassIntegrator(*im));
-   a->AddDomainIntegrator(new VectorFEMassIntegrator(*sigma), NULL);
-   a->AddBoundaryIntegrator(NULL, new VectorFEMassIntegrator(*im));  // im part
-#else
-   // Real version
-   ParBilinearForm *a = new ParBilinearForm(fespace);
-   if (pa) { a->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
-   a->AddDomainIntegrator(new CurlCurlIntegrator(*muinv));
-   //a->AddDomainIntegrator(new VectorFEMassIntegrator(*sigma), new VectorFEMassIntegrator(*im));
-   a->AddDomainIntegrator(new VectorFEMassIntegrator(*sigma));
-#endif
-
-   // 12. Assemble the parallel bilinear form and the corresponding linear
-   //     system, applying any necessary transformations such as: parallel
-   //     assembly, 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);
-
-   ParBilinearForm a_Re(fespace);
-   a_Re.AddDomainIntegrator(new CurlCurlIntegrator(*muinv));
-   a_Re.AddDomainIntegrator(new VectorFEMassIntegrator(*abssigma));
-
-   //if (pa) { a_Re.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
-   a_Re.SetAssemblyLevel(AssemblyLevel::PARTIAL);
-   a_Re.Assemble();
-
-   OperatorPtr A_Re;
-   a_Re.FormSystemMatrix(ess_tdof_list, A_Re);
-
-   ParBilinearForm a_Im(fespace);
-   a_Im.AddBoundaryIntegrator(new VectorFEMassIntegrator(*imabs));
-   a_Im.Assemble();
-
-   OperatorPtr A_Im;
-   a_Im.FormSystemMatrix(ess_tdof_list, A_Im);
-
-   // 13. Solve the system AX=B using PCG with the AMS preconditioner from hypre
-   //     (in the full assembly case) or CG with Jacobi preconditioner (in the
-   //     partial assembly case).
-
-   Array<int> offsets(3);
-   offsets[0] = 0;
-   offsets[1] = fespace->GetTrueVSize();
-   offsets[2] = fespace->GetTrueVSize();
-   offsets.PartialSum();
-
-   //OperatorJacobiSmoother massJacobi(a_Im, ess_tdof_list);
-
-   StopWatch sw;
-   sw.Clear();
-   sw.Start();
-
-   if (pa) // Jacobi preconditioning in partial assembly mode
-   {
-      MFEM_VERIFY(false, "TODO");
-      //OperatorJacobiSmoother Jacobi(*a, ess_tdof_list);
-
-      CGSolver cg(MPI_COMM_WORLD);
-      cg.SetRelTol(1e-12);
-      cg.SetMaxIter(1000);
-      cg.SetPrintLevel(1);
-      cg.SetOperator(*A);
-      //cg.SetPreconditioner(Jacobi);
-      cg.Mult(B, X);
-   }
-   else
-   {
-      if (myid == 0)
-      {
-         cout << "Size of linear system: "
-              << A.As<HypreParMatrix>()->GetGlobalNumRows() << endl;
-      }
-
-      //HypreAMS ams(*A_Re.As<HypreParMatrix>(), fespace);
-
-      // One option is to use the standard real-valued MatrixFreeAMS to precondition
-      // the real part of the complex system in the PMHSS preconditioner (BlockDiagonalPreconditioner).
-      // Another option is to use complex MatrixFreeAMS to precondition the
-      // complex system without PMHSS and without a BlockDiagonalPreconditioner.
-      //#define COMPLEX_AMS
-
-#ifdef MFEM_USE_AMGX
-      bool useAmgX = false;
-      cout << "Built with AMGX, using AMGX " << useAmgX << endl;
-      MatrixFreeAMS ams(a_Re, *A_Re, *fespace, muinv, abssigma, im, imabs, NULL,
-                        ess_bdr, useAmgX);
-      MatrixFreeAMS ams(a_Re, *A_Re, *fespace, muinv, abssigma, NULL, NULL, ess_bdr,
-                        useAmgX);
-#ifdef COMPLEX_AMS
-      MFEM_VERIFY(false, "TODO");
-#endif
-
-#else
-      cout << "Not built with AMGX" << endl;
-#ifdef COMPLEX_AMS
-      MatrixFreeAMS ams(a_Re, *A_Re, A.Ptr(), *fespace, muinv, abssigma, im, imabs,
-                        NULL, ess_bdr);
-#else
-      MatrixFreeAMS ams(a_Re, *A_Re, NULL, *fespace, muinv, abssigma, NULL, NULL,
-                        NULL, ess_bdr);
-#endif
-#endif
-
-#ifdef COMPLEX_VERSION
-
-#ifdef COMPLEX_AMS
-      //MFEM_VERIFY(false, "TODO");
-#else
-      BlockDiagonalPreconditioner BlockDP(offsets);
-      BlockDP.SetDiagonalBlock(0, &ams);
-      BlockDP.SetDiagonalBlock(1, &ams);
-
-      /*
-      BlockDiagonalPreconditioner BlockDP_Im(offsets);
-      BlockDP_Im.SetDiagonalBlock(0, &massJacobi); // TODO: this won't work if it has zeros on diagonal
-      BlockDP_Im.SetDiagonalBlock(1, &massJacobi);
-      */
-
-      //Complex_PMHSS PMHSS(A_Re, A_Im, &BlockDP, &BlockDP_Im);
-      //Complex_PMHSS PMHSS(A_Re, A_Im, &BlockDP, NULL, 2.0 * omega);
-      //Complex_PMHSS PMHSS(A_Re, A_Im, &BlockDP, NULL, omega);
-      Complex_PMHSS PMHSS(A_Re.Ptr(), A_Im.Ptr(), &BlockDP, NULL, 1.0);
-
-      ComplexOperator AspdComplex(A_Re.Ptr(), A_Im.Ptr(), false, false);
-
-      GMRESSolver PMHSSgmres(MPI_COMM_WORLD);
-      PMHSSgmres.SetPrintLevel(1);
-      PMHSSgmres.SetKDim(100);
-      PMHSSgmres.SetMaxIter(100);
-      PMHSSgmres.SetRelTol(1e-6);
-      PMHSSgmres.SetAbsTol(0.0);
-      PMHSSgmres.SetOperator(AspdComplex);
-      PMHSSgmres.SetPreconditioner(PMHSS);
-#endif
-
-      GMRESSolver gmres(MPI_COMM_WORLD);
-      gmres.SetPrintLevel(1);
-      gmres.SetKDim(1000);
-      gmres.SetMaxIter(100);
-      gmres.SetRelTol(1e-8);
-      gmres.SetAbsTol(0.0);
-      gmres.SetOperator(*A);
-      //gmres.SetPreconditioner(BlockDP);
-#ifdef COMPLEX_AMS
-      //MFEM_VERIFY(false, "TODO");
-      gmres.SetPreconditioner(ams);
-#else
-      gmres.SetPreconditioner(PMHSS);
-      //gmres.SetPreconditioner(PMHSSgmres);
-#endif
-
-#else
-      GMRESSolver gmres(MPI_COMM_WORLD);
-      gmres.SetPrintLevel(1);
-      gmres.SetKDim(1000);
-      gmres.SetMaxIter(100);
-      gmres.SetRelTol(1e-8);
-      gmres.SetAbsTol(0.0);
-      gmres.SetOperator(*A);
-      gmres.SetPreconditioner(ams);
-#endif
-
-      gmres.Mult(B, X);
-   }
-
-   sw.Stop();
-   mfem::out << "Total solve time " <<sw.RealTime() << endl;
-
-   // 14. Recover the parallel grid function corresponding to X. This is the
-   //     local finite element solution on each processor.
-   a->RecoverFEMSolution(X, *b, x);
-
-   // 15. Compute and print the L^2 norm of the error.
-   {
-#ifdef COMPLEX_VERSION
-      double err = x.real().ComputeL2Error(E_Re);
-#else
-      double err = x.ComputeL2Error(E_Re);
-#endif
-      if (myid == 0)
-      {
-         cout << "\n|| E_h - E ||_{L^2} = " << err << '\n' << endl;
-      }
-   }
-
-   // 16. Save the refined mesh and the solution in parallel. This output can
-   //     be viewed later using GLVis: "glvis -np <np> -m mesh -g sol".
-   {
-      ostringstream mesh_name, sol_name;
-      mesh_name << "mesh." << setfill('0') << setw(6) << myid;
-      sol_name << "sol." << setfill('0') << setw(6) << myid;
-
-      ofstream mesh_ofs(mesh_name.str().c_str());
-      mesh_ofs.precision(8);
-      pmesh->Print(mesh_ofs);
-
-      ofstream sol_ofs(sol_name.str().c_str());
-      sol_ofs.precision(8);
-#ifdef COMPLEX_VERSION
-      x.real().Save(sol_ofs);
-#else
-      x.Save(sol_ofs);
-#endif
-   }
-
-   // 17. Send the solution by socket to a GLVis server.
-   if (visualization)
-   {
-      char vishost[] = "localhost";
-      int  visport   = 19916;
-      socketstream sol_sock(vishost, visport);
-      sol_sock << "parallel " << num_procs << " " << myid << "\n";
-      sol_sock.precision(8);
-#ifdef COMPLEX_VERSION
-      sol_sock << "solution\n" << *pmesh << x.real() << flush;
-#else
-      sol_sock << "solution\n" << *pmesh << x << flush;
-#endif
-   }
-
-   // 18. Free the used memory.
-   delete a;
-   delete sigma;
-   delete muinv;
-   delete b;
-   delete fespace;
-   delete fec;
-   delete pmesh;
-
-   MPI_Finalize();
-
-   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 curlE_exact(const Vector &x, Vector &curl)
-{
-   if (dim == 3)
-   {
-      curl(0) = kappa * cos(kappa * x(2));
-      curl(1) = kappa * cos(kappa * x(0));
-      curl(2) = kappa * cos(kappa * x(1));
-   }
-   else
-   {
-      MFEM_VERIFY(false, "");
-   }
-}
-
-void f_exact(const Vector &x, Vector &f)
-{
-   if (dim == 3)
-   {
-      // indefinite -m, definite +m
-      const double c = kappa * kappa;
-#ifdef INDEFINITE
-      const double m = -omega * omega;
-#else
-      const double m = omega * omega;
-#endif
-      f(0) = (c + m) * sin(kappa * x(1));
-      f(1) = (c + m) * sin(kappa * x(2));
-      f(2) = (c + m) * 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; }
-   }
-}
@@ -13,7 +13,6 @@ set(SRCS
  bilinearform.cpp
  bilinearform_ext.cpp
  bilininteg.cpp
-  bilininteg_br2.cpp
  bilininteg_convection_pa.cpp
  bilininteg_convection_ea.cpp
  bilininteg_dgtrace_pa.cpp
@@ -1437,9 +1437,9 @@ void MixedBilinearForm::Assemble (int skip_zeros)
         ftr = mesh->GetBdrFaceTransformations(i);
         if (ftr)
         {
-            trial_fes->GetFaceVDofs(ftr->ElementNo, tr_vdofs);
+            trial_fes->GetFaceVDofs(i, tr_vdofs);
            test_fes->GetElementVDofs(ftr->Elem1No, te_vdofs);
-            trial_face_fe = trial_fes->GetFaceElement(ftr->ElementNo);
+            trial_face_fe = trial_fes->GetFaceElement(i);
            test_fe1 = test_fes->GetFE(ftr->Elem1No);
            // The test_fe2 object is really a dummy and not used on the
            // boundaries, but we can't dereference a NULL pointer, and we don't
@@ -1770,41 +1770,9 @@ MixedBilinearForm::~MixedBilinearForm()
   delete ext;
 }

-void DiscreteLinearOperator::SetAssemblyLevel(AssemblyLevel assembly_level)
-{
-   if (ext)
-   {
-      MFEM_ABORT("the assembly level has already been set!");
-   }
-   assembly = assembly_level;
-   switch (assembly)
-   {
-      case AssemblyLevel::LEGACYFULL:
-      case AssemblyLevel::FULL:
-         // Use the original implementation for now
-         break;
-      case AssemblyLevel::ELEMENT:
-         mfem_error("Element assembly not supported yet... stay tuned!");
-         break;
-      case AssemblyLevel::PARTIAL:
-         ext = new PADiscreteLinearOperatorExtension(this);
-         break;
-      case AssemblyLevel::NONE:
-         mfem_error("Matrix-free action not supported yet... stay tuned!");
-         break;
-      default:
-         mfem_error("Unknown assembly level");
-   }
-}

 void DiscreteLinearOperator::Assemble(int skip_zeros)
 {
-   if (ext)
-   {
-      ext->Assemble();
-      return;
-   }
-
   Array<int> dom_vdofs, ran_vdofs;
   ElementTransformation *T;
   const FiniteElement *dom_fe, *ran_fe;
@@ -376,13 +376,6 @@ public:
   /// Get the output finite element space prolongation matrix
   virtual const Operator *GetOutputProlongation() const
   { return GetProlongation(); }
-   /** @brief Returns the output fe space restriction matrix, transposed
-
-       Logically, this is the transpose of GetOutputRestriction, but in
-       practice it is convenient to have it in transposed form for
-       construction of RAP operators in matrix-free methods. */
-   virtual const Operator *GetOutputRestrictionTranspose() const
-   { return GetOutputProlongation(); }
   /// Get the output finite element space restriction matrix
   virtual const Operator *GetOutputRestriction() const
   { return GetRestriction(); }
@@ -984,18 +977,9 @@ public:
   /// Access all interpolators added with AddDomainInterpolator().
   Array<BilinearFormIntegrator*> *GetDI() { return &dbfi; }

-   /// Set the desired assembly level. The default is AssemblyLevel::FULL.
-   /** This method must be called before assembly. */
-   void SetAssemblyLevel(AssemblyLevel assembly_level);
-
   /** @brief Construct the internal matrix representation of the discrete
       linear operator. */
   virtual void Assemble(int skip_zeros = 1);
-
-   /** @brief Get the output finite element space restriction matrix in
-       transposed form. */
-   virtual const Operator *GetOutputRestrictionTranspose() const
-   { return test_fes->GetRestrictionTransposeOperator(); }
 };

 }
@@ -1021,6 +1021,7 @@ void PAMixedBilinearFormExtension::Update()
      localTrial.UseDevice(true);
      localTrial.SetSize(elem_restrict_trial->Height(),
                         Device::GetMemoryType());
+
   }
   if (elem_restrict_test)
   {
@@ -1220,134 +1221,4 @@ void PAMixedBilinearFormExtension::AssembleDiagonal_ADAt(const Vector &D,
   }
 }

-PADiscreteLinearOperatorExtension::PADiscreteLinearOperatorExtension(
-   DiscreteLinearOperator *linop) :
-   PAMixedBilinearFormExtension(linop)
-{
-}
-
-const
-Operator *PADiscreteLinearOperatorExtension::GetOutputRestrictionTranspose()
-const
-{
-   return a->GetOutputRestrictionTranspose();
-}
-
-void PADiscreteLinearOperatorExtension::Assemble()
-{
-   Array<BilinearFormIntegrator*> &integrators = *a->GetDBFI();
-   const int integratorCount = integrators.Size();
-   for (int i = 0; i < integratorCount; ++i)
-   {
-      integrators[i]->AssemblePA(*trialFes, *testFes);
-   }
-
-   test_multiplicity.UseDevice(true);
-   test_multiplicity.SetSize(elem_restrict_test->Width()); // l-vector
-   Vector ones(elem_restrict_test->Height()); // e-vector
-   ones = 1.0;
-
-   const ElementRestriction* elem_restrict =
-      dynamic_cast<const ElementRestriction*>(elem_restrict_test);
-   if (elem_restrict)
-   {
-      elem_restrict->MultTransposeUnsigned(ones, test_multiplicity);
-   }
-   else
-   {
-      mfem_error("A real ElementRestriction is required in this setting!");
-   }
-
-   auto tm = test_multiplicity.ReadWrite();
-   MFEM_FORALL(i, test_multiplicity.Size(),
-   {
-      tm[i] = 1.0 / tm[i];
-   });
-}
-
-void PADiscreteLinearOperatorExtension::AddMult(
-   const Vector &x, Vector &y, const double c) const
-{
-   Array<BilinearFormIntegrator*> &integrators = *a->GetDBFI();
-   const int iSz = integrators.Size();
-
-   // * G operation
-   SetupMultInputs(elem_restrict_trial, x, localTrial,
-                   elem_restrict_test, y, localTest, c);
-
-   // * B^TDB operation
-   for (int i = 0; i < iSz; ++i)
-   {
-      integrators[i]->AddMultPA(localTrial, localTest);
-   }
-
-   // do a kind of "set" rather than "add" in the below
-   // operation as compared to the BilinearForm case
-   // * G^T operation (kind of...)
-   const ElementRestriction* elem_restrict =
-      dynamic_cast<const ElementRestriction*>(elem_restrict_test);
-   if (elem_restrict)
-   {
-      tempY.SetSize(y.Size());
-      elem_restrict->MultLeftInverse(localTest, tempY);
-      y += tempY;
-   }
-   else
-   {
-      mfem_error("In this setting you need a real ElementRestriction!");
-   }
-}
-
-void PADiscreteLinearOperatorExtension::AddMultTranspose(
-   const Vector &x, Vector &y, const double c) const
-{
-   Array<BilinearFormIntegrator*> &integrators = *a->GetDBFI();
-   const int iSz = integrators.Size();
-
-   // do a kind of "set" rather than "add" in the below
-   // operation as compared to the BilinearForm case
-   // * G operation (kinda)
-   Vector xscaled(x);
-   MFEM_VERIFY(x.Size() == test_multiplicity.Size(), "Input vector of wrong size");
-   auto xs = xscaled.ReadWrite();
-   auto tm = test_multiplicity.Read();
-   MFEM_FORALL(i, x.Size(),
-   {
-      xs[i] *= tm[i];
-   });
-   SetupMultInputs(elem_restrict_test, xscaled, localTest,
-                   elem_restrict_trial, y, localTrial, c);
-
-   // * B^TD^TB operation
-   for (int i = 0; i < iSz; ++i)
-   {
-      integrators[i]->AddMultTransposePA(localTest, localTrial);
-   }
-
-   // * G^T operation
-   if (elem_restrict_trial)
-   {
-      tempY.SetSize(y.Size());
-      elem_restrict_trial->MultTranspose(localTrial, tempY);
-      y += tempY;
-   }
-   else
-   {
-      mfem_error("Trial ElementRestriction not defined");
-   }
-}
-
-void PADiscreteLinearOperatorExtension::FormRectangularSystemOperator(
-   const Array<int>& ess1, const Array<int>& ess2, OperatorHandle &A)
-{
-   const Operator *Pi = this->GetProlongation();
-   const Operator *RoT = this->GetOutputRestrictionTranspose();
-   Operator *rap = SetupRAP(Pi, RoT);
-
-   RectangularConstrainedOperator *Arco
-      = new RectangularConstrainedOperator(rap, ess1, ess2, rap != this);
-
-   A.Reset(Arco);
-}
-
 } // namespace mfem
@@ -21,7 +21,6 @@ namespace mfem

 class BilinearForm;
 class MixedBilinearForm;
-class DiscreteLinearOperator;

 /// Class extending the BilinearForm class to support different AssemblyLevels.
 /**  FA - Full Assembly
@@ -213,7 +212,7 @@ protected:
   mutable Vector localTrial, localTest, tempY;
   const Operator *elem_restrict_trial; // Not owned
   const Operator *elem_restrict_test;  // Not owned
-
+private:
   /// Helper function to set up inputs/outputs for Mult or MultTranspose
   void SetupMultInputs(const Operator *elem_restrict_x,
                        const Vector &x, Vector &localX,
@@ -259,35 +258,6 @@ public:
   void Update();
 };

-
-/**
-   @brief Partial assembly extension for DiscreteLinearOperator
-
-   This acts very much like PAMixedBilinearFormExtension, but its
-   FormRectangularSystemOperator implementation emulates 'Set' rather than
-   'Add' in the assembly case.
-*/
-class PADiscreteLinearOperatorExtension : public PAMixedBilinearFormExtension
-{
-public:
-   PADiscreteLinearOperatorExtension(DiscreteLinearOperator *linop);
-
-   /// Partial assembly of all internal integrators
-   void Assemble();
-
-   void AddMult(const Vector &x, Vector &y, const double c) const;
-
-   void AddMultTranspose(const Vector &x, Vector &y, const double c=1.0) const;
-
-   void FormRectangularSystemOperator(const Array<int>&, const Array<int>&,
-                                      OperatorHandle& A);
-
-   const Operator * GetOutputRestrictionTranspose() const;
-
-private:
-   Vector test_multiplicity;
-};
-
 }

 #endif
@@ -22,14 +22,14 @@ namespace mfem

 void BilinearFormIntegrator::AssemblePA(const FiniteElementSpace&)
 {
-   mfem_error ("BilinearFormIntegrator::AssemblePA(fes)\n"
+   mfem_error ("BilinearFormIntegrator::AssemblePA(...)\n"
               "   is not implemented for this class.");
 }

 void BilinearFormIntegrator::AssemblePA(const FiniteElementSpace&,
                                        const FiniteElementSpace&)
 {
-   mfem_error ("BilinearFormIntegrator::AssemblePA(fes, fes)\n"
+   mfem_error ("BilinearFormIntegrator::AssemblePA(...)\n"
               "   is not implemented for this class.");
 }

@@ -92,7 +92,7 @@ void BilinearFormIntegrator::AddMultPA(const Vector &, Vector &) const

 void BilinearFormIntegrator::AddMultTransposePA(const Vector &, Vector &) const
 {
-   mfem_error ("BilinearFormIntegrator::AddMultTransposePA(...)\n"
+   mfem_error ("BilinearFormIntegrator::MultAssembledTranspose(...)\n"
               "   is not implemented for this class.");
 }

@@ -1913,12 +1913,12 @@ void VectorFEMassIntegrator::AssembleElementMatrix(
   double w;

 #ifdef MFEM_THREAD_SAFE
-   Vector D(DQ ? DQ->GetVDim() : 0);
+   Vector D(VQ ? VQ->GetVDim() : 0);
   DenseMatrix trial_vshape(dof, spaceDim);
   DenseMatrix K(MQ ? MQ->GetVDim() : 0, MQ ? MQ->GetVDim() : 0);
 #else
   trial_vshape.SetSize(dof, spaceDim);
-   D.SetSize(DQ ? DQ->GetVDim() : 0);
+   D.SetSize(VQ ? VQ->GetVDim() : 0);
   K.SetSize(MQ ? MQ->GetVDim() : 0, MQ ? MQ->GetVDim() : 0);
 #endif
   DenseMatrix tmp(trial_vshape.Height(), K.Width());
@@ -1950,9 +1950,9 @@ void VectorFEMassIntegrator::AssembleElementMatrix(
         Mult(trial_vshape,K,tmp);
         AddMultABt(tmp,trial_vshape,elmat);
      }
-      else if (DQ)
+      else if (VQ)
      {
-         DQ->Eval(D, Trans, ip);
+         VQ->Eval(D, Trans, ip);
         D *= w;
         AddMultADAt(trial_vshape, D, elmat);
      }
@@ -1984,12 +1984,12 @@ void VectorFEMassIntegrator::AssembleElementMatrix2(
 #ifdef MFEM_THREAD_SAFE
      DenseMatrix trial_vshape(trial_dof, spaceDim);
      Vector shape(test_dof);
-      Vector D(DQ ? DQ->GetVDim() : 0);
+      Vector D(VQ ? VQ->GetVDim() : 0);
      DenseMatrix K(MQ ? MQ->GetVDim() : 0, MQ ? MQ->GetVDim() : 0);
 #else
      trial_vshape.SetSize(trial_dof, spaceDim);
      shape.SetSize(test_dof);
-      D.SetSize(DQ ? DQ->GetVDim() : 0);
+      D.SetSize(VQ ? VQ->GetVDim() : 0);
      K.SetSize(MQ ? MQ->GetVDim() : 0, MQ ? MQ->GetVDim() : 0);
 #endif

@@ -2013,9 +2013,9 @@ void VectorFEMassIntegrator::AssembleElementMatrix2(
         test_fe.CalcShape(ip, shape);

         w = ip.weight * Trans.Weight();
-         if (DQ)
+         if (VQ)
         {
-            DQ->Eval(D, Trans, ip);
+            VQ->Eval(D, Trans, ip);
            D *= w;
            for (int d = 0; d < vdim; d++)
            {
@@ -2081,12 +2081,12 @@ void VectorFEMassIntegrator::AssembleElementMatrix2(
 #ifdef MFEM_THREAD_SAFE
      DenseMatrix trial_vshape(trial_dof,spaceDim);
      DenseMatrix test_vshape(test_dof,spaceDim);
-      Vector D(DQ ? DQ->GetVDim() : 0);
+      Vector D(VQ ? VQ->GetVDim() : 0);
      DenseMatrix K(MQ ? MQ->GetVDim() : 0, MQ ? MQ->GetVDim() : 0);
 #else
      trial_vshape.SetSize(trial_dof,spaceDim);
      test_vshape.SetSize(test_dof,spaceDim);
-      D.SetSize(DQ ? DQ->GetVDim() : 0);
+      D.SetSize(VQ ? VQ->GetVDim() : 0);
      K.SetSize(MQ ? MQ->GetVDim() : 0, MQ ? MQ->GetVDim() : 0);
 #endif
      DenseMatrix tmp(test_vshape.Height(), K.Width());
@@ -2118,9 +2118,9 @@ void VectorFEMassIntegrator::AssembleElementMatrix2(
            Mult(test_vshape,K,tmp);
            AddMultABt(tmp,trial_vshape,elmat);
         }
-         else if (DQ)
+         else if (VQ)
         {
-            DQ->Eval(D, Trans, ip);
+            VQ->Eval(D, Trans, ip);
            D *= w;
            AddMultADBt(test_vshape,D,trial_vshape,elmat);
         }
@@ -3515,47 +3515,6 @@ VectorScalarProductInterpolator::AssembleElementMatrix2(
 }


-void
-ScalarCrossProductInterpolator::AssembleElementMatrix2(
-   const FiniteElement &dom_fe,
-   const FiniteElement &ran_fe,
-   ElementTransformation &Trans,
-   DenseMatrix &elmat)
-{
-   // Vector coefficient product with vector shape functions
-   struct VCrossVShapeCoefficient : public VectorCoefficient
-   {
-      VectorCoefficient &VQ;
-      const FiniteElement &fe;
-      DenseMatrix vshape;
-      Vector vc;
-
-      VCrossVShapeCoefficient(VectorCoefficient &vq, const FiniteElement &fe_)
-         : VectorCoefficient(fe_.GetDof()), VQ(vq), fe(fe_),
-           vshape(vdim, vq.GetVDim()), vc(vq.GetVDim()) { }
-
-      virtual void Eval(Vector &V, ElementTransformation &T,
-                        const IntegrationPoint &ip)
-      {
-         V.SetSize(vdim);
-         VQ.Eval(vc, T, ip);
-         fe.CalcPhysVShape(T, vshape);
-         for (int k = 0; k < vdim; k++)
-         {
-            V(k) = vc(0) * vshape(k,1) - vc(1) * vshape(k,0);
-         }
-      }
-   };
-
-   VCrossVShapeCoefficient dom_shape_coeff(*VQ, dom_fe);
-
-   elmat.SetSize(ran_fe.GetDof(),dom_fe.GetDof());
-
-   Vector elmat_as_vec(elmat.Data(), elmat.Height()*elmat.Width());
-
-   ran_fe.Project(dom_shape_coeff, Trans, elmat_as_vec);
-}
-
 void
 VectorCrossProductInterpolator::AssembleElementMatrix2(
   const FiniteElement &dom_fe,
@@ -465,8 +465,8 @@ protected:
      : same_calc_shape(false), Q(NULL), VQ(NULL), DQ(NULL), MQ(NULL) {}
   MixedVectorIntegrator(Coefficient &q)
      : same_calc_shape(false), Q(&q), VQ(NULL), DQ(NULL), MQ(NULL) {}
-   MixedVectorIntegrator(VectorCoefficient &vq, bool diag = true)
-      : same_calc_shape(false), Q(NULL), VQ(diag?NULL:&vq), DQ(diag?&vq:NULL),
+   MixedVectorIntegrator(VectorCoefficient &dq, bool diag = true)
+      : same_calc_shape(false), Q(NULL), VQ(diag?NULL:&dq), DQ(diag?&dq:NULL),
        MQ(NULL) {}
   MixedVectorIntegrator(MatrixCoefficient &mq)
      : same_calc_shape(false), Q(NULL), VQ(NULL), DQ(NULL), MQ(&mq) {}
@@ -503,7 +503,7 @@ protected:

   Coefficient *Q;
   VectorCoefficient *VQ;
-   DiagonalMatrixCoefficient *DQ;
+   VectorCoefficient *DQ;
   MatrixCoefficient *MQ;

 private:
@@ -901,7 +901,7 @@ public:
   MixedVectorMassIntegrator() { same_calc_shape = true; }
   MixedVectorMassIntegrator(Coefficient &q)
      : MixedVectorIntegrator(q) { same_calc_shape = true; }
-   MixedVectorMassIntegrator(DiagonalMatrixCoefficient &dq)
+   MixedVectorMassIntegrator(VectorCoefficient &dq)
      : MixedVectorIntegrator(dq, true) { same_calc_shape = true; }
   MixedVectorMassIntegrator(MatrixCoefficient &mq)
      : MixedVectorIntegrator(mq) { same_calc_shape = true; }
@@ -1019,7 +1019,7 @@ public:
   MixedGradGradIntegrator() { same_calc_shape = true; }
   MixedGradGradIntegrator(Coefficient &q)
      : MixedVectorIntegrator(q) { same_calc_shape = true; }
-   MixedGradGradIntegrator(DiagonalMatrixCoefficient &dq)
+   MixedGradGradIntegrator(VectorCoefficient &dq)
      : MixedVectorIntegrator(dq, true) { same_calc_shape = true; }
   MixedGradGradIntegrator(MatrixCoefficient &mq)
      : MixedVectorIntegrator(mq) { same_calc_shape = true; }
@@ -1107,7 +1107,7 @@ public:
   MixedCurlCurlIntegrator() { same_calc_shape = true; }
   MixedCurlCurlIntegrator(Coefficient &q)
      : MixedVectorIntegrator(q) { same_calc_shape = true; }
-   MixedCurlCurlIntegrator(DiagonalMatrixCoefficient &dq)
+   MixedCurlCurlIntegrator(VectorCoefficient &dq)
      : MixedVectorIntegrator(dq, true) { same_calc_shape = true; }
   MixedCurlCurlIntegrator(MatrixCoefficient &mq)
      : MixedVectorIntegrator(mq) { same_calc_shape = true; }
@@ -1651,7 +1651,7 @@ public:
   MixedVectorGradientIntegrator() {}
   MixedVectorGradientIntegrator(Coefficient &q)
      : MixedVectorIntegrator(q) {}
-   MixedVectorGradientIntegrator(DiagonalMatrixCoefficient &dq)
+   MixedVectorGradientIntegrator(VectorCoefficient &dq)
      : MixedVectorIntegrator(dq, true) {}
   MixedVectorGradientIntegrator(MatrixCoefficient &mq)
      : MixedVectorIntegrator(mq) {}
@@ -1705,7 +1705,7 @@ public:
   MixedVectorCurlIntegrator() {}
   MixedVectorCurlIntegrator(Coefficient &q)
      : MixedVectorIntegrator(q) {}
-   MixedVectorCurlIntegrator(DiagonalMatrixCoefficient &dq)
+   MixedVectorCurlIntegrator(VectorCoefficient &dq)
      : MixedVectorIntegrator(dq, true) {}
   MixedVectorCurlIntegrator(MatrixCoefficient &mq)
      : MixedVectorIntegrator(mq) {}
@@ -1760,7 +1760,7 @@ public:
   MixedVectorWeakCurlIntegrator() {}
   MixedVectorWeakCurlIntegrator(Coefficient &q)
      : MixedVectorIntegrator(q) {}
-   MixedVectorWeakCurlIntegrator(DiagonalMatrixCoefficient &dq)
+   MixedVectorWeakCurlIntegrator(VectorCoefficient &dq)
      : MixedVectorIntegrator(dq, true) {}
   MixedVectorWeakCurlIntegrator(MatrixCoefficient &mq)
      : MixedVectorIntegrator(mq) {}
@@ -1813,7 +1813,7 @@ public:
   MixedVectorWeakDivergenceIntegrator() {}
   MixedVectorWeakDivergenceIntegrator(Coefficient &q)
      : MixedVectorIntegrator(q) {}
-   MixedVectorWeakDivergenceIntegrator(DiagonalMatrixCoefficient &dq)
+   MixedVectorWeakDivergenceIntegrator(VectorCoefficient &dq)
      : MixedVectorIntegrator(dq, true) {}
   MixedVectorWeakDivergenceIntegrator(MatrixCoefficient &mq)
      : MixedVectorIntegrator(mq) {}
@@ -1844,10 +1844,8 @@ protected:
 };

 /** Class for integrating the bilinear form a(u,v) := (Q grad u, v) where Q is a
-    scalar coefficient, and v is a vector with components v_i in the same (H1) space
-    as u.
-
-    See also MixedVectorGradientIntegrator when v is in H(curl). */
+    scalar coefficient, and v is a vector with components v_i in the same space
+    as u. */
 class GradientIntegrator : public BilinearFormIntegrator
 {
 protected:
@@ -1902,7 +1900,6 @@ protected:
   Coefficient *Q;
   VectorCoefficient *VQ;
   MatrixCoefficient *MQ;
-   SymmetricMatrixCoefficient *SMQ;

 private:
   Vector vec, pointflux, shape;
@@ -1925,28 +1922,19 @@ private:
 public:
   /// Construct a diffusion integrator with coefficient Q = 1
   DiffusionIntegrator()
-      : Q(NULL), VQ(NULL), MQ(NULL), SMQ(NULL), maps(NULL), geom(NULL),
-        ceedDataPtr(NULL) { }
+      : Q(NULL), VQ(NULL), MQ(NULL), maps(NULL), geom(NULL), ceedDataPtr(NULL) { }

   /// Construct a diffusion integrator with a scalar coefficient q
   DiffusionIntegrator(Coefficient &q)
-      : Q(&q), VQ(NULL), MQ(NULL), SMQ(NULL), maps(NULL), geom(NULL),
-        ceedDataPtr(NULL) { }
+      : Q(&q), VQ(NULL), MQ(NULL), maps(NULL), geom(NULL), ceedDataPtr(NULL) { }

   /// Construct a diffusion integrator with a vector coefficient q
   DiffusionIntegrator(VectorCoefficient &q)
-      : Q(NULL), VQ(&q), MQ(NULL), SMQ(NULL), maps(NULL), geom(NULL),
-        ceedDataPtr(NULL) { }
+      : Q(NULL), VQ(&q), MQ(NULL), maps(NULL), geom(NULL), ceedDataPtr(NULL) { }

   /// Construct a diffusion integrator with a matrix coefficient q
   DiffusionIntegrator(MatrixCoefficient &q)
-      : Q(NULL), VQ(NULL), MQ(&q), SMQ(NULL), maps(NULL), geom(NULL),
-        ceedDataPtr(NULL) { }
-
-   /// Construct a diffusion integrator with a symmetric matrix coefficient q
-   DiffusionIntegrator(SymmetricMatrixCoefficient &q)
-      : Q(NULL), VQ(NULL), MQ(NULL), SMQ(&q), maps(NULL), geom(NULL),
-        ceedDataPtr(NULL) { }
+      : Q(NULL), VQ(NULL), MQ(&q), maps(NULL), geom(NULL), ceedDataPtr(NULL) { }

   virtual ~DiffusionIntegrator()
   {
@@ -2345,9 +2333,8 @@ private:

 protected:
   Coefficient *Q;
-   DiagonalMatrixCoefficient *DQ;
+   VectorCoefficient *DQ;
   MatrixCoefficient *MQ;
-   SymmetricMatrixCoefficient *SMQ;

   // PA extension
   Vector pa_data;
@@ -2358,18 +2345,14 @@ protected:
   bool symmetric = true; ///< False if using a nonsymmetric matrix coefficient

 public:
-   CurlCurlIntegrator() { Q = NULL; DQ = NULL; MQ = NULL; SMQ = NULL; }
+   CurlCurlIntegrator() { Q = NULL; DQ = NULL; MQ = NULL; }
   /// Construct a bilinear form integrator for Nedelec elements
   CurlCurlIntegrator(Coefficient &q, const IntegrationRule *ir = NULL) :
-      BilinearFormIntegrator(ir), Q(&q), DQ(NULL), MQ(NULL), SMQ(NULL) { }
-   CurlCurlIntegrator(DiagonalMatrixCoefficient &dq,
-                      const IntegrationRule *ir = NULL) :
-      BilinearFormIntegrator(ir), Q(NULL), DQ(&dq), MQ(NULL), SMQ(NULL) { }
+      BilinearFormIntegrator(ir), Q(&q) { DQ = NULL; MQ = NULL; }
+   CurlCurlIntegrator(VectorCoefficient &dq, const IntegrationRule *ir = NULL) :
+      BilinearFormIntegrator(ir), DQ(&dq) { Q = NULL; MQ = NULL; }
   CurlCurlIntegrator(MatrixCoefficient &mq, const IntegrationRule *ir = NULL) :
-      BilinearFormIntegrator(ir), Q(NULL), DQ(NULL), MQ(&mq), SMQ(NULL) { }
-   CurlCurlIntegrator(SymmetricMatrixCoefficient &smq,
-                      const IntegrationRule *ir = NULL) :
-      BilinearFormIntegrator(ir), Q(NULL), DQ(NULL), MQ(NULL), SMQ(&smq) { }
+      BilinearFormIntegrator(ir), MQ(&mq) { Q = NULL; DQ = NULL; }

   /* Given a particular Finite Element, compute the
      element curl-curl matrix elmat */
@@ -2426,9 +2409,8 @@ public:
 class VectorFEMassIntegrator: public BilinearFormIntegrator
 {
 private:
-   void Init(Coefficient *q, DiagonalMatrixCoefficient *dq, MatrixCoefficient *mq,
-             SymmetricMatrixCoefficient *smq)
-   { Q = q; DQ = dq; MQ = mq; SMQ = smq; }
+   void Init(Coefficient *q, VectorCoefficient *vq, MatrixCoefficient *mq)
+   { Q = q; VQ = vq; MQ = mq; }

 #ifndef MFEM_THREAD_SAFE
   Vector shape;
@@ -2441,9 +2423,8 @@ private:

 protected:
   Coefficient *Q;
-   DiagonalMatrixCoefficient *DQ;
+   VectorCoefficient *VQ;
   MatrixCoefficient *MQ;
-   SymmetricMatrixCoefficient *SMQ;

   // PA extension
   Vector pa_data;
@@ -2456,15 +2437,13 @@ protected:
   bool symmetric = true; ///< False if using a nonsymmetric matrix coefficient

 public:
-   VectorFEMassIntegrator() { Init(NULL, NULL, NULL, NULL); }
-   VectorFEMassIntegrator(Coefficient *_q) { Init(_q, NULL, NULL, NULL); }
-   VectorFEMassIntegrator(Coefficient &q) { Init(&q, NULL, NULL, NULL); }
-   VectorFEMassIntegrator(DiagonalMatrixCoefficient *_dq) { Init(NULL, _dq, NULL, NULL); }
-   VectorFEMassIntegrator(DiagonalMatrixCoefficient &dq) { Init(NULL, &dq, NULL, NULL); }
-   VectorFEMassIntegrator(MatrixCoefficient *_mq) { Init(NULL, NULL, _mq, NULL); }
-   VectorFEMassIntegrator(MatrixCoefficient &mq) { Init(NULL, NULL, &mq, NULL); }
-   VectorFEMassIntegrator(SymmetricMatrixCoefficient &smq) { Init(NULL, NULL, NULL, &smq); }
-   VectorFEMassIntegrator(SymmetricMatrixCoefficient *smq) { Init(NULL, NULL, NULL, smq); }
+   VectorFEMassIntegrator() { Init(NULL, NULL, NULL); }
+   VectorFEMassIntegrator(Coefficient *_q) { Init(_q, NULL, NULL); }
+   VectorFEMassIntegrator(Coefficient &q) { Init(&q, NULL, NULL); }
+   VectorFEMassIntegrator(VectorCoefficient *_vq) { Init(NULL, _vq, NULL); }
+   VectorFEMassIntegrator(VectorCoefficient &vq) { Init(NULL, &vq, NULL); }
+   VectorFEMassIntegrator(MatrixCoefficient *_mq) { Init(NULL, NULL, _mq); }
+   VectorFEMassIntegrator(MatrixCoefficient &mq) { Init(NULL, NULL, &mq); }

   virtual void AssembleElementMatrix(const FiniteElement &el,
                                      ElementTransformation &Trans,
@@ -2673,28 +2652,11 @@ public:
 /** Integrator for the DG form:
    alpha < rho_u (u.n) {v},[w] > + beta < rho_u |u.n| [v],[w] >,
    where v and w are the trial and test variables, respectively, and rho/u are
-    given scalar/vector coefficients. {v} represents the average value of v on
-    the face and [v] is the jump such that {v}=(v1+v2)/2 and [v]=(v1-v2) for the
-    face between elements 1 and 2. For boundary elements, v2=0. The vector
-    coefficient, u, is assumed to be continuous across the faces and when given
-    the scalar coefficient, rho, is assumed to be discontinuous. The integrator
-    uses the upwind value of rho, rho_u, which is value from the side into which
-    the vector coefficient, u, points.
-
-    One use case for this integrator is to discretize the operator -u.grad(v)
-    with a DG formulation. The resulting formulation uses the
-    ConvectionIntegrator (with coefficient u, and parameter alpha = -1) and the
-    transpose of the DGTraceIntegrator (with coefficient u, and parameters
-    alpha = 1, beta = -1/2 to use the upwind face flux). This discretization and
-    the handling of the inflow and outflow boundaries is illustrated in Example
-    9/9p.
-
-    Another use case for this integrator is to discretize the operator -div(u v)
-    with a DG formulation. The resulting formulation is conservative and
-    consists of the transpose of the ConvectionIntegrator (with coefficient u,
-    and parameter alpha = 1) plus the DGTraceIntegrator (with coefficient u, and
-    parameters alpha = -1, beta = -1/2 to use the upwind face flux).
-    */
+    given scalar/vector coefficients. The vector coefficient, u, is assumed to
+    be continuous across the faces and when given the scalar coefficient, rho,
+    is assumed to be discontinuous. The integrator uses the upwind value of rho,
+    rho_u, which is value from the side into which the vector coefficient, u,
+    points. */
 class DGTraceIntegrator : public BilinearFormIntegrator
 {
 protected:
@@ -2788,51 +2750,6 @@ public:
                                   DenseMatrix &elmat);
 };

-/** Integrator for the "BR2" diffusion stabilization term
-
-    sum_e eta (r_e([u]), r_e([v]))
-
-    where r_e is the lifting operator defined on each edge e. The parameter eta
-    can be chosen to be one to obtain a stable discretization. The constructor
-    for this integrator requires the finite element space because the lifting
-    operator depends on the element-wise inverse mass matrix.
-
-    BR2 stands for the second method of Bassi and Rebay:
-
-    - F. Bassi and S. Rebay. A high order discontinuous Galerkin method for
-      compressible turbulent flows. In B. Cockburn, G. E. Karniadakis, and
-      C.-W. Shu, editors, Discontinuous Galerkin Methods, pages 77–88. Springer
-      Berlin Heidelberg, 2000.
-    - D. N. Arnold, F. Brezzi, B. Cockburn, and L. D. Marini. Unified analysis
-      of discontinuous Galerkin methods for elliptic problems. SIAM Journal on
-      Numerical Analysis, 39(5):1749–1779, 2002.
-*/
-class DGDiffusionBR2Integrator : public BilinearFormIntegrator
-{
-protected:
-   double eta;
-
-   // Block factorizations of local mass matrices, with offsets for the case of
-   // not equally sized blocks (mixed meshes, p-refinement)
-   Array<double> Minv;
-   Array<int> ipiv;
-   Array<int> ipiv_offsets, Minv_offsets;
-
-   Vector shape1, shape2;
-
-   DenseMatrix R11, R12, R21, R22;
-   DenseMatrix MinvR11, MinvR12, MinvR21, MinvR22;
-   DenseMatrix Re, MinvRe;
-
-public:
-   DGDiffusionBR2Integrator(class FiniteElementSpace *fes, double e = 1.0);
-   using BilinearFormIntegrator::AssembleFaceMatrix;
-   virtual void AssembleFaceMatrix(const FiniteElement &el1,
-                                   const FiniteElement &el2,
-                                   FaceElementTransformations &Trans,
-                                   DenseMatrix &elmat);
-};
-
 /** Integrator for the DG elasticity form, for the formulations see:
    - PhD Thesis of Jonas De Basabe, High-Order Finite %Element Methods for
      Seismic Wave Propagation, UT Austin, 2009, p. 23, and references therein
@@ -2991,36 +2908,11 @@ class DiscreteInterpolator : public BilinearFormIntegrator { };
 class GradientInterpolator : public DiscreteInterpolator
 {
 public:
-   GradientInterpolator() : dofquad_fe(NULL) { }
-   virtual ~GradientInterpolator() { delete dofquad_fe; }
-
   virtual void AssembleElementMatrix2(const FiniteElement &h1_fe,
                                       const FiniteElement &nd_fe,
                                       ElementTransformation &Trans,
                                       DenseMatrix &elmat)
   { nd_fe.ProjectGrad(h1_fe, Trans, elmat); }
-
-   using BilinearFormIntegrator::AssemblePA;
-
-   /** @brief Setup method for PA data.
-
-       @param[in] trial_fes   H1 Lagrange space
-       @param[in] test_fes    H(curl) Nedelec space
-    */
-   virtual void AssemblePA(const FiniteElementSpace &trial_fes,
-                           const FiniteElementSpace &test_fes);
-
-   virtual void AddMultPA(const Vector &x, Vector &y) const;
-   virtual void AddMultTransposePA(const Vector &x, Vector &y) const;
-
-private:
-   /// 1D finite element that generates and owns the 1D DofToQuad maps below
-   FiniteElement * dofquad_fe;
-
-   bool B_id; // is the B basis operator (maps_C_C) the identity?
-   const DofToQuad *maps_C_C; // one-d map with Lobatto rows, Lobatto columns
-   const DofToQuad *maps_O_C; // one-d map with Legendre rows, Lobatto columns
-   int dim, ne, o_dofs1D, c_dofs1D;
 };


@@ -3035,24 +2927,6 @@ public:
                                       ElementTransformation &Trans,
                                       DenseMatrix &elmat)
   { ran_fe.Project(dom_fe, Trans, elmat); }
-
-   using BilinearFormIntegrator::AssemblePA;
-
-   virtual void AssemblePA(const FiniteElementSpace &trial_fes,
-                           const FiniteElementSpace &test_fes);
-
-   virtual void AddMultPA(const Vector &x, Vector &y) const;
-   virtual void AddMultTransposePA(const Vector &x, Vector &y) const;
-
-private:
-   /// 1D finite element that generates and owns the 1D DofToQuad maps below
-   FiniteElement * dofquad_fe;
-
-   const DofToQuad *maps_C_C; // one-d map with Lobatto rows, Lobatto columns
-   const DofToQuad *maps_O_C; // one-d map with Legendre rows, Lobatto columns
-   int dim, ne, o_dofs1D, c_dofs1D;
-
-   Vector pa_data;
 };


@@ -3152,22 +3026,6 @@ protected:
   VectorCoefficient *VQ;
 };

-/** Interpolator of the 2D cross product between a vector coefficient and an
-    H(curl)-conforming field onto an L2-conforming field. */
-class ScalarCrossProductInterpolator : public DiscreteInterpolator
-{
-public:
-   ScalarCrossProductInterpolator(VectorCoefficient & vc)
-      : VQ(&vc) { }
-
-   virtual void AssembleElementMatrix2(const FiniteElement &nd_fe,
-                                       const FiniteElement &l2_fe,
-                                       ElementTransformation &Trans,
-                                       DenseMatrix &elmat);
-protected:
-   VectorCoefficient *VQ;
-};
-
 /** Interpolator of the cross product between a vector coefficient and an
    H(curl)-conforming field onto an H(div)-conforming field. The range space
    can also be vector L2. */
@@ -1,242 +0,0 @@
-// Copyright (c) 2010-2020, 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 "bilininteg.hpp"
-#include "pfespace.hpp"
-#include <algorithm>
-
-namespace mfem
-{
-
-DGDiffusionBR2Integrator::DGDiffusionBR2Integrator(FiniteElementSpace *fes,
-                                                   double e) : eta(e)
-{
-   // Precompute local mass matrix inverses needed for the lifting operators
-   // First compute offsets and total size needed (e.g. for mixed meshes or
-   // p-refinement)
-   int nel = fes->GetNE();
-   Minv_offsets.SetSize(nel+1);
-   ipiv_offsets.SetSize(nel+1);
-   ipiv_offsets[0] = 0;
-   Minv_offsets[0] = 0;
-   for (int i=0; i<nel; ++i)
-   {
-      int dof = fes->GetFE(i)->GetDof();
-      ipiv_offsets[i+1] = ipiv_offsets[i] + dof;
-      Minv_offsets[i+1] = Minv_offsets[i] + dof*dof;
-   }
-
-#ifdef MFEM_USE_MPI
-   // When running in parallel, we also need to compute the local mass matrices
-   // of face neighbor elements
-   ParFiniteElementSpace *pfes = dynamic_cast<ParFiniteElementSpace *>(fes);
-   if (pfes != NULL)
-   {
-      ParMesh *pmesh = pfes->GetParMesh();
-      pfes->ExchangeFaceNbrData();
-      int nel_nbr = pmesh->GetNFaceNeighborElements();
-      Minv_offsets.SetSize(nel+nel_nbr+1);
-      ipiv_offsets.SetSize(nel+nel_nbr+1);
-      for (int i=0; i<nel_nbr; ++i)
-      {
-         int dof = pfes->GetFaceNbrFE(i)->GetDof();
-         ipiv_offsets[nel+i+1] = ipiv_offsets[nel+i] + dof;
-         Minv_offsets[nel+i+1] = Minv_offsets[nel+i] + dof*dof;
-      }
-      nel += nel_nbr;
-   }
-#endif
-   // The final "offset" is the total size of all the blocks
-   Minv.SetSize(Minv_offsets[nel]);
-   ipiv.SetSize(ipiv_offsets[nel]);
-
-   // Assemble the local mass matrices and compute LU factorization
-   MassIntegrator mi;
-   for (int i=0; i<nel; ++i)
-   {
-      const FiniteElement *fe = NULL;
-      ElementTransformation *tr = NULL;
-      if (i < fes->GetNE())
-      {
-         fe = fes->GetFE(i);
-         tr = fes->GetElementTransformation(i);
-      }
-      else
-      {
-#ifdef MFEM_USE_MPI
-         int inbr = i - fes->GetNE();
-         fe = pfes->GetFaceNbrFE(inbr);
-         tr = pfes->GetParMesh()->GetFaceNbrElementTransformation(inbr);
-#endif
-      }
-      int dof = fe->GetDof();
-      double *Minv_el = &Minv[Minv_offsets[i]];
-      int *ipiv_el = &ipiv[ipiv_offsets[i]];
-      DenseMatrix Me(Minv_el, dof, dof);
-      mi.AssembleElementMatrix(*fe, *tr, Me);
-      LUFactors lu(Minv_el, ipiv_el);
-      lu.Factor(dof);
-   }
-}
-
-void DGDiffusionBR2Integrator::AssembleFaceMatrix(
-   const FiniteElement &el1, const FiniteElement &el2,
-   FaceElementTransformations &Trans, DenseMatrix &elmat)
-{
-   int ndof1 = el1.GetDof();
-   shape1.SetSize(ndof1);
-
-   R11.SetSize(ndof1, ndof1);
-   R11 = 0.0;
-   LUFactors M1inv(&Minv[Minv_offsets[Trans.Elem1No]],
-                   &ipiv[ipiv_offsets[Trans.Elem1No]]);
-   LUFactors M2inv;
-
-   double factor = Geometries.NumBdr(Trans.Elem1->GetGeometryType());
-
-   int ndof2;
-   if (Trans.Elem2No >= 0)
-   {
-      ndof2 = el2.GetDof();
-      shape2.SetSize(ndof2);
-      R12.SetSize(ndof1, ndof2);
-      R21.SetSize(ndof2, ndof1);
-      R22.SetSize(ndof2, ndof2);
-      M2inv.data = &Minv[Minv_offsets[Trans.Elem2No]];
-      M2inv.ipiv = &ipiv[ipiv_offsets[Trans.Elem2No]];
-
-      R12 = 0.0;
-      R21 = 0.0;
-      R22 = 0.0;
-
-      Geometry::Type geom2 = Trans.Elem2->GetGeometryType();
-      factor = std::max(factor, double(Geometries.NumBdr(geom2)));
-   }
-   else
-   {
-      ndof2 = 0;
-   }
-
-   int ndofs = ndof1 + ndof2;
-
-   Re.SetSize(ndofs, ndofs);
-   MinvRe.SetSize(ndofs, ndofs);
-
-   elmat.SetSize(ndofs);
-   elmat = 0.0;
-
-   const IntegrationRule *ir = IntRule;
-   if (ir == NULL)
-   {
-      int order;
-      if (ndof2)
-      {
-         order = 2*std::max(el1.GetOrder(), el2.GetOrder());
-      }
-      else
-      {
-         order = 2*el1.GetOrder();
-      }
-      ir = &IntRules.Get(Trans.FaceGeom, order);
-   }
-
-   for (int p = 0; p < ir->GetNPoints(); p++)
-   {
-      const IntegrationPoint &ip = ir->IntPoint(p);
-      IntegrationPoint eip1, eip2;
-
-      Trans.Loc1.Transform(ip, eip1);
-      el1.CalcShape(eip1, shape1);
-      if (ndof2)
-      {
-         Trans.Loc2.Transform(ip, eip2);
-         el2.CalcShape(eip2, shape2);
-      }
-
-      double w = factor*sqrt(eta)*ip.weight*Trans.Face->Weight();
-      if (ndof2)
-      {
-         w /= 2;
-      }
-
-      for (int i = 0; i < ndof1; i++)
-      {
-         const double wsi = w*shape1(i);
-         for (int j = 0; j < ndof1; j++)
-         {
-            R11(i, j) += wsi*shape1(j);
-         }
-      }
-
-      if (ndof2)
-      {
-         for (int i = 0; i < ndof2; i++)
-         {
-            const double wsi = w*shape2(i);
-            for (int j = 0; j < ndof1; j++)
-            {
-               R21(i, j) += wsi*shape1(j);
-               R12(j, i) -= wsi*shape1(j);
-            }
-            for (int j = 0; j < ndof2; j++)
-            {
-               R22(i, j) -= wsi*shape2(j);
-            }
-         }
-      }
-   }
-
-   MinvR11 = R11;
-   M1inv.Solve(ndof1, ndof1, MinvR11.Data());
-   for (int i = 0; i < ndof1; i++)
-   {
-      for (int j = 0; j < ndof1; j++)
-      {
-         Re(i, j) = R11(i, j);
-         MinvRe(i, j) = MinvR11(i, j);
-      }
-   }
-
-   if (ndof2)
-   {
-      MinvR12 = R12;
-      MinvR21 = R21;
-      MinvR22 = R22;
-      M1inv.Solve(ndof1, ndof2, MinvR12.Data());
-      M2inv.Solve(ndof2, ndof1, MinvR21.Data());
-      M2inv.Solve(ndof2, ndof2, MinvR22.Data());
-
-      for (int i = 0; i < ndof2; i++)
-      {
-         for (int j = 0; j < ndof1; j++)
-         {
-            Re(ndof1 + i, j) = R21(i, j);
-            MinvRe(ndof1 + i, j) = MinvR21(i, j);
-
-            Re(j, ndof1 + i) = R12(j, i);
-            MinvRe(j, ndof1 + i) = MinvR12(j, i);
-         }
-         for (int j = 0; j < ndof2; j++)
-         {
-            Re(ndof1 + i, ndof1 + j) = R22(i, j);
-            MinvRe(ndof1 + i, ndof1 + j) = MinvR22(i, j);
-         }
-      }
-   }
-
-   // Compute the matrix associated with (r_e([u]), r_e([u])).
-   // The matrix for r_e([u]) is `MinvRe`, and so we need to form the product
-   // `(MinvRe)^T M MinvRe`. Using `Minv^T M = Minv M = I`, we obtain
-   // `Re^T MinvRe`.
-   MultAtB(Re, MinvRe, elmat);
-}
-
-}
@@ -379,54 +379,51 @@ void DiffusionIntegrator::AssemblePA(const FiniteElementSpace &fes)
   const int MQfullDim = MQ ? MQ->GetHeight() * MQ->GetWidth() : 0;
   if (MQ)
   {
-      symmetric = false;
      MFEM_VERIFY(MQ->GetHeight() == dim && MQ->GetWidth() == dim, "");
+      const int MQsymmDim = MQ->GetWidth() * (MQ->GetWidth() + 1) / 2;

-      coeffDim = MQfullDim;
+      const int MQdim = MQ->IsSymmetric() ? MQsymmDim : MQfullDim;
+      coeffDim = MQdim;

-      coeff.SetSize(MQfullDim * nq * ne);
+      coeff.SetSize(MQdim * nq * ne);
+      symmetric = MQ ? MQ->IsSymmetric() : true;

      DenseMatrix M;
-      M.SetSize(dim);
-
-      auto C = Reshape(coeff.HostWrite(), MQfullDim, nq, ne);
-      for (int e=0; e<ne; ++e)
+      Vector Msymm;
+      if (symmetric)
      {
-         ElementTransformation *tr = mesh->GetElementTransformation(e);
-         for (int p=0; p<nq; ++p)
-         {
-            MQ->Eval(M, *tr, ir->IntPoint(p));
-            for (int i=0; i<dim; ++i)
-               for (int j=0; j<dim; ++j)
-               {
-                  C(j+(i*dim), p, e) = M(i,j);
-               }
-         }
+         Msymm.SetSize(MQsymmDim);
+      }
+      else
+      {
+         M.SetSize(dim);
      }
-   }
-   else if (SMQ)
-   {
-      MFEM_VERIFY(SMQ->GetSize() == dim, "");
-      coeffDim = symmDims;
-      coeff.SetSize(symmDims * nq * ne);
-
-      DenseSymmetricMatrix M;
-      M.SetSize(dim);
-
-      auto C = Reshape(coeff.HostWrite(), symmDims, nq, ne);

+      auto C = Reshape(coeff.HostWrite(), MQdim, nq, ne);
      for (int e=0; e<ne; ++e)
      {
         ElementTransformation *tr = mesh->GetElementTransformation(e);
         for (int p=0; p<nq; ++p)
         {
-            SMQ->Eval(M, *tr, ir->IntPoint(p));
-            int cnt = 0;
-            for (int i=0; i<dim; ++i)
-               for (int j=i; j<dim; ++j, ++cnt)
+            if (MQ->IsSymmetric())
+            {
+               MQ->EvalSymmetric(Msymm, *tr, ir->IntPoint(p));
+
+               for (int i=0; i<MQsymmDim; ++i)
               {
-                  C(cnt, p, e) = M(i,j);
+                  C(i, p, e) = Msymm[i];
               }
+            }
+            else
+            {
+               MQ->Eval(M, *tr, ir->IntPoint(p));
+
+               for (int i=0; i<dim; ++i)
+                  for (int j=0; j<dim; ++j)
+                  {
+                     C(j+(i*dim), p, e) = M(i,j);
+                  }
+            }
         }
      }
   }
@@ -334,7 +334,7 @@ static void PAGradientApplyTranspose2D(const int NE,
                                       const int q1d = 0)
 {
   // TODO
-   MFEM_ASSERT(false, "PAGradientApplyTranspose2D not implemented.");
+   MFEM_ASSERT(false, "GradientPAApplyTranspose 3D not implemented.");
 }

 // PA Gradient Apply 3D kernel
@@ -761,12 +761,12 @@ void VectorFEMassIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,
   trial_fetype = trial_el->GetDerivType();
   test_fetype = test_el->GetDerivType();

-   const int MQsymmDim = SMQ ? (SMQ->GetSize() * (SMQ->GetSize() + 1)) / 2 : 0;
+   const int MQsymmDim = MQ ? (MQ->GetWidth() * (MQ->GetWidth() + 1)) / 2 : 0;
   const int MQfullDim = MQ ? (MQ->GetHeight() * MQ->GetWidth()) : 0;
-   const int MQdim = MQ ? MQfullDim : MQsymmDim;
-   const int coeffDim = (MQ || SMQ) ? MQdim : (DQ ? DQ->GetVDim() : 1);
+   const int MQdim = MQ ? (MQ->IsSymmetric() ? MQsymmDim : MQfullDim) : 0;
+   const int coeffDim = MQ ? MQdim : (VQ ? VQ->GetVDim() : 1);

-   symmetric = (MQ == NULL);
+   symmetric = MQ ? MQ->IsSymmetric() : true;

   const bool trial_curl = (trial_fetype == mfem::FiniteElement::CURL);
   const bool trial_div = (trial_fetype == mfem::FiniteElement::DIV);
@@ -783,13 +783,24 @@ void VectorFEMassIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,
   Vector coeff(coeffDim * ne * nq);
   coeff = 1.0;
   auto coeffh = Reshape(coeff.HostWrite(), coeffDim, nq, ne);
-   if (Q || DQ || MQ || SMQ)
+   if (Q || VQ || MQ)
   {
-      Vector D(DQ ? coeffDim : 0);
+      Vector D(VQ ? coeffDim : 0);
      DenseMatrix M;
-      DenseSymmetricMatrix SM;
+      Vector Msymm;
+      if (MQ)
+      {
+         if (symmetric)
+         {
+            Msymm.SetSize(MQsymmDim);
+         }
+         else
+         {
+            M.SetSize(dim);
+         }
+      }

-      if (DQ)
+      if (VQ)
      {
         MFEM_VERIFY(coeffDim == dim, "");
      }
@@ -797,12 +808,6 @@ void VectorFEMassIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,
      {
         MFEM_VERIFY(coeffDim == MQdim, "");
         MFEM_VERIFY(MQ->GetHeight() == dim && MQ->GetWidth() == dim, "");
-         M.SetSize(dim);
-      }
-      if (SMQ)
-      {
-         MFEM_VERIFY(SMQ->GetSize() == dim, "");
-         SM.SetSize(dim);
      }

      for (int e=0; e<ne; ++e)
@@ -812,27 +817,29 @@ void VectorFEMassIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,
         {
            if (MQ)
            {
-               MQ->Eval(M, *tr, ir->IntPoint(p));
+               if (MQ->IsSymmetric())
+               {
+                  MQ->EvalSymmetric(Msymm, *tr, ir->IntPoint(p));

-               for (int i=0; i<dim; ++i)
-                  for (int j=0; j<dim; ++j)
+                  for (int i=0; i<MQsymmDim; ++i)
                  {
-                     coeffh(j+(i*dim), p, e) = M(i,j);
+                     coeffh(i, p, e) = Msymm[i];
                  }
+               }
+               else
+               {
+                  MQ->Eval(M, *tr, ir->IntPoint(p));
+
+                  for (int i=0; i<dim; ++i)
+                     for (int j=0; j<dim; ++j)
+                     {
+                        coeffh(j+(i*dim), p, e) = M(i,j);
+                     }
+               }
            }
-            else if (SMQ)
+            else if (VQ)
            {
-               SMQ->Eval(SM, *tr, ir->IntPoint(p));
-               int cnt = 0;
-               for (int i=0; i<dim; ++i)
-                  for (int j=i; j<dim; ++j, ++cnt)
-                  {
-                     coeffh(cnt, p, e) = SM(i,j);
-                  }
-            }
-            else if (DQ)
-            {
-               DQ->Eval(D, *tr, ir->IntPoint(p));
+               VQ->Eval(D, *tr, ir->IntPoint(p));
               for (int i=0; i<coeffDim; ++i)
               {
                  coeffh(i, p, e) = D[i];
@@ -1000,14 +1007,14 @@ void VectorFEMassIntegrator::AddMultPA(const Vector &x, Vector &y) const
      }
      else if (trial_curl && test_div)
      {
-         const bool scalarCoeff = !(DQ || MQ || SMQ);
+         const bool scalarCoeff = !(VQ || MQ);
         PAHcurlHdivMassApply3D(dofs1D, dofs1Dtest, quad1D, ne, scalarCoeff,
                                true, mapsO->B, mapsC->B, mapsOtest->Bt,
                                mapsCtest->Bt, pa_data, x, y);
      }
      else if (trial_div && test_curl)
      {
-         const bool scalarCoeff = !(DQ || MQ || SMQ);
+         const bool scalarCoeff = !(VQ || MQ);
         PAHcurlHdivMassApply3D(dofs1D, dofs1Dtest, quad1D, ne, scalarCoeff,
                                false, mapsO->B, mapsC->B, mapsOtest->Bt,
                                mapsCtest->Bt, pa_data, x, y);
@@ -1031,7 +1038,7 @@ void VectorFEMassIntegrator::AddMultPA(const Vector &x, Vector &y) const
      }
      else if ((trial_curl && test_div) || (trial_div && test_curl))
      {
-         const bool scalarCoeff = !(DQ || MQ || SMQ);
+         const bool scalarCoeff = !(VQ || MQ);
         PAHcurlHdivMassApply2D(dofs1D, dofs1Dtest, quad1D, ne, scalarCoeff,
                                trial_curl, mapsO->B, mapsC->B, mapsOtest->Bt,
                                mapsCtest->Bt, pa_data, x, y);
@@ -301,7 +301,7 @@ void MatrixFunctionCoefficient::Eval(DenseMatrix &K, ElementTransformation &T,

   K.SetSize(height, width);

-   if (symmetric) // Use SymmFunction (deprecated version)
+   if (symmetric) // Use SymmFunction
   {
      MFEM_VERIFY(height == width && SymmFunction,
                  "MatrixFunctionCoefficient is not symmetric");
@@ -371,36 +371,6 @@ void MatrixFunctionCoefficient::EvalSymmetric(Vector &K,
   }
 }

-void SymmetricMatrixFunctionCoefficient::Eval(DenseSymmetricMatrix &K,
-                                              ElementTransformation &T,
-                                              const IntegrationPoint &ip)
-{
-   double x[3];
-   Vector transip(x, 3);
-
-   T.Transform(ip, transip);
-
-   K.SetSize(dim);
-
-   if (Function)
-   {
-      Function(transip, K);
-   }
-   else if (TDFunction)
-   {
-      TDFunction(transip, GetTime(), K);
-   }
-   else
-   {
-      K = mat;
-   }
-
-   if (Q)
-   {
-      K *= Q->Eval(T, ip, GetTime());
-   }
-}
-
 MatrixArrayCoefficient::MatrixArrayCoefficient (int dim)
   : MatrixCoefficient (dim)
 {
@@ -625,7 +595,6 @@ void MatrixVectorProductCoefficient::Eval(Vector &V, ElementTransformation &T,
 {
   a->Eval(ma, T, ip);
   b->Eval(vb, T, ip);
-   V.SetSize(vdim);
   ma.Mult(vb, V);
 }

@@ -688,7 +688,6 @@ public:
                     const IntegrationRule &ir);
 };

-typedef VectorCoefficient DiagonalMatrixCoefficient;

 /// Base class for Matrix Coefficients that optionally depend on time and space.
 class MatrixCoefficient
@@ -696,7 +695,7 @@ class MatrixCoefficient
 protected:
   int height, width;
   double time;
-   bool symmetric;  // deprecated
+   bool symmetric;

 public:
   /// Construct a dim x dim matrix coefficient.
@@ -722,7 +721,6 @@ public:
   /// For backward compatibility get the width of the matrix.
   int GetVDim() const { return width; }

-   /** @deprecated Use SymmetricMatrixCoefficient instead */
   bool IsSymmetric() const { return symmetric; }

   /** @brief Evaluate the matrix coefficient in the element described by @a T
@@ -733,13 +731,11 @@ public:
   virtual void Eval(DenseMatrix &K, ElementTransformation &T,
                     const IntegrationPoint &ip) = 0;

-   /// (DEPRECATED) Evaluate a symmetric matrix coefficient.
   /** @brief Evaluate the upper triangular entries of the matrix coefficient
       in the symmetric case, similarly to Eval. Matrix entry (i,j) is stored
       in K[j - i + os_i] for 0 <= i <= j < width, os_0 = 0,
       os_{i+1} = os_i + width - i. That is, K = {M(0,0), ..., M(0,w-1),
-       M(1,1), ..., M(1,w-1), ..., M(w-1,w-1) with w = width.
-       @deprecated Use Eval() instead. */
+       M(1,1), ..., M(1,w-1), ..., M(w-1,w-1) with w = width. */
   virtual void EvalSymmetric(Vector &K, ElementTransformation &T,
                              const IntegrationPoint &ip)
   { mfem_error("MatrixCoefficient::EvalSymmetric"); }
@@ -771,7 +767,7 @@ class MatrixFunctionCoefficient : public MatrixCoefficient
 {
 private:
   std::function<void(const Vector &, DenseMatrix &)> Function;
-   std::function<void(const Vector &, Vector &)> SymmFunction;  // deprecated
+   std::function<void(const Vector &, Vector &)> SymmFunction;
   std::function<void(const Vector &, double, DenseMatrix &)> TDFunction;

   Coefficient *Q;
@@ -795,18 +791,6 @@ public:
      : MatrixCoefficient(m.Height(), m.Width()), Q(&q), mat(m)
   { }

-   /** @brief Define a time-independent symmetric square matrix coefficient from
-       a std function */
-   /** \param dim - the size of the matrix
-       \param SymmF - function used in EvalSymmetric
-       \param q - optional scalar Coefficient to scale the matrix coefficient
-       @deprecated Use another constructor without setting SymmFunction. */
-   MatrixFunctionCoefficient(int dim,
-                             std::function<void(const Vector &, Vector &)> SymmF,
-                             Coefficient *q = NULL)
-      : MatrixCoefficient(dim, true), SymmFunction(std::move(SymmF)), Q(q), mat(0)
-   { }
-
   /// Define a time-dependent square matrix coefficient from a std function
   /** \param dim - the size of the matrix
       \param TDF - time-dependent function
@@ -817,12 +801,22 @@ public:
      : MatrixCoefficient(dim), TDFunction(std::move(TDF)), Q(q)
   { }

+   /** @brief Define a time-independent symmetric square matrix coefficient from
+       a std function */
+   /** \param dim - the size of the matrix
+       \param SymmF - function used in EvalSymmetric
+       \param q - optional scalar Coefficient to scale the matrix coefficient */
+   MatrixFunctionCoefficient(int dim,
+                             std::function<void(const Vector &, Vector &)> SymmF,
+                             Coefficient *q = NULL)
+      : MatrixCoefficient(dim, true), SymmFunction(std::move(SymmF)), Q(q), mat(0)
+   { }
+
   /// Evaluate the matrix coefficient at @a ip.
   virtual void Eval(DenseMatrix &K, ElementTransformation &T,
                     const IntegrationPoint &ip);

-   /// (DEPRECATED) Evaluate the symmetric matrix coefficient at @a ip.
-   /** @deprecated Use Eval() instead. */
+   /// Evaluate the symmetric matrix coefficient at @a ip.
   virtual void EvalSymmetric(Vector &K, ElementTransformation &T,
                              const IntegrationPoint &ip);

@@ -830,6 +824,7 @@ public:
 };


+
 /** @brief Matrix coefficient defined by a matrix of scalar coefficients.
    Coefficients that are not set will evaluate to zero in the vector. The
    coefficient is stored as a flat Array with indexing (i,j) -> i*width+j. */
@@ -945,106 +940,6 @@ public:
   }
 };

-
-/// Base class for symmetric matrix coefficients that optionally depend on time and space.
-class SymmetricMatrixCoefficient
-{
-protected:
-   int dim;
-   double time;
-
-public:
-   /// Construct a dim x dim matrix coefficient.
-   explicit SymmetricMatrixCoefficient(int dimension)
-   { dim = dimension; time = 0.; }
-
-   /// Set the time for time dependent coefficients
-   void SetTime(double t) { time = t; }
-
-   /// Get the time for time dependent coefficients
-   double GetTime() { return time; }
-
-   /// Get the size of the matrix.
-   int GetSize() const { return dim; }
-
-   /** @brief Evaluate the matrix coefficient in the element described by @a T
-       at the point @a ip, storing the result in @a K. */
-   /** @note When this method is called, the caller must make sure that the
-       IntegrationPoint associated with @a T is the same as @a ip. This can be
-       achieved by calling T.SetIntPoint(&ip). */
-   virtual void Eval(DenseSymmetricMatrix &K, ElementTransformation &T,
-                     const IntegrationPoint &ip) = 0;
-
-   virtual ~SymmetricMatrixCoefficient() { }
-};
-
-
-/// A matrix coefficient that is constant in space and time.
-class SymmetricMatrixConstantCoefficient : public SymmetricMatrixCoefficient
-{
-private:
-   DenseSymmetricMatrix mat;
-
-public:
-   ///Construct using matrix @a m for the constant.
-   SymmetricMatrixConstantCoefficient(const DenseSymmetricMatrix &m)
-      : SymmetricMatrixCoefficient(m.Height()), mat(m) { }
-   using SymmetricMatrixCoefficient::Eval;
-   /// Evaluate the matrix coefficient at @a ip.
-   virtual void Eval(DenseSymmetricMatrix &M, ElementTransformation &T,
-                     const IntegrationPoint &ip) { M = mat; }
-};
-
-
-/** @brief A matrix coefficient with an optional scalar coefficient multiplier
-    \a q.  The matrix function can either be represented by a std function or
-    a constant matrix provided when constructing this object.  */
-class SymmetricMatrixFunctionCoefficient : public SymmetricMatrixCoefficient
-{
-private:
-   std::function<void(const Vector &, DenseSymmetricMatrix &)> Function;
-   std::function<void(const Vector &, double, DenseSymmetricMatrix &)> TDFunction;
-
-   Coefficient *Q;
-   DenseSymmetricMatrix mat;
-
-public:
-   /// Define a time-independent symmetric matrix coefficient from a std function
-   /** \param dim - the size of the matrix
-       \param F - time-independent function
-       \param q - optional scalar Coefficient to scale the matrix coefficient */
-   SymmetricMatrixFunctionCoefficient(int dim,
-                                      std::function<void(const Vector &, DenseSymmetricMatrix &)> F,
-                                      Coefficient *q = nullptr)
-      : SymmetricMatrixCoefficient(dim), Function(std::move(F)), Q(q), mat(0)
-   { }
-
-   /// Define a constant matrix coefficient times a scalar Coefficient
-   /** \param m - constant matrix
-       \param q - optional scalar Coefficient to scale the matrix coefficient */
-   SymmetricMatrixFunctionCoefficient(const DenseSymmetricMatrix &m,
-                                      Coefficient &q)
-      : SymmetricMatrixCoefficient(m.Height()), Q(&q), mat(m)
-   { }
-
-   /// Define a time-dependent square matrix coefficient from a std function
-   /** \param dim - the size of the matrix
-       \param TDF - time-dependent function
-       \param q - optional scalar Coefficient to scale the matrix coefficient */
-   SymmetricMatrixFunctionCoefficient(int dim,
-                                      std::function<void(const Vector &, double, DenseSymmetricMatrix &)> TDF,
-                                      Coefficient *q = nullptr)
-      : SymmetricMatrixCoefficient(dim), TDFunction(std::move(TDF)), Q(q)
-   { }
-
-   /// Evaluate the matrix coefficient at @a ip.
-   virtual void Eval(DenseSymmetricMatrix &K, ElementTransformation &T,
-                     const IntegrationPoint &ip);
-
-   virtual ~SymmetricMatrixFunctionCoefficient() { }
-};
-
-
 /** @brief Scalar coefficient defined as the product of two scalar coefficients
    or a scalar and a scalar coefficient. */
 class ProductCoefficient : public Coefficient
@@ -1204,7 +1204,6 @@ ParSesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
      });
      // Modify offdiagonal blocks (imaginary parts of the matrix) to conform
      // with standard essential BC treatment
-      ess_tdof_list.HostRead();
      if (A_i.Type() == Operator::Hypre_ParCSR)
      {
         HypreParMatrix * Ah;
@@ -48,373 +48,8 @@ void L2ZienkiewiczZhuEstimator::ComputeEstimates()
   current_sequence = solution->FESpace()->GetMesh()->GetSequence();
 }

-
-KellyErrorEstimator::KellyErrorEstimator(BilinearFormIntegrator& di_,
-                                         ParGridFunction& sol_,
-                                         ParFiniteElementSpace& flux_fespace_,
-                                         const Array<int> &attributes_)
-   : attributes(attributes_)
-   , flux_integrator(&di_)
-   , solution(&sol_)
-   , flux_space(&flux_fespace_)
-   , own_flux_fespace(false)
-{
-   ResetCoefficientFunctions();
-}
-
-KellyErrorEstimator::KellyErrorEstimator(BilinearFormIntegrator& di_,
-                                         ParGridFunction& sol_,
-                                         ParFiniteElementSpace* flux_fespace_,
-                                         const Array<int> &attributes_)
-   : attributes(attributes_)
-   , flux_integrator(&di_)
-   , solution(&sol_)
-   , flux_space(flux_fespace_)
-   , own_flux_fespace(true)
-{
-   ResetCoefficientFunctions();
-}
-
-KellyErrorEstimator::~KellyErrorEstimator()
-{
-   if (own_flux_fespace)
-   {
-      delete flux_space;
-   }
-}
-
-void KellyErrorEstimator::ResetCoefficientFunctions()
-{
-   compute_element_coefficient = [](ParMesh* pmesh, const int e)
-   {
-      return 1.0;
-   };
-
-   compute_face_coefficient = [](ParMesh* pmesh, const int f,
-                                 const bool shared_face)
-   {
-      auto FT = [&]()
-      {
-         if (shared_face)
-         {
-            return pmesh->GetSharedFaceTransformations(f);
-         }
-         return pmesh->GetFaceElementTransformations(f);
-      }();
-      const auto order = FT->GetFE()->GetOrder();
-
-      // Poor man's face diameter.
-      double diameter = 0.0;
-
-      Vector p1(pmesh->SpaceDimension());
-      Vector p2(pmesh->SpaceDimension());
-      // NOTE: We have no direct access to vertices for shared faces,
-      // so we fall back to compute the positions from the element.
-      // This can also be modified to compute the diameter for non-linear
-      // geometries by sampling along geometry-specific lines.
-      auto vtx_intrule = Geometries.GetVertices(FT->GetGeometryType());
-      const auto nip = vtx_intrule->GetNPoints();
-      for (int i = 0; i < nip; i++)
-      {
-         // Evaluate flux vector at integration point
-         auto fip1 = vtx_intrule->IntPoint(i);
-         FT->Transform(fip1, p1);
-
-         for (int j = 0; j < nip; j++)
-         {
-            auto fip2 = vtx_intrule->IntPoint(j);
-            FT->Transform(fip2, p2);
-
-            diameter = std::max<double>(diameter, p2.DistanceTo(p1));
-         }
-      }
-      return diameter/(2.0*order);
-   };
-}
-
-void KellyErrorEstimator::ComputeEstimates()
-{
-   // Remarks:
-   // For some context you may have to consult the documentation of
-   // the FaceInfo class [1]. Also, the FaceElementTransformations
-   // documentation [2] may be helpful to grasp what is going on. Note
-   // that the FaceElementTransformations also works in the non-
-   // conforming case to transfer the gauss points from the slave to
-   // the master element.
-   // [1]
-   // https://github.com/mfem/mfem/blob/02d0bfe9c18ce049c3c93a6a4208080fcfc96991/mesh/mesh.hpp#L94
-   // [2]
-   // https://github.com/mfem/mfem/blob/02d0bfe9c18ce049c3c93a6a4208080fcfc96991/fem/eltrans.hpp#L435
-
-   flux_space->Update(false);
-
-   auto xfes      = solution->ParFESpace();
-   MFEM_ASSERT(xfes->GetVDim() == 1,
-               "Estimation for vector-valued problems not implemented yet.");
-   auto pmesh     = xfes->GetParMesh();
-
-   this->error_estimates.SetSize(xfes->GetNE());
-   this->error_estimates = 0.0;
-
-   // 1. Compute fluxes in discontinuous space
-   ParGridFunction flux(flux_space);
-   flux = 0.0;
-
-   // We pre-sort the array to speed up the search in the following loops.
-   if (attributes.Size())
-   {
-      attributes.Sort();
-   }
-
-   Array<int> xdofs, fdofs;
-   Vector el_x, el_f;
-   for (int e = 0; e < xfes->GetNE(); e++)
-   {
-      auto attr = xfes->GetAttribute(e);
-      if (attributes.Size() && attributes.FindSorted(attr) == -1)
-      {
-         continue;
-      }
-
-      xfes->GetElementVDofs(e, xdofs);
-      solution->GetSubVector(xdofs, el_x);
-
-      ElementTransformation* Transf = xfes->GetElementTransformation(e);
-      flux_integrator->ComputeElementFlux(*xfes->GetFE(e), *Transf, el_x,
-                                          *flux_space->GetFE(e), el_f, true);
-
-      flux_space->GetElementVDofs(e, fdofs);
-      flux.AddElementVector(fdofs, el_f);
-   }
-
-   // 2. Add error contribution from local interior faces
-   for (int f = 0; f < pmesh->GetNumFaces(); f++)
-   {
-      auto FT = pmesh->GetFaceElementTransformations(f);
-
-      auto &int_rule = IntRules.Get(FT->FaceGeom, 2 * xfes->GetFaceOrder(f));
-      const auto nip = int_rule.GetNPoints();
-
-      if (pmesh->FaceIsInterior(f))
-      {
-         int Inf1, Inf2, NCFace;
-         pmesh->GetFaceInfos(f, &Inf1, &Inf2, &NCFace);
-
-         // Convention
-         // * Conforming face: Face side with smaller element id handles
-         // the integration
-         // * Non-conforming face: The slave handles the integration.
-         // See FaceInfo documentation for details.
-         bool isNCSlave    = FT->Elem2No >= 0 && NCFace >= 0;
-         bool isConforming = FT->Elem2No >= 0 && NCFace == -1;
-         if ((FT->Elem1No < FT->Elem2No && isConforming) || isNCSlave)
-         {
-            if (attributes.Size() &&
-                (attributes.FindSorted(FT->Elem1->Attribute) == -1
-                 || attributes.FindSorted(FT->Elem2->Attribute) == -1))
-            {
-               continue;
-            }
-
-            IntegrationRule eir;
-            Vector jumps(nip);
-
-            // Integral over local half face on the side of e₁
-            // i.e. the numerical integration of ∫ flux ⋅ n dS₁
-            for (int i = 0; i < nip; i++)
-            {
-               // Evaluate flux at IP
-               auto &fip = int_rule.IntPoint(i);
-               IntegrationPoint ip;
-               FT->Loc1.Transform(fip, ip);
-
-               Vector val(flux_space->GetVDim());
-               flux.GetVectorValue(FT->Elem1No, ip, val);
-
-               // And build scalar product with normal
-               Vector normal(pmesh->SpaceDimension());
-               FT->Face->SetIntPoint(&fip);
-               if (pmesh->Dimension() == pmesh->SpaceDimension())
-               {
-                  CalcOrtho(FT->Face->Jacobian(), normal);
-               }
-               else
-               {
-                  Vector ref_normal(pmesh->Dimension());
-                  FT->Loc1.Transf.SetIntPoint(&fip);
-                  CalcOrtho(FT->Loc1.Transf.Jacobian(), ref_normal);
-                  auto &e1 = FT->GetElement1Transformation();
-                  e1.AdjugateJacobian().MultTranspose(ref_normal, normal);
-                  normal /= e1.Weight();
-               }
-               jumps(i) = val * normal * fip.weight * FT->Face->Weight();
-            }
-
-            // Subtract integral over half face of e₂
-            // i.e. the numerical integration of ∫ flux ⋅ n dS₂
-            for (int i = 0; i < nip; i++)
-            {
-               // Evaluate flux vector at IP
-               auto &fip = int_rule.IntPoint(i);
-               IntegrationPoint ip;
-               FT->Loc2.Transform(fip, ip);
-
-               Vector val(flux_space->GetVDim());
-               flux.GetVectorValue(FT->Elem2No, ip, val);
-
-               // And build scalar product with normal
-               Vector normal(pmesh->SpaceDimension());
-               FT->Face->SetIntPoint(&fip);
-               if (pmesh->Dimension() == pmesh->SpaceDimension())
-               {
-                  CalcOrtho(FT->Face->Jacobian(), normal);
-               }
-               else
-               {
-                  Vector ref_normal(pmesh->Dimension());
-                  FT->Loc1.Transf.SetIntPoint(&fip);
-                  CalcOrtho(FT->Loc1.Transf.Jacobian(), ref_normal);
-                  auto &e1 = FT->GetElement1Transformation();
-                  e1.AdjugateJacobian().MultTranspose(ref_normal, normal);
-                  normal /= e1.Weight();
-               }
-
-               jumps(i) -= val * normal * fip.weight * FT->Face->Weight();
-            }
-
-            // Finalize "local" L₂ contribution
-            for (int i = 0; i < nip; i++)
-            {
-               jumps(i) *= jumps(i);
-            }
-            auto h_k_face = compute_face_coefficient(pmesh, f, false);
-            double jump_integral = h_k_face*jumps.Sum();
-
-            // A local face is shared between two local elements, so we
-            // can get away with integrating the jump only once and add
-            // it to both elements. To minimize communication, the jump
-            // of shared faces is computed locally by each process.
-            error_estimates(FT->Elem1No) += jump_integral;
-            error_estimates(FT->Elem2No) += jump_integral;
-         }
-      }
-   }
-
-   // 3. Add error contribution from shared interior faces
-   // Synchronize face data.
-   flux.ExchangeFaceNbrData();
-
-   for (int sf = 0; sf < pmesh->GetNSharedFaces(); sf++)
-   {
-      auto FT = pmesh->GetSharedFaceTransformations(sf, true);
-      if (attributes.Size() &&
-          (attributes.FindSorted(FT->Elem1->Attribute) == -1
-           || attributes.FindSorted(FT->Elem2->Attribute) == -1))
-      {
-         continue;
-      }
-
-      auto &int_rule = IntRules.Get(FT->FaceGeom, 2 * xfes->GetFaceOrder(0));
-      const auto nip = int_rule.GetNPoints();
-
-      IntegrationRule eir;
-      Vector jumps(nip);
-
-      // Integral over local half face on the side of e₁
-      // i.e. the numerical integration of ∫ flux ⋅ n dS₁
-      for (int i = 0; i < nip; i++)
-      {
-         // Evaluate flux vector at integration point
-         auto &fip = int_rule.IntPoint(i);
-         IntegrationPoint ip;
-         FT->Loc1.Transform(fip, ip);
-
-         Vector val(flux_space->GetVDim());
-         flux.GetVectorValue(FT->Elem1No, ip, val);
-
-         Vector normal(pmesh->SpaceDimension());
-         FT->Face->SetIntPoint(&fip);
-         if (pmesh->Dimension() == pmesh->SpaceDimension())
-         {
-            CalcOrtho(FT->Face->Jacobian(), normal);
-         }
-         else
-         {
-            Vector ref_normal(pmesh->Dimension());
-            FT->Loc1.Transf.SetIntPoint(&fip);
-            CalcOrtho(FT->Loc1.Transf.Jacobian(), ref_normal);
-            auto &e1 = FT->GetElement1Transformation();
-            e1.AdjugateJacobian().MultTranspose(ref_normal, normal);
-            normal /= e1.Weight();
-         }
-
-         jumps(i) = val * normal * fip.weight * FT->Face->Weight();
-      }
-
-      // Subtract integral over non-local half face of e₂
-      // i.e. the numerical integration of ∫ flux ⋅ n dS₂
-      for (int i = 0; i < nip; i++)
-      {
-         // Evaluate flux vector at integration point
-         auto &fip = int_rule.IntPoint(i);
-         IntegrationPoint ip;
-         FT->Loc2.Transform(fip, ip);
-
-         Vector val(flux_space->GetVDim());
-         flux.GetVectorValue(FT->Elem2No, ip, val);
-
-         // Evaluate gauss point
-         Vector normal(pmesh->SpaceDimension());
-         FT->Face->SetIntPoint(&fip);
-         if (pmesh->Dimension() == pmesh->SpaceDimension())
-         {
-            CalcOrtho(FT->Face->Jacobian(), normal);
-         }
-         else
-         {
-            Vector ref_normal(pmesh->Dimension());
-            CalcOrtho(FT->Loc1.Transf.Jacobian(), ref_normal);
-            auto &e1 = FT->GetElement1Transformation();
-            e1.AdjugateJacobian().MultTranspose(ref_normal, normal);
-            normal /= e1.Weight();
-         }
-
-         jumps(i) -= val * normal * fip.weight * FT->Face->Weight();
-      }
-
-      // Finalize "local" L₂ contribution
-      for (int i = 0; i < nip; i++)
-      {
-         jumps(i) *= jumps(i);
-      }
-      auto h_k_face = compute_face_coefficient(pmesh, sf, true);
-      double jump_integral = h_k_face*jumps.Sum();
-
-      error_estimates(FT->Elem1No) += jump_integral;
-      // We skip "error_estimates(FT->Elem2No) += jump_integral"
-      // because the error is stored on the remote process and
-      // recomputed there.
-   }
-
-   // Finalize element errors
-   for (int e = 0; e < xfes->GetNE(); e++)
-   {
-      auto factor = compute_element_coefficient(pmesh, e);
-      // The sqrt belongs to the norm and hₑ to the indicator.
-      error_estimates(e) = sqrt(factor * error_estimates(e));
-   }
-
-   current_sequence = solution->FESpace()->GetMesh()->GetSequence();
-
-   // Finish by computing the global error.
-   double process_local_error = error_estimates.Sum();
-   MPI_Allreduce(&process_local_error, &total_error, 1, MPI_DOUBLE,
-                 MPI_SUM, xfes->GetComm());
-}
-
 #endif // MFEM_USE_MPI

-
 void LpErrorEstimator::ComputeEstimates()
 {
   MFEM_VERIFY(coef != NULL || vcoef != NULL,
@@ -429,17 +64,6 @@ void LpErrorEstimator::ComputeEstimates()
   {
      sol->ComputeElementLpErrors(local_norm_p, *vcoef, error_estimates);
   }
-#ifdef MFEM_USE_MPI
-   total_error = error_estimates.Sum();
-   auto pfes = dynamic_cast<ParFiniteElementSpace*>(sol->FESpace());
-   if (pfes)
-   {
-      auto process_local_error = total_error;
-      MPI_Allreduce(&process_local_error, &total_error, 1, MPI_DOUBLE,
-                    MPI_SUM, pfes->GetComm());
-   }
-#endif // MFEM_USE_MPI
-   total_error = pow(total_error, 1.0/local_norm_p);
   current_sequence = sol->FESpace()->GetMesh()->GetSequence();
 }

@@ -12,8 +12,6 @@
 #ifndef MFEM_ERROR_ESTIMATORS
 #define MFEM_ERROR_ESTIMATORS

-#include <functional>
-
 #include "../config/config.hpp"
 #include "../linalg/vector.hpp"
 #include "bilinearform.hpp"
@@ -41,11 +39,6 @@ public:
 class ErrorEstimator : public AbstractErrorEstimator
 {
 public:
-   /// Return the total error from the last error estimate.
-   /** @note This method is optional for derived classes to override and the
-       base class implementation simply returns 0. */
-   virtual double GetTotalError() const { return 0.0; }
-
   /// Get a Vector with all element errors.
   virtual const Vector &GetLocalErrors() = 0;

@@ -155,8 +148,8 @@ public:
        own_flux_fes(false)
   { }

-   /** @brief Consider the coefficient in BilinearFormIntegrator to calculate
-       the fluxes for the error estimator.*/
+   /** @brief Consider the coefficient in BilinearFormIntegrator to calculate the
+       fluxes for the error estimator.*/
   void SetWithCoeff(bool w_coeff = true) { with_coeff = w_coeff; }

   /** @brief Enable/disable anisotropic estimates. To enable this option, the
@@ -173,10 +166,10 @@ public:
   void SetFluxAveraging(int fa) { flux_averaging = fa; }

   /// Return the total error from the last error estimate.
-   virtual double GetTotalError() const override { return total_error; }
+   double GetTotalError() const { return total_error; }

   /// Get a Vector with all element errors.
-   virtual const Vector &GetLocalErrors() override
+   virtual const Vector &GetLocalErrors()
   {
      if (MeshIsModified()) { ComputeEstimates(); }
      return error_estimates;
@@ -185,14 +178,14 @@ public:
   /** @brief Get an Array<int> with anisotropic flags for all mesh elements.
       Return an empty array when anisotropic estimates are not available or
       enabled. */
-   virtual const Array<int> &GetAnisotropicFlags() override
+   virtual const Array<int> &GetAnisotropicFlags()
   {
      if (MeshIsModified()) { ComputeEstimates(); }
      return aniso_flags;
   }

   /// Reset the error estimator.
-   virtual void Reset() override { current_sequence = -1; }
+   virtual void Reset() { current_sequence = -1; }

   /** @brief Destroy a ZienkiewiczZhuEstimator object. Destroys, if owned, the
       FiniteElementSpace, flux_space. */
@@ -299,17 +292,17 @@ public:
   void SetLocalErrorNormP(int p) { local_norm_p = p; }

   /// Return the total error from the last error estimate.
-   virtual double GetTotalError() const override { return total_error; }
+   double GetTotalError() const { return total_error; }

   /// Get a Vector with all element errors.
-   virtual const Vector &GetLocalErrors() override
+   virtual const Vector &GetLocalErrors()
   {
      if (MeshIsModified()) { ComputeEstimates(); }
      return error_estimates;
   }

   /// Reset the error estimator.
-   virtual void Reset() override { current_sequence = -1; }
+   virtual void Reset() { current_sequence = -1; }

   /** @brief Destroy a L2ZienkiewiczZhuEstimator object. Destroys, if owned,
       the FiniteElementSpace, flux_space. */
@@ -321,7 +314,6 @@ public:

 #endif // MFEM_USE_MPI

-
 /** @brief The LpErrorEstimator class compares the solution to a known
    coefficient.

@@ -340,8 +332,6 @@ protected:
   int local_norm_p;
   Vector error_estimates;

-   double total_error = 0.0;
-
   Coefficient * coef;
   VectorCoefficient * vcoef;
   GridFunction * sol;
@@ -393,10 +383,10 @@ public:
   void SetCoef(VectorCoefficient &A) { vcoef = &A; }

   /// Reset the error estimator.
-   virtual void Reset() override { current_sequence = -1; }
+   virtual void Reset() { current_sequence = -1; }

   /// Get a Vector with all element errors.
-   virtual const Vector &GetLocalErrors() override
+   virtual const Vector &GetLocalErrors()
   {
      if (MeshIsModified()) { ComputeEstimates(); }
      return error_estimates;
@@ -406,174 +396,6 @@ public:
   virtual ~LpErrorEstimator() {}
 };

-
-#ifdef MFEM_USE_MPI
-/** @brief The KellyErrorEstimator class provides a fast error indication
-    strategy for smooth scalar parallel problems.
-
-    The Kelly error indicator is based on the following papers:
-
-    Kelly, D. W., et al. "A posteriori error analysis and adaptive processes in
-    the finite element method: Part I—Error analysis." International journal for
-    numerical methods in engineering 19.11 (1983): 1593-1619.
-
-    De SR Gago, J. P., et al. "A posteriori error analysis and adaptive
-    processes in the finite element method: Part II—Adaptive mesh refinement."
-    International journal for numerical methods in engineering 19.11 (1983):
-    1621-1656.
-
-    It can be roughly described by:
-        ||∇(u-uₕ)||ₑ ≦ √( C hₑ ∑ₖ (hₖ ∫ |J[∇uₕ]|²) dS )
-    where "e" denotes an element, ||⋅||ₑ the corresponding local norm and k the
-    corresponding faces. u is the analytic solution and uₕ the discretized
-    solution. hₖ and hₑ are factors dependend on the face and element geometry.
-    J is the jump function, i.e. the difference between the limits at each point
-    for each side of the face. A custom method to compute hₖ can be provided. It
-    is also possible to estimate the error only on a subspace by feeding this
-    class an attribute array describing the subspace.
-
-    @note This algorithm is only for Poisson problems a proper error esimator.
-    The current implementation does not reflect this, because the "C" factor is
-    not included.
-    It further assumes that the approximation error at the boundary is small
-    enough, as the implementation ignores boundary faces.
-*/
-class KellyErrorEstimator final : public ErrorEstimator
-{
-public:
-   /// Function type to compute the local coefficient hₑ of an element.
-   using ElementCoefficientFunction =
-      std::function<double(ParMesh*, const int)>;
-   /** @brief Function type to compute the local coefficient hₖ of a face. The
-       third argument is true for shared faces and false for local faces. */
-   using FaceCoefficientFunction =
-      std::function<double(ParMesh*, const int, const bool)>;
-
-private:
-   int current_sequence = -1;
-
-   Vector error_estimates;
-
-   double total_error = 0.0;
-
-   Array<int> attributes;
-
-   /** @brief A method to compute hₑ on per-element basis.
-
-       This method weights the error approximation on the element level.
-
-       Defaults to hₑ=1.0.
-   */
-   ElementCoefficientFunction compute_element_coefficient;
-
-   /** @brief A method to compute hₖ on per-face basis.
-
-       This method weights the error approximation on the face level. The
-       background here is that classical Kelly error estimator implementations
-       approximate the geometrical characteristic hₖ with the face diameter,
-       which should be also be a possibility in this implementation.
-
-       Defaults to hₖ=diameter/2p.
-   */
-   FaceCoefficientFunction compute_face_coefficient;
-
-   BilinearFormIntegrator* flux_integrator; ///< Not owned.
-   ParGridFunction* solution;               ///< Not owned.
-
-   ParFiniteElementSpace*
-   flux_space; /**< @brief Ownership based on own_flux_fes. */
-   bool own_flux_fespace; ///< Ownership flag for flux_space.
-
-   /// Check if the mesh of the solution was modified.
-   bool MeshIsModified()
-   {
-      long mesh_sequence = solution->FESpace()->GetMesh()->GetSequence();
-      MFEM_ASSERT(mesh_sequence >= current_sequence,
-                  "improper mesh update sequence");
-      return (mesh_sequence > current_sequence);
-   }
-
-   /** @brief Compute the element error estimates.
-
-       Algorithm outline:
-       1. Compute flux field for each element
-       2. Add error contribution from local interior faces
-       3. Add error contribution from shared interior faces
-       4. Finalize by computing hₖ and scale errors.
-   */
-   void ComputeEstimates();
-
-public:
-   /** @brief Construct a new KellyErrorEstimator object for a scalar field.
-       @param di_         The bilinearform to compute the interface flux.
-       @param sol_        The solution field whose error is to be estimated.
-       @param flux_fes_   The finite element space for the interface flux.
-       @param attributes_ The attributes of the subdomain(s) for which the
-                          error should be estimated. An empty array results in
-                          estimating the error over the complete domain.
-   */
-   KellyErrorEstimator(BilinearFormIntegrator& di_, ParGridFunction& sol_,
-                       ParFiniteElementSpace& flux_fes_,
-                       const Array<int> &attributes_ = Array<int>());
-
-   /** @brief Construct a new KellyErrorEstimator object for a scalar field.
-       @param di_         The bilinearform to compute the interface flux.
-       @param sol_        The solution field whose error is to be estimated.
-       @param flux_fes_   The finite element space for the interface flux.
-       @param attributes_ The attributes of the subdomain(s) for which the
-                          error should be estimated. An empty array results in
-                          estimating the error over the complete domain.
-   */
-   KellyErrorEstimator(BilinearFormIntegrator& di_, ParGridFunction& sol_,
-                       ParFiniteElementSpace* flux_fes_,
-                       const Array<int> &attributes_ = Array<int>());
-
-   ~KellyErrorEstimator();
-
-   /// Get a Vector with all element errors.
-   const Vector& GetLocalErrors() override
-   {
-      if (MeshIsModified())
-      {
-         ComputeEstimates();
-      }
-      return error_estimates;
-   }
-
-   /// Reset the error estimator.
-   void Reset() override { current_sequence = -1; };
-
-   virtual double GetTotalError() const override { return total_error; }
-
-   /** @brief Change the method to compute hₑ on a per-element basis.
-       @param compute_element_coefficient_
-                        A function taking a mesh and an element index to
-                        compute the local hₑ for the element.
-   */
-   void SetElementCoefficientFunction(ElementCoefficientFunction
-                                      compute_element_coefficient_)
-   {
-      compute_element_coefficient = compute_element_coefficient_;
-   }
-
-   /** @brief Change the method to compute hₖ on a per-element basis.
-       @param compute_face_coefficient_
-                        A function taking a mesh and a face index to
-                        compute the local hₖ for the face.
-   */
-   void SetFaceCoefficientFunction(
-      FaceCoefficientFunction
-      compute_face_coefficient_)
-   {
-      compute_face_coefficient = compute_face_coefficient_;
-   }
-
-   /// Change the coefficients back to default as described above.
-   void ResetCoefficientFunctions();
-};
-
-#endif // MFEM_USE_MPI
-
 } // namespace mfem

 #endif // MFEM_ERROR_ESTIMATORS
@@ -435,7 +435,7 @@ void ScalarFiniteElement::ScalarLocalInterpolation(
   IntegrationPoint f_ip;

   const int fs = fine_fe.GetDof(), cs = this->GetDof();
-   I.SetSize(fs, cs);
+   I.SetSize(fs, cs );
   Vector fine_shape(fs), coarse_shape(cs);
   DenseMatrix fine_mass(fs), fine_coarse_mass(fs, cs); // initialized with 0
   const int ir_order = GetOrder() + fine_fe.GetOrder();
@@ -464,44 +464,6 @@ void ScalarFiniteElement::ScalarLocalInterpolation(
   }
 }

-void ScalarFiniteElement::ScalarLocalRestriction(
-   ElementTransformation &Trans, DenseMatrix &R,
-   const ScalarFiniteElement &coarse_fe) const
-{
-   // General "restriction", defined by L2 projection
-   double v[Geometry::MaxDim];
-   Vector vv (v, dim);
-   IntegrationPoint f_ip;
-
-   const int cs = coarse_fe.GetDof(), fs = this->GetDof();
-   R.SetSize(cs, fs);
-   Vector fine_shape(fs), coarse_shape(cs);
-   DenseMatrix coarse_mass(cs), coarse_fine_mass(cs, fs); // initialized with 0
-   const int ir_order = GetOrder() + coarse_fe.GetOrder();
-   const IntegrationRule &ir = IntRules.Get(coarse_fe.GetGeomType(), ir_order);
-
-   for (int i = 0; i < ir.GetNPoints(); i++)
-   {
-      const IntegrationPoint &ip = ir.IntPoint(i);
-      this->CalcShape(ip, fine_shape);
-      Trans.Transform(ip, vv);
-      f_ip.Set(v, dim);
-      coarse_fe.CalcShape(f_ip, coarse_shape);
-
-      AddMult_a_VVt(ip.weight, coarse_shape, coarse_mass);
-      AddMult_a_VWt(ip.weight, coarse_shape, fine_shape, coarse_fine_mass);
-   }
-
-   DenseMatrixInverse coarse_mass_inv(coarse_mass);
-   coarse_mass_inv.Mult(coarse_fine_mass, R);
-
-   if (map_type == INTEGRAL)
-   {
-      // assuming Trans is linear; this should be ok for all refinement types
-      Trans.SetIntPoint(&Geometries.GetCenter(geom_type));
-      R *= 1.0 / Trans.Weight();
-   }
-}
 const DofToQuad &ScalarFiniteElement::GetDofToQuad(const IntegrationRule &ir,
                                                   DofToQuad::Mode mode) const
 {
@@ -596,22 +558,17 @@ void NodalFiniteElement::ProjectCurl_2D(
   const FiniteElement &fe, ElementTransformation &Trans,
   DenseMatrix &curl) const
 {
+   MFEM_ASSERT(GetMapType() == FiniteElement::INTEGRAL, "");
+
   DenseMatrix curl_shape(fe.GetDof(), 1);

   curl.SetSize(dof, fe.GetDof());
   for (int i = 0; i < dof; i++)
   {
      fe.CalcCurlShape(Nodes.IntPoint(i), curl_shape);
-
-      double w = 1.0;
-      if (GetMapType() == FiniteElement::VALUE)
-      {
-         Trans.SetIntPoint(&Nodes.IntPoint(i));
-         w /= Trans.Weight();
-      }
      for (int j = 0; j < fe.GetDof(); j++)
      {
-         curl(i,j) = w * curl_shape(j,0);
+         curl(i,j) = curl_shape(j,0);
      }
   }
 }
@@ -727,34 +684,17 @@ void NodalFiniteElement::Project(
 {
   if (fe.GetRangeType() == SCALAR)
   {
+      MFEM_ASSERT(map_type == fe.GetMapType(), "");
+
      Vector shape(fe.GetDof());

      I.SetSize(dof, fe.GetDof());
-      if (map_type == fe.GetMapType())
+      for (int k = 0; k < dof; k++)
      {
-         for (int k = 0; k < dof; k++)
+         fe.CalcShape(Nodes.IntPoint(k), shape);
+         for (int j = 0; j < shape.Size(); j++)
         {
-            fe.CalcShape(Nodes.IntPoint(k), shape);
-            for (int j = 0; j < shape.Size(); j++)
-            {
-               I(k,j) = (fabs(shape(j)) < 1e-12) ? 0.0 : shape(j);
-            }
-         }
-      }
-      else
-      {
-         for (int k = 0; k < dof; k++)
-         {
-            Trans.SetIntPoint(&Nodes.IntPoint(k));
-            fe.CalcPhysShape(Trans, shape);
-            if (map_type == INTEGRAL)
-            {
-               shape *= Trans.Weight();
-            }
-            for (int j = 0; j < shape.Size(); j++)
-            {
-               I(k,j) = (fabs(shape(j)) < 1e-12) ? 0.0 : shape(j);
-            }
+            I(k,j) = (fabs(shape(j)) < 1e-12) ? 0.0 : shape(j);
         }
      }
   }
@@ -1053,8 +993,6 @@ void VectorFiniteElement::Project_RT(

         fe.CalcShape(ip, shape);
         Trans.SetIntPoint(&ip);
-         // Transform RT face normals from reference to physical space
-         // vk = adj(J)^T nk
         Trans.AdjugateJacobian().MultTranspose(nk + d2n[k]*dim, vk);
         if (fe.GetMapType() == INTEGRAL)
         {
@@ -1072,8 +1010,6 @@ void VectorFiniteElement::Project_RT(
            {
               s = 0.0;
            }
-            // Project scalar basis function multiplied by each coordinate
-            // direction onto the transformed face normals
            for (int d = 0; d < sdim; d++)
            {
               I(k,j+d*shape.Size()) = s*vk[d];
@@ -1083,31 +1019,7 @@ void VectorFiniteElement::Project_RT(
   }
   else
   {
-      int sdim = Trans.GetSpaceDim();
-      double vk[Geometry::MaxDim];
-      DenseMatrix vshape(fe.GetDof(), sdim);
-      Vector vshapenk(fe.GetDof());
-      const bool square_J = (dim == sdim);
-
-      I.SetSize(dof, fe.GetDof());
-      for (int k = 0; k < dof; k++)
-      {
-         const IntegrationPoint &ip = Nodes.IntPoint(k);
-
-         Trans.SetIntPoint(&ip);
-         // Transform RT face normals from reference to physical space
-         // vk = adj(J)^T nk
-         Trans.AdjugateJacobian().MultTranspose(nk + d2n[k]*dim, vk);
-         // Compute fe basis functions in physical space
-         fe.CalcVShape(Trans, vshape);
-         // Project fe basis functions onto transformed face normals
-         vshape.Mult(vk, vshapenk);
-         if (!square_J) { vshapenk /= Trans.Weight(); }
-         for (int j=0; j<vshapenk.Size(); j++)
-         {
-            I(k,j) = vshapenk(j);
-         }
-      }
+      mfem_error("VectorFiniteElement::Project_RT (fe version)");
   }
 }

@@ -1268,8 +1180,6 @@ void VectorFiniteElement::Project_ND(

         fe.CalcShape(ip, shape);
         Trans.SetIntPoint(&ip);
-         // Transform ND edge tengents from reference to physical space
-         // vk = J tk
         Trans.Jacobian().Mult(tk + d2t[k]*dim, vk);
         if (fe.GetMapType() == INTEGRAL)
         {
@@ -1287,8 +1197,6 @@ void VectorFiniteElement::Project_ND(
            {
               s = 0.0;
            }
-            // Project scalar basis function multiplied by each coordinate
-            // direction onto the transformed edge tangents
            for (int d = 0; d < sdim; d++)
            {
               I(k, j + d*shape.Size()) = s*vk[d];
@@ -1298,29 +1206,7 @@ void VectorFiniteElement::Project_ND(
   }
   else
   {
-      int sdim = Trans.GetSpaceDim();
-      double vk[Geometry::MaxDim];
-      DenseMatrix vshape(fe.GetDof(), sdim);
-      Vector vshapetk(fe.GetDof());
-
-      I.SetSize(dof, fe.GetDof());
-      for (int k = 0; k < dof; k++)
-      {
-         const IntegrationPoint &ip = Nodes.IntPoint(k);
-
-         Trans.SetIntPoint(&ip);
-         // Transform ND edge tangents from reference to physical space
-         // vk = J tk
-         Trans.Jacobian().Mult(tk + d2t[k]*dim, vk);
-         // Compute fe basis functions in physical space
-         fe.CalcVShape(Trans, vshape);
-         // Project fe basis functions onto transformed edge tangents
-         vshape.Mult(vk, vshapetk);
-         for (int j=0; j<vshapetk.Size(); j++)
-         {
-            I(k, j) = vshapetk(j);
-         }
-      }
+      mfem_error("VectorFiniteElement::Project_ND (fe version)");
   }
 }

@@ -504,18 +504,12 @@ public:
   /** @brief Given a coefficient and a transformation, compute its projection
       (approximation) in the local finite dimensional space in terms
       of the degrees of freedom. */
-   /** The approximation used to project is usually local interpolation of
-       degrees of freedom. The derived class could use other methods not
-       implemented yet, e.g. local L2 projection. */
   virtual void Project(Coefficient &coeff,
                        ElementTransformation &Trans, Vector &dofs) const;

   /** @brief Given a vector coefficient and a transformation, compute its
       projection (approximation) in the local finite dimensional space
       in terms of the degrees of freedom. (VectorFiniteElements) */
-   /** The approximation used to project is usually local interpolation of
-       degrees of freedom. The derived class could use other methods not
-       implemented yet, e.g. local L2 projection. */
   virtual void Project(VectorCoefficient &vc,
                        ElementTransformation &Trans, Vector &dofs) const;

@@ -672,7 +666,7 @@ public:
                                const ScalarFiniteElement &fine_fe) const;

   /** @brief Get matrix @a I "Interpolation" defined through local
-       L2-projection in the space defined by the @a fine_fe. */
+       L2-projection in the space defined by the @a fine_fe.  */
   /** If the "fine" elements cannot represent all basis functions of the
       "coarse" element, then boundary values from different sub-elements are
       generally different. */
@@ -680,15 +674,6 @@ public:
                                 DenseMatrix &I,
                                 const ScalarFiniteElement &fine_fe) const;

-   /** @brief Get restriction matrix @a R defined through local L2-projection
-        in the space defined by the @a coarse_fe. */
-   /** If the "fine" elements cannot represent all basis functions of the
-       "coarse" element, then boundary values from different sub-elements are
-       generally different. */
-   void ScalarLocalRestriction(ElementTransformation &Trans,
-                               DenseMatrix &R,
-                               const ScalarFiniteElement &coarse_fe) const;
-
   virtual const DofToQuad &GetDofToQuad(const IntegrationRule &ir,
                                         DofToQuad::Mode mode) const;
 };
@@ -769,10 +754,6 @@ public:
                                      DenseMatrix &I) const
   { ScalarLocalInterpolation(Trans, I, *this); }

-   virtual void GetLocalRestriction(ElementTransformation &Trans,
-                                    DenseMatrix &R) const
-   { ScalarLocalRestriction(Trans, R, *this); }
-
   virtual void GetTransferMatrix(const FiniteElement &fe,
                                  ElementTransformation &Trans,
                                  DenseMatrix &I) const
@@ -819,25 +800,11 @@ protected:
   void CalcVShape_ND(ElementTransformation &Trans,
                      DenseMatrix &shape) const;

-   /** @brief Project a vector coefficient onto the RT basis functions
-       @param nk    Face normal vectors for this element type
-       @param d2n   Offset into nk for each degree of freedom
-       @param vc    Vector coefficient to be projected
-       @param Trans Transformation from reference to physical coordinates
-       @param dofs  Expansion coefficients for the approximation of vc
-   */
   void Project_RT(const double *nk, const Array<int> &d2n,
                   VectorCoefficient &vc, ElementTransformation &Trans,
                   Vector &dofs) const;

   /// Projects the vector of values given at FE nodes to RT space
-   /** Project vector values onto the RT basis functions
-       @param nk    Face normal vectors for this element type
-       @param d2n   Offset into nk for each degree of freedom
-       @param vc    Vector values at each interpolation point
-       @param Trans Transformation from reference to physical coordinates
-       @param dofs  Expansion coefficients for the approximation of vc
-   */
   void Project_RT(const double *nk, const Array<int> &d2n,
                   Vector &vc, ElementTransformation &Trans,
                   Vector &dofs) const;
@@ -847,19 +814,6 @@ protected:
      const double *nk, const Array<int> &d2n,
      MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const;

-   /** @brief Project vector-valued basis functions onto the RT basis functions
-       @param nk    Face normal vectors for this element type
-       @param d2n   Offset into nk for each degree of freedom
-       @param fe    Vector-valued finite element basis
-       @param Trans Transformation from reference to physical coordinates
-       @param I     Expansion coefficients for the approximation of each basis
-                    function
-
-       Note: If the FiniteElement, fe, is scalar-valued the projection will
-             assume that a FiniteElementSpace is being used to define a vector
-             field using the scalar basis functions for each component of the
-             vector field.
-   */
   void Project_RT(const double *nk, const Array<int> &d2n,
                   const FiniteElement &fe, ElementTransformation &Trans,
                   DenseMatrix &I) const;
@@ -879,25 +833,11 @@ protected:
                       const FiniteElement &fe, ElementTransformation &Trans,
                       DenseMatrix &curl) const;

-   /** @brief Project a vector coefficient onto the ND basis functions
-       @param tk    Edge tangent vectors for this element type
-       @param d2t   Offset into tk for each degree of freedom
-       @param vc    Vector coefficient to be projected
-       @param Trans Transformation from reference to physical coordinates
-       @param dofs  Expansion coefficients for the approximation of vc
-   */
   void Project_ND(const double *tk, const Array<int> &d2t,
                   VectorCoefficient &vc, ElementTransformation &Trans,
                   Vector &dofs) const;

   /// Projects the vector of values given at FE nodes to ND space
-   /** Project vector values onto the ND basis functions
-       @param tk    Edge tangent vectors for this element type
-       @param d2t   Offset into tk for each degree of freedom
-       @param vc    Vector values at each interpolation point
-       @param Trans Transformation from reference to physical coordinates
-       @param dofs  Expansion coefficients for the approximation of vc
-   */
   void Project_ND(const double *tk, const Array<int> &d2t,
                   Vector &vc, ElementTransformation &Trans,
                   Vector &dofs) const;
@@ -907,19 +847,6 @@ protected:
      const double *tk, const Array<int> &d2t,
      MatrixCoefficient &mc, ElementTransformation &T, Vector &dofs) const;

-   /** @brief Project vector-valued basis functions onto the ND basis functions
-       @param tk    Edge tangent vectors for this element type
-       @param d2t   Offset into tk for each degree of freedom
-       @param fe    Vector-valued finite element basis
-       @param Trans Transformation from reference to physical coordinates
-       @param I     Expansion coefficients for the approximation of each basis
-                    function
-
-       Note: If the FiniteElement, fe, is scalar-valued the projection will
-             assume that a FiniteElementSpace is being used to define a vector
-             field using the scalar basis functions for each component of the
-             vector field.
-   */
   void Project_ND(const double *tk, const Array<int> &d2t,
                   const FiniteElement &fe, ElementTransformation &Trans,
                   DenseMatrix &I) const;
@@ -2608,9 +2608,9 @@ const Operator &GridTransfer::MakeTrueOperator(
   else // Parallel() == true
   {
 #ifdef MFEM_USE_MPI
+      const SparseMatrix *out_R = fes_out.GetRestrictionMatrix();
      if (oper_type == Operator::Hypre_ParCSR)
      {
-         const SparseMatrix *out_R = fes_out.GetRestrictionMatrix();
         const ParFiniteElementSpace *pfes_in =
            dynamic_cast<const ParFiniteElementSpace *>(&fes_in);
         const ParFiniteElementSpace *pfes_out =
@@ -2638,7 +2638,6 @@ const Operator &GridTransfer::MakeTrueOperator(
      }
      else if (oper_type == Operator::ANY_TYPE)
      {
-         const Operator *out_R = fes_out.GetRestrictionOperator();
         t_oper.Reset(new TripleProductOperator(
                         out_R, &oper, fes_in.GetProlongationMatrix(),
                         false, false, false));
@@ -330,18 +330,6 @@ public:
   virtual const Operator *GetProlongationMatrix() const
   { return GetConformingProlongation(); }

-   /// Return an operator that performs the transpose of GetRestrictionOperator
-   /** The returned operator is owned by the FiniteElementSpace. In serial this
-       is the same as GetProlongationMatrix() */
-   virtual const Operator *GetRestrictionTransposeOperator() const
-   { return GetConformingProlongation(); }
-
-   /// An abstract operator that performs the same action as GetRestrictionMatrix
-   /** In some cases this is an optimized matrix-free implementation. The
-       returned operator is owned by the FiniteElementSpace. */
-   virtual const Operator *GetRestrictionOperator() const
-   { return GetConformingRestriction(); }
-
   /// The returned SparseMatrix is owned by the FiniteElementSpace.
   virtual const SparseMatrix *GetRestrictionMatrix() const
   { return GetConformingRestriction(); }
@@ -583,7 +571,7 @@ public:

   /** @brief Returns pointer to the FiniteElement in the FiniteElementCollection
        associated with i'th element in the mesh object. */
-   virtual const FiniteElement *GetFE(int i) const;
+   const FiniteElement *GetFE(int i) const;

   /** @brief Returns pointer to the FiniteElement in the FiniteElementCollection
        associated with i'th boundary face in the mesh object. */
@@ -768,9 +756,6 @@ public:
   /// Return the total number of quadrature points.
   int GetSize() const { return size; }

-   /// Return the order of the quadrature rule(s) used by all elements.
-   int GetOrder() const { return order; }
-
   /// Returns the mesh
   inline Mesh *GetMesh() const { return mesh; }

@@ -15,10 +15,6 @@
 #include "../mesh/nurbs.hpp"
 #include "../general/text.hpp"

-#ifdef MFEM_USE_MPI
-#include "pfespace.hpp"
-#endif
-
 #include <limits>
 #include <cstring>
 #include <string>
@@ -26,7 +22,6 @@
 #include <iostream>
 #include <algorithm>

-
 namespace mfem
 {

@@ -467,26 +462,15 @@ const
   fes->GetElementDofs(i, dofs);
   fes->DofsToVDofs(vdim-1, dofs);
   const FiniteElement *FElem = fes->GetFE(i);
+   MFEM_ASSERT(FElem->GetMapType() == FiniteElement::VALUE,
+               "invalid FE map type");
   int dof = FElem->GetDof();
   Vector DofVal(dof), loc_data(dof);
   GetSubVector(dofs, loc_data);
-   if (FElem->GetMapType() == FiniteElement::VALUE)
+   for (int k = 0; k < n; k++)
   {
-      for (int k = 0; k < n; k++)
-      {
-         FElem->CalcShape(ir.IntPoint(k), DofVal);
-         vals(k) = DofVal * loc_data;
-      }
-   }
-   else
-   {
-      ElementTransformation *Tr = fes->GetElementTransformation(i);
-      for (int k = 0; k < n; k++)
-      {
-         Tr->SetIntPoint(&ir.IntPoint(k));
-         FElem->CalcPhysShape(*Tr, DofVal);
-         vals(k) = DofVal * loc_data;
-      }
+      FElem->CalcShape(ir.IntPoint(k), DofVal);
+      vals(k) = DofVal * loc_data;
   }
 }

@@ -1000,14 +984,15 @@ void GridFunction::GetVectorValues(ElementTransformation &T,

   if (FElem->GetRangeType() == FiniteElement::SCALAR)
   {
+      MFEM_ASSERT(FElem->GetMapType() == FiniteElement::VALUE,
+                  "invalid FE map type");
      Vector shape(dof);
      int vdim = fes->GetVDim();
      vals.SetSize(vdim, nip);
      for (int j = 0; j < nip; j++)
      {
         const IntegrationPoint &ip = ir.IntPoint(j);
-         T.SetIntPoint(&ip);
-         FElem->CalcPhysShape(T, shape);
+         FElem->CalcShape(ip, shape);

         for (int k = 0; k < vdim; k++)
         {
@@ -1565,16 +1550,18 @@ void GridFunction::GetGradient(ElementTransformation &T, Vector &grad) const
   {
      case ElementTransformation::ELEMENT:
      {
-         const FiniteElement *fe = fes->GetFE(T.ElementNo);
+         const FiniteElement * fe = fes->GetFE(T.ElementNo);
         MFEM_ASSERT(fe->GetMapType() == FiniteElement::VALUE,
                     "invalid FE map type");
         int spaceDim = fes->GetMesh()->SpaceDimension();
         int dim = fe->GetDim(), dof = fe->GetDof();
         DenseMatrix dshape(dof, dim);
         Vector lval, gh(dim);
+         Array<int> dofs;

         grad.SetSize(spaceDim);
-         GetElementDofValues(T.ElementNo, lval);
+         fes->GetElementDofs(T.ElementNo, dofs);
+         GetSubVector(dofs, lval);
         fe->CalcDShape(T.GetIntPoint(), dshape);
         dshape.MultTranspose(lval, gh);
         T.InverseJacobian().MultTranspose(gh, grad);
@@ -1744,13 +1731,6 @@ void GridFunction::GetElementAverages(GridFunction &avgs) const
   }
 }

-void GridFunction::GetElementDofValues(int el, Vector &dof_vals) const
-{
-   Array<int> dof_idx;
-   fes->GetElementVDofs(el, dof_idx);
-   GetSubVector(dof_idx, dof_vals);
-}
-
 void GridFunction::ProjectGridFunction(const GridFunction &src)
 {
   Mesh *mesh = fes->GetMesh();
@@ -3813,15 +3793,7 @@ double ZZErrorEstimator(BilinearFormIntegrator &blfi,
         }
      }
   }
-#ifdef MFEM_USE_MPI
-   auto pfes = dynamic_cast<ParFiniteElementSpace*>(ufes);
-   if (pfes)
-   {
-      auto process_local_error = total_error;
-      MPI_Allreduce(&process_local_error, &total_error, 1, MPI_DOUBLE,
-                    MPI_SUM, pfes->GetComm());
-   }
-#endif // MFEM_USE_MPI
+
   return std::sqrt(total_error);
 }

@@ -325,10 +325,6 @@ public:
       Both FE spaces should be scalar and on the same mesh. */
   void GetElementAverages(GridFunction &avgs) const;

-   /** Sets the output vector @a dof_vals to the values of the degrees of
-       freedom of element @a el. */
-   virtual void GetElementDofValues(int el, Vector &dof_vals) const;
-
   /** Impose the given bounds on the function's DOFs while preserving its local
    *  integral (described in terms of the given weights) on the i'th element
    *  through SLBPQ optimization.
@@ -349,30 +345,14 @@ public:
       projection matrix. */
   void ProjectGridFunction(const GridFunction &src);

-   /** @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). */
   virtual void ProjectCoefficient(Coefficient &coeff);

-   /** @brief Project @a coeff Coefficient to @a this GridFunction, using one
-       element for each degree of freedom in @a dofs and nodal interpolation on
-       that element. */
   void ProjectCoefficient(Coefficient &coeff, Array<int> &dofs, int vd = 0);

-   /** @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).*/
   void ProjectCoefficient(VectorCoefficient &vcoeff);

-   /** @brief Project @a vcoeff VectorCoefficient to @a this GridFunction, using
-       one element for each degree of freedom in @a dofs and nodal interpolation
-       on that element. */
   void ProjectCoefficient(VectorCoefficient &vcoeff, Array<int> &dofs);

-   /** @brief Analogous to the version with argument @a vcoeff VectorCoefficient
-       but using an array of scalar coefficients for each component. */
   void ProjectCoefficient(Coefficient *coeff[]);

   /** @brief Project a discontinuous vector coefficient as a grid function on
@@ -482,10 +462,11 @@ public:

   /// Returns the Face Jumps error for L2 elements, with 1/h scaling.
   MFEM_DEPRECATED
-   double ComputeDGFaceJumpError(Coefficient *exsol,
-                                 Coefficient *ell_coeff,
-                                 double Nu,
-                                 const IntegrationRule *irs[] = NULL) const;
+   virtual double ComputeDGFaceJumpError(Coefficient *exsol,
+                                         Coefficient *ell_coeff,
+                                         double Nu,
+                                         const IntegrationRule *irs[] = NULL)
+   const;

   /** This method is kept for backward compatibility.

@@ -14,40 +14,11 @@
 namespace mfem
 {

-Multigrid::Multigrid()
-   : cycleType(CycleType::VCYCLE), preSmoothingSteps(1), postSmoothingSteps(1)
+Multigrid::Multigrid(const FiniteElementSpaceHierarchy& fespaces_)
+   : fespaces(fespaces_), cycleType(CycleType::VCYCLE), preSmoothingSteps(1),
+     postSmoothingSteps(1)
 {}

-Multigrid::Multigrid(const Array<Operator*>& operators_,
-                     const Array<Solver*>& smoothers_,
-                     const Array<Operator*>& prolongations_,
-                     const Array<bool>& ownedOperators_,
-                     const Array<bool>& ownedSmoothers_,
-                     const Array<bool>& ownedProlongations_)
-   : Solver(operators_.Last()->NumRows()), cycleType(CycleType::VCYCLE),
-     preSmoothingSteps(1), postSmoothingSteps(1),
-     X(operators_.Size()), Y(X.Size()), R(X.Size()), Z(X.Size())
-{
-   operators_.Copy(operators);
-   smoothers_.Copy(smoothers);
-   prolongations_.Copy(prolongations);
-   ownedOperators_.Copy(ownedOperators);
-   ownedSmoothers_.Copy(ownedSmoothers);
-   ownedProlongations_.Copy(ownedProlongations);
-
-   for (int level = 0; level < operators.Size(); ++level)
-   {
-      X[level] = new Vector(operators[level]->NumRows());
-      *X[level] = 0.0;
-      Y[level] = new Vector(operators[level]->NumRows());
-      *Y[level] = 0.0;
-      R[level] = new Vector(operators[level]->NumRows());
-      *R[level] = 0.0;
-      Z[level] = new Vector(operators[level]->NumRows());
-      *Z[level] = 0.0;
-   }
-}
-
 Multigrid::~Multigrid()
 {
   for (int i = 0; i < operators.Size(); ++i)
@@ -66,21 +37,26 @@ Multigrid::~Multigrid()
      delete Z[i];
   }

-   for (int i = 0; i < prolongations.Size(); ++i)
-   {
-      if (ownedProlongations[i])
-      {
-         delete prolongations[i];
-      }
-   }
-
   operators.DeleteAll();
   smoothers.DeleteAll();
-   prolongations.DeleteAll();
   X.DeleteAll();
   Y.DeleteAll();
   R.DeleteAll();
   Z.DeleteAll();
+
+   for (int i = 0; i < bfs.Size(); ++i)
+   {
+      delete bfs[i];
+   }
+
+   bfs.DeleteAll();
+
+   for (int i = 0; i < essentialTrueDofs.Size(); ++i)
+   {
+      delete essentialTrueDofs[i];
+   }
+
+   essentialTrueDofs.DeleteAll();
 }

 void Multigrid::AddLevel(Operator* opr, Solver* smoother, bool ownOperator,
@@ -192,7 +168,8 @@ void Multigrid::Cycle(int level) const
   subtract(*X[level], *R[level], *R[level]);

   // Restrict residual
-   GetProlongationAtLevel(level - 1)->MultTranspose(*R[level], *X[level - 1]);
+   fespaces.GetProlongationAtLevel(level - 1)->MultTranspose(*R[level],
+                                                             *X[level - 1]);

   // Init zeros
   *Y[level - 1] = 0.0;
@@ -209,7 +186,7 @@ void Multigrid::Cycle(int level) const
   }

   // Prolongate
-   GetProlongationAtLevel(level - 1)->Mult(*Y[level - 1], *R[level]);
+   fespaces.GetProlongationAtLevel(level - 1)->Mult(*Y[level - 1], *R[level]);

   // Add update
   *Y[level] += *R[level];
@@ -221,44 +198,16 @@ void Multigrid::Cycle(int level) const
   }
 }

-const Operator* Multigrid::GetProlongationAtLevel(int level) const
-{
-   return prolongations[level];
-}
-
-GeometricMultigrid::~GeometricMultigrid()
-{
-   for (int i = 0; i < bfs.Size(); ++i)
-   {
-      delete bfs[i];
-   }
-
-   bfs.DeleteAll();
-
-   for (int i = 0; i < essentialTrueDofs.Size(); ++i)
-   {
-      delete essentialTrueDofs[i];
-   }
-
-   essentialTrueDofs.DeleteAll();
-}
-
-void GeometricMultigrid::FormFineLinearSystem(Vector& x, Vector& b,
-                                              OperatorHandle& A,
-                                              Vector& X, Vector& B)
+void Multigrid::FormFineLinearSystem(Vector& x, Vector& b, OperatorHandle& A,
+                                     Vector& X, Vector& B)
 {
   bfs.Last()->FormLinearSystem(*essentialTrueDofs.Last(), x, b, A, X, B);
 }

-void GeometricMultigrid::RecoverFineFEMSolution(const Vector& X,
-                                                const Vector& b, Vector& x)
+void Multigrid::RecoverFineFEMSolution(const Vector& X, const Vector& b,
+                                       Vector& x)
 {
   bfs.Last()->RecoverFEMSolution(X, b, x);
 }

-const Operator* GeometricMultigrid::GetProlongationAtLevel(int level) const
-{
-   return fespaces.GetProlongationAtLevel(level);
-}
-
 } // namespace mfem
@@ -32,13 +32,16 @@ public:
   };

 protected:
+   const FiniteElementSpaceHierarchy& fespaces;
+   Array<Array<int>*> essentialTrueDofs;
+   Array<BilinearForm*> bfs;
+
+private:
   Array<Operator*> operators;
   Array<Solver*> smoothers;
-   Array<Operator*> prolongations;

   Array<bool> ownedOperators;
   Array<bool> ownedSmoothers;
-   Array<bool> ownedProlongations;

   CycleType cycleType;
   int preSmoothingSteps;
@@ -50,16 +53,8 @@ protected:
   mutable Array<Vector*> Z;

 public:
-   /// Constructs an empty multigrid hierarchy.
-   Multigrid();
-
-   /// Constructs a multigrid hierarchy from the given inputs.
-   /** Inputs include operators and smoothers on all levels, prolongation
-       operators that go from coarser to finer levels, and ownership of the
-       given operators, smoothers, and prolongations. */
-   Multigrid(const Array<Operator*>& operators_, const Array<Solver*>& smoothers_,
-             const Array<Operator*>& prolongations_, const Array<bool>& ownedOperators_,
-             const Array<bool>& ownedSmoothers_, const Array<bool>& ownedProlongations_);
+   /// Constructs an empty multigrid for the given FiniteElementSpaceHierarchy
+   Multigrid(const FiniteElementSpaceHierarchy& fespaces_);

   /// Destructor
   virtual ~Multigrid();
@@ -94,7 +89,7 @@ public:
   /// Returns smoother at given level
   Solver* GetSmootherAtLevel(int level);

-   /// Set cycle type and number of pre- and post-smoothing steps used by Mult
+   /// Set the cycle type and number of pre- and post-smoothing steps used by Mult
   void SetCycleType(CycleType cycleType_, int preSmoothingSteps_,
                     int postSmoothingSteps_);

@@ -104,36 +99,7 @@ public:
   /// Not supported for multigrid
   virtual void SetOperator(const Operator& op) override;

-private:
-   /// Application of a smoothing step at particular level
-   void SmoothingStep(int level, bool transpose) const;
-
-   /// Application of a multigrid cycle at particular level
-   void Cycle(int level) const;
-
-   /// Returns prolongation operator at given level
-   virtual const Operator* GetProlongationAtLevel(int level) const;
-};
-
-/// Geometric multigrid associated with a hierarchy of finite element spaces
-class GeometricMultigrid : public Multigrid
-{
-protected:
-   const FiniteElementSpaceHierarchy& fespaces;
-   Array<Array<int>*> essentialTrueDofs;
-   Array<BilinearForm*> bfs;
-
-public:
-   /** Construct an empty multigrid object for the given finite element space
-       hierarchy @a fespaces_ */
-   GeometricMultigrid(const FiniteElementSpaceHierarchy& fespaces_)
-      : Multigrid(), fespaces(fespaces_) { }
-
-   /// Destructor
-   virtual ~GeometricMultigrid();
-
-   /** Form the linear system A X = B, corresponding to the operator on the
-       finest level of the geometric multigrid hierarchy */
+   /// Form the linear system A X = B, corresponding to the operator on the finest level
   void FormFineLinearSystem(Vector& x, Vector& b, OperatorHandle& A, Vector& X,
                             Vector& B);

@@ -141,8 +107,11 @@ public:
   void RecoverFineFEMSolution(const Vector& X, const Vector& b, Vector& x);

 private:
-   /// Returns prolongation operator at given level
-   virtual const Operator* GetProlongationAtLevel(int level) const override;
+   /// Application of a smoothing step at particular level
+   void SmoothingStep(int level, bool transpose) const;
+
+   /// Application of a cycle at particular level
+   void Cycle(int level) const;
 };

 } // namespace mfem
@@ -565,38 +565,6 @@ HypreParMatrix* ParDiscreteLinearOperator::ParallelAssemble() const
   return RAP;
 }

-void ParDiscreteLinearOperator::ParallelAssemble(OperatorHandle &A)
-{
-   // construct the rectangular block-diagonal matrix dA
-   OperatorHandle dA(A.Type());
-   dA.MakeRectangularBlockDiag(domain_fes->GetComm(),
-                               range_fes->GlobalVSize(),
-                               domain_fes->GlobalVSize(),
-                               range_fes->GetDofOffsets(),
-                               domain_fes->GetDofOffsets(),
-                               mat);
-
-   OperatorHandle R_test_transpose(A.Type()), P_trial(A.Type());
-
-   // TODO - construct the Dof_TrueDof_Matrix directly in the required format.
-   R_test_transpose.ConvertFrom(range_fes->Dof_TrueDof_Matrix());
-   P_trial.ConvertFrom(domain_fes->Dof_TrueDof_Matrix());
-
-   A.MakeRAP(R_test_transpose, dA, P_trial);
-}
-
-void ParDiscreteLinearOperator::FormRectangularSystemMatrix(OperatorHandle &A)
-{
-   if (ext)
-   {
-      Array<int> empty;
-      ext->FormRectangularSystemOperator(empty, empty, A);
-      return;
-   }
-
-   mfem_error("not implemented!");
-}
-
 void ParDiscreteLinearOperator::GetParBlocks(Array2D<HypreParMatrix *> &blocks)
 const
 {
@@ -160,9 +160,6 @@ public:
   /// Get the parallel finite element space prolongation matrix
   virtual const Operator *GetProlongation() const
   { return pfes->GetProlongationMatrix(); }
-   /// Get the transpose of GetRestriction, useful for matrix-free RAP
-   virtual const Operator *GetRestrictionTranspose() const
-   { return pfes->GetRestrictionTransposeOperator(); }
   /// Get the parallel finite element space restriction matrix
   virtual const Operator *GetRestriction() const
   { return pfes->GetRestrictionMatrix(); }
@@ -304,18 +301,10 @@ public:
   /// Returns the matrix "assembled" on the true dofs
   HypreParMatrix *ParallelAssemble() const;

-   /** @brief Returns the matrix assembled on the true dofs, i.e.
-       @a A = R_test A_local P_trial, in the format (type id) specified by
-       @a A. */
-   void ParallelAssemble(OperatorHandle &A);
-
   /** Extract the parallel blocks corresponding to the vector dimensions of the
       domain and range parallel finite element spaces */
   void GetParBlocks(Array2D<HypreParMatrix *> &blocks) const;

-   /** @brief Return in @a A a parallel (on truedofs) version of this operator. */
-   virtual void FormRectangularSystemMatrix(OperatorHandle &A);
-
   virtual ~ParDiscreteLinearOperator() { }
 };

@@ -101,8 +101,6 @@ void ParFiniteElementSpace::ParInit(ParMesh *pm)

   P = NULL;
   Pconf = NULL;
-   Rconf = NULL;
-   R_transpose = NULL;
   R = NULL;

   num_face_nbr_dofs = -1;
@@ -501,12 +499,6 @@ void ParFiniteElementSpace::GetFaceDofs(int i, Array<int> &dofs) const
   }
 }

-const FiniteElement *ParFiniteElementSpace::GetFE(int i) const
-{
-   int ne = mesh->GetNE();
-   if (i >= ne) { return GetFaceNbrFE(i - ne); }
-   else { return FiniteElementSpace::GetFE(i); }
-}

 const Operator *ParFiniteElementSpace::GetFaceRestriction(
   ElementDofOrdering e_ordering, FaceType type, L2FaceValues mul) const
@@ -929,45 +921,6 @@ const Operator *ParFiniteElementSpace::GetProlongationMatrix() const
   }
 }

-const Operator *ParFiniteElementSpace::GetRestrictionOperator() const
-{
-   if (Conforming())
-   {
-      if (Rconf) { return Rconf; }
-
-      if (NRanks == 1)
-      {
-         R_transpose = new IdentityOperator(GetTrueVSize());
-      }
-      else
-      {
-         if (!Device::Allows(Backend::DEVICE_MASK))
-         {
-            R_transpose = new ConformingProlongationOperator(*this, true);
-         }
-         else
-         {
-            R_transpose =
-               new DeviceConformingProlongationOperator(*this, true);
-         }
-      }
-      Rconf = new TransposeOperator(R_transpose);
-      return Rconf;
-   }
-   else
-   {
-      Dof_TrueDof_Matrix();
-      R_transpose = new TransposeOperator(R);
-      return R;
-   }
-}
-
-const Operator *ParFiniteElementSpace::GetRestrictionTransposeOperator() const
-{
-   GetRestrictionOperator();
-   return R_transpose;
-}
-
 void ParFiniteElementSpace::ExchangeFaceNbrData()
 {
   if (num_face_nbr_dofs >= 0) { return; }
@@ -2881,8 +2834,6 @@ void ParFiniteElementSpace::Destroy()

   delete P; P = NULL;
   delete Pconf; Pconf = NULL;
-   delete Rconf; Rconf = NULL;
-   delete R_transpose; R_transpose = NULL;
   delete R; R = NULL;

   delete gcomm; gcomm = NULL;
@@ -3008,12 +2959,12 @@ void ParFiniteElementSpace::Update(bool want_transform)
   }
 }

+
 ConformingProlongationOperator::ConformingProlongationOperator(
-   const ParFiniteElementSpace &pfes, bool local_)
+   const ParFiniteElementSpace &pfes)
   : Operator(pfes.GetVSize(), pfes.GetTrueVSize()),
     external_ldofs(),
-     gc(pfes.GroupComm()),
-     local(local_)
+     gc(pfes.GroupComm())
 {
   MFEM_VERIFY(pfes.Conforming(), "");
   const Table &group_ldof = gc.GroupLDofTable();
@@ -3062,14 +3013,7 @@ void ConformingProlongationOperator::Mult(const Vector &x, Vector &y) const
   const int m = external_ldofs.Size();

   const int in_layout = 2; // 2 - input is ltdofs array
-   if (local)
-   {
-      y = 0.0;
-   }
-   else
-   {
-      gc.BcastBegin(const_cast<double*>(xdata), in_layout);
-   }
+   gc.BcastBegin(const_cast<double*>(xdata), in_layout);

   int j = 0;
   for (int i = 0; i < m; i++)
@@ -3081,10 +3025,7 @@ void ConformingProlongationOperator::Mult(const Vector &x, Vector &y) const
   std::copy(xdata+j-m, xdata+Width(), ydata+j);

   const int out_layout = 0; // 0 - output is ldofs array
-   if (!local)
-   {
-      gc.BcastEnd(ydata, out_layout);
-   }
+   gc.BcastEnd(ydata, out_layout);
 }

 void ConformingProlongationOperator::MultTranspose(
@@ -3097,10 +3038,7 @@ void ConformingProlongationOperator::MultTranspose(
   double *ydata = y.HostWrite();
   const int m = external_ldofs.Size();

-   if (!local)
-   {
-      gc.ReduceBegin(xdata);
-   }
+   gc.ReduceBegin(xdata);

   int j = 0;
   for (int i = 0; i < m; i++)
@@ -3112,18 +3050,13 @@ void ConformingProlongationOperator::MultTranspose(
   std::copy(xdata+j, xdata+Height(), ydata+j-m);

   const int out_layout = 2; // 2 - output is an array on all ltdofs
-   if (!local)
-   {
-      gc.ReduceEnd<double>(ydata, out_layout, GroupCommunicator::Sum);
-   }
+   gc.ReduceEnd<double>(ydata, out_layout, GroupCommunicator::Sum);
 }

 DeviceConformingProlongationOperator::DeviceConformingProlongationOperator(
-   const ParFiniteElementSpace &pfes,
-   bool local_) :
+   const ParFiniteElementSpace &pfes) :
   ConformingProlongationOperator(pfes),
-   mpi_gpu_aware(Device::GetGPUAwareMPI()),
-   local(local_)
+   mpi_gpu_aware(Device::GetGPUAwareMPI())
 {
   MFEM_ASSERT(pfes.Conforming(), "internal error");
   const SparseMatrix *R = pfes.GetRestrictionMatrix();
@@ -3240,42 +3173,32 @@ void DeviceConformingProlongationOperator::Mult(const Vector &x,
                                                Vector &y) const
 {
   const GroupTopology &gtopo = gc.GetGroupTopology();
+   BcastBeginCopy(x); // copy to 'shr_buf'
   int req_counter = 0;
-   if (local)
+   for (int nbr = 1; nbr < gtopo.GetNumNeighbors(); nbr++)
   {
-      y = 0.0;
-   }
-   else
-   {
-      BcastBeginCopy(x); // copy to 'shr_buf'
-      for (int nbr = 1; nbr < gtopo.GetNumNeighbors(); nbr++)
+      const int send_offset = shr_buf_offsets[nbr];
+      const int send_size = shr_buf_offsets[nbr+1] - send_offset;
+      if (send_size > 0)
      {
-         const int send_offset = shr_buf_offsets[nbr];
-         const int send_size = shr_buf_offsets[nbr+1] - send_offset;
-         if (send_size > 0)
-         {
-            auto send_buf = mpi_gpu_aware ? shr_buf.Read() : shr_buf.HostRead();
-            MPI_Isend(send_buf + send_offset, send_size, MPI_DOUBLE,
-                      gtopo.GetNeighborRank(nbr), 41822,
-                      gtopo.GetComm(), &requests[req_counter++]);
-         }
-         const int recv_offset = ext_buf_offsets[nbr];
-         const int recv_size = ext_buf_offsets[nbr+1] - recv_offset;
-         if (recv_size > 0)
-         {
-            auto recv_buf = mpi_gpu_aware ? ext_buf.Write() : ext_buf.HostWrite();
-            MPI_Irecv(recv_buf + recv_offset, recv_size, MPI_DOUBLE,
-                      gtopo.GetNeighborRank(nbr), 41822,
-                      gtopo.GetComm(), &requests[req_counter++]);
-         }
+         auto send_buf = mpi_gpu_aware ? shr_buf.Read() : shr_buf.HostRead();
+         MPI_Isend(send_buf + send_offset, send_size, MPI_DOUBLE,
+                   gtopo.GetNeighborRank(nbr), 41822,
+                   gtopo.GetComm(), &requests[req_counter++]);
+      }
+      const int recv_offset = ext_buf_offsets[nbr];
+      const int recv_size = ext_buf_offsets[nbr+1] - recv_offset;
+      if (recv_size > 0)
+      {
+         auto recv_buf = mpi_gpu_aware ? ext_buf.Write() : ext_buf.HostWrite();
+         MPI_Irecv(recv_buf + recv_offset, recv_size, MPI_DOUBLE,
+                   gtopo.GetNeighborRank(nbr), 41822,
+                   gtopo.GetComm(), &requests[req_counter++]);
      }
   }
   BcastLocalCopy(x, y);
-   if (!local)
-   {
-      MPI_Waitall(req_counter, requests, MPI_STATUSES_IGNORE);
-      BcastEndCopy(y); // copy from 'ext_buf'
-   }
+   MPI_Waitall(req_counter, requests, MPI_STATUSES_IGNORE);
+   BcastEndCopy(y); // copy from 'ext_buf'
 }

 DeviceConformingProlongationOperator::~DeviceConformingProlongationOperator()
@@ -3338,38 +3261,32 @@ void DeviceConformingProlongationOperator::MultTranspose(const Vector &x,
                                                         Vector &y) const
 {
   const GroupTopology &gtopo = gc.GetGroupTopology();
+   ReduceBeginCopy(x); // copy to 'ext_buf'
   int req_counter = 0;
-   if (!local)
+   for (int nbr = 1; nbr < gtopo.GetNumNeighbors(); nbr++)
   {
-      ReduceBeginCopy(x); // copy to 'ext_buf'
-      for (int nbr = 1; nbr < gtopo.GetNumNeighbors(); nbr++)
+      const int send_offset = ext_buf_offsets[nbr];
+      const int send_size = ext_buf_offsets[nbr+1] - send_offset;
+      if (send_size > 0)
      {
-         const int send_offset = ext_buf_offsets[nbr];
-         const int send_size = ext_buf_offsets[nbr+1] - send_offset;
-         if (send_size > 0)
-         {
-            auto send_buf = mpi_gpu_aware ? ext_buf.Read() : ext_buf.HostRead();
-            MPI_Isend(send_buf + send_offset, send_size, MPI_DOUBLE,
-                      gtopo.GetNeighborRank(nbr), 41823,
-                      gtopo.GetComm(), &requests[req_counter++]);
-         }
-         const int recv_offset = shr_buf_offsets[nbr];
-         const int recv_size = shr_buf_offsets[nbr+1] - recv_offset;
-         if (recv_size > 0)
-         {
-            auto recv_buf = mpi_gpu_aware ? shr_buf.Write() : shr_buf.HostWrite();
-            MPI_Irecv(recv_buf + recv_offset, recv_size, MPI_DOUBLE,
-                      gtopo.GetNeighborRank(nbr), 41823,
-                      gtopo.GetComm(), &requests[req_counter++]);
-         }
+         auto send_buf = mpi_gpu_aware ? ext_buf.Read() : ext_buf.HostRead();
+         MPI_Isend(send_buf + send_offset, send_size, MPI_DOUBLE,
+                   gtopo.GetNeighborRank(nbr), 41823,
+                   gtopo.GetComm(), &requests[req_counter++]);
+      }
+      const int recv_offset = shr_buf_offsets[nbr];
+      const int recv_size = shr_buf_offsets[nbr+1] - recv_offset;
+      if (recv_size > 0)
+      {
+         auto recv_buf = mpi_gpu_aware ? shr_buf.Write() : shr_buf.HostWrite();
+         MPI_Irecv(recv_buf + recv_offset, recv_size, MPI_DOUBLE,
+                   gtopo.GetNeighborRank(nbr), 41823,
+                   gtopo.GetComm(), &requests[req_counter++]);
      }
   }
   ReduceLocalCopy(x, y);
-   if (!local)
-   {
-      MPI_Waitall(req_counter, requests, MPI_STATUSES_IGNORE);
-      ReduceEndAssemble(y); // assemble from 'shr_buf'
-   }
+   MPI_Waitall(req_counter, requests, MPI_STATUSES_IGNORE);
+   ReduceEndAssemble(y); // assemble from 'shr_buf'
 }

 } // namespace mfem
@@ -75,12 +75,6 @@ private:

   /// The (block-diagonal) matrix R (restriction of dof to true dof). Owned.
   mutable SparseMatrix *R;
-   /// Optimized action-only restriction operator for conforming meshes. Owned.
-   mutable Operator *Rconf;
-   /** Transpose of R or Rconf. For conforming mesh, this is a matrix-free
-       (Device)ConformingProlongationOperator, for a non-conforming mesh
-       this is a TransposeOperator wrapping R. */
-   mutable Operator *R_transpose;

   ParNURBSExtension *pNURBSext() const
   { return dynamic_cast<ParNURBSExtension *>(NURBSext); }
@@ -270,12 +264,6 @@ public:
       including the dofs for the edges and the vertices of the face. */
   virtual void GetFaceDofs(int i, Array<int> &dofs) const;

-   /** Returns pointer to the FiniteElement in the FiniteElementCollection
-       associated with i'th element in the mesh object. If @a i is greater than
-       or equal to the number of local mesh elements, @a i will be interpreted
-       as a shifted index of a face neigbor element. */
-   virtual const FiniteElement *GetFE(int i) const;
-
   /** Returns an Operator that converts L-vectors to E-vectors on each face.
       The parallel version is different from the serial one because of the
       presence of shared faces. Shared faces are treated as interior faces,
@@ -347,16 +335,6 @@ public:
   HYPRE_Int GetMyTDofOffset() const;

   virtual const Operator *GetProlongationMatrix() const;
-   /** @brief Return logical transpose of restriction matrix, but in
-       non-assembled optimized matrix-free form.
-
-       The implementation is like GetProlongationMatrix, but it sets local
-       DOFs to the true DOF values if owned locally, otherwise zero. */
-   virtual const Operator *GetRestrictionTransposeOperator() const;
-   /** Get an Operator that performs the action of GetRestrictionMatrix(),
-       but potentially with a non-assembled optimized matrix-free
-       implementation. */
-   virtual const Operator *GetRestrictionOperator() const;
   /// Get the R matrix which restricts a local dof vector to true dof vector.
   virtual const SparseMatrix *GetRestrictionMatrix() const
   { Dof_TrueDof_Matrix(); return R; }
@@ -411,11 +389,9 @@ class ConformingProlongationOperator : public Operator
 protected:
   Array<int> external_ldofs;
   const GroupCommunicator &gc;
-   bool local;

 public:
-   ConformingProlongationOperator(const ParFiniteElementSpace &pfes,
-                                  bool local_=false);
+   ConformingProlongationOperator(const ParFiniteElementSpace &pfes);

   virtual void Mult(const Vector &x, Vector &y) const;

@@ -434,8 +410,6 @@ protected:
   Array<int> ltdof_ldof, unq_ltdof;
   Array<int> unq_shr_i, unq_shr_j;
   MPI_Request *requests;
-   bool local;
-
   // Kernel: copy ltdofs from 'src' to 'shr_buf' - prepare for send.
   //         shr_buf[i] = src[shr_ltdof[i]]
   void BcastBeginCopy(const Vector &src) const;
@@ -461,8 +435,7 @@ protected:
   void ReduceEndAssemble(Vector &dst) const;

 public:
-   DeviceConformingProlongationOperator(const ParFiniteElementSpace &pfes,
-                                        bool local_=false);
+   DeviceConformingProlongationOperator(const ParFiniteElementSpace &pfes);

   virtual ~DeviceConformingProlongationOperator();

@@ -471,25 +471,6 @@ void ParGridFunction::GetVectorValue(ElementTransformation &T,
   }
 }

-void ParGridFunction::GetElementDofValues(int el, Vector &dof_vals) const
-{
-   int ne = fes->GetNE();
-   if (el >= ne)
-   {
-      MFEM_ASSERT(face_nbr_data.Size() > 0,
-                  "ParGridFunction::GetElementDofValues: ExchangeFaceNbrData "
-                  "must be called before accessing face neighbor elements.");
-      // Face neighbor element
-      Array<int> dof_idx;
-      pfes->GetFaceNbrElementVDofs(el - ne, dof_idx);
-      face_nbr_data.GetSubVector(dof_idx, dof_vals);
-   }
-   else
-   {
-      GridFunction::GetElementDofValues(el, dof_vals);
-   }
-}
-
 void ParGridFunction::ProjectCoefficient(Coefficient &coeff)
 {
   DeltaCoefficient *delta_c = dynamic_cast<DeltaCoefficient *>(&coeff);
@@ -221,12 +221,6 @@ public:
                               const IntegrationPoint &ip,
                               Vector &val, Vector *tr = NULL) 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
-       of local elements, it will be interpreted as a shifted index of a face
-       neighbor element. */
-   virtual void GetElementDofValues(int el, Vector &dof_vals) const;
-
   using GridFunction::ProjectCoefficient;
   virtual void ProjectCoefficient(Coefficient &coeff);

@@ -420,35 +420,6 @@ void QuadratureInterpolator::MultTranspose(
   MFEM_ABORT("this method is not implemented yet");
 }

-static void D2QValues1D(const int NE,
-                        const Array<double> &b_,
-                        const Vector &x_,
-                        Vector &y_,
-                        const int vdim = 1,
-                        const int d1d = 0,
-                        const int q1d = 0)
-{
-   auto b = Reshape(b_.Read(), q1d, d1d);
-   auto x = Reshape(x_.Read(), d1d, vdim, NE);
-   auto y = Reshape(y_.Write(), vdim, q1d, NE);
-
-   MFEM_FORALL(e, NE,
-   {
-      for (int c = 0; c < vdim; c++)
-      {
-         for (int q = 0; q < q1d; ++q)
-         {
-            double val = 0.0;
-            for (int d = 0; d < d1d; ++d)
-            {
-               val += b(q, d) * x(d, c, e);
-            }
-            y(c, q, e) = val;
-         }
-      }
-   });
-}
-

 template<int T_VDIM = 0, int T_D1D = 0, int T_Q1D = 0, int T_NBZ = 0>
 static void D2QValues2D(const int NE,
@@ -660,15 +631,6 @@ static void D2QValues(const FiniteElementSpace &fes,
   const int Q1D = maps->nqpt;
   const int id = (vdim<<8) | (D1D<<4) | Q1D;

-   if (dim == 1)
-   {
-      MFEM_VERIFY(D1D <= MAX_D1D, "Orders higher than " << MAX_D1D-1
-                  << " are not supported!");
-      MFEM_VERIFY(Q1D <= MAX_Q1D, "Quadrature rules with more than "
-                  << MAX_Q1D << " 1D points are not supported!");
-      D2QValues1D(NE, maps->B, e_vec, q_val, vdim, D1D, Q1D);
-      return;
-   }
   if (dim == 2)
   {
      switch (id)
@@ -195,31 +195,6 @@ void ElementRestriction::MultTransposeUnsigned(const Vector& x, Vector& y) const
   });
 }

-void ElementRestriction::MultLeftInverse(const Vector& x, Vector& y) const
-{
-   // Assumes all elements have the same number of dofs
-   const int nd = dof;
-   const int vd = vdim;
-   const bool t = byvdim;
-   auto d_offsets = offsets.Read();
-   auto d_indices = indices.Read();
-   auto d_x = Reshape(x.Read(), nd, vd, ne);
-   auto d_y = Reshape(y.Write(), t?vd:ndofs, t?ndofs:vd);
-   MFEM_FORALL(i, ndofs,
-   {
-      const int nextOffset = d_offsets[i + 1];
-      for (int c = 0; c < vd; ++c)
-      {
-         double dofValue = 0;
-         const int j = nextOffset - 1;
-         const int idx_j = (d_indices[j] >= 0) ? d_indices[j] : -1 - d_indices[j];
-         dofValue = (d_indices[j] >= 0) ? d_x(idx_j % nd, c, idx_j / nd) :
-         -d_x(idx_j % nd, c, idx_j / nd);
-         d_y(t?c:i,t?i:c) = dofValue;
-      }
-   });
-}
-
 void ElementRestriction::BooleanMask(Vector& y) const
 {
   // Assumes all elements have the same number of dofs
@@ -57,10 +57,6 @@ public:
   /// Compute MultTranspose without applying signs based on DOF orientations.
   void MultTransposeUnsigned(const Vector &x, Vector &y) const;

-   /// Compute MultTranspose by setting (rather than adding) element
-   /// contributions; this is a left inverse of the Mult() operation
-   void MultLeftInverse(const Vector &x, Vector &y) const;
-
   /// @brief Fills the E-vector y with `boolean` values 0.0 and 1.0 such that each
   /// each entry of the L-vector is uniquely represented in `y`.
   /** This means, the sum of the E-vector `y` is equal to the sum of the
@@ -19,42 +19,6 @@ namespace mfem

 // Target-matrix optimization paradigm (TMOP) mesh quality metrics.

-double TMOP_Combo_QualityMetric::EvalW(const DenseMatrix &Jpt) const
-{
-   double metric = 0.;
-   for (int i = 0; i < tmop_q_arr.Size(); i++)
-   {
-      metric += wt_arr[i]*tmop_q_arr[i]->EvalW(Jpt);
-   }
-   return metric;
-}
-
-void TMOP_Combo_QualityMetric::EvalP(const DenseMatrix &Jpt,
-                                     DenseMatrix &P) const
-{
-   DenseMatrix Pt(P.Size());
-   for (int i = 0; i < tmop_q_arr.Size(); i++)
-   {
-      tmop_q_arr[i]->EvalP(Jpt, Pt);
-      Pt *= wt_arr[i];
-      P += Pt;
-   }
-}
-
-void TMOP_Combo_QualityMetric::AssembleH(const DenseMatrix &Jpt,
-                                         const DenseMatrix &DS,
-                                         const double weight,
-                                         DenseMatrix &A) const
-{
-   DenseMatrix At(A.Size());
-   for (int i = 0; i < tmop_q_arr.Size(); i++)
-   {
-      tmop_q_arr[i]->AssembleH(Jpt, DS, weight, At);
-      At *= wt_arr[i];
-      A += At;
-   }
-}
-
 double TMOP_Metric_001::EvalW(const DenseMatrix &Jpt) const
 {
   ie.SetJacobian(Jpt.GetData());
@@ -196,6 +160,23 @@ double TMOP_Metric_aspratio3D::EvalW(const DenseMatrix &Jpt) const
          ) / 3.0;
 }

+// mu_14 = |T-I|^2
+double TMOP_Metric_SSA2D::EvalW(const DenseMatrix &Jpt) const
+{
+   MFEM_VERIFY(Jtr != NULL,
+               "Requires a target Jacobian, use SetTargetJacobian().");
+
+   DenseMatrix Id(2,2);
+
+   Id(0,0) = 1; Id(0,1) = 0;
+   Id(1,0) = 0; Id(1,1) = 1;
+
+   DenseMatrix Mat(2,2);
+   Mat = Jpt;
+   Mat.Add(-1,Id);
+   return Mat.FNorm2();
+}
+
 double TMOP_Metric_002::EvalW(const DenseMatrix &Jpt) const
 {
   ie.SetJacobian(Jpt.GetData());
@@ -291,41 +272,13 @@ void TMOP_Metric_009::AssembleH(const DenseMatrix &Jpt,
   ie.Assemble_ddI1b(weight, A.GetData());
 }

-// mu_14 = |T-I|^2
-double TMOP_Metric_014::EvalW(const DenseMatrix &Jpt) const
-{
-   MFEM_VERIFY(Jtr != NULL,
-               "Requires a target Jacobian, use SetTargetJacobian().");
-
-   DenseMatrix Id(2,2);
-
-   Id(0,0) = 1; Id(0,1) = 0;
-   Id(1,0) = 0; Id(1,1) = 1;
-
-   DenseMatrix Mat(2,2);
-   Mat = Jpt;
-   Mat.Add(-1,Id);
-   return Mat.FNorm2();
-}
-
 double TMOP_Metric_022::EvalW(const DenseMatrix &Jpt) const
 {
   // mu_22 = (0.5*|J|^2 - det(J)) / (det(J) - tau0)
   //       = (0.5*I1 - I2b) / (I2b - tau0)
   ie.SetJacobian(Jpt.GetData());
   const double I2b = ie.Get_I2b();
-
-   double d = I2b - min_detT;
-   if (d < 0.0 && min_detT == 0.0)
-   {
-      // The mesh has been untangled, but it's still possible to get negative
-      // detJ in FD calculations, as they move the nodes around with some small
-      // increments and can produce negative determinants. Thus we put a small
-      // value in the denominator. Note that here I2b < 0.
-      d = - I2b * 0.1;
-   }
-
-   return (0.5*ie.Get_I1() - I2b) / d;
+   return (0.5*ie.Get_I1() - I2b) / (I2b - tau0);
 }

 void TMOP_Metric_022::EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const
@@ -334,8 +287,8 @@ void TMOP_Metric_022::EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const
   // P = 1/(I2b - tau0)*(0.5*dI1 - dI2b) - (0.5*I1 - I2b)/(I2b - tau0)^2*dI2b
   //   = 0.5/(I2b - tau0)*dI1 + (tau0 - 0.5*I1)/(I2b - tau0)^2*dI2b
   ie.SetJacobian(Jpt.GetData());
-   const double c1 = 1.0/(ie.Get_I2b() - min_detT);
-   Add(c1/2, ie.Get_dI1(), (min_detT - ie.Get_I1()/2)*c1*c1, ie.Get_dI2b(), P);
+   const double c1 = 1.0/(ie.Get_I2b() - tau0);
+   Add(c1/2, ie.Get_dI1(), (tau0 - ie.Get_I1()/2)*c1*c1, ie.Get_dI2b(), P);
 }

 void TMOP_Metric_022::AssembleH(const DenseMatrix &Jpt,
@@ -355,10 +308,10 @@ void TMOP_Metric_022::AssembleH(const DenseMatrix &Jpt,
   //      +0.5/(I2b - tau0)*ddI1 + z*ddI2b
   ie.SetJacobian(Jpt.GetData());
   ie.SetDerivativeMatrix(DS.Height(), DS.GetData());
-   const double c1 = 1.0/(ie.Get_I2b() - min_detT);
+   const double c1 = 1.0/(ie.Get_I2b() - tau0);
   const double c2 = weight*c1/2;
   const double c3 = c1*c2;
-   const double c4 = (2*min_detT - ie.Get_I1())*c3; // weight*z
+   const double c4 = (2*tau0 - ie.Get_I1())*c3; // weight*z
   ie.Assemble_TProd(-c3, ie.Get_dI1(), ie.Get_dI2b(), A.GetData());
   ie.Assemble_TProd(-2*c1*c4, ie.Get_dI2b(), A.GetData());
   ie.Assemble_ddI1(c2, A.GetData());
@@ -531,23 +484,6 @@ double TMOP_Metric_085::EvalW(const DenseMatrix &Jpt) const
   return Mat.FNorm2();
 }

-// mu_98 = 1/(tau)|T-I|^2
-double TMOP_Metric_098::EvalW(const DenseMatrix &Jpt) const
-{
-   MFEM_VERIFY(Jtr != NULL,
-               "Requires a target Jacobian, use SetTargetJacobian().");
-
-   DenseMatrix Id(2,2);
-
-   Id(0,0) = 1; Id(0,1) = 0;
-   Id(1,0) = 0; Id(1,1) = 1;
-
-   DenseMatrix Mat(2,2);
-   Mat = Jpt;
-   Mat.Add(-1,Id);
-   return Mat.FNorm2()/Jtr->Det();
-}
-
 double TMOP_Metric_211::EvalW(const DenseMatrix &Jpt) const
 {
   // mu_211 = (det(J) - 1)^2 - det(J) + (det(J)^2 + eps)^{1/2}
@@ -714,71 +650,6 @@ void TMOP_Metric_303::AssembleH(const DenseMatrix &Jpt,
   ie.Assemble_ddI1b(weight/3., A.GetData());
 }

-double TMOP_Metric_311::EvalW(const DenseMatrix &Jpt) const
-{
-   // mu_311 = (det(J) - 1)^2 - det(J) + (det(J)^2 + eps)^{1/2}
-   //        = (I3b - 1)^2 - I3b + sqrt(I3b^2 + eps)
-   ie.SetJacobian(Jpt.GetData());
-   const double I3b = ie.Get_I3b();
-   return (I3b - 1.0)*(I3b - 1.0) - I3b + std::sqrt(I3b*I3b + eps);
-}
-
-void TMOP_Metric_311::EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const
-{
-   ie.SetJacobian(Jpt.GetData());
-   const double I3b = ie.Get_I3b();
-   const double c = 2*I3b-3+(I3b)/(std::pow((I3b*I3b+eps),0.5));
-   P.Set(c, ie.Get_dI3b());
-}
-
-void TMOP_Metric_311::AssembleH(const DenseMatrix &Jpt,
-                                const DenseMatrix &DS,
-                                const double weight,
-                                DenseMatrix &A) const
-{
-   ie.SetJacobian(Jpt.GetData());
-   ie.SetDerivativeMatrix(DS.Height(), DS.GetData());
-   const double I3b = ie.Get_I3b();
-   const double c0 = I3b*I3b+eps;
-   const double c1 = 2 + 1/(pow(c0,0.5)) - I3b*I3b/(pow(c0,1.5));
-   const double c2 = 2*I3b - 3 + I3b/(pow(c0,0.5));
-   ie.Assemble_TProd(weight*c1, ie.Get_dI3b(), A.GetData());
-   ie.Assemble_ddI3b(c2*weight, A.GetData());
-}
-
-double TMOP_Metric_313::EvalW(const DenseMatrix &Jpt) const
-{
-   ie.SetJacobian(Jpt.GetData());
-
-   const double I3b = ie.Get_I3b();
-   double d = I3b - min_detT;
-   if (d < 0.0 && min_detT == 0.0)
-   {
-      // The mesh has been untangled, but it's still possible to get negative
-      // detJ in FD calculations, as they move the nodes around with some small
-      // increments and can produce negative determinants. Thus we put a small
-      // value in the denominator. Note that here I3b < 0.
-      d = - I3b * 0.1;
-   }
-
-   const double c = std::pow(d, -2.0/3.0);
-
-   return ie.Get_I1() * c / 3.0;
-}
-
-void TMOP_Metric_313::EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const
-{
-   MFEM_ABORT("Metric not implemented yet.");
-}
-
-void TMOP_Metric_313::AssembleH(const DenseMatrix &Jpt,
-                                const DenseMatrix &DS,
-                                const double weight,
-                                DenseMatrix &A) const
-{
-   MFEM_ABORT("Metric not implemented yet.");
-}
-
 double TMOP_Metric_315::EvalW(const DenseMatrix &Jpt) const
 {
   // mu_315 = mu_15_3D = (det(J) - 1)^2
@@ -929,84 +800,6 @@ void TMOP_Metric_352::AssembleH(const DenseMatrix &Jpt,
   ie.Assemble_ddI3b(weight*(c - 0.5*c*c), A.GetData());
 }

-double TMOP_AMetric_011::EvalW(const DenseMatrix &Jpt) const
-{
-   MFEM_VERIFY(Jtr != NULL,
-               "Requires a target Jacobian, use SetTargetJacobian().");
-
-   int dim = Jpt.Size();
-
-   DenseMatrix Jpr(dim, dim);
-   Mult(Jpt, *Jtr, Jpr);
-
-   double alpha = Jpr.Det(),
-          omega = Jtr->Det();
-
-   DenseMatrix AdjAt(dim), WtW(dim), WRK(dim), Jtrt(dim);
-   CalcAdjugateTranspose(Jpr, AdjAt);
-   Jtrt.Transpose(*Jtr);
-   MultAAt(Jtrt, WtW);
-   WtW *= 1./omega;
-   Mult(AdjAt, WtW, WRK);
-
-   WRK -= Jpr;
-   WRK *= -1.;
-
-   return (0.25/alpha)*WRK.FNorm2();
-}
-
-double TMOP_AMetric_014a::EvalW(const DenseMatrix &Jpt) const
-{
-   MFEM_VERIFY(Jtr != NULL,
-               "Requires a target Jacobian, use SetTargetJacobian().");
-
-   int dim = Jpt.Size();
-
-   DenseMatrix Jpr(dim, dim);
-   Mult(Jpt, *Jtr, Jpr);
-
-   double sqalpha = pow(Jpr.Det(), 0.5),
-          sqomega = pow(Jtr->Det(), 0.5);
-
-   return 0.5*pow(sqalpha/sqomega - sqomega/sqalpha, 2.);
-}
-
-double TMOP_AMetric_036::EvalW(const DenseMatrix &Jpt) const
-{
-   MFEM_VERIFY(Jtr != NULL,
-               "Requires a target Jacobian, use SetTargetJacobian().");
-
-   int dim = Jpt.Size();
-
-   DenseMatrix Jpr(dim, dim);
-   Mult(Jpt, *Jtr, Jpr); // T*W = A
-
-   double alpha = Jpr.Det(); // det(A)
-   Jpr -= *Jtr; // A-W
-
-   return (1./alpha)*(Jpr.FNorm2()); //(1/alpha)*(|A-W|^2)
-}
-
-double TMOP_AMetric_107a::EvalW(const DenseMatrix &Jpt) const
-{
-   MFEM_VERIFY(Jtr != NULL,
-               "Requires a target Jacobian, use SetTargetJacobian().");
-
-   int dim = Jpt.Size();
-
-   DenseMatrix Jpr(dim, dim);
-   Mult(Jpt, *Jtr, Jpr);
-
-   double alpha = Jpr.Det(),
-          aw = Jpr.FNorm()/Jtr->FNorm();
-
-   DenseMatrix W = *Jtr;
-   W *= aw;
-   Jpr -= W;
-
-   return (0.5/alpha)*Jpr.FNorm2();
-}
-

 void TargetConstructor::ComputeAvgVolume() const
 {
@@ -40,7 +40,7 @@ public:
       The specified Jacobian matrix, #Jtr, can be used by metrics that cannot
       be written just as a function of the target->physical Jacobian matrix,
       Jpt. */
-   virtual void SetTargetJacobian(const DenseMatrix &_Jtr) { Jtr = &_Jtr; }
+   void SetTargetJacobian(const DenseMatrix &_Jtr) { Jtr = &_Jtr; }

   /** @brief Evaluate the strain energy density function, W = W(Jpt).
       @param[in] Jpt  Represents the target->physical transformation
@@ -70,37 +70,8 @@ public:
                          const double weight, DenseMatrix &A) const = 0;
 };

-/// Abstract class used to define combination of metrics with constant coefficients.
-class TMOP_Combo_QualityMetric : public TMOP_QualityMetric
-{
-protected:
-   Array<TMOP_QualityMetric *> tmop_q_arr; //not owned
-   Array<double> wt_arr;

-public:
-   virtual void AddQualityMetric(TMOP_QualityMetric *tq, double wt = 1.0)
-   {
-      tmop_q_arr.Append(tq);
-      wt_arr.Append(wt);
-   }
-
-   virtual void SetTargetJacobian(const DenseMatrix &_Jtr)
-   {
-      for (int i = 0; i < tmop_q_arr.Size(); i++)
-      {
-         tmop_q_arr[i]->SetTargetJacobian(_Jtr);
-      }
-   }
-
-   virtual double EvalW(const DenseMatrix &Jpt) const;
-
-   virtual void EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const;
-
-   virtual void AssembleH(const DenseMatrix &Jpt, const DenseMatrix &DS,
-                          const double weight, DenseMatrix &A) const;
-};
-
-/// 2D non-barrier metric without a type.
+/// Metric without a type, 2D
 class TMOP_Metric_001 : public TMOP_QualityMetric
 {
 protected:
@@ -116,7 +87,7 @@ public:
                          const double weight, DenseMatrix &A) const;
 };

-/// 2D non-barrier Skew metric.
+/// Skew metric, 2D.
 class TMOP_Metric_skew2D : public TMOP_QualityMetric
 {
 public:
@@ -131,7 +102,7 @@ public:
   { MFEM_ABORT("Not implemented"); }
 };

-/// 3D non-barrier Skew metric.
+/// Skew metric, 3D.
 class TMOP_Metric_skew3D : public TMOP_QualityMetric
 {
 public:
@@ -146,7 +117,7 @@ public:
   { MFEM_ABORT("Not implemented"); }
 };

-/// 2D non-barrier Aspect ratio metric.
+/// Aspect ratio metric, 2D.
 class TMOP_Metric_aspratio2D : public TMOP_QualityMetric
 {
 public:
@@ -161,7 +132,7 @@ public:
   { MFEM_ABORT("Not implemented"); }
 };

-/// 3D non-barrier Aspect ratio metric.
+/// Aspect ratio metric, 3D.
 class TMOP_Metric_aspratio3D : public TMOP_QualityMetric
 {
 public:
@@ -176,7 +147,22 @@ public:
   { MFEM_ABORT("Not implemented"); }
 };

-/// 2D barrier shape (S) metric (polyconvex).
+/// Shape+Size+Orientation metric, 2D.
+class TMOP_Metric_SSA2D : public TMOP_QualityMetric
+{
+public:
+   // W = 0.5 (1 - cos(theta_Jpr - theta_Jtr)).
+   virtual double EvalW(const DenseMatrix &Jpt) const;
+
+   virtual void EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const
+   { MFEM_ABORT("Not implemented"); }
+
+   virtual void AssembleH(const DenseMatrix &Jpt, const DenseMatrix &DS,
+                          const double weight, DenseMatrix &A) const
+   { MFEM_ABORT("Not implemented"); }
+};
+
+/// Shape, ideal barrier metric, 2D
 class TMOP_Metric_002 : public TMOP_QualityMetric
 {
 protected:
@@ -192,7 +178,7 @@ public:
                          const double weight, DenseMatrix &A) const;
 };

-/// 2D barrier Shape+Size (VS) metric (not polyconvex).
+/// Shape & area, ideal barrier metric, 2D
 class TMOP_Metric_007 : public TMOP_QualityMetric
 {
 protected:
@@ -208,7 +194,7 @@ public:
                          const double weight, DenseMatrix &A) const;
 };

-/// 2D barrier Shape+Size (VS) metric (not polyconvex).
+/// Shape & area metric, 2D
 class TMOP_Metric_009 : public TMOP_QualityMetric
 {
 protected:
@@ -224,30 +210,15 @@ public:
                          const double weight, DenseMatrix &A) const;
 };

-/// 2D non-barrier Shape+Size+Orientation (VOS) metric (polyconvex).
-class TMOP_Metric_014 : public TMOP_QualityMetric
-{
-public:
-   // W = |T-I|^2.
-   virtual double EvalW(const DenseMatrix &Jpt) const;
-
-   virtual void EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const
-   { MFEM_ABORT("Not implemented"); }
-
-   virtual void AssembleH(const DenseMatrix &Jpt, const DenseMatrix &DS,
-                          const double weight, DenseMatrix &A) const
-   { MFEM_ABORT("Not implemented"); }
-};
-
-/// 2D Shifted barrier form of shape metric (mu_2).
+/// Shifted barrier form of metric 2 (shape, ideal barrier metric), 2D
 class TMOP_Metric_022 : public TMOP_QualityMetric
 {
 protected:
-   double &min_detT;
+   double &tau0;
   mutable InvariantsEvaluator2D<double> ie;

 public:
-   TMOP_Metric_022(double &t0): min_detT(t0) {}
+   TMOP_Metric_022(double &t0): tau0(t0) {}

   // W = 0.5(|J|^2 - 2det(J)) / (det(J) - tau0).
   virtual double EvalW(const DenseMatrix &Jpt) const;
@@ -258,7 +229,7 @@ public:
                          const double weight, DenseMatrix &A) const;
 };

-/// 2D barrier (not a shape) metric (polyconvex).
+/// Shape, ideal barrier metric, 2D
 class TMOP_Metric_050 : public TMOP_QualityMetric
 {
 protected:
@@ -274,7 +245,7 @@ public:
                          const double weight, DenseMatrix &A) const;
 };

-/// 2D non-barrier size (V) metric (not polyconvex).
+/// Area metric, 2D
 class TMOP_Metric_055 : public TMOP_QualityMetric
 {
 protected:
@@ -291,7 +262,7 @@ public:

 };

-/// 2D barrier size (V) metric (polyconvex).
+/// Area, ideal barrier metric, 2D
 class TMOP_Metric_056 : public TMOP_QualityMetric
 {
 protected:
@@ -310,7 +281,7 @@ public:

 };

-/// 2D barrier shape (S) metric (not polyconvex).
+/// Shape, ideal barrier metric, 2D
 class TMOP_Metric_058 : public TMOP_QualityMetric
 {
 protected:
@@ -328,7 +299,7 @@ public:

 };

-/// 2D barrier size (V) metric (polyconvex).
+/// Area, ideal barrier metric, 2D
 class TMOP_Metric_077 : public TMOP_QualityMetric
 {
 protected:
@@ -345,28 +316,7 @@ public:

 };

-/// 2D barrier Shape+Size (VS) metric (polyconvex).
-class TMOP_Metric_080 : public TMOP_Combo_QualityMetric
-{
-protected:
-   mutable InvariantsEvaluator2D<double> ie;
-   double gamma;
-   TMOP_QualityMetric *sh_metric, *sz_metric;
-
-public:
-   TMOP_Metric_080(double gamma_) : gamma(gamma_),
-      sh_metric(new TMOP_Metric_002),
-      sz_metric(new TMOP_Metric_077)
-   {
-      // (1-gamma) mu_2 + gamma mu_77
-      AddQualityMetric(sh_metric, 1.-gamma_);
-      AddQualityMetric(sz_metric, gamma_);
-   }
-
-   virtual ~TMOP_Metric_080() { delete sh_metric; delete sz_metric; }
-};
-
-/// 2D barrier Shape+Orientation (OS) metric (polyconvex).
+/// Shape & orientation metric, 2D.
 class TMOP_Metric_085 : public TMOP_QualityMetric
 {
 public:
@@ -381,22 +331,7 @@ public:
   { MFEM_ABORT("Not implemented"); }
 };

-/// 2D barrier Shape+Size+Orientation (VOS) metric (polyconvex).
-class TMOP_Metric_098 : public TMOP_QualityMetric
-{
-public:
-   // W = 1/tau |T-I|^2.
-   virtual double EvalW(const DenseMatrix &Jpt) const;
-
-   virtual void EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const
-   { MFEM_ABORT("Not implemented"); }
-
-   virtual void AssembleH(const DenseMatrix &Jpt, const DenseMatrix &DS,
-                          const double weight, DenseMatrix &A) const
-   { MFEM_ABORT("Not implemented"); }
-};
-
-/// 2D untangling metric.
+/// Untangling metric, 2D
 class TMOP_Metric_211 : public TMOP_QualityMetric
 {
 protected:
@@ -435,7 +370,7 @@ public:
                          const double weight, DenseMatrix &A) const;
 };

-/// 3D barrier Shape (S) metric.
+/// Shape, ideal barrier metric, 3D
 class TMOP_Metric_301 : public TMOP_QualityMetric
 {
 protected:
@@ -451,7 +386,7 @@ public:
                          const double weight, DenseMatrix &A) const;
 };

-/// 3D barrier Shape (S) metric.
+/// Shape, ideal barrier metric, 3D
 class TMOP_Metric_302 : public TMOP_QualityMetric
 {
 protected:
@@ -467,14 +402,14 @@ public:
                          const double weight, DenseMatrix &A) const;
 };

-/// 3D barrier Shape (S) metric.
+/// Shape, ideal barrier metric, 3D
 class TMOP_Metric_303 : public TMOP_QualityMetric
 {
 protected:
   mutable InvariantsEvaluator3D<double> ie;

 public:
-   // W = |J|^2 / 3 * det(J)^(-2/3) - 1.
+   // W = |J|^2 / 3 * det(J)^(2/3) - 1.
   virtual double EvalW(const DenseMatrix &Jpt) const;

   virtual void EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const;
@@ -483,45 +418,7 @@ public:
                          const double weight, DenseMatrix &A) const;
 };

-/// 3D Size (V) untangling metric.
-class TMOP_Metric_311 : public TMOP_QualityMetric
-{
-protected:
-   const double eps;
-   mutable InvariantsEvaluator3D<double> ie;
-
-public:
-   TMOP_Metric_311(double epsilon = 1e-4) : eps(epsilon) { }
-
-   // W = (det(J) - 1)^2 - det(J)  + (det(J)^2 + eps)^(1/2).
-   virtual double EvalW(const DenseMatrix &Jpt) const;
-
-   virtual void EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const;
-
-   virtual void AssembleH(const DenseMatrix &Jpt, const DenseMatrix &DS,
-                          const double weight, DenseMatrix &A) const;
-};
-
-/// 3D Shape (S) metric, untangling version of 303.
-class TMOP_Metric_313 : public TMOP_QualityMetric
-{
-protected:
-   double &min_detT;
-   mutable InvariantsEvaluator3D<double> ie;
-
-public:
-   TMOP_Metric_313(double &mindet) : min_detT(mindet) { }
-
-   // W = 1/3 |J|^2 / [det(J)-tau0]^(-2/3).
-   virtual double EvalW(const DenseMatrix &Jpt) const;
-
-   virtual void EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const;
-
-   virtual void AssembleH(const DenseMatrix &Jpt, const DenseMatrix &DS,
-                          const double weight, DenseMatrix &A) const;
-};
-
-/// 3D non-barrier Size (V) metric.
+/// Volume metric, 3D
 class TMOP_Metric_315 : public TMOP_QualityMetric
 {
 protected:
@@ -537,7 +434,7 @@ public:
                          const double weight, DenseMatrix &A) const;
 };

-/// 3D barrier Size (V) metric.
+/// Volume, ideal barrier metric, 3D
 class TMOP_Metric_316 : public TMOP_QualityMetric
 {
 protected:
@@ -555,7 +452,7 @@ public:
                          const double weight, DenseMatrix &A) const;
 };

-/// 3D barrier Shape+Size (VS) metric.
+/// Shape & volume, ideal barrier metric, 3D
 class TMOP_Metric_321 : public TMOP_QualityMetric
 {
 protected:
@@ -590,99 +487,6 @@ public:
                          const double weight, DenseMatrix &A) const;
 };

-/// A-metrics
-/// 2D barrier Shape (S) metric (polyconvex).
-class TMOP_AMetric_011 : public TMOP_QualityMetric
-{
-protected:
-   mutable InvariantsEvaluator3D<double> ie;
-
-public:
-   // (1/4 alpha) | A - (adj A)^t W^t W / omega |^2
-   virtual double EvalW(const DenseMatrix &Jpt) const;
-
-   virtual void EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const
-   { MFEM_ABORT("Not implemented"); }
-
-   virtual void AssembleH(const DenseMatrix &Jpt, const DenseMatrix &DS,
-                          const double weight, DenseMatrix &A) const
-   { MFEM_ABORT("Not implemented"); }
-};
-
-/// 2D barrier Size (V) metric (polyconvex).
-class TMOP_AMetric_014a : public TMOP_QualityMetric
-{
-protected:
-   mutable InvariantsEvaluator3D<double> ie;
-
-public:
-   // 0.5 * ( sqrt(alpha/omega) - sqrt(omega/alpha) )^2
-   virtual double EvalW(const DenseMatrix &Jpt) const;
-
-   virtual void EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const
-   { MFEM_ABORT("Not implemented"); }
-
-   virtual void AssembleH(const DenseMatrix &Jpt, const DenseMatrix &DS,
-                          const double weight, DenseMatrix &A) const
-   { MFEM_ABORT("Not implemented"); }
-};
-
-/// 2D barrier Shape+Size+Orientation (VOS) metric (polyconvex).
-class TMOP_AMetric_036 : public TMOP_QualityMetric
-{
-protected:
-   mutable InvariantsEvaluator3D<double> ie;
-
-public:
-   // (1/alpha) | A - W |^2
-   virtual double EvalW(const DenseMatrix &Jpt) const;
-
-   virtual void EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const
-   { MFEM_ABORT("Not implemented"); }
-
-   virtual void AssembleH(const DenseMatrix &Jpt, const DenseMatrix &DS,
-                          const double weight, DenseMatrix &A) const
-   { MFEM_ABORT("Not implemented"); }
-};
-
-/// 2D barrier Shape+Orientation (OS) metric (polyconvex).
-class TMOP_AMetric_107a : public TMOP_QualityMetric
-{
-protected:
-   mutable InvariantsEvaluator3D<double> ie;
-
-public:
-   // (1/2 alpha) | A - (|A|/|W|) W |^2
-   virtual double EvalW(const DenseMatrix &Jpt) const;
-
-   virtual void EvalP(const DenseMatrix &Jpt, DenseMatrix &P) const
-   { MFEM_ABORT("Not implemented"); }
-
-   virtual void AssembleH(const DenseMatrix &Jpt, const DenseMatrix &DS,
-                          const double weight, DenseMatrix &A) const
-   { MFEM_ABORT("Not implemented"); }
-};
-
-/// 2D barrier Shape+Size (VS) metric (polyconvex).
-class TMOP_AMetric_126 : public TMOP_Combo_QualityMetric
-{
-protected:
-   mutable InvariantsEvaluator2D<double> ie;
-   double gamma;
-   TMOP_QualityMetric *sh_metric, *sz_metric;
-
-public:
-   TMOP_AMetric_126(double gamma_) : gamma(gamma_),
-      sh_metric(new TMOP_AMetric_011),
-      sz_metric(new TMOP_AMetric_014a)
-   {
-      // (1-gamma) nu_11 + gamma nu_14
-      AddQualityMetric(sh_metric, 1.-gamma_);
-      AddQualityMetric(sz_metric, gamma_);
-   }
-
-   virtual ~TMOP_AMetric_126() { delete sh_metric; delete sz_metric; }
-};

 /// Base class for limiting functions to be used in class TMOP_Integrator.
 /** This class represents a scalar function f(x, x0, d), where x and x0 are
@@ -331,6 +331,10 @@ double TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
      energy_in = nlf->GetEnergy(x);
   }

+   const int NE = fes->GetMesh()->GetNE(), dim = fes->GetMesh()->Dimension();
+   Array<int> xdofs;
+   DenseMatrix Jpr(dim);
+
   // Get the local prolongation of the solution vector.
   Vector x_out_loc(fes->GetVSize());
   if (serial)
@@ -346,32 +350,51 @@ double TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
   }
 #endif

-   // Check if the starting mesh (given by x) is inverted. Note that x hasn't
-   // been modified by the Newton update yet.
-   const double min_detT_in = ComputeMinDet(x_out_loc, *fes);
-   const bool untangling = (min_detT_in <= 0.0) ? true : false;
-   const double untangle_factor = 1.5;
-   if (untangling)
+   // Check if the starting mesh (given by x) is inverted.
+   // Note that x hasn't been modified by the Newton update yet.
+   double min_detJ = infinity();
+   for (int i = 0; i < NE; i++)
   {
-      // Needed for the line search below. The untangling metrics see this
-      // reference to detect deteriorations.
-      *min_det_ptr = untangle_factor * min_detT_in;
+      const int dof = fes->GetFE(i)->GetDof();
+      DenseMatrix dshape(dof, dim), pos(dof, dim);
+      Vector posV(pos.Data(), dof * dim);
+
+      fes->GetElementVDofs(i, xdofs);
+      x_out_loc.GetSubVector(xdofs, posV);
+
+      const IntegrationRule &irule = GetIntegrationRule(*fes->GetFE(i));
+      const int nsp = irule.GetNPoints();
+      for (int j = 0; j < nsp; j++)
+      {
+         fes->GetFE(i)->CalcDShape(irule.IntPoint(j), dshape);
+         MultAtB(pos, dshape, Jpr);
+         min_detJ = std::min(min_detJ, Jpr.Det());
+      }
   }
+   double min_detJ_all = min_detJ;
+#ifdef MFEM_USE_MPI
+   if (parallel)
+   {
+      MPI_Allreduce(&min_detJ, &min_detJ_all, 1, MPI_DOUBLE, MPI_MIN,
+                    p_nlf->ParFESpace()->GetComm());
+   }
+#endif
+   const bool untangling = (min_detJ_all <= 0) ? true : false;

   const bool have_b = (b.Size() == Height());

   Vector x_out(x.Size());
   bool x_out_ok = false;
-   double scale = 1.0, energy_out = 0.0, min_detT_out;
-   const double norm_in = Norm(r);
+   double scale = 1.0, energy_out = 0.0;
+   const double norm0 = Norm(r);

   const double detJ_factor = (solver_type == 1) ? 0.25 : 0.5;

   // Perform the line search.
   for (int i = 0; i < 12; i++)
   {
-      // Update the mesh and get the L-vector in x_out_loc.
      add(x, -scale, c, x_out);
+
      if (serial)
      {
         const SparseMatrix *cP = fes->GetConformingProlongation();
@@ -385,29 +408,47 @@ double TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
      }
 #endif

-      // Check the changes in detJ.
-      min_detT_out = ComputeMinDet(x_out_loc, *fes);
-      if (untangling == false && min_detT_out < 0.0)
+      // Check det(Jpr) > 0.
+      if (!untangling)
      {
-         // No untangling, and detJ got negative -- no good.
-         if (print_level >= 0)
-         { mfem::out << "Scale = " << scale << " Neg det(J) found.\n"; }
-         scale *= detJ_factor; continue;
-      }
-      if (untangling == true && min_detT_out < *min_det_ptr)
-      {
-         // Untangling, and detJ got even more negative -- no good.
-         if (print_level >= 0)
-         { mfem::out << "Scale = " << scale << " Neg det(J) decreased.\n"; }
-         scale *= detJ_factor; continue;
-      }
+         int jac_ok = 1;
+         for (int i = 0; i < NE; i++)
+         {
+            const int dof = fes->GetFE(i)->GetDof();
+            DenseMatrix dshape(dof, dim), pos(dof, dim);
+            Vector posV(pos.Data(), dof * dim);

-      // Skip the energy and residual checks when we're untangling. The
-      // untangling metrics change their denominators, which can affect the
-      // energy and residual, so their increase/decrease is not relevant.
-      if (untangling) { x_out_ok = true; break; }
+            fes->GetElementVDofs(i, xdofs);
+            x_out_loc.GetSubVector(xdofs, posV);
+
+            const IntegrationRule &irule = GetIntegrationRule(*fes->GetFE(i));
+            const int nsp = irule.GetNPoints();
+            for (int j = 0; j < nsp; j++)
+            {
+               fes->GetFE(i)->CalcDShape(irule.IntPoint(j), dshape);
+               MultAtB(pos, dshape, Jpr);
+               if (Jpr.Det() <= 0.0) { jac_ok = 0; goto break2; }
+            }
+         }
+
+      break2:
+         int jac_ok_all = jac_ok;
+#ifdef MFEM_USE_MPI
+         if (parallel)
+         {
+            MPI_Allreduce(&jac_ok, &jac_ok_all, 1, MPI_INT, MPI_LAND,
+                          p_nlf->ParFESpace()->GetComm());
+         }
+#endif
+
+         if (jac_ok_all == 0)
+         {
+            if (print_level >= 0)
+            { mfem::out << "Scale = " << scale << " Neg det(J) found.\n"; }
+            scale *= detJ_factor; continue;
+         }
+      } // endif(!untangling)

-      // Check the changes in total energy.
      ProcessNewState(x_out);
      if (serial)
      {
@@ -419,55 +460,43 @@ double TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
         energy_out = p_nlf->GetParGridFunctionEnergy(x_out_loc);
      }
 #endif
-      if (energy_out > 1.2*energy_in || std::isnan(energy_out) != 0)
-      {
-         if (print_level >= 0)
-         {
-            mfem::out << "Scale = " << scale << " Increasing energy.\n";
-         }
-         scale *= 0.5; continue;
-      }

-      // Check the changes in the Newton residual.
-      oper->Mult(x_out, r);
-      if (have_b) { r -= b; }
-      double norm_out = Norm(r);
-
-      if (norm_out > 1.2*norm_in)
-      {
-         if (print_level >= 0)
-         { mfem::out << "Scale = " << scale << " Norm increased.\n"; }
-         scale *= 0.5; continue;
-      }
-      else { x_out_ok = true; break; }
-   } // end line search
-
-   if (untangling)
-   {
-      // Update the global min detJ. Untangling metrics see this min_det_ptr.
-      if (min_detT_out > 0.0)
-      {
-         *min_det_ptr = 0.0;
-         if (print_level >= 0)
-         { mfem::out << "The mesh has been untangled at the used points!\n"; }
-      }
-      else { *min_det_ptr = untangle_factor * min_detT_out; }
-   }
-
-   if (print_level >= 0)
-   {
      if (untangling)
      {
-         mfem::out << "Min det(T) change: "
-                   << min_detT_in << " -> " << min_detT_out
-                   << " with " << scale << " scaling.\n";
+         if (energy_out > energy_in || std::isnan(energy_out) != 0)
+         {
+            scale *= 0.5;
+         }
+         else { x_out_ok = true; break; }
      }
      else
      {
-         mfem::out << "Energy decrease: "
-                   << (energy_in - energy_out) / energy_in * 100.0
-                   << "% with " << scale << " scaling.\n";
-      }
+         if (energy_out > 1.2*energy_in || std::isnan(energy_out) != 0)
+         {
+            if (print_level >= 0)
+            { mfem::out << "Scale = " << scale << " Increasing energy.\n"; }
+            scale *= 0.5; continue;
+         }
+
+         oper->Mult(x_out, r);
+         if (have_b) { r -= b; }
+         double norm = Norm(r);
+
+         if (norm > 1.2*norm0)
+         {
+            if (print_level >= 0)
+            { mfem::out << "Scale = " << scale << " Norm increased.\n"; }
+            scale *= 0.5; continue;
+         }
+         else { x_out_ok = true; break; }
+      } // endif (untangling)
+   } // enddo (i)
+
+   if (print_level >= 0)
+   {
+      mfem::out << "Energy decrease: "
+                << (energy_in - energy_out) / energy_in * 100.0
+                << "% with " << scale << " scaling.\n";
   }

   if (x_out_ok == false) { scale = 0.0; }
@@ -479,8 +508,8 @@ void TMOPNewtonSolver::ProcessNewState(const Vector &x) const
   const NonlinearForm *nlf = dynamic_cast<const NonlinearForm *>(oper);
   const Array<NonlinearFormIntegrator*> &integs = *nlf->GetDNFI();

-   // Reset the update flags of all TargetConstructors. This is done to avoid
-   // repeated updates of shared TargetConstructors.
+   // Reset the update flags of all TargetConstructors.
+   // This is done to avoid repeated updates of shared TargetConstructors.
   TMOP_Integrator *ti  = NULL;
   TMOPComboIntegrator *co = NULL;
   DiscreteAdaptTC *dtc = NULL;
@@ -588,47 +617,6 @@ void TMOPNewtonSolver::UpdateDiscreteTC(const TMOP_Integrator &ti,
   }
 }

-double TMOPNewtonSolver::ComputeMinDet(const Vector &x_loc,
-                                       const FiniteElementSpace &fes) const
-{
-   double min_detJ = infinity();
-   const int NE = fes.GetNE(), dim = fes.GetMesh()->Dimension();
-   Array<int> xdofs;
-   DenseMatrix Jpr(dim);
-   for (int i = 0; i < NE; i++)
-   {
-      const int dof = fes.GetFE(i)->GetDof();
-      DenseMatrix dshape(dof, dim), pos(dof, dim);
-      Vector posV(pos.Data(), dof * dim);
-
-      fes.GetElementVDofs(i, xdofs);
-      x_loc.GetSubVector(xdofs, posV);
-
-      const IntegrationRule &irule = GetIntegrationRule(*fes.GetFE(i));
-      const int nsp = irule.GetNPoints();
-      for (int j = 0; j < nsp; j++)
-      {
-         fes.GetFE(i)->CalcDShape(irule.IntPoint(j), dshape);
-         MultAtB(pos, dshape, Jpr);
-         min_detJ = std::min(min_detJ, Jpr.Det());
-      }
-   }
-   double min_detT_all = min_detJ;
-#ifdef MFEM_USE_MPI
-   if (parallel)
-   {
-      auto p_nlf = dynamic_cast<const ParNonlinearForm *>(oper);
-      MPI_Allreduce(&min_detJ, &min_detT_all, 1, MPI_DOUBLE, MPI_MIN,
-                    p_nlf->ParFESpace()->GetComm());
-   }
-#endif
-   const DenseMatrix &Wideal =
-      Geometries.GetGeomToPerfGeomJac(fes.GetFE(0)->GetGeomType());
-   min_detT_all /= Wideal.Det();
-
-   return min_detT_all;
-}
-
 #ifdef MFEM_USE_MPI
 // Metric values are visualized by creating an L2 finite element functions and
 // computing the metric values at the nodes.
@@ -114,9 +114,6 @@ protected:
   int solver_type;
   bool parallel;

-   // Minimum determinant over the whole mesh. Used for mesh untangling.
-   double *min_det_ptr = nullptr;
-
   // Quadrature points that are checked for negative Jacobians etc.
   const IntegrationRule &ir;
   // These fields are relevant for mixed meshes.
@@ -134,9 +131,6 @@ protected:

   void UpdateDiscreteTC(const TMOP_Integrator &ti, const Vector &x_new) const;

-   double ComputeMinDet(const Vector &x_loc,
-                        const FiniteElementSpace &fes) const;
-
 public:
 #ifdef MFEM_USE_MPI
   TMOPNewtonSolver(MPI_Comm comm, const IntegrationRule &irule, int type = 0)
@@ -156,8 +150,6 @@ public:
      integ_order = order;
   }

-   void SetMinDetPtr(double *md_ptr) { min_det_ptr = md_ptr; }
-
   virtual double ComputeScalingFactor(const Vector &x, const Vector &b) const;

   virtual void ProcessNewState(const Vector &x) const;
@@ -43,7 +43,7 @@ CeedRestrMap ceed_restr_map;
 static const Backend::Id backend_list[Backend::NUM_BACKENDS] =
 {
   Backend::CEED_CUDA, Backend::OCCA_CUDA, Backend::RAJA_CUDA, Backend::CUDA,
-   Backend::CEED_HIP, Backend::RAJA_HIP, Backend::HIP, Backend::DEBUG_DEVICE,
+   Backend::CEED_HIP, Backend::HIP, Backend::DEBUG_DEVICE,
   Backend::OCCA_OMP, Backend::RAJA_OMP, Backend::OMP,
   Backend::CEED_CPU, Backend::OCCA_CPU, Backend::RAJA_CPU, Backend::CPU
 };
@@ -52,7 +52,7 @@ static const Backend::Id backend_list[Backend::NUM_BACKENDS] =
 static const char *backend_name[Backend::NUM_BACKENDS] =
 {
   "ceed-cuda", "occa-cuda", "raja-cuda", "cuda",
-   "ceed-hip", "raja-hip", "hip", "debug",
+   "ceed-hip", "hip", "debug",
   "occa-omp", "raja-omp", "omp",
   "ceed-cpu", "occa-cpu", "raja-cpu", "cpu"
 };
@@ -394,8 +394,6 @@ static void RajaDeviceSetup(const int dev, int &ngpu)
 {
 #ifdef MFEM_USE_CUDA
   if (ngpu <= 0) { DeviceSetup(dev, ngpu); }
-#elif defined(MFEM_USE_HIP)
-   HipDeviceSetup(dev, ngpu);
 #else
   MFEM_CONTRACT_VAR(dev);
   MFEM_CONTRACT_VAR(ngpu);
@@ -509,8 +507,7 @@ void Device::Setup(const int device)
 #endif
   if (Allows(Backend::CUDA)) { CudaDeviceSetup(dev, ngpu); }
   if (Allows(Backend::HIP)) { HipDeviceSetup(dev, ngpu); }
-   if (Allows(Backend::RAJA_CUDA) || Allows(Backend::RAJA_HIP))
-   { RajaDeviceSetup(dev, ngpu); }
+   if (Allows(Backend::RAJA_CUDA)) { RajaDeviceSetup(dev, ngpu); }
   // The check for MFEM_USE_OCCA is in the function OccaDeviceSetup().
   if (Allows(Backend::OCCA_MASK)) { OccaDeviceSetup(dev); }
   if (Allows(Backend::CEED_CPU))
@@ -46,33 +46,30 @@ struct Backend
      /** @brief [device] RAJA CUDA backend. Enabled when MFEM_USE_RAJA = YES
          and MFEM_USE_CUDA = YES. */
      RAJA_CUDA = 1 << 6,
-      /** @brief [device] RAJA HIP backend. Enabled when MFEM_USE_RAJA = YES
-          and MFEM_USE_HIP = YES. */
-      RAJA_HIP = 1 << 7,
      /** @brief [host] OCCA CPU backend: sequential execution on each MPI rank.
          Enabled when MFEM_USE_OCCA = YES. */
-      OCCA_CPU = 1 << 8,
+      OCCA_CPU = 1 << 7,
      /// [host] OCCA OpenMP backend. Enabled when MFEM_USE_OCCA = YES.
-      OCCA_OMP = 1 << 9,
+      OCCA_OMP = 1 << 8,
      /** @brief [device] OCCA CUDA backend. Enabled when MFEM_USE_OCCA = YES
          and MFEM_USE_CUDA = YES. */
-      OCCA_CUDA = 1 << 10,
+      OCCA_CUDA = 1 << 9,
      /** @brief [host] CEED CPU backend. GPU backends can still be used, but
          with expensive memory transfers. Enabled when MFEM_USE_CEED = YES. */
-      CEED_CPU  = 1 << 11,
+      CEED_CPU  = 1 << 10,
      /** @brief [device] CEED CUDA backend working together with the CUDA
          backend. Enabled when MFEM_USE_CEED = YES and MFEM_USE_CUDA = YES.
          NOTE: The current default libCEED CUDA backend is non-deterministic! */
-      CEED_CUDA = 1 << 12,
+      CEED_CUDA = 1 << 11,
      /** @brief [device] CEED HIP backend working together with the HIP
          backend. Enabled when MFEM_USE_CEED = YES and MFEM_USE_HIP = YES. */
-      CEED_HIP = 1 << 13,
+      CEED_HIP = 1 << 12,
      /** @brief [device] Debug backend: host memory is READ/WRITE protected
          while a device is in use. It allows to test the "device" code-path
          (using separate host/device memory pools and host <-> device
          transfers) without any GPU hardware. As 'DEBUG' is sometimes used
          as a macro, `_DEVICE` has been added to avoid conflicts. */
-      DEBUG_DEVICE = 1 << 14
+      DEBUG_DEVICE = 1 << 13
   };

   /** @brief Additional useful constants. For example, the *_MASK constants can
@@ -80,14 +77,14 @@ struct Backend
   enum
   {
      /// Number of backends: from (1 << 0) to (1 << (NUM_BACKENDS-1)).
-      NUM_BACKENDS = 15,
+      NUM_BACKENDS = 14,

      /// Biwise-OR of all CPU backends
      CPU_MASK = CPU | RAJA_CPU | OCCA_CPU | CEED_CPU,
      /// Biwise-OR of all CUDA backends
      CUDA_MASK = CUDA | RAJA_CUDA | OCCA_CUDA | CEED_CUDA,
      /// Biwise-OR of all HIP backends
-      HIP_MASK = HIP | RAJA_HIP | CEED_HIP,
+      HIP_MASK = HIP | CEED_HIP,
      /// Biwise-OR of all OpenMP backends
      OMP_MASK = OMP | RAJA_OMP | OCCA_OMP,
      /// Bitwise-OR of all CEED backends
@@ -96,7 +93,7 @@ struct Backend
      DEVICE_MASK = CUDA_MASK | HIP_MASK | DEBUG_DEVICE,

      /// Biwise-OR of all RAJA backends
-      RAJA_MASK = RAJA_CPU | RAJA_OMP | RAJA_CUDA | RAJA_HIP,
+      RAJA_MASK = RAJA_CPU | RAJA_OMP | RAJA_CUDA,
      /// Biwise-OR of all OCCA backends
      OCCA_MASK = OCCA_CPU | OCCA_OMP | OCCA_CUDA
   };
@@ -86,157 +86,77 @@ void OmpWrap(const int N, HBODY &&h_body)
 }


-/// RAJA Cuda and Hip backends
+/// RAJA Cuda backend
 #if defined(MFEM_USE_RAJA) && defined(RAJA_ENABLE_CUDA)
-using cuda_launch_policy =
-   RAJA::expt::LaunchPolicy<RAJA::expt::null_launch_t, RAJA::expt::cuda_launch_t<false>>;
-using cuda_teams_x =
-   RAJA::expt::LoopPolicy<RAJA::loop_exec,RAJA::cuda_block_x_direct>;
-using cuda_threads_z =
-   RAJA::expt::LoopPolicy<RAJA::loop_exec,RAJA::cuda_thread_z_direct>;
+
+#if RAJA_VERSION_MAJOR == 0 && RAJA_VERSION_MINOR < 12
+using RAJA::statement::Segs;
+#else
+using RAJA::Segs;
 #endif

-#if defined(MFEM_USE_RAJA) && defined(RAJA_ENABLE_HIP)
-using hip_launch_policy =
-   RAJA::expt::LaunchPolicy<RAJA::expt::null_launch_t, RAJA::expt::hip_launch_t<false>>;
-using hip_teams_x =
-   RAJA::expt::LoopPolicy<RAJA::loop_exec,RAJA::hip_block_x_direct>;
-using hip_threads_z =
-   RAJA::expt::LoopPolicy<RAJA::loop_exec,RAJA::hip_thread_z_direct>;
-#endif
-
-#if defined(MFEM_USE_RAJA) && defined(RAJA_ENABLE_CUDA)
 template <const int BLOCKS = MFEM_CUDA_BLOCKS, typename DBODY>
-void RajaCuWrap1D(const int N, DBODY &&d_body)
+void RajaCudaWrap1D(const int N, DBODY &&d_body)
 {
-   //true denotes asynchronous kernel
+   // true denotes asynchronous kernel
   RAJA::forall<RAJA::cuda_exec<BLOCKS,true>>(RAJA::RangeSegment(0,N),d_body);
 }

 template <typename DBODY>
-void RajaCuWrap2D(const int N, DBODY &&d_body,
-                  const int X, const int Y, const int BZ)
+void RajaCudaWrap2D(const int N, DBODY &&d_body,
+                    const int X, const int Y, const int BZ)
 {
   MFEM_VERIFY(N>0, "");
   MFEM_VERIFY(BZ>0, "");
   const int G = (N+BZ-1)/BZ;
-
-   using namespace RAJA::expt;
-   using RAJA::RangeSegment;
-
-   launch<cuda_launch_policy>
-   (DEVICE, Resources(Teams(G), Threads(X, Y, BZ)),
-    [=] RAJA_DEVICE (LaunchContext ctx)
+   RAJA::kernel<RAJA::KernelPolicy<
+   RAJA::statement::CudaKernelAsync<
+   RAJA::statement::For<0, RAJA::cuda_block_x_direct,
+        RAJA::statement::For<1, RAJA::cuda_thread_x_direct,
+        RAJA::statement::For<2, RAJA::cuda_thread_y_direct,
+        RAJA::statement::For<3, RAJA::cuda_thread_z_direct,
+        RAJA::statement::Lambda<0, Segs<0>>>>>>>>>
+        (RAJA::make_tuple(RAJA::RangeSegment(0,G), RAJA::RangeSegment(0,X),
+                          RAJA::RangeSegment(0,Y), RAJA::RangeSegment(0,BZ)),
+         [=] RAJA_DEVICE (const int n)
   {
-
-      loop<cuda_teams_x>(ctx, RangeSegment(0, G), [&] (const int n)
-      {
-
-         loop<cuda_threads_z>(ctx, RangeSegment(0, BZ), [&] (const int tz)
-         {
-
-            const int k = n*BZ + tz;
-            if (k >= N) { return; }
-            d_body(k);
-
-         });
-
-      });
-
+      const int k = n*BZ + threadIdx.z;
+      if (k >= N) { return; }
+      d_body(k);
   });
-
   MFEM_GPU_CHECK(cudaGetLastError());
 }

 template <typename DBODY>
-void RajaCuWrap3D(const int N, DBODY &&d_body,
-                  const int X, const int Y, const int Z)
+void RajaCudaWrap3D(const int N, DBODY &&d_body,
+                    const int X, const int Y, const int Z)
 {
   MFEM_VERIFY(N>0, "");
-   using namespace RAJA::expt;
-   using RAJA::RangeSegment;
-
-   launch<cuda_launch_policy>
-   (DEVICE, Resources(Teams(N), Threads(X, Y, Z)),
-    [=] RAJA_DEVICE (LaunchContext ctx)
-   {
-
-      loop<cuda_teams_x>(ctx, RangeSegment(0, N), d_body);
-
-   });
-
+   RAJA::kernel<RAJA::KernelPolicy<
+   RAJA::statement::CudaKernelAsync<
+   RAJA::statement::For<0, RAJA::cuda_block_x_direct,
+        RAJA::statement::For<1, RAJA::cuda_thread_x_direct,
+        RAJA::statement::For<2, RAJA::cuda_thread_y_direct,
+        RAJA::statement::For<3, RAJA::cuda_thread_z_direct,
+        RAJA::statement::Lambda<0, Segs<0>>>>>>>>>
+        (RAJA::make_tuple(RAJA::RangeSegment(0,N), RAJA::RangeSegment(0,X),
+                          RAJA::RangeSegment(0,Y), RAJA::RangeSegment(0,Z)),
+   [=] RAJA_DEVICE (const int k) { d_body(k); });
   MFEM_GPU_CHECK(cudaGetLastError());
 }

 #endif

-#if defined(MFEM_USE_RAJA) && defined(RAJA_ENABLE_HIP)
-template <const int BLOCKS = MFEM_HIP_BLOCKS, typename DBODY>
-void RajaHipWrap1D(const int N, DBODY &&d_body)
-{
-   //true denotes asynchronous kernel
-   RAJA::forall<RAJA::hip_exec<BLOCKS,true>>(RAJA::RangeSegment(0,N),d_body);
-}
-
-template <typename DBODY>
-void RajaHipWrap2D(const int N, DBODY &&d_body,
-                   const int X, const int Y, const int BZ)
-{
-   MFEM_VERIFY(N>0, "");
-   MFEM_VERIFY(BZ>0, "");
-   const int G = (N+BZ-1)/BZ;
-
-   using namespace RAJA::expt;
-   using RAJA::RangeSegment;
-
-   launch<hip_launch_policy>
-   (DEVICE, Resources(Teams(G), Threads(X, Y, BZ)),
-    [=] RAJA_DEVICE (LaunchContext ctx)
-   {
-
-      loop<hip_teams_x>(ctx, RangeSegment(0, G), [&] (const int n)
-      {
-
-         loop<hip_threads_z>(ctx, RangeSegment(0, BZ), [&] (const int tz)
-         {
-
-            const int k = n*BZ + tz;
-            if (k >= N) { return; }
-            d_body(k);
-
-         });
-
-      });
-
-   });
-
-   MFEM_GPU_CHECK(hipGetLastError());
-}
-
-template <typename DBODY>
-void RajaHipWrap3D(const int N, DBODY &&d_body,
-                   const int X, const int Y, const int Z)
-{
-   MFEM_VERIFY(N>0, "");
-   using namespace RAJA::expt;
-   using RAJA::RangeSegment;
-
-   launch<hip_launch_policy>
-   (DEVICE, Resources(Teams(N), Threads(X, Y, Z)),
-    [=] RAJA_DEVICE (LaunchContext ctx)
-   {
-
-      loop<hip_teams_x>(ctx, RangeSegment(0, N), d_body);
-
-   });
-
-   MFEM_GPU_CHECK(hipGetLastError());
-}
-#endif

 /// RAJA OpenMP backend
 #if defined(MFEM_USE_RAJA) && defined(RAJA_ENABLE_OPENMP)

+#if RAJA_VERSION_MAJOR == 0 && RAJA_VERSION_MINOR < 12
+using RAJA::statement::Segs;
+#else
+using RAJA::Segs;
+#endif
+
 template <typename HBODY>
 void RajaOmpWrap(const int N, HBODY &&h_body)
 {
@@ -399,19 +319,9 @@ inline void ForallWrap(const bool use_dev, const int N,
   // If Backend::RAJA_CUDA is allowed, use it
   if (Device::Allows(Backend::RAJA_CUDA))
   {
-      if (DIM == 1) { return RajaCuWrap1D(N, d_body); }
-      if (DIM == 2) { return RajaCuWrap2D(N, d_body, X, Y, Z); }
-      if (DIM == 3) { return RajaCuWrap3D(N, d_body, X, Y, Z); }
-   }
-#endif
-
-#if defined(MFEM_USE_RAJA) && defined(RAJA_ENABLE_HIP)
-   // If Backend::RAJA_HIP is allowed, use it
-   if (Device::Allows(Backend::RAJA_HIP))
-   {
-      if (DIM == 1) { return RajaHipWrap1D(N, d_body); }
-      if (DIM == 2) { return RajaHipWrap2D(N, d_body, X, Y, Z); }
-      if (DIM == 3) { return RajaHipWrap3D(N, d_body, X, Y, Z); }
+      if (DIM == 1) { return RajaCudaWrap1D(N, d_body); }
+      if (DIM == 2) { return RajaCudaWrap2D(N, d_body, X, Y, Z); }
+      if (DIM == 3) { return RajaCudaWrap3D(N, d_body, X, Y, Z); }
   }
 #endif

@@ -125,7 +125,7 @@ void* HipMemcpyDtoDAsync(void* dst, const void *src, size_t bytes)
 void* HipMemcpyDtoH(void *dst, const void *src, size_t bytes)
 {
 #ifdef MFEM_USE_HIP
-#ifdef MFEM_TRACK_HIP_MEM
+#ifdef MFEM_TRACK_HPI_MEM
   mfem::out << "HipMemcpyDtoH(): copying " << bytes << " bytes from "
             << src << " to " << dst << " ... " << std::flush;
 #endif
@@ -71,7 +71,7 @@ static std::string strerror()
   }
 #elif (_POSIX_C_SOURCE >= 200112L || _XOPEN_SOURCE >= 600) && ! _GNU_SOURCE || \
      defined(__APPLE__) || defined(__FreeBSD__) || defined(__OpenBSD__) || \
-      defined(__NetBSD__) || defined(__DragonFly__) || defined(__EMSCRIPTEN__)
+      defined(__NetBSD__) || defined(__DragonFly__)
   // XSI-compliant strerror_r()
   if (strerror_r(errno, &buff[0], buff.size()) != 0)
   {
@@ -10,13 +10,11 @@
 # CONTRIBUTING.md for details.

 list(APPEND SRCS
-  auxiliary.cpp
  blockmatrix.cpp
  blockoperator.cpp
  blockvector.cpp
  complex_operator.cpp
  densemat.cpp
-  symmat.cpp
  handle.cpp
  matrix.cpp
  ode.cpp
@@ -28,13 +26,11 @@ list(APPEND SRCS
  )

 list(APPEND HDRS
-  auxiliary.hpp
  blockmatrix.hpp
  blockoperator.hpp
  blockvector.hpp
  complex_operator.hpp
  densemat.hpp
-  symmat.hpp
  dtensor.hpp
  handle.hpp
  invariants.hpp
@@ -30,16 +30,10 @@ int AmgXSolver::count = 0;

 AMGX_resources_handle AmgXSolver::rsrc = nullptr;

-AmgXSolver::AmgXSolver()
-   : ConvergenceCheck(false) {};
-
 AmgXSolver::AmgXSolver(const AMGX_MODE amgxMode_, const bool verbose)
 {
   amgxMode = amgxMode_;

-   if (amgxMode == AmgXSolver::SOLVER) { ConvergenceCheck = true;}
-   else { ConvergenceCheck = false;}
-
   DefaultParameters(amgxMode, verbose);

   InitSerial();
@@ -53,9 +47,6 @@ AmgXSolver::AmgXSolver(const MPI_Comm &comm,
   std::string config;
   amgxMode = amgxMode_;

-   if (amgxMode == AmgXSolver::SOLVER) { ConvergenceCheck = true;}
-   else { ConvergenceCheck = false;}
-
   DefaultParameters(amgxMode, verbose);

   InitExclusiveGPU(comm);
@@ -67,9 +58,6 @@ AmgXSolver::AmgXSolver(const MPI_Comm &comm, const int nDevs,
   std::string config;
   amgxMode = amgxMode_;

-   if (amgxMode == AmgXSolver::SOLVER) { ConvergenceCheck = true;}
-   else { ConvergenceCheck = false;}
-
   DefaultParameters(amgxMode_, verbose);

   InitMPITeams(comm, nDevs);
@@ -190,11 +178,6 @@ void AmgXSolver::ReadParameters(const std::string config,
   configSrc = source;
 }

-void AmgXSolver::SetConvergenceCheck(bool setConvergenceCheck_)
-{
-   ConvergenceCheck = setConvergenceCheck_;
-}
-
 void AmgXSolver::DefaultParameters(const AMGX_MODE amgxMode_,
                                   const bool verbose)
 {
@@ -218,8 +201,8 @@ void AmgXSolver::DefaultParameters(const AMGX_MODE amgxMode_,
      {
         amgx_config = amgx_config + ",\n"
                       "   \"obtain_timings\": 1, \n"
-                       "   \"print_grid_stats\": 1, \n"
                       "   \"monitor_residual\": 1, \n"
+                       "   \"print_grid_stats\": 1, \n"
                       "   \"print_solve_stats\": 1 \n";
      }
      else
@@ -255,12 +238,12 @@ void AmgXSolver::DefaultParameters(const AMGX_MODE amgxMode_,
                    "  \"convergence\": \"RELATIVE_MAX\", \n"
                    "  \"scope\": \"main\", \n"
                    "  \"tolerance\": 1e-12, \n"
-                    "  \"monitor_residual\": 1, \n"
                    "  \"norm\": \"L2\" ";
      if (verbose)
      {
         amgx_config = amgx_config + ", \n"
                       "        \"obtain_timings\": 1, \n"
+                       "        \"monitor_residual\": 1, \n"
                       "        \"print_grid_stats\": 1, \n"
                       "        \"print_solve_stats\": 1 \n";
      }
@@ -901,7 +884,7 @@ void AmgXSolver::Mult(const Vector& B, Vector& X) const

      AMGX_SOLVE_STATUS   status;
      AMGX_solver_get_status(solver, &status);
-      if (status != AMGX_SOLVE_SUCCESS && ConvergenceCheck)
+      if (status != AMGX_SOLVE_SUCCESS && amgxMode == SOLVER)
      {
         if (status == AMGX_SOLVE_DIVERGED)
         {
@@ -73,16 +73,13 @@ public:
   /// Flags to configure AmgXSolver as a solver or preconditioner
   enum AMGX_MODE {SOLVER, PRECONDITIONER};

-   /// Flag to check for convergence
-   bool ConvergenceCheck;
-
   /**
      Flags to determine whether user solver settings are defined internally in
      the source code or will be read through an external JSON file.
   */
   enum CONFIG_SRC {INTERNAL, EXTERNAL, UNDEFINED};

-   AmgXSolver();
+   AmgXSolver() = default;

   /**
      Configures AmgX with a default configuration based on the AmgX mode, and
@@ -165,9 +162,6 @@ public:
   */
   void DefaultParameters(const AMGX_MODE amgxMode_, const bool verbose);

-   /// Add a check for convergence after applying Mult.
-   void SetConvergenceCheck(bool setConvergenceCheck_=true);
-
   ~AmgXSolver();

   void Finalize();
@@ -1,285 +0,0 @@
-// Copyright (c) 2010-2020, 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_AUXILIARY
-#define MFEM_AUXILIARY
-
-#include "../config/config.hpp"
-
-#ifdef MFEM_USE_MPI
-
-#include "../general/tic_toc.hpp"
-#include "solvers.hpp"
-
-namespace mfem
-{
-
-// forward declarations
-class Coefficient;
-class MatrixCoefficient;
-class ParMesh;
-class ParBilinearForm;
-class ParDiscreteLinearOperator;
-
-/** @brief Auxiliary space solvers for MatrixFreeAMS preconditioner
-
-    Given an operator A and a transfer G, this will create a solver
-    that approximates (G^T A G)^{-1}. Used for two different
-    auxiliary spaces in the AMS cycle.
-
-    The produced solver is based on a low-order refined discretization
-    for the high-order H1 problem. */
-class MatrixFreeAuxiliarySpace : public Solver
-{
-public:
-   /** @brief Pi space constructor
-
-       In the AMS framework this auxiliary space has two coefficients.
-
-       @param mesh_lor       Low-order refined auxiliary mesh
-       @param alpha_coeff    coefficient on curl-curl term (1 if null)
-       @param beta_coeff     coefficient on mass term (1 if null)
-       @param beta_mcoeff    matrix coefficient on mass term
-       @param ess_bdr        attributes for essential boundaries
-       @param curlcurl_oper  High-order operator for the system
-       @param pi             Intentity interpolation operator
-       @param useAmgX_       Use AmgX instead of hypre for auxiliary solves
-       @param cg_iterations  number of CG iterations used to invert
-                             auxiliary system, choosing 0 means to use a
-                             single V-cycle
-   */
-   MatrixFreeAuxiliarySpace(
-      ParMesh& mesh_lor, Coefficient* alpha_coeff, Coefficient* beta_coeff,
-      MatrixCoefficient* beta_mcoeff,
-      Array<int>& ess_bdr, Operator& curlcurl_oper, Operator& pi,
-#ifdef MFEM_USE_AMGX
-      bool useAmgX_,
-#endif
-      int cg_iterations = 0);
-
-   // Complex Pi space constructor
-   MatrixFreeAuxiliarySpace(
-      ParMesh& mesh_lor, Coefficient* alpha_coeff, Coefficient* beta_coeff,
-      Coefficient* beta_imag, Coefficient* abs_beta_imag,
-      MatrixCoefficient* beta_mcoeff,
-      Array<int>& ess_bdr, Operator& curlcurl_oper, Operator *oper_complex,
-      Operator& pi,
-#ifdef MFEM_USE_AMGX
-      bool useAmgX_,
-#endif
-      int cg_iterations = 0);
-
-   /** @brief G space constructor
-
-       This has one coefficient in the AMS framework.
-
-       @param mesh_lor       Low-order refined auxiliary mesh
-       @param beta_coeff     coefficient on mass term (1 if null)
-       @param beta_mcoeff    matrix coefficient on mass term
-       @param ess_bdr        attributes for essential boundaries
-       @param curlcurl_oper  High-order operator for the system
-       @param g              Gradient interpolation operator
-       @param useAmgX_       Use AmgX instead of hypre for auxiliary solves
-       @param cg_iterations  number of CG iterations used to invert
-                             auxiliary system, choosing 0 means to
-                             use a single V-cycle
-   */
-   MatrixFreeAuxiliarySpace(
-      ParMesh& mesh_lor, Coefficient* beta_coeff,
-      MatrixCoefficient* beta_mcoeff, Array<int>& ess_bdr,
-      Operator& curlcurl_oper, Operator& g,
-#ifdef MFEM_USE_AMGX
-      bool useAmgX_,
-#endif
-      int cg_iterations = 1);
-
-   // Complex G space constructor
-   MatrixFreeAuxiliarySpace(
-      ParMesh& mesh_lor, Coefficient* beta_coeff, Coefficient* beta_imag,
-      Coefficient* abs_beta_imag,
-      MatrixCoefficient* beta_mcoeff, Array<int>& ess_bdr,
-      Operator& curlcurl_oper, Operator *oper_complex, Operator& g,
-#ifdef MFEM_USE_AMGX
-      bool useAmgX_,
-#endif
-      int cg_iterations = 1);
-
-   ~MatrixFreeAuxiliarySpace();
-
-   void Mult(const Vector& x, Vector& y) const;
-
-   void SetOperator(const Operator& op) {}
-
-private:
-   /** @brief Helper routine for constructors.
-
-       @param system_dimension is passed to HypreBoomerAMG::SetSystemsOptions
-   */
-   void SetupAMG(int system_dimension);
-   void SetupVCycle();
-
-   /// inner_cg_iterations > 99 applies an exact solve here
-   void SetupCG(Operator& curlcurl_oper, Operator& conn,
-                int inner_cg_iterations);
-
-   void SetupGMRES(Operator& curlcurl_oper, Operator& conn);
-
-   void SetupPMHSS();
-
-   MPI_Comm comm;
-   Array<int> ess_tdof_list;
-   HypreParMatrix * aspacematrix;
-   HypreParMatrix * aspacematrix_complex;
-   HypreParMatrix * aspacematrix_imag;
-   Solver * aspacepc;
-   Operator* matfree;
-   CGSolver* cg;
-   GMRESSolver* gmres;
-   GMRESSolver* gmres_PMHSS;
-   Operator* aspacewrapper;
-#ifdef MFEM_USE_AMGX
-   const bool useAmgX;
-#endif
-   mutable int inner_aux_iterations;
-
-   const bool imagBdry;
-
-   Complex_PMHSS *PMHSS = NULL;
-
-   Array<int> offsets;
-   Array<int> offsets_nd;
-   BlockDiagonalPreconditioner *BlockDP;
-
-   BlockOperator *conn_block;
-};
-
-
-/** @brief Perform AMS cycle with generic Operator objects.
-
-    Most users should use MatrixFreeAMS, which wraps this. */
-class GeneralAMS : public Solver
-{
-public:
-   /** @brief Constructor.
-
-       Most of these arguments just need a Mult() operation,
-       but pi and g also require MultTranspose() */
-   GeneralAMS(const Operator& curlcurl_op_,
-              Operator *oper_complex,
-              const Operator& pi_,
-              const Operator& gradient_,
-              const Operator& pispacesolver_,
-              const Operator& gspacesolver_,
-              const Operator& smoother_,
-              const Array<int>& ess_tdof_list_);
-   virtual ~GeneralAMS();
-
-   /// in principle this should set A_ = op;
-   void SetOperator(const Operator &op) {}
-
-   virtual void Mult(const Vector& x, Vector& y) const;
-
-private:
-   const Operator& curlcurl_op;
-   Operator *oper_complex;
-   const Operator& pi;
-   const Operator& gradient;
-   const Operator& pispacesolver;
-   const Operator& gspacesolver;
-   const Operator& smoother;
-   const Array<int> ess_tdof_list;
-
-   void FormResidual(const Vector& rhs, const Vector& x,
-                     Vector& residual) const;
-};
-
-
-/** @brief An auxiliary Maxwell solver for a high-order curl-curl
-    system without high-order assembly.
-
-    The auxiliary space solves are done using a low-order refined approach,
-    but all the interpolation operators, residuals, etc. are done in a
-    matrix-free manner.
-
-    See Barker and Kolev, Matrix-free preconditioning for high-order H(curl)
-    discretizations (https://doi.org/10.1002/nla.2348) */
-class MatrixFreeAMS : public Solver
-{
-public:
-   /** @brief Construct matrix-free AMS preconditioner
-
-       @param aform        BilinearForm for curl-curl problem, generally will
-                           have a CurlCurlIntegrator and possibly a
-                           VectorFEMassIntegrator.
-       @param oper         Operator to precondition.
-       @param nd_fespace   Underlying Nedelec finite element space.
-       @param alpha_coeff  coefficient on curl-curl term in Maxwell problem
-                           (can be null, in which case constant 1 is assumed)
-       @param beta_coeff   (scalar) coefficient on mass term in Maxwell problem
-       @param beta_mcoeff  (matrix) coefficient on mass term
-       @param ess_bdr      boundary *attributes* that are marked essential. In
-                           contrast to other MFEM cases, these are *attributes*
-                           not dofs, because we need to apply these boundary
-                           conditions to different bilinear forms.
-       @param useAmgX      use AmgX (instead of hypre) for LOR problems
-       @param inner_pi_its number of CG iterations on auxiliary pi space,
-                           may need more for difficult coefficients
-       @param inner_g_its  number of CG iterations on auxiliary g space,
-                           may need more for difficult coefficients
-       @param nd_smoother  optional user-provided smoother for Nedelec space,
-                           this object takes ownership and will delete.
-   */
-   MatrixFreeAMS(ParBilinearForm& aform, Operator& oper, Operator *oper_complex,
-                 ParFiniteElementSpace& nd_fespace, Coefficient* alpha_coeff,
-                 Coefficient* beta_coeff, Coefficient* beta_imag,
-                 Coefficient* abs_beta_imag, MatrixCoefficient* beta_mcoeff,
-                 Array<int>& ess_bdr,
-#ifdef MFEM_USE_AMGX
-                 bool useAmgX = false,
-#endif
-                 int inner_pi_its = 0, int inner_g_its = 1,
-                 Solver* nd_smoother = NULL);
-
-   ~MatrixFreeAMS();
-
-   void SetOperator(const Operator &op) {}
-
-   void Mult(const Vector& x, Vector& y) const { general_ams->Mult(x, y); }
-
-private:
-   GeneralAMS * general_ams;
-
-   Solver * smoother;
-   ParDiscreteLinearOperator * pa_grad;
-   OperatorPtr Gradient;
-   ParDiscreteLinearOperator * pa_interp;
-   OperatorPtr Pi;
-
-   Solver * Gspacesolver;
-   Solver * Pispacesolver;
-
-   ParFiniteElementSpace * h1_fespace;
-   ParFiniteElementSpace * h1_fespace_d;
-
-   Array<int> offsets_nd;
-   Array<int> offsets_vector;
-   Array<int> offsets_scalar;
-
-   BlockOperator *Pi_block;
-   BlockOperator *Gradient_block;
-   BlockOperator *smoother_block;
-};
-
-} // namespace mfem
-
-#endif // MFEM_USE_MPI
-
-#endif
@@ -101,7 +101,6 @@ void BlockVector::Update(Vector & data, const Array<int> & bOffsets)
   {
      blocks[i].MakeRef(data, blockOffsets[i], BlockSize(i));
   }
-   MakeRef(data, 0, blockOffsets[numBlocks]);
 }

 void BlockVector::Update(const Array<int> &bOffsets)
@@ -287,7 +287,10 @@ void ComplexUMFPackSolver::Init()

 void ComplexUMFPackSolver::SetOperator(const Operator &op)
 {
+   int *Ap, *Ai;
   void *Symbolic;
+   double *Ax;
+   double *Az;

   if (Numeric)
   {
@@ -319,11 +322,10 @@ void ComplexUMFPackSolver::SetOperator(const Operator &op)
   width = mat->real().Width();
   MFEM_VERIFY(width == height, "not a square matrix");

-   const int * Ap =
-      mat->real().HostReadI(); // assuming real and imag have the same sparsity
-   const int * Ai = mat->real().HostReadJ();
-   const double * Ax = mat->real().HostReadData();
-   const double * Az = mat->imag().HostReadData();
+   Ap = mat->real().GetI(); // assuming real and imag have the same sparsity
+   Ai = mat->real().GetJ();
+   Ax = mat->real().GetData();
+   Az = mat->imag().GetData();

   if (!use_long_ints)
   {
@@ -393,10 +395,6 @@ void ComplexUMFPackSolver::Mult(const Vector &b, Vector &x) const
   if (mat == NULL)
      mfem_error("ComplexUMFPackSolver::Mult : matrix is not set!"
                 " Call SetOperator first!");
-
-   b.HostRead();
-   x.HostReadWrite();
-
   int n = b.Size()/2;
   double * datax = x.GetData();
   double * datab = b.GetData();
@@ -415,8 +413,8 @@ void ComplexUMFPackSolver::Mult(const Vector &b, Vector &x) const
   if (!use_long_ints)
   {
      int status =
-         umfpack_zi_solve(UMFPACK_Aat, mat->real().HostReadI(), mat->real().HostReadJ(),
-                          mat->real().HostReadData(), mat->imag().HostReadData(),
+         umfpack_zi_solve(UMFPACK_Aat, mat->real().GetI(), mat->real().GetJ(),
+                          mat->real().GetData(), mat->imag().GetData(),
                          datax, &datax[n], datab, &datab[n], Numeric, Control, Info);
      umfpack_zi_report_info(Control, Info);
      if (status < 0)
@@ -428,8 +426,8 @@ void ComplexUMFPackSolver::Mult(const Vector &b, Vector &x) const
   else
   {
      SuiteSparse_long status =
-         umfpack_zl_solve(UMFPACK_Aat,AI,AJ,mat->real().HostReadData(),
-                          mat->imag().HostReadData(),
+         umfpack_zl_solve(UMFPACK_Aat,AI,AJ,mat->real().GetData(),
+                          mat->imag().GetData(),
                          datax,&datax[n],datab,&datab[n],Numeric,Control,Info);

      umfpack_zl_report_info(Control, Info);
@@ -450,8 +448,6 @@ void ComplexUMFPackSolver::MultTranspose(const Vector &b, Vector &x) const
   if (mat == NULL)
      mfem_error("ComplexUMFPackSolver::Mult : matrix is not set!"
                 " Call SetOperator first!");
-   b.HostRead();
-   x.HostReadWrite();
   int n = b.Size()/2;
   double * datax = x.GetData();
   double * datab = b.GetData();
@@ -471,8 +467,8 @@ void ComplexUMFPackSolver::MultTranspose(const Vector &b, Vector &x) const
   if (!use_long_ints)
   {
      int status =
-         umfpack_zi_solve(UMFPACK_A, mat->real().HostReadI(), mat->real().HostReadJ(),
-                          mat->real().HostReadData(), mat->imag().HostReadData(),
+         umfpack_zi_solve(UMFPACK_A, mat->real().GetI(), mat->real().GetJ(),
+                          mat->real().GetData(), mat->imag().GetData(),
                          datax, &datax[n], datab, &datab[n], Numeric, Control, Info);
      umfpack_zi_report_info(Control, Info);
      if (status < 0)
@@ -484,8 +480,8 @@ void ComplexUMFPackSolver::MultTranspose(const Vector &b, Vector &x) const
   else
   {
      SuiteSparse_long status =
-         umfpack_zl_solve(UMFPACK_A,AI,AJ,mat->real().HostReadData(),
-                          mat->imag().HostReadData(),
+         umfpack_zl_solve(UMFPACK_A,AI,AJ,mat->real().GetData(),
+                          mat->imag().GetData(),
                          datax,&datax[n],datab,&datab[n],Numeric,Control,Info);

      umfpack_zl_report_info(Control, Info);
@@ -1279,7 +1279,6 @@ void HypreParMatrix::operator*=(double s)
 static void get_sorted_rows_cols(const Array<int> &rows_cols,
                                 Array<HYPRE_Int> &hypre_sorted)
 {
-   rows_cols.HostRead();
   hypre_sorted.SetSize(rows_cols.Size());
   bool sorted = true;
   for (int i = 0; i < rows_cols.Size(); i++)
@@ -1996,8 +1995,6 @@ void EliminateBC(HypreParMatrix &A, HypreParMatrix &Ae,
   double *data_offd = hypre_CSRMatrixData(A_offd);
 #endif

-   ess_dof_list.HostRead();
-
   for (int i = 0; i < ess_dof_list.Size(); i++)
   {
      int r = ess_dof_list[i];
@@ -2164,7 +2161,6 @@ HypreSmoother::HypreSmoother() : Solver()
   B = X = V = Z = NULL;
   X0 = X1 = NULL;
   fir_coeffs = NULL;
-   A_is_symmetric = false;
 }

 HypreSmoother::HypreSmoother(HypreParMatrix &_A, int _type,
@@ -2184,7 +2180,6 @@ HypreSmoother::HypreSmoother(HypreParMatrix &_A, int _type,
   B = X = V = Z = NULL;
   X0 = X1 = NULL;
   fir_coeffs = NULL;
-   A_is_symmetric = false;

   SetOperator(_A);
 }
@@ -2472,16 +2467,6 @@ void HypreSmoother::Mult(const Vector &b, Vector &x) const
   Mult(*B, *X);
 }

-void HypreSmoother::MultTranspose(const Vector &b, Vector &x) const
-{
-   if (A_is_symmetric || type == 0 || type == 1 || type == 5)
-   {
-      Mult(b, x);
-      return;
-   }
-   mfem_error("HypreSmoother::MultTranspose (...) : undefined!\n");
-}
-
 HypreSmoother::~HypreSmoother()
 {
   if (B) { delete B; }
@@ -2526,14 +2511,6 @@ void HypreSolver::Mult(const HypreParVector &b, HypreParVector &x) const
      mfem_error("HypreSolver::Mult (...) : HypreParMatrix A is missing");
      return;
   }
-
-   if (!iterative_mode)
-   {
-      x = 0.0;
-   }
-
-   b.HostRead();
-   x.HostReadWrite();
   if (!setup_called)
   {
      err = SetupFcn()(*this, *A, b, x);
@@ -2549,6 +2526,10 @@ void HypreSolver::Mult(const HypreParVector &b, HypreParVector &x) const
      setup_called = 1;
   }

+   if (!iterative_mode)
+   {
+      x = 0.0;
+   }
   err = SolveFcn()(*this, *A, b, x);
   if (error_mode == WARN_HYPRE_ERRORS)
   {
@@ -2569,7 +2550,7 @@ void HypreSolver::Mult(const Vector &b, Vector &x) const
      return;
   }
   auto b_data = b.HostRead();
-   auto x_data = iterative_mode ? x.HostReadWrite() : x.HostWrite();
+   auto x_data = x.HostWrite();
   if (B == NULL)
   {
      B = new HypreParVector(A->GetComm(),
@@ -2692,11 +2673,6 @@ void HyprePCG::Mult(const HypreParVector &b, HypreParVector &x) const

   HYPRE_ParCSRMatrixGetComm(*A, &comm);

-   if (!iterative_mode)
-   {
-      x = 0.0;
-   }
-
   if (!setup_called)
   {
      if (print_level > 0 && print_level < 3)
@@ -2705,8 +2681,6 @@ void HyprePCG::Mult(const HypreParVector &b, HypreParVector &x) const
         hypre_BeginTiming(time_index);
      }

-      b.HostRead();
-      x.HostReadWrite();
      HYPRE_ParCSRPCGSetup(pcg_solver, *A, b, x);
      setup_called = 1;

@@ -2725,8 +2699,14 @@ void HyprePCG::Mult(const HypreParVector &b, HypreParVector &x) const
      hypre_BeginTiming(time_index);
   }

+   if (!iterative_mode)
+   {
+      x = 0.0;
+   }
+
   b.HostRead();
   x.HostReadWrite();
+
   HYPRE_ParCSRPCGSolve(pcg_solver, *A, b, x);

   if (print_level > 0)
@@ -3723,9 +3703,6 @@ void HypreAMS::Init(ParFiniteElementSpace *edge_fespace)
      }
      x = x_coord.ParallelProject();
      y = y_coord.ParallelProject();
-
-      x->HostReadWrite();
-      y->HostReadWrite();
      if (sdim == 2)
      {
         z = NULL;
@@ -3734,7 +3711,6 @@ void HypreAMS::Init(ParFiniteElementSpace *edge_fespace)
      else
      {
         z = z_coord.ParallelProject();
-         z->HostReadWrite();
         HYPRE_AMSSetCoordinateVectors(ams, *x, *y, *z);
      }
   }
@@ -638,9 +638,6 @@ protected:
   /// Combined coefficients for windowing and Chebyshev polynomials.
   double* fir_coeffs;

-   /// A flag that indicates whether the linear system matrix A is symmetric
-   bool A_is_symmetric;
-
 public:
   /** Hypre smoother types:
       0    = Jacobi
@@ -687,12 +684,6 @@ public:
       entries in the associated matrix. */
   void SetPositiveDiagonal(bool pos = true) { pos_l1_norms = pos; }

-   /** Explicitly indicate whether the linear system matrix A is symmetric. If A
-       is symmetric, the smoother will also be symmetric. In this case, calling
-       MultTranspose will be redirected to Mult. (This is also done if the
-       smoother is diagonal.) By default, A is assumed to be nonsymmetric. */
-   void SetOperatorSymmetry(bool is_sym) { A_is_symmetric = is_sym; }
-
   /** Set/update the associated operator. Must be called after setting the
       HypreSmoother type and options. */
   virtual void SetOperator(const Operator &op);
@@ -701,9 +692,6 @@ public:
   virtual void Mult(const HypreParVector &b, HypreParVector &x) const;
   virtual void Mult(const Vector &b, Vector &x) const;

-   /// Apply transpose of the smoother to relax the linear system Ax=b
-   virtual void MultTranspose(const Vector &b, Vector &x) const;
-
   virtual ~HypreSmoother();
 };

@@ -62,6 +62,8 @@ protected:
   const scalar_t *D; // Always points to external data or is empty
   scalar_t *DaJ, *DJt, *DXt, *DYt;

+   scalar_t sign_detJ;
+
   enum EvalMasks
   {
      HAVE_I1   = 1,
@@ -94,7 +96,8 @@ protected:
   {
      eval_state |= HAVE_I2b;
      const scalar_t det = J[0]*J[3] - J[1]*J[2];
-      I2b = det;
+      sign_detJ = scalar_ops::sign(det);
+      I2b = sign_detJ*det;
   }
   void Eval_dI1()
   {
@@ -133,10 +136,10 @@ protected:
      // I2b = det(J)
      // dI2b = adj(J)^T
      Get_I2b();
-      dI2b[0] =  J[3];
-      dI2b[1] = -J[2];
-      dI2b[2] = -J[1];
-      dI2b[3] =  J[0];
+      dI2b[0] =  sign_detJ*J[3];
+      dI2b[1] = -sign_detJ*J[2];
+      dI2b[2] = -sign_detJ*J[1];
+      dI2b[3] =  sign_detJ*J[0];
   }
   void Eval_DaJ() // D adj(J) = D dI2b^t
   {
@@ -513,6 +516,8 @@ protected:
   const scalar_t *D; // Always points to external data or is empty
   scalar_t *DaJ, *DJt, *DdI2t, *DXt, *DYt;

+   scalar_t sign_detJ;
+
   enum EvalMasks
   {
      HAVE_I1     = 1,
@@ -580,6 +585,8 @@ protected:
      eval_state |= HAVE_I3b;
      I3b = J[0]*(J[4]*J[8] - J[7]*J[5]) - J[1]*(J[3]*J[8] - J[5]*J[6]) +
            J[2]*(J[3]*J[7] - J[4]*J[6]);
+      sign_detJ = scalar_ops::sign(I3b);
+      I3b = sign_detJ*I3b;
   }
   scalar_t Get_I3b_p()  // I3b^{-2/3}
   {
@@ -587,7 +594,7 @@ protected:
      {
         eval_state |= HAVE_I3b_p;
         const scalar_t i3b = Get_I3b();
-         I3b_p = scalar_ops::pow(i3b, -2, 3);
+         I3b_p = sign_detJ*scalar_ops::pow(i3b, -2, 3);
      }
      return I3b_p;
   }
@@ -673,15 +680,15 @@ protected:
      eval_state |= HAVE_dI3b;
      // I3b = det(J)
      // dI3b = adj(J)^T
-      dI3b[0] = J[4]*J[8] - J[5]*J[7];  // 0  3  6
-      dI3b[1] = J[5]*J[6] - J[3]*J[8];  // 1  4  7
-      dI3b[2] = J[3]*J[7] - J[4]*J[6];  // 2  5  8
-      dI3b[3] = J[2]*J[7] - J[1]*J[8];
-      dI3b[4] = J[0]*J[8] - J[2]*J[6];
-      dI3b[5] = J[1]*J[6] - J[0]*J[7];
-      dI3b[6] = J[1]*J[5] - J[2]*J[4];
-      dI3b[7] = J[2]*J[3] - J[0]*J[5];
-      dI3b[8] = J[0]*J[4] - J[1]*J[3];
+      dI3b[0] = sign_detJ*(J[4]*J[8] - J[5]*J[7]);  // 0  3  6
+      dI3b[1] = sign_detJ*(J[5]*J[6] - J[3]*J[8]);  // 1  4  7
+      dI3b[2] = sign_detJ*(J[3]*J[7] - J[4]*J[6]);  // 2  5  8
+      dI3b[3] = sign_detJ*(J[2]*J[7] - J[1]*J[8]);
+      dI3b[4] = sign_detJ*(J[0]*J[8] - J[2]*J[6]);
+      dI3b[5] = sign_detJ*(J[1]*J[6] - J[0]*J[7]);
+      dI3b[6] = sign_detJ*(J[1]*J[5] - J[2]*J[4]);
+      dI3b[7] = sign_detJ*(J[2]*J[3] - J[0]*J[5]);
+      dI3b[8] = sign_detJ*(J[0]*J[4] - J[1]*J[3]);
   }
   void Eval_DZt(const scalar_t *Z, scalar_t **DZt_ptr)
   {
@@ -24,12 +24,10 @@
 #include "blockoperator.hpp"
 #include "sparsesmoothers.hpp"
 #include "densemat.hpp"
-#include "symmat.hpp"
 #include "ode.hpp"
 #include "solvers.hpp"
 #include "handle.hpp"
 #include "invariants.hpp"
-#include "auxiliary.hpp"

 #ifdef MFEM_USE_AMGX
 #include "amgxsolver.hpp"
@@ -196,8 +196,6 @@ void MUMPSSolver::SetOperator(const Operator &op)

 void MUMPSSolver::Mult(const Vector &x, Vector &y) const
 {
-   x.HostRead();
-   y.HostReadWrite();
 #if MFEM_MUMPS_VERSION >= 530

   id->nloc_rhs = x.Size();
@@ -37,8 +37,7 @@ protected:
      const Array<int> &test_tdof_list,
      RectangularConstrainedOperator* &Aout);

-   /** @brief Returns RAP Operator of this, using input/output Prolongation matrices
-       @a Pi corresponds to "P", @a Po corresponds to "Rt" */
+   /// Returns RAP Operator of this, taking in input/output Prolongation matrices
   Operator *SetupRAP(const Operator *Pi, const Operator *Po);

 public:
@@ -113,11 +112,6 @@ public:
   {
      return GetProlongation(); // Assume square unless specialized
   }
-   /** @brief Transpose of GetOutputRestriction, directly available in this
-       form to facilitate matrix-free RAP-type operators.
-
-       `NULL` means identity. */
-   virtual const Operator *GetOutputRestrictionTranspose() const { return NULL; }
   /** @brief Restriction operator from output vectors for the operator to linear
       algebra (linear system) vectors. `NULL` means identity. */
   virtual const Operator *GetOutputRestriction() const
@@ -612,22 +606,23 @@ public:

   using TimeDependentOperator::ImplicitSolve;
   /** @brief Solve the equation:
-       @a k = f(@a x + @a fac0 @a k, @a dxdt + @a fac1 @a k, t), for the
+       @a k = f(@a x + 1/2 @a dt0^2 @a k, @a dxdt + @a dt1 @a k, t), for the
       unknown @a k at the current time t.

       For general F and G, the equation for @a k becomes:
-       F(@a x +  @a fac0 @a k, @a dxdt + @a fac1 @a k, t)
-                        = G(@a x +  @a fac0 @a k, @a dxdt + @a fac1 @a k, t).
+       F(@a x + 1/2 @a dt0^2 @a k, @a dxdt + @a dt1 @a k, t)
+                        = G(@a x + 1/2 @a dt0^2 @a k, @a dxdt + @a dt1 @a k, t).

-       The input vectors @a x and @a dxdt corresponds to time index (or cycle) n, while the
+       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.
+       index n+1. The time step @a dt corresponds to the time interval between
+       cycles n and n+1.

       This method allows for the abstract implementation of some time
       integration methods.

       If not re-implemented, this method simply generates an error. */
-   virtual void ImplicitSolve(const double fac0, const double fac1,
+   virtual void ImplicitSolve(const double dt0, const double dt1,
                              const Vector &x, const Vector &dxdt, Vector &k);


@@ -696,42 +691,6 @@ public:
   { A_.Mult(x, y); y *= a_; }
 };

-/// General sum operator: x -> A(x)+B(x)
-class SumOperator : public Operator
-{
-   const Operator *A, *B;
-   bool ownA, ownB;
-   mutable Vector z, w;
-   double cA, cB;
-
-public:
-   SumOperator(const Operator *A_, const Operator *B_,
-               bool ownA_, bool ownB_, double cA_, double cB_)
-      : Operator(A_->Height(), B_->Width()),
-        A(A_), B(B_), ownA(ownA_), ownB(ownB_), z(A_->Height()), w(A_->Width()),
-        cA(cA_), cB(cB_)
-   {
-      MFEM_VERIFY(A->Width() == B->Width() && A->Height() == B->Height(),
-                  "incompatible Operators: A->Width() = " << A->Width()
-                  << ", B->Height() = " << B->Height());
-
-      z.UseDevice(true);
-      w.UseDevice(true);
-   }
-
-   ~SumOperator()
-   {
-      if (ownA) { delete A; }
-      if (ownB) { delete B; }
-   }
-
-   virtual void Mult(const Vector &x, Vector &y) const
-   { B->Mult(x, z); A->Mult(x, y); y *= cA; z *= cB; y += z;}
-
-   virtual void MultTranspose(const Vector &x, Vector &y) const
-   { B->MultTranspose(x, w); A->MultTranspose(x, y); y *= cA; w *= cB; y += w;}
-
-};

 /** @brief The transpose of a given operator. Switches the roles of the methods
    Mult() and MultTranspose(). */
@@ -1556,7 +1556,6 @@ void NewtonSolver::SetOperator(const Operator &op)
   width = op.Width();
   MFEM_ASSERT(height == width, "square Operator is required.");

-   xcur.SetSize(width);
   r.SetSize(width);
   c.SetSize(width);
 }
@@ -1616,20 +1615,9 @@ void NewtonSolver::Mult(const Vector &b, Vector &x) const
         break;
      }

-      grad = &oper->GetGradient(x);
-      prec->SetOperator(*grad);
+      prec->SetOperator(oper->GetGradient(x));

-      if (lin_rtol_type)
-      {
-         AdaptiveLinRtolPreSolve(x, it, norm);
-      }
-
-      prec->Mult(r, c); // c = [DF(x_i)]^{-1} [F(x_i)-b]
-
-      if (lin_rtol_type)
-      {
-         AdaptiveLinRtolPostSolve(c, r, it, norm);
-      }
+      prec->Mult(r, c);  // c = [DF(x_i)]^{-1} [F(x_i)-b]

      const double c_scale = ComputeScalingFactor(x, b);
      if (c_scale == 0.0)
@@ -1653,86 +1641,6 @@ void NewtonSolver::Mult(const Vector &b, Vector &x) const
   final_norm = norm;
 }

-void NewtonSolver::SetAdaptiveLinRtol(const int type,
-                                      const double rtol0,
-                                      const double rtol_max,
-                                      const double alpha,
-                                      const double gamma)
-{
-   lin_rtol_type = type;
-   lin_rtol0 = rtol0;
-   lin_rtol_max = rtol_max;
-   this->alpha = alpha;
-   this->gamma = gamma;
-}
-
-void NewtonSolver::AdaptiveLinRtolPreSolve(const Vector &x,
-                                           const int it,
-                                           const double fnorm) const
-{
-   // Assume that when adaptive linear solver relative tolerance is activated,
-   // we are working with an iterative solver.
-   auto iterative_solver = static_cast<IterativeSolver *>(prec);
-   // Adaptive linear solver relative tolerance
-   double eta;
-   // Safeguard threshold
-   double sg_threshold = 0.1;
-
-   if (it == 0)
-   {
-      eta = lin_rtol0;
-   }
-   else
-   {
-      if (lin_rtol_type == 1)
-      {
-         // eta = gamma * abs(||F(x1)|| - ||F(x0) + DF(x0) s0||) / ||F(x0)||
-         eta = gamma * abs(fnorm - lnorm_last) / fnorm_last;
-      }
-      else if (lin_rtol_type == 2)
-      {
-         // eta = gamma * (||F(x1)|| / ||F(x0)||)^alpha
-         eta = gamma * pow(fnorm / fnorm_last, alpha);
-      }
-      else
-      {
-         MFEM_ABORT("Unknown adaptive linear solver rtol version");
-      }
-
-      // Safeguard rtol from "oversolving" ?!
-      const double sg_eta = gamma * pow(eta_last, alpha);
-      if (sg_eta > sg_threshold) { eta = std::max(eta, sg_eta); }
-   }
-
-   eta = std::min(eta, lin_rtol_max);
-   iterative_solver->SetRelTol(eta);
-   eta_last = eta;
-   if (print_level >= 0)
-   {
-      mfem::out << "Eisenstat-Walker rtol = " << eta << "\n";
-   }
-}
-
-void NewtonSolver::AdaptiveLinRtolPostSolve(const Vector &x,
-                                            const Vector &b,
-                                            const int it,
-                                            const double fnorm) const
-{
-   fnorm_last = fnorm;
-
-   // If version 1 is chosen, the true linear residual norm has to be computed
-   // and in most cases we can only retrieve the preconditioned linear residual
-   // norm.
-   if (lin_rtol_type == 1)
-   {
-      // lnorm_last = ||F(x0) + DF(x0) s0||
-      Vector linres(x.Size());
-      grad->Mult(x, linres);
-      linres -= b;
-      lnorm_last = Norm(linres);
-   }
-}
-
 void LBFGSSolver::Mult(const Vector &b, Vector &x) const
 {
   MFEM_VERIFY(oper != NULL, "the Operator is not set (use SetOperator).");
@@ -2812,7 +2720,9 @@ void UMFPackSolver::Init()

 void UMFPackSolver::SetOperator(const Operator &op)
 {
+   int *Ap, *Ai;
   void *Symbolic;
+   double *Ax;

   if (Numeric)
   {
@@ -2838,9 +2748,9 @@ void UMFPackSolver::SetOperator(const Operator &op)
   width = mat->Width();
   MFEM_VERIFY(width == height, "not a square matrix");

-   const int * Ap = mat->HostReadI();
-   const int * Ai = mat->HostReadJ();
-   const double * Ax = mat->HostReadData();
+   Ap = mat->GetI();
+   Ai = mat->GetJ();
+   Ax = mat->GetData();

   if (!use_long_ints)
   {
@@ -2910,13 +2820,12 @@ void UMFPackSolver::Mult(const Vector &b, Vector &x) const
   if (mat == NULL)
      mfem_error("UMFPackSolver::Mult : matrix is not set!"
                 " Call SetOperator first!");
-   b.HostRead();
-   x.HostReadWrite();
+
   if (!use_long_ints)
   {
      int status =
-         umfpack_di_solve(UMFPACK_At, mat->HostReadI(), mat->HostReadJ(),
-                          mat->HostReadData(), x, b, Numeric, Control, Info);
+         umfpack_di_solve(UMFPACK_At, mat->GetI(), mat->GetJ(),
+                          mat->GetData(), x, b, Numeric, Control, Info);
      umfpack_di_report_info(Control, Info);
      if (status < 0)
      {
@@ -2927,7 +2836,7 @@ void UMFPackSolver::Mult(const Vector &b, Vector &x) const
   else
   {
      SuiteSparse_long status =
-         umfpack_dl_solve(UMFPACK_At, AI, AJ, mat->HostReadData(), x, b,
+         umfpack_dl_solve(UMFPACK_At, AI, AJ, mat->GetData(), x, b,
                          Numeric, Control, Info);
      umfpack_dl_report_info(Control, Info);
      if (status < 0)
@@ -2943,13 +2852,12 @@ void UMFPackSolver::MultTranspose(const Vector &b, Vector &x) const
   if (mat == NULL)
      mfem_error("UMFPackSolver::MultTranspose : matrix is not set!"
                 " Call SetOperator first!");
-   b.HostRead();
-   x.HostReadWrite();
+
   if (!use_long_ints)
   {
      int status =
-         umfpack_di_solve(UMFPACK_A, mat->HostReadI(), mat->HostReadJ(),
-                          mat->HostReadData(), x, b, Numeric, Control, Info);
+         umfpack_di_solve(UMFPACK_A, mat->GetI(), mat->GetJ(),
+                          mat->GetData(), x, b, Numeric, Control, Info);
      umfpack_di_report_info(Control, Info);
      if (status < 0)
      {
@@ -2961,7 +2869,7 @@ void UMFPackSolver::MultTranspose(const Vector &b, Vector &x) const
   else
   {
      SuiteSparse_long status =
-         umfpack_dl_solve(UMFPACK_A, AI, AJ, mat->HostReadData(), x, b,
+         umfpack_dl_solve(UMFPACK_A, AI, AJ, mat->GetData(), x, b,
                          Numeric, Control, Info);
      umfpack_dl_report_info(Control, Info);
      if (status < 0)
@@ -406,40 +406,7 @@ void MINRES(const Operator &A, Solver &B, const Vector &b, Vector &x,
 class NewtonSolver : public IterativeSolver
 {
 protected:
-   mutable Vector xcur, r, c;
-   mutable Operator *grad;
-
-   // Adaptive linear solver rtol variables
-
-   // Method to determine rtol, 0 means the adaptive algorithm is deactivated.
-   int lin_rtol_type = 0;
-   // rtol to use in first iteration
-   double lin_rtol0;
-   // Maximum rtol
-   double lin_rtol_max;
-   // Function norm ||F(x)|| of the previous iterate
-   mutable double fnorm_last = 0.0;
-   // Linear residual norm of the previous iterate
-   mutable double lnorm_last = 0.0;
-   // Forcing term (linear residual rtol) from the previous iterate
-   mutable double eta_last = 0.0;
-   // Eisenstat-Walker factor gamma
-   double gamma;
-   // Eisenstat-Walker factor alpha
-   double alpha;
-
-   /** @brief Method for the adaptive linear solver rtol invoked before the
-       linear solve. */
-   void AdaptiveLinRtolPreSolve(const Vector &x,
-                                const int it,
-                                const double fnorm) const;
-
-   /** @brief Method for the adaptive linear solver rtol invoked after the
-       linear solve. */
-   void AdaptiveLinRtolPostSolve(const Vector &x,
-                                 const Vector &b,
-                                 const int it,
-                                 const double fnorm) const;
+   mutable Vector r, c;

 public:
   NewtonSolver() { }
@@ -467,26 +434,6 @@ public:
   /** @brief This method can be overloaded in derived classes to perform
       computations that need knowledge of the newest Newton state. */
   virtual void ProcessNewState(const Vector &x) const { }
-
-   const Vector &GetCurrentResidual() const { return r; }
-   const Vector &GetCurrentIterate() const { return xcur; }
-
-   /// Enable adaptive linear solver relative tolerance algorithm.
-   /** Compute a relative tolerance for the Krylov method after each nonlinear
-    iteration, based on the algorithm presented in [1].
-
-    The maximum linear solver relative tolerance @a rtol_max should be < 1. For
-    @a type 1 the parameters @a alpha and @a gamma are ignored. For @a type 2
-    @a alpha has to be between 0 and 1 and @a gamma between 1 and 2.
-
-    [1] Eisenstat, Stanley C., and Homer F. Walker. "Choosing the forcing terms
-    in an inexact Newton method."
-    */
-   void SetAdaptiveLinRtol(const int type = 2,
-                           const double rtol0 = 0.5,
-                           const double rtol_max = 0.9,
-                           const double alpha = 0.5 * (1.0 + sqrt(5.0)),
-                           const double gamma = 1.0);
 };

 /** L-BFGS method for solving F(x)=b for a given operator F, by minimizing
@@ -852,169 +799,6 @@ public:

 #endif // MFEM_USE_SUITESPARSE

-class Complex_PMHSS : public Solver
-{
-public:
-   Complex_PMHSS(Operator *Re, Operator *Im, Solver *prec_Re, Solver *prec_Im,
-                 double a_)
-      : Solver(2*Re->Height()), a(a_), A(Re, Im, false, false),
-        A_Re(Re, NULL, false, false),
-        A_Im(Im, NULL, false, false), u(2*Re->Height()), rhs(2*Re->Height()),
-        n(Re->Height())
-   {
-      MFEM_VERIFY(Re->Height() == Im->Height() && Re->Height() == Re->Width() &&
-                  Im->Height() == Im->Width(), "");
-      MFEM_VERIFY(this->Height() == A.Height(), "");
-
-      // Create CG solver for real operator aV + A_Re in complex space.
-
-      V = useIdentityV ? (Operator*) new IdentityOperator(this->Height()) :
-          (Operator*) &A_Re;
-
-      // In the case V = A_Re, it is faster to use a scaled operator than a SumOperator
-      Operator *sumOpRe = useIdentityV ? (Operator*) new SumOperator(V, &A_Re, false,
-                                                                     false, a, 1.0)
-                          : (Operator*) new ScaledOperator(&A_Re, a + 1.0);
-
-      SumOperator *sumOpIm = new SumOperator(V, &A_Im, false, false, a, 1.0);
-
-      CGSolver *cg = new CGSolver(MPI_COMM_WORLD);
-      cg->SetRelTol(1e-6);
-      cg->SetMaxIter(1000);
-      cg->SetPrintLevel(0);
-      cg->SetOperator(*sumOpRe);
-      cg->SetPreconditioner(*prec_Re);
-      cg->iterative_mode = false;
-
-      SRe = cg;
-
-      CGSolver *cgi = new CGSolver(MPI_COMM_WORLD);
-      cgi->SetRelTol(1e-6);
-      cgi->SetMaxIter(1000);
-      cgi->SetPrintLevel(0);
-      cgi->SetOperator(*sumOpIm);
-      if (prec_Im && useIdentityV) { cgi->SetPreconditioner(*prec_Im); }
-      if (!useIdentityV) { cgi->SetPreconditioner(*prec_Re); }
-      cgi->iterative_mode = false;
-
-      /*
-      // For negative definite imaginary part, but then PMHSS does not work?
-      MINRESSolver *cgi = new MINRESSolver(MPI_COMM_WORLD);
-      cgi->SetRelTol(1e-12);
-      cgi->SetMaxIter(1000);
-      cgi->SetPrintLevel(0);
-      cgi->SetOperator(*sumOpIm);
-      if (prec_Im) cgi->SetPreconditioner(*prec_Im);
-      */
-
-      SIm = cgi;
-   }
-
-   void SetOperator(const Operator &op)
-   {
-      MFEM_VERIFY(false, "Don't call SetOperator");
-   }
-
-   void ComputeResidual(const Vector &b, const Vector &sol, Vector &res) const
-   {
-      A.Mult(sol, res);
-      res -= b;
-   }
-
-   void Mult(const Vector &x, Vector &y) const
-   {
-      if (!(x.Size() == Height() && y.Size() == Height()))
-      {
-         std::cout << "bug";
-      }
-
-      MFEM_VERIFY(x.Size() == Height() && y.Size() == Height(), "");
-
-      const double initNorm = x.Norml2();
-      mfem::out << "MHSS RHS norm " << initNorm << '\n';
-
-      // With V = I, use modified HSS (MHSS) from Bai, Benzi, Chen 2010.
-      y = 0.0;
-
-      for (int it=0; it<maxiter; ++it)
-      {
-         // Solve (aI + Re) u = (aI - i Im) y + x
-
-         if (it == 0)
-         {
-            // Optimize the first iteration, when the initial guess is y=0.
-            SRe->Mult(x, u);
-         }
-         else
-         {
-            A_Im.Mult(y, u);  // u = Im y
-            // Set rhs = -i Im y = -i u
-            for (int j=0; j<n; ++j)
-            {
-               rhs[j] = u[n+j];
-               rhs[n+j] = -u[j];
-            }
-
-            rhs += x;
-
-            V->Mult(y, u);
-            rhs.Add(a, u);
-
-            SRe->Mult(rhs, u);
-         }
-
-         // Solve (aI + Im) y = (aI + i Re) u - i x
-
-         A_Re.Mult(u, y);  // y = Re u
-         // Set rhs = i (Re u - x) = i (y - x)
-         for (int j=0; j<n; ++j)
-         {
-            rhs[j] = -(y[n+j] - x[n+j]);
-            rhs[n+j] = y[j] - x[j];
-         }
-
-         if (useIdentityV)
-         {
-            //V->Mult(u, y);
-            //rhs.Add(a, y);
-            rhs.Add(a, u);
-         }
-         else
-         {
-            // Using V = A_Re
-            rhs.Add(a, y);
-         }
-
-         SIm->Mult(rhs, y);
-
-         ComputeResidual(x, y, rhs);
-         const double resNorm = rhs.Norml2();
-         mfem::out << "MHSS iter " << it << " residual norm " << resNorm << '\n';
-
-         if (resNorm / initNorm < tol)
-         {
-            mfem::out << "MHSS converged\n";
-            break;
-         }
-      }
-   }
-
-private:
-   const double a;
-   const int maxiter = 1;
-   ComplexOperator A, A_Re, A_Im;
-   mutable Vector u, rhs;
-   const int n;
-
-   const double tol = 1.0e-8;
-
-   const bool useIdentityV = false;
-   Operator *V = NULL;
-
-   Solver *SRe = NULL;
-   Solver *SIm = NULL;
-};
-
 }

 #endif // MFEM_SOLVERS
@@ -415,14 +415,10 @@ void SparseMatrix::SortColumnIndices()
      return;
   }

-   const int * Ip=HostReadI();
-   HostReadWriteJ();
-   HostReadWriteData();
-
   Array<Pair<int,double> > row;
   for (int j = 0, i = 0; i < height; i++)
   {
-      int end = Ip[i+1];
+      int end = I[i+1];
      row.SetSize(end - j);
      for (int k = 0; k < row.Size(); k++)
      {
@@ -3362,12 +3358,12 @@ SparseMatrix *Mult (const SparseMatrix &A, const SparseMatrix &B,
               "number of columns of A (" << ncolsA
               << ") must equal number of rows of B (" << nrowsB << ")");

-   A_i    = A.HostReadI();
-   A_j    = A.HostReadJ();
-   A_data = A.HostReadData();
-   B_i    = B.HostReadI();
-   B_j    = B.HostReadJ();
-   B_data = B.HostReadData();
+   A_i    = A.GetI();
+   A_j    = A.GetJ();
+   A_data = A.GetData();
+   B_i    = B.GetI();
+   B_j    = B.GetJ();
+   B_data = B.GetData();

   B_marker = new int[ncolsB];

@@ -3413,16 +3409,16 @@ SparseMatrix *Mult (const SparseMatrix &A, const SparseMatrix &B,
   {
      C = OAB;

-      MFEM_VERIFY(nrowsA == C->Height() && ncolsB == C->Width(),
+      MFEM_VERIFY(nrowsA == C -> Height() && ncolsB == C -> Width(),
                  "Input matrix sizes do not match output sizes"
                  << " nrowsA = " << nrowsA
                  << ", C->Height() = " << C->Height()
                  << " ncolsB = " << ncolsB
                  << ", C->Width() = " << C->Width());

-      // C_i    = C->HostReadI(); // not used
-      C_j    = C->HostWriteJ();
-      C_data = C->HostWriteData();
+      // C_i    = C -> GetI(); // not used
+      C_j    = C -> GetJ();
+      C_data = C -> GetData();
   }

   counter = 0;
@@ -523,10 +523,9 @@ void SuperLUSolver::Mult( const Vector & x, Vector & y ) const
   // SuperLU overwrites x with y, so copy x to y and pass that to the solve
   // routine.

-   const double *xPtr = x.HostRead();
-   y = xPtr;
-   double * yPtr = y.HostReadWrite();
+   y = x;

+   double*  yPtr = (double*)y;
   int      info = -1, locSize = y.Size();

   // Solve the system
@@ -1,109 +0,0 @@
-// Copyright (c) 2010-2020, 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.
-
-
-// Implementation of data type DenseSymmetricMatrix
-
-#include "symmat.hpp"
-
-namespace mfem
-{
-
-DenseSymmetricMatrix::DenseSymmetricMatrix() : Matrix(0)
-{
-   data.Reset();
-}
-
-DenseSymmetricMatrix::DenseSymmetricMatrix(int s) : Matrix(s)
-{
-   MFEM_ASSERT(s >= 0, "invalid DenseSymmetricMatrix size: " << s);
-   if (s > 0)
-   {
-      data.New((s*(s+1))/2);
-      *this = 0.0; // init with zeroes
-   }
-   else
-   {
-      data.Reset();
-   }
-}
-
-void DenseSymmetricMatrix::SetSize(int s)
-{
-   MFEM_ASSERT(s >= 0,
-               "invalid DenseSymmetricMatrix size: " << s);
-   if (Height() == s)
-   {
-      return;
-   }
-   height = s;
-   width = s;
-   const int s2 = (s*(s+1))/2;
-   if (s2 > data.Capacity())
-   {
-      data.Delete();
-      data.New(s2);
-      *this = 0.0; // init with zeroes
-   }
-}
-
-DenseSymmetricMatrix &DenseSymmetricMatrix::operator=(double c)
-{
-   const int s = (Height()*(Height()+1))/2;
-   for (int i = 0; i < s; i++)
-   {
-      data[i] = c;
-   }
-   return *this;
-}
-
-double &DenseSymmetricMatrix::Elem(int i, int j)
-{
-   return (*this)(i,j);
-}
-
-const double &DenseSymmetricMatrix::Elem(int i, int j) const
-{
-   return (*this)(i,j);
-}
-
-DenseSymmetricMatrix &DenseSymmetricMatrix::operator*=(double c)
-{
-   int s = Height()*(Height()+1)/2;
-   for (int i = 0; i < s; i++)
-   {
-      data[i] *= c;
-   }
-   return *this;
-}
-
-void DenseSymmetricMatrix::Mult(const Vector &x, Vector &y) const
-{
-   mfem_error("DenseSymmetricMatrix::Mult() not implemented!");
-}
-
-MatrixInverse *DenseSymmetricMatrix::Inverse() const
-{
-   mfem_error("DenseSymmetricMatrix::Inverse() not implemented!");
-   return nullptr;
-}
-
-void DenseSymmetricMatrix::Print (std::ostream & out, int width_) const
-{
-   mfem_error("DenseSymmetricMatrix::Print() not implemented!");
-}
-
-DenseSymmetricMatrix::~DenseSymmetricMatrix()
-{
-   data.Delete();
-}
-
-}
@@ -1,175 +0,0 @@
-// Copyright (c) 2010-2020, 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_SYMMETRICMAT
-#define MFEM_SYMMETRICMAT
-
-#include "../config/config.hpp"
-#include "../general/globals.hpp"
-#include "matrix.hpp"
-
-namespace mfem
-{
-
-/// Dense symmetric matrix storing the upper triangular part. This class so far
-/// has little functionality beyond storage.
-class DenseSymmetricMatrix : public Matrix
-{
-private:
-   Memory<double> data;
-
-public:
-
-   /** Default constructor for DenseSymmetricMatrix.
-       Sets data = NULL and height = width = 0. */
-   DenseSymmetricMatrix();
-
-   /// Creates square matrix of size s.
-   explicit DenseSymmetricMatrix(int s);
-
-   /// Construct a DenseSymmetricMatrix using an existing data array.
-   /** The DenseSymmetricMatrix does not assume ownership of the data array, i.e. it will
-       not delete the array. */
-   DenseSymmetricMatrix(double *d, int s)
-      : Matrix(s, s) { UseExternalData(d, s); }
-
-   /// Change the data array and the size of the DenseSymmetricMatrix.
-   /** The DenseSymmetricMatrix does not assume ownership of the data array, i.e. it will
-       not delete the data array @a d. This method should not be used with
-       DenseSymmetricMatrix that owns its current data array. */
-   void UseExternalData(double *d, int s)
-   {
-      data.Wrap(d, (s*(s+1))/2, false);
-      height = s; width = s;
-   }
-
-   /// Change the data array and the size of the DenseSymmetricMatrix.
-   /** The DenseSymmetricMatrix does not assume ownership of the data array, i.e. it will
-       not delete the new array @a d. This method will delete the current data
-       array, if owned. */
-   void Reset(double *d, int s)
-   { if (OwnsData()) { data.Delete(); } UseExternalData(d, s); }
-
-   /** Clear the data array and the dimensions of the DenseSymmetricMatrix. This method
-       should not be used with DenseSymmetricMatrix that owns its current data array. */
-   void ClearExternalData() { data.Reset(); height = width = 0; }
-
-   /// Delete the matrix data array (if owned) and reset the matrix state.
-   void Clear()
-   { if (OwnsData()) { data.Delete(); } ClearExternalData(); }
-
-   /// Change the size of the DenseSymmetricMatrix to s x s.
-   void SetSize(int s);
-
-   /// Returns the matrix data array.
-   inline double *Data() const
-   { return const_cast<double*>((const double*)data);}
-
-   /// Returns the matrix data array.
-   inline double *GetData() const { return Data(); }
-
-   Memory<double> &GetMemory() { return data; }
-   const Memory<double> &GetMemory() const { return data; }
-
-   /// Return the DenseSymmetricMatrix data (host pointer) ownership flag.
-   inline bool OwnsData() const { return data.OwnsHostPtr(); }
-
-   /// Returns reference to a_{ij}.
-   inline double &operator()(int i, int j);
-
-   /// Returns constant reference to a_{ij}.
-   inline const double &operator()(int i, int j) const;
-
-   /// Returns reference to a_{ij}.
-   virtual double &Elem(int i, int j);
-
-   /// Returns constant reference to a_{ij}.
-   virtual const double &Elem(int i, int j) const;
-
-   /// Sets the matrix elements equal to constant c
-   DenseSymmetricMatrix &operator=(double c);
-
-   DenseSymmetricMatrix &operator*=(double c);
-
-   long MemoryUsage() const { return data.Capacity() * sizeof(double); }
-
-   /// Shortcut for mfem::Read( GetMemory(), TotalSize(), on_dev).
-   const double *Read(bool on_dev = true) const
-   { return mfem::Read(data, Height()*Width(), on_dev); }
-
-   /// Shortcut for mfem::Read(GetMemory(), TotalSize(), false).
-   const double *HostRead() const
-   { return mfem::Read(data, Height()*Width(), false); }
-
-   /// Shortcut for mfem::Write(GetMemory(), TotalSize(), on_dev).
-   double *Write(bool on_dev = true)
-   { return mfem::Write(data, Height()*Width(), on_dev); }
-
-   /// Shortcut for mfem::Write(GetMemory(), TotalSize(), false).
-   double *HostWrite()
-   { return mfem::Write(data, Height()*Width(), false); }
-
-   /// Shortcut for mfem::ReadWrite(GetMemory(), TotalSize(), on_dev).
-   double *ReadWrite(bool on_dev = true)
-   { return mfem::ReadWrite(data, Height()*Width(), on_dev); }
-
-   /// Shortcut for mfem::ReadWrite(GetMemory(), TotalSize(), false).
-   double *HostReadWrite()
-   { return mfem::ReadWrite(data, Height()*Width(), false); }
-
-   /// Matrix vector multiplication.
-   virtual void Mult(const Vector &x, Vector &y) const;
-
-   /// 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();
-};
-
-// Inline methods
-
-// The number of entries stored in rows 1,...,k is
-// n + n-1 + n-2 + ... + n-k+1, where there are k terms. This equals
-// kn - sum_{i=1}^{k-1} i = kn - (k-1)k/2
-// This formula is used for the offset for each row.
-inline double &DenseSymmetricMatrix::operator()(int i, int j)
-{
-   MFEM_ASSERT(data && i >= 0 && i < height && j >= 0 && j < width, "");
-   if (i > j)  // reverse i and j
-   {
-      return data[(j*height) - (((j-1)*j)/2) + i - j];
-   }
-   else
-   {
-      return data[(i*height) - (((i-1)*i)/2) + j - i];
-   }
-}
-
-inline const double &DenseSymmetricMatrix::operator()(int i, int j) const
-{
-   MFEM_ASSERT(data && i >= 0 && i < height && j >= 0 && j < width, "");
-   if (i > j)  // reverse i and j
-   {
-      return data[(j*height) - (((j-1)*j)/2) + i - j];
-   }
-   else
-   {
-      return data[(i*height) - (((i-1)*i)/2) + j - i];
-   }
-}
-
-} // namespace mfem
-
-#endif
@@ -536,7 +536,7 @@ clean: $(addsuffix /clean,$(EM_DIRS) $(TEST_DIRS))
 distclean: clean config/clean doc/clean
 	rm -rf mfem/

-INSTALL_SHARED_LIB = $(MFEM_CXX) $(MFEM_LINK_FLAGS) $(INSTALL_SOFLAGS)\
+INSTALL_SHARED_LIB = $(MFEM_CXX) $(MFEM_BUILD_FLAGS) $(INSTALL_SOFLAGS)\
   $(OBJECT_FILES) $(EXT_LIBS) -o $(PREFIX_LIB)/libmfem.$(SO_VER) && \
   cd $(PREFIX_LIB) && ln -sf libmfem.$(SO_VER) libmfem.$(SO_EXT)

@@ -29,6 +29,7 @@
 #include <cstring>
 #include <ctime>
 #include <functional>
+#include <unordered_set>

 // Include the METIS header, if using version 5. If using METIS 4, the needed
 // declarations are inlined below, i.e. no header is needed.
@@ -1049,13 +1050,6 @@ void Mesh::GetFaceInfos(int Face, int *Inf1, int *Inf2) const
   *Inf2 = faces_info[Face].Elem2Inf;
 }

-void Mesh::GetFaceInfos(int Face, int *Inf1, int *Inf2, int *NCFace) const
-{
-   *Inf1   = faces_info[Face].Elem1Inf;
-   *Inf2   = faces_info[Face].Elem2Inf;
-   *NCFace = faces_info[Face].NCFace;
-}
-
 Geometry::Type Mesh::GetFaceGeometryType(int Face) const
 {
   switch (Dim)
@@ -5990,7 +5984,6 @@ int *Mesh::GeneratePartitioning(int nparts, int part_method)
   el_to_el = NULL;

   // Check for empty partitionings (a "feature" in METIS)
-   if (nparts > 1 && NumOfElements > nparts)
   {
      Array< Pair<int,int> > psize(nparts);
      int empty_parts;
@@ -8937,11 +8930,9 @@ void Mesh::PrintVTK(std::ostream &out)
         const int *v = elements[i]->GetVertices();
         const int nv = elements[i]->GetNVertices();
         out << nv;
-         Geometry::Type geom = elements[i]->GetGeometryType();
-         const int *perm = (geom == Geometry::PRISM) ? vtk_prism_perm : NULL;
         for (int j = 0; j < nv; j++)
         {
-            out << ' ' << v[perm ? perm[j] : j];
+            out << ' ' << v[j];
         }
         out << '\n';
      }
@@ -9269,10 +9260,9 @@ void Mesh::PrintVTU(std::ostream &out, int ref, VTKFormat format,
         {
            coff = coff+nv;
            offset.push_back(coff);
-            const int *p = (geom == Geometry::PRISM) ? vtk_prism_perm : NULL;
            for (int k = 0; k < nv; k++, j++)
            {
-               WriteBinaryOrASCII(out, buf, np + RG[p ? p[j] : j], " ", format);
+               WriteBinaryOrASCII(out, buf, np + RG[j], " ", format);
            }
            if (format == VTKFormat::ASCII) { out << '\n'; }
         }
@@ -10560,6 +10550,227 @@ void Mesh::RemoveInternalBoundaries()
   attribs.Copy(bdr_attributes);
 }

+void Mesh::RemoveInternalBoundaries(const Array<int> &_keep)
+{
+   if (NURBSext || ncmesh) { return; }
+
+   std::unordered_set<int> keep(_keep.GetData(),
+                                _keep.GetData() + _keep.Size());
+
+   int num_bdr_elem = 0;
+   int new_bel_to_edge_nnz = 0;
+   for (int i = 0; i < GetNBE(); i++)
+   {
+      if (FaceIsInterior(GetBdrElementEdgeIndex(i)) && 
+         (keep.count(boundary[i]->GetAttribute()) == 0))
+      {
+         FreeElement(boundary[i]);
+      }
+      else
+      {
+         num_bdr_elem++;
+         if (Dim == 3)
+         {
+            new_bel_to_edge_nnz += bel_to_edge->RowSize(i);
+         }
+      }
+   }
+
+   if (num_bdr_elem == GetNBE()) { return; }
+
+   Array<Element *> new_boundary(num_bdr_elem);
+   Array<int> new_be_to_edge, new_be_to_face;
+   Table *new_bel_to_edge = NULL;
+   new_boundary.SetSize(0);
+   if (Dim == 2)
+   {
+      new_be_to_edge.Reserve(num_bdr_elem);
+   }
+   else if (Dim == 3)
+   {
+      new_be_to_face.Reserve(num_bdr_elem);
+      new_bel_to_edge = new Table;
+      new_bel_to_edge->SetDims(num_bdr_elem, new_bel_to_edge_nnz);
+   }
+   for (int i = 0; i < GetNBE(); i++)
+   {
+      /// if it's not interior or is in the keep boundary list
+      if (!FaceIsInterior(GetBdrElementEdgeIndex(i)) || 
+          keep.count(boundary[i]->GetAttribute()))
+      {
+         new_boundary.Append(boundary[i]);
+         if (Dim == 2)
+         {
+            new_be_to_edge.Append(be_to_edge[i]);
+         }
+         else if (Dim == 3)
+         {
+            int row = new_be_to_face.Size();
+            new_be_to_face.Append(be_to_face[i]);
+            int *e = bel_to_edge->GetRow(i);
+            int ne = bel_to_edge->RowSize(i);
+            int *new_e = new_bel_to_edge->GetRow(row);
+            for (int j = 0; j < ne; j++)
+            {
+               new_e[j] = e[j];
+            }
+            new_bel_to_edge->GetI()[row+1] = new_bel_to_edge->GetI()[row] + ne;
+         }
+      }
+   }
+
+   NumOfBdrElements = new_boundary.Size();
+   mfem::Swap(boundary, new_boundary);
+
+   if (Dim == 2)
+   {
+      mfem::Swap(be_to_edge, new_be_to_edge);
+   }
+   else if (Dim == 3)
+   {
+      mfem::Swap(be_to_face, new_be_to_face);
+      delete bel_to_edge;
+      bel_to_edge = new_bel_to_edge;
+   }
+
+   Array<int> attribs(num_bdr_elem);
+   for (int i = 0; i < attribs.Size(); i++)
+   {
+      attribs[i] = GetBdrAttribute(i);
+   }
+   attribs.Sort();
+   attribs.Unique();
+   bdr_attributes.DeleteAll();
+   attribs.Copy(bdr_attributes);
+}
+
+void Mesh::RemoveInternalBoundariesNotAdjacentTo(const Array<int> &_regions)
+{
+   if (NURBSext || ncmesh) { return; }
+
+   std::unordered_set<int> regions(_regions.GetData(),
+                                   _regions.GetData() + _regions.Size());
+
+   int num_bdr_elem = 0;
+   int new_bel_to_edge_nnz = 0;
+   for (int i = 0; i < GetNBE(); i++)
+   {
+      auto faceNo = GetBdrElementEdgeIndex(i);
+      bool interior = FaceIsInterior(faceNo);
+
+      auto face_info = faces_info[faceNo];
+      auto adj = regions.count(elements[face_info.Elem1No]->GetAttribute());
+      if (face_info.Elem2No > 0)
+         adj += regions.count(elements[face_info.Elem2No]->GetAttribute());
+
+      bool remove = interior & !adj;
+
+      // if (remove)
+      // {
+      //    std::cout << "remove face on: " << boundary[i]->GetAttribute()
+      //              << " adj to ("
+      //              << elements[face_info.Elem1No]->GetAttribute();
+      //    if (face_info.Elem2No > 0)
+      //    {
+      //       std::cout << ", "
+      //                 << elements[face_info.Elem2No]->GetAttribute() << ")\n";
+      //    }
+      //    else
+      //    {
+      //       std::cout << ")\n";
+      //    }
+      // }
+
+      if (remove)
+      {
+         FreeElement(boundary[i]);
+      }
+      else
+      {
+         num_bdr_elem++;
+         if (Dim == 3)
+         {
+            new_bel_to_edge_nnz += bel_to_edge->RowSize(i);
+         }
+      }
+   }
+
+   if (num_bdr_elem == GetNBE()) { return; }
+
+   Array<Element *> new_boundary(num_bdr_elem);
+   Array<int> new_be_to_edge, new_be_to_face;
+   Table *new_bel_to_edge = NULL;
+   new_boundary.SetSize(0);
+   if (Dim == 2)
+   {
+      new_be_to_edge.Reserve(num_bdr_elem);
+   }
+   else if (Dim == 3)
+   {
+      new_be_to_face.Reserve(num_bdr_elem);
+      new_bel_to_edge = new Table;
+      new_bel_to_edge->SetDims(num_bdr_elem, new_bel_to_edge_nnz);
+   }
+   for (int i = 0; i < GetNBE(); i++)
+   {
+      auto faceNo = GetBdrElementEdgeIndex(i);
+      bool interior = FaceIsInterior(faceNo);
+
+      auto face_info = faces_info[faceNo];
+      auto adj = regions.count(elements[face_info.Elem1No]->GetAttribute());
+      if (face_info.Elem2No > 0)
+         adj += regions.count(elements[face_info.Elem2No]->GetAttribute());
+
+      bool keep = !interior || adj;
+
+      if (keep)
+      {
+         new_boundary.Append(boundary[i]);
+         if (Dim == 2)
+         {
+            new_be_to_edge.Append(be_to_edge[i]);
+         }
+         else if (Dim == 3)
+         {
+            int row = new_be_to_face.Size();
+            new_be_to_face.Append(be_to_face[i]);
+            int *e = bel_to_edge->GetRow(i);
+            int ne = bel_to_edge->RowSize(i);
+            int *new_e = new_bel_to_edge->GetRow(row);
+            for (int j = 0; j < ne; j++)
+            {
+               new_e[j] = e[j];
+            }
+            new_bel_to_edge->GetI()[row+1] = new_bel_to_edge->GetI()[row] + ne;
+         }
+      }
+   }
+
+   NumOfBdrElements = new_boundary.Size();
+   mfem::Swap(boundary, new_boundary);
+
+   if (Dim == 2)
+   {
+      mfem::Swap(be_to_edge, new_be_to_edge);
+   }
+   else if (Dim == 3)
+   {
+      mfem::Swap(be_to_face, new_be_to_face);
+      delete bel_to_edge;
+      bel_to_edge = new_bel_to_edge;
+   }
+
+   Array<int> attribs(num_bdr_elem);
+   for (int i = 0; i < attribs.Size(); i++)
+   {
+      attribs[i] = GetBdrAttribute(i);
+   }
+   attribs.Sort();
+   attribs.Unique();
+   bdr_attributes.DeleteAll();
+   attribs.Copy(bdr_attributes);
+}
+
 void Mesh::FreeElement(Element *E)
 {
 #ifdef MFEM_USE_MEMALLOC
@@ -1036,7 +1036,6 @@ public:
   }
   void GetFaceElements (int Face, int *Elem1, int *Elem2) const;
   void GetFaceInfos (int Face, int *Inf1, int *Inf2) const;
-   void GetFaceInfos (int Face, int *Inf1, int *Inf2, int *NCFace) const;

   Geometry::Type GetFaceGeometryType(int Face) const;
   Element::Type  GetFaceElementType(int Face) const;
@@ -1307,6 +1306,12 @@ public:
       have two adjacent faces in 3D, or edges in 2D. */
   void RemoveInternalBoundaries();

+   /** Remove boundary elements that lie in the interior of the mesh, except 
+       for those with boundary attributes in `keep` */
+   void RemoveInternalBoundaries(const Array<int> &keep);
+
+   void RemoveInternalBoundariesNotAdjacentTo(const Array<int> &regions);
+
   /** @brief Get the size of the i-th element relative to the perfect
       reference element. */
   double GetElementSize(int i, int type = 0);
@@ -284,7 +284,6 @@ public:
                             int ordering = 1);

   int GetNFaceNeighbors() const { return face_nbr_group.Size(); }
-   int GetNFaceNeighborElements() const { return face_nbr_elements.Size(); }
   int GetFaceNbrGroup(int fn) const { return face_nbr_group[fn]; }
   int GetFaceNbrRank(int fn) const;

@@ -18,8 +18,6 @@
 namespace mfem
 {

-const int vtk_prism_perm[6] = {0, 2, 1, 3, 5, 4};
-
 int BarycentricToVTKTriangle(int *b, int ref)
 {
   // Cf. https://git.io/JvW8f
@@ -26,8 +26,6 @@ enum class VTKFormat
   BINARY32
 };

-extern const int vtk_prism_perm[6];
-
 /// Create the VTK element connectivity array for a given element geometry and
 /// refinement level. Converts node numbers from MFEM to VTK ordering.
 void CreateVTKElementConnectivity(Array<int> &con, Geometry::Type geom,
@@ -19,10 +19,6 @@ if (MFEM_USE_GSLIB)
    MAIN findpts.cpp
    LIBRARIES mfem)

-  add_mfem_miniapp(field-interp
-    MAIN field-interp.cpp
-    LIBRARIES mfem)
-
  # Parallel apps.
  if (MFEM_USE_MPI)
    add_mfem_miniapp(pfindpts
@@ -1,717 +0,0 @@
-MFEM mesh v1.0
-
-#
-# MFEM Geometry Types (see mesh/geom.hpp):
-#
-# POINT       = 0
-# SEGMENT     = 1
-# TRIANGLE    = 2
-# SQUARE      = 3
-# TETRAHEDRON = 4
-# CUBE        = 5
-#
-
-dimension
-2
-
-elements
-18
-1 3 0 1 8 7
-1 3 1 2 9 8
-1 3 2 3 10 9
-1 3 3 4 11 10
-1 3 4 5 12 11
-1 3 5 6 13 12
-1 3 7 8 15 14
-1 3 8 9 16 15
-1 3 9 10 17 16
-1 3 10 11 18 17
-1 3 11 12 19 18
-1 3 12 13 20 19
-1 3 14 15 22 21
-1 3 15 16 23 22
-1 3 16 17 24 23
-1 3 17 18 25 24
-1 3 18 19 26 25
-1 3 19 20 27 26
-
-boundary
-18
-4 1 0 1
-4 1 1 2
-4 1 2 3
-4 1 3 4
-4 1 4 5
-4 1 5 6
-2 1 22 21
-2 1 23 22
-2 1 24 23
-2 1 25 24
-2 1 26 25
-2 1 27 26
-4 1 7 0
-4 1 14 7
-4 1 21 14
-4 1 6 13
-4 1 13 20
-4 1 20 27
-
-vertices
-28
-
-nodes
-FiniteElementSpace
-FiniteElementCollection: H1_2D_P4
-VDim: 2
-Ordering: 0
-
-0
-1
-2
-3
-3.1666667
-3.3333333
-3.5
-0.33333333
-0.61111111
-1.5555556
-2.5
-2.8888889
-3.2777778
-3.6666667
-0.66666667
-0.22222222
-1.1111111
-2
-2.6111111
-3.2222222
-3.8333333
-1
-0.16666667
-0.66666667
-1.5
-2.3333333
-3.1666667
-4
-0.17267316
-0.5
-0.82732684
-0.93284932
-0.80555556
-0.67826179
-0.17025312
-0.13888889
-0.4480309
-0.057557722
-0.16666667
-0.27577561
-1.1726732
-1.5
-1.8273268
-1.9232564
-1.7777778
-1.6322992
-0.77419132
-1.0833333
-1.3924753
-2.1726732
-2.5
-2.8273268
-2.9136634
-2.75
-2.5863366
-1.7186358
-2.0277778
-2.3369198
-3.0287789
-3.0833333
-3.1378878
-3.1187019
-3.0277778
-2.9368537
-2.5671507
-2.6944444
-2.8217382
-3.1954455
-3.25
-3.3045545
-3.3237404
-3.3055556
-3.2873707
-2.9560396
-3.0833333
-3.2106271
-3.3621122
-3.4166667
-3.4712211
-3.5287789
-3.5833333
-3.6378878
-3.3449285
-3.4722222
-3.599516
-0.54396044
-0.41666667
-0.2893729
-0.51317941
-0.22222222
-0.068734965
-0.39089105
-0.5
-0.60910895
-1.4788119
-1.3333333
-1.1878547
-0.37570948
-0.66666667
-0.95762385
-2.4136634
-2.25
-2.0863366
-1.2645984
-1.5555556
-1.8465127
-2.8409241
-2.75
-2.6590759
-2.1055225
-2.3055556
-2.5055886
-3.2681848
-3.25
-3.2318152
-2.7166336
-2.9166667
-3.1166997
-3.6954455
-3.75
-3.8045545
-3.3277447
-3.5277778
-3.7278108
-0.15507155
-0.027777778
-0.099515992
-0.8561057
-0.58333333
-0.31056097
-0.72422439
-0.83333333
-0.94244228
-1.0343675
-0.88888889
-0.7434103
-0.022772363
-0.25
-0.52277236
-1.9136634
-1.75
-1.5863366
-0.81056097
-1.0833333
-1.3561057
-2.5631463
-2.4722222
-2.3812981
-1.6438943
-1.9166667
-2.189439
-3.2126293
-3.1944444
-3.1762596
-2.4772276
-2.75
-3.0227724
-3.8621122
-3.9166667
-3.9712211
-3.310561
-3.5833333
-3.8561057
-0.113459
-0.4376458
-0.76183261
-0.0012100212
-0.31944444
-0.63767887
-0.11103896
-0.20124309
-0.51352513
-1.103866
-1.4280528
-1.7522397
-0.97343224
-1.2916667
-1.6099011
-0.84299844
-1.1552805
-1.4675625
-2.0942731
-2.4184599
-2.7426467
-1.9456545
-2.2638889
-2.5821233
-1.7970358
-2.1093179
-2.4215999
-2.9490681
-3.0161827
-3.0832973
-2.7979648
-2.8888889
-2.979813
-2.6468615
-2.7615951
-2.8763288
-3.1541065
-3.2212211
-3.2883357
-3.0757425
-3.1666667
-3.2575908
-2.9973785
-3.1121122
-3.2268458
-3.359145
-3.4262596
-3.4933742
-3.3535203
-3.4444444
-3.5353686
-3.3478956
-3.4626293
-3.5773629
-0.22946729
-0.076534691
-0.38253667
-0.34171627
-0.041666667
-0.25838293
-0.45396524
-0.15986802
-0.13422919
-0.7053842
-1.0113862
-1.3173882
-0.5749504
-0.875
-1.1750496
-0.4445166
-0.73861382
-1.032711
-1.6402357
-1.9462377
-2.2522397
-1.4916171
-1.7916667
-2.0917163
-1.3429984
-1.6370957
-1.9311929
-2.4874399
-2.6272938
-2.7671477
-2.3363366
-2.5
-2.6636634
-2.1852333
-2.3727062
-2.5601792
-2.9147006
-3.0545545
-3.1944084
-2.8363366
-3
-3.1636634
-2.7579726
-2.9454455
-3.1329185
-3.3419613
-3.4818152
-3.6216691
-3.3363366
-3.5
-3.6636634
-3.3307119
-3.5181848
-3.7056578
-0.57239358
-0.28457642
-0.0032407353
-0.68464255
-0.40277778
-0.120913
-0.79689153
-0.52097913
-0.24506674
-0.30690236
-0.59471951
-0.88253667
-0.17646856
-0.45833333
-0.74019811
-0.046034758
-0.32194715
-0.59785955
-1.1861983
-1.4740155
-1.7618326
-1.0375797
-1.3194444
-1.6013092
-0.88896104
-1.1648734
-1.4407858
-2.0258117
-2.2384049
-2.4509981
-1.8747084
-2.1111111
-2.3475138
-1.7236051
-1.9838173
-2.2440296
-2.6752946
-2.8878878
-3.100481
-2.5969306
-2.8333333
-3.069736
-2.5185666
-2.7787789
-3.0389911
-3.3247775
-3.5373707
-3.7499639
-3.3191528
-3.5555556
-3.7919583
-3.3135281
-3.5737404
-3.8339526
-0
-0.5
-1
-1.5
-1
-0.5
-0
-0.66666667
-1
-1.3333333
-1.6666667
-1.3333333
-1
-0.66666667
-1.3333333
-1.5
-1.6666667
-1.8333333
-1.6666667
-1.5
-1.3333333
-2
-2
-2
-2
-2
-2
-2
-0.086336582
-0.25
-0.41366342
-0.58633658
-0.75
-0.91366342
-0.72422439
-0.83333333
-0.94244228
-0.11511544
-0.33333333
-0.55155122
-0.58633658
-0.75
-0.91366342
-1.0575577
-1.1666667
-1.2757756
-1.0575577
-1.1666667
-1.2757756
-1.0863366
-1.25
-1.4136634
-1.5287789
-1.5833333
-1.6378878
-1.3908911
-1.5
-1.6091089
-1.4136634
-1.25
-1.0863366
-1.0575577
-1.1666667
-1.2757756
-1.6091089
-1.5
-1.3908911
-0.91366342
-0.75
-0.58633658
-0.58633658
-0.75
-0.91366342
-1.2757756
-1.1666667
-1.0575577
-0.41366342
-0.25
-0.086336582
-0.11511544
-0.33333333
-0.55155122
-0.94244228
-0.83333333
-0.72422439
-1.0863366
-1.25
-1.4136634
-1.3621122
-1.4166667
-1.4712211
-0.78178211
-1
-1.2182179
-1.3908911
-1.5
-1.6091089
-1.5287789
-1.5833333
-1.6378878
-1.6954455
-1.75
-1.8045545
-1.6954455
-1.75
-1.8045545
-1.3908911
-1.5
-1.6091089
-1.8045545
-1.75
-1.6954455
-1.0863366
-1.25
-1.4136634
-1.6378878
-1.5833333
-1.5287789
-0.78178211
-1
-1.2182179
-1.4712211
-1.4166667
-1.3621122
-1.5863366
-1.75
-1.9136634
-2
-2
-2
-1.4484488
-1.6666667
-1.8848846
-1.7242244
-1.8333333
-1.9424423
-2
-2
-2
-1.8621122
-1.9166667
-1.9712211
-2
-2
-2
-1.7242244
-1.8333333
-1.9424423
-2
-2
-2
-1.5863366
-1.75
-1.9136634
-2
-2
-2
-1.4484488
-1.6666667
-1.8848846
-2
-2
-2
-0.19648269
-0.35072601
-0.50496934
-0.40528049
-0.54166667
-0.67805285
-0.61407828
-0.73260732
-0.85113636
-0.66770383
-0.82194715
-0.97619048
-0.82194715
-0.95833333
-1.0947195
-0.97619048
-1.0947195
-1.2132486
-1.138925
-1.2931683
-1.4474116
-1.2386138
-1.375
-1.5113862
-1.3383027
-1.4568317
-1.5753607
-1.4474116
-1.2931683
-1.138925
-1.5113862
-1.375
-1.2386138
-1.5753607
-1.4568317
-1.3383027
-0.97619048
-0.82194715
-0.66770383
-1.0947195
-0.95833333
-0.82194715
-1.2132486
-1.0947195
-0.97619048
-0.50496934
-0.35072601
-0.19648269
-0.67805285
-0.54166667
-0.40528049
-0.85113636
-0.73260732
-0.61407828
-0.83437049
-0.93405935
-1.0337482
-1.0431683
-1.125
-1.2068317
-1.2519661
-1.3159407
-1.3799152
-1.138925
-1.2386138
-1.3383027
-1.2931683
-1.375
-1.4568317
-1.4474116
-1.5113862
-1.5753607
-1.4434794
-1.5431683
-1.6428571
-1.5431683
-1.625
-1.7068317
-1.6428571
-1.7068317
-1.7708063
-1.6428571
-1.5431683
-1.4434794
-1.7068317
-1.625
-1.5431683
-1.7708063
-1.7068317
-1.6428571
-1.3383027
-1.2386138
-1.138925
-1.4568317
-1.375
-1.2931683
-1.5753607
-1.5113862
-1.4474116
-1.0337482
-0.93405935
-0.83437049
-1.2068317
-1.125
-1.0431683
-1.3799152
-1.3159407
-1.2519661
-1.4722583
-1.5173927
-1.5625271
-1.6810561
-1.7083333
-1.7356106
-1.8898539
-1.899274
-1.9086941
-1.6101461
-1.6552805
-1.7004149
-1.7643894
-1.7916667
-1.8189439
-1.9186328
-1.9280528
-1.9374729
-1.7480339
-1.7931683
-1.8383027
-1.8477228
-1.875
-1.9022772
-1.9474116
-1.9568317
-1.9662518
-1.8383027
-1.7931683
-1.7480339
-1.9022772
-1.875
-1.8477228
-1.9662518
-1.9568317
-1.9474116
-1.7004149
-1.6552805
-1.6101461
-1.8189439
-1.7916667
-1.7643894
-1.9374729
-1.9280528
-1.9186328
-1.5625271
-1.5173927
-1.4722583
-1.7356106
-1.7083333
-1.6810561
-1.9086941
-1.899274
-1.8898539
@@ -33,25 +33,25 @@
 //
 // Sample runs:
 //   Adapted analytic shape:
-//     mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 2 -tid 4 -ni 200 -bnd -qt 1 -qo 8
+//     mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 2 -tid 4 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8
 //   Adapted analytic size+orientation:
-//     mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 14 -tid 4 -ni 100 -bnd -qt 1 -qo 8 -fd
+//     mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 14 -tid 4 -ni 100 -ls 2 -li 100 -bnd -qt 1 -qo 8 -fd
 //   Adapted analytic shape+orientation:
-//     mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 85 -tid 4 -ni 100 -bnd -qt 1 -qo 8 -fd
+//     mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 85 -tid 4 -ni 100 -ls 2 -li 100 -bnd -qt 1 -qo 8 -fd
 //
 //   Adapted discrete size:
-//     mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 80 -tid 5 -ni 50 -qo 4 -nor
+//     mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 7 -tid 5 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8
 //   Adapted discrete size; explicit combo of metrics; mixed tri/quad mesh:
 //     mesh-optimizer -m ../../data/square-mixed.mesh -o 2 -rs 2 -mid 2 -tid 5 -ni 200 -bnd -qo 6 -cmb 2 -nor
 //   Adapted discrete size+aspect_ratio:
 //     mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 7 -tid 6 -ni 100
 //     mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 7 -tid 6 -ni 100 -qo 6 -ex -st 1 -nor
 //   Adapted discrete size+orientation (requires GSLIB):
-//   * mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 36 -tid 8 -qo 4 -fd -ae 1 -nor
+//   * mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 14 -tid 8 -ni 100  -ls 2 -li 100 -bnd -qt 1 -qo 8 -fd -ae 1
 //   Adapted discrete aspect-ratio+orientation (requires GSLIB):
-//   * mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 85 -tid 8 -ni 10 -bnd -qt 1 -qo 8 -fd -ae 1
+//   * mesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 85 -tid 8 -ni 10  -ls 2 -li 100 -bnd -qt 1 -qo 8 -fd -ae 1
 //   Adapted discrete aspect ratio (3D):
-//     mesh-optimizer -m cube.mesh -o 2 -rs 2 -mid 302 -tid 7 -ni 20 -bnd -qt 1 -qo 8
+//     mesh-optimizer -m cube.mesh -o 2 -rs 2 -mid 302 -tid 7 -ni 20  -ls 2 -li 100 -bnd -qt 1 -qo 8
 //
 //   Adaptive limiting:
 //     mesh-optimizer -m stretched2D.mesh -o 2 -mid 2 -tid 1 -ni 50 -qo 5 -nor -vl 1 -alc 0.5
@@ -61,33 +61,28 @@
 //   * mesh-optimizer -m stretched2D.mesh -o 2 -mid 2 -tid 1 -ni 50 -qo 5 -nor -vl 1 -alc 0.5 -fd -ae 1
 //
 //   Blade shape:
-//     mesh-optimizer -m blade.mesh -o 4 -mid 2 -tid 1 -ni 30 -ls 3 -art 1 -bnd -qt 1 -qo 8
+//     mesh-optimizer -m blade.mesh -o 4 -rs 0 -mid 2 -tid 1 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8
 //   Blade shape with FD-based solver:
-//     mesh-optimizer -m blade.mesh -o 4 -mid 2 -tid 1 -ni 30 -ls 4 -bnd -qt 1 -qo 8 -fd
+//     mesh-optimizer -m blade.mesh -o 4 -rs 0 -mid 2 -tid 1 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8 -fd
 //   Blade limited shape:
-//     mesh-optimizer -m blade.mesh -o 4 -mid 2 -tid 1 -bnd -qt 1 -qo 8 -lc 5000
+//     mesh-optimizer -m blade.mesh -o 4 -rs 0 -mid 2 -tid 1 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8 -lc 5000
 //   ICF shape and equal size:
-//     mesh-optimizer -o 3 -mid 9 -tid 2 -ni 25 -ls 3 -art 2 -qo 5
+//     mesh-optimizer -o 3 -rs 0 -mid 9 -tid 2 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8
 //   ICF shape and initial size:
-//     mesh-optimizer -o 3 -mid 9 -tid 3 -ni 30 -ls 3 -bnd -qt 1 -qo 8
+//     mesh-optimizer -o 3 -rs 0 -mid 9 -tid 3 -ni 100 -ls 2 -li 100 -bnd -qt 1 -qo 8
 //   ICF shape:
-//     mesh-optimizer -o 3 -mid 1 -tid 1 -ni 100 -bnd -qt 1 -qo 8
+//     mesh-optimizer -o 3 -rs 0 -mid 1 -tid 1 -ni 100 -ls 2 -li 100 -bnd -qt 1 -qo 8
 //   ICF limited shape:
-//     mesh-optimizer -o 3 -mid 1 -tid 1 -ni 100 -bnd -qt 1 -qo 8 -lc 10
+//     mesh-optimizer -o 3 -rs 0 -mid 1 -tid 1 -ni 100 -ls 2 -li 100 -bnd -qt 1 -qo 8 -lc 10
 //   ICF combo shape + size (rings, slow convergence):
-//     mesh-optimizer -o 3 -mid 1 -tid 1 -ni 1000 -bnd -qt 1 -qo 8 -cmb 1
+//     mesh-optimizer -o 3 -rs 0 -mid 1 -tid 1 -ni 1000 -ls 2 -li 100 -bnd -qt 1 -qo 8 -cmb 1
 //   Mixed tet / cube / hex mesh with limiting:
 //     mesh-optimizer -m ../../data/fichera-mixed-p2.mesh -o 4 -rs 1 -mid 301 -tid 1 -fix-bnd -qo 6 -nor -lc 0.25
 //   3D pinched sphere shape (the mesh is in the mfem/data GitHub repository):
-//   * mesh-optimizer -m ../../../mfem_data/ball-pert.mesh -o 4 -mid 303 -tid 1 -ni 20 -li 500 -fix-bnd
+//   * mesh-optimizer -m ../../../mfem_data/ball-pert.mesh -o 4 -rs 0 -mid 303 -tid 1 -ni 20 -ls 2 -li 500 -fix-bnd
 //   2D non-conforming shape and equal size:
-//     mesh-optimizer -m ./amr-quad-q2.mesh -o 2 -rs 1 -mid 9 -tid 2 -ni 200 -bnd -qt 1 -qo 8
-//
-//   2D untangling:
-//     mesh-optimizer -m jagged.mesh -o 2 -mid 22 -tid 1 -ni 50 -li 50 -qo 4 -fd -vl 1
-//   3D untangling (the mesh is in the mfem/data GitHub repository):
-//   * mesh-optimizer -m ../../../mfem_data/cube-holes-inv.mesh -o 3 -mid 313 -tid 1 -rtol 1e-5 -li 50 -qo 4 -fd -vl 1
-//
+//     mesh-optimizer -m ./amr-quad-q2.mesh -o 2 -rs 1 -mid 9 -tid 2 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8
+

 #include "mfem.hpp"
 #include "../common/mfem-common.hpp"
@@ -112,9 +107,8 @@ int main(int argc, char *argv[])
   int quad_type         = 1;
   int quad_order        = 8;
   int solver_type       = 0;
-   int solver_iter       = 20;
+   int solver_iter       = 10;
   double solver_rtol    = 1e-10;
-   int solver_art_type   = 0;
   int lin_solver        = 2;
   int max_lin_iter      = 100;
   bool move_bnd         = true;
@@ -138,38 +132,26 @@ int main(int argc, char *argv[])
                  "Random perturbation scaling factor.");
   args.AddOption(&metric_id, "-mid", "--metric-id",
                  "Mesh optimization metric:\n\t"
-                  "T-metrics\n\t"
                  "1  : |T|^2                          -- 2D shape\n\t"
                  "2  : 0.5|T|^2/tau-1                 -- 2D shape (condition number)\n\t"
                  "7  : |T-T^-t|^2                     -- 2D shape+size\n\t"
                  "9  : tau*|T-T^-t|^2                 -- 2D shape+size\n\t"
-                  "14 : |T-I|^2                        -- 2D shape+size+orientation\n\t"
+                  "14: 0.5*(1-cos(theta_A - theta_W)   -- 2D Sh+Sz+Alignment\n\t"
                  "22 : 0.5(|T|^2-2*tau)/(tau-tau_0)   -- 2D untangling\n\t"
                  "50 : 0.5|T^tT|^2/tau^2-1            -- 2D shape\n\t"
                  "55 : (tau-1)^2                      -- 2D size\n\t"
                  "56 : 0.5(sqrt(tau)-1/sqrt(tau))^2   -- 2D size\n\t"
                  "58 : |T^tT|^2/(tau^2)-2*|T|^2/tau+2 -- 2D shape\n\t"
                  "77 : 0.5(tau-1/tau)^2               -- 2D size\n\t"
-                  "80 : (1-gamma)mu_2 + gamma mu_77    -- 2D shape+size\n\t"
-                  "85 : |T-|T|/sqrt(2)I|^2             -- 2D shape+orientation\n\t"
-                  "98 : (1/tau)|T-I|^2                 -- 2D shape+size+orientation\n\t"
-                  // "211: (tau-1)^2-tau+sqrt(tau^2)      -- 2D untangling\n\t"
-                  // "252: 0.5(tau-1)^2/(tau-tau_0)       -- 2D untangling\n\t"
+                  "211: (tau-1)^2-tau+sqrt(tau^2)      -- 2D untangling\n\t"
+                  "252: 0.5(tau-1)^2/(tau-tau_0)       -- 2D untangling\n\t"
                  "301: (|T||T^-1|)/3-1              -- 3D shape\n\t"
                  "302: (|T|^2|T^-1|^2)/9-1          -- 3D shape\n\t"
                  "303: (|T|^2)/3*tau^(2/3)-1        -- 3D shape\n\t"
-                  //"311: (tau-1)^2-tau+sqrt(tau^2+eps)-- 3D untangling\n\t"
-                  "313: (|T|^2)(tau-tau0)^(-2/3)/3   -- 3D untangling\n\t"
                  "315: (tau-1)^2                    -- 3D size\n\t"
                  "316: 0.5(sqrt(tau)-1/sqrt(tau))^2 -- 3D size\n\t"
                  "321: |T-T^-t|^2                   -- 3D shape+size\n\t"
-                  // "352: 0.5(tau-1)^2/(tau-tau_0)     -- 3D untangling\n\t"
-                  "A-metrics\n\t"
-                  "11 : (1/4*alpha)|A-(adjA)^T(W^TW)/omega|^2 -- 2D shape\n\t"
-                  "36 : (1/alpha)|A-W|^2                      -- 2D shape+size+orientation\n\t"
-                  "107: (1/2*alpha)|A-|A|/|W|W|^2             -- 2D shape+orientation\n\t"
-                  "126: (1-gamma)nu_11 + gamma*nu_14a         -- 2D shape+size\n\t"
-                 );
+                  "352: 0.5(tau-1)^2/(tau-tau_0)     -- 3D untangling");
   args.AddOption(&target_id, "-tid", "--target-id",
                  "Target (ideal element) type:\n\t"
                  "1: Ideal shape, unit size\n\t"
@@ -193,25 +175,15 @@ 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"
-                  "1: CG\n\t"
-                  "2: MINRES\n\t"
-                  "3: MINRES + Jacobi preconditioner\n\t"
-                  "4: MINRES + l1-Jacobi preconditioner");
+                  "Linear solver: 0 - l1-Jacobi, 1 - CG, 2 - MINRES.");
   args.AddOption(&max_lin_iter, "-li", "--lin-iter",
                  "Maximum number of iterations in the linear solve.");
   args.AddOption(&move_bnd, "-bnd", "--move-boundary", "-fix-bnd",
                  "--fix-boundary",
                  "Enable motion along horizontal and vertical boundaries.");
   args.AddOption(&combomet, "-cmb", "--combo-type",
-                  "Combination of metrics options:\n\t"
+                  "Combination of metrics options:"
                  "0: Use single metric\n\t"
                  "1: Shape + space-dependent size given analytically\n\t"
                  "2: Shape + adapted size given discretely; shared target");
@@ -343,40 +315,28 @@ int main(int argc, char *argv[])
   TMOP_QualityMetric *metric = NULL;
   switch (metric_id)
   {
-      // T-metrics
      case 1: metric = new TMOP_Metric_001; break;
      case 2: metric = new TMOP_Metric_002; break;
      case 7: metric = new TMOP_Metric_007; break;
      case 9: metric = new TMOP_Metric_009; break;
-      case 14: metric = new TMOP_Metric_014; break;
+      case 14: metric = new TMOP_Metric_SSA2D; break;
      case 22: metric = new TMOP_Metric_022(tauval); break;
      case 50: metric = new TMOP_Metric_050; break;
      case 55: metric = new TMOP_Metric_055; break;
      case 56: metric = new TMOP_Metric_056; break;
      case 58: metric = new TMOP_Metric_058; break;
      case 77: metric = new TMOP_Metric_077; break;
-      case 80: metric = new TMOP_Metric_080(0.5); break;
      case 85: metric = new TMOP_Metric_085; break;
-      case 98: metric = new TMOP_Metric_098; break;
-      // case 211: metric = new TMOP_Metric_211; break;
-      // case 252: metric = new TMOP_Metric_252(tauval); break;
+      case 211: metric = new TMOP_Metric_211; break;
+      case 252: metric = new TMOP_Metric_252(tauval); break;
      case 301: metric = new TMOP_Metric_301; break;
      case 302: metric = new TMOP_Metric_302; break;
      case 303: metric = new TMOP_Metric_303; break;
-      // case 311: metric = new TMOP_Metric_311; break;
-      case 313: metric = new TMOP_Metric_313(tauval); break;
      case 315: metric = new TMOP_Metric_315; break;
      case 316: metric = new TMOP_Metric_316; break;
      case 321: metric = new TMOP_Metric_321; break;
-      // case 352: metric = new TMOP_Metric_352(tauval); break;
-      // A-metrics
-      case 11: metric = new TMOP_AMetric_011; break;
-      case 36: metric = new TMOP_AMetric_036; break;
-      case 107: metric = new TMOP_AMetric_107a; break;
-      case 126: metric = new TMOP_AMetric_126(0.9); break;
-      default:
-         cout << "Unknown metric_id: " << metric_id << endl;
-         return 3;
+      case 352: metric = new TMOP_Metric_352(tauval); break;
+      default: cout << "Unknown metric_id: " << metric_id << endl; return 3;
   }
   TargetConstructor::TargetType target_t;
   TargetConstructor *target_c = NULL;
@@ -385,7 +345,7 @@ int main(int argc, char *argv[])
   FiniteElementSpace ind_fes(mesh, &ind_fec);
   FiniteElementSpace ind_fesv(mesh, &ind_fec, dim);
   GridFunction size(&ind_fes), aspr(&ind_fes), disc(&ind_fes), ori(&ind_fes);
-   GridFunction aspr3d(&ind_fesv);
+   GridFunction aspr3d(&ind_fesv), size3d(&ind_fesv);
   switch (target_id)
   {
      case 1: target_t = TargetConstructor::IDEAL_SHAPE_UNIT_SIZE; break;
@@ -443,14 +403,14 @@ int main(int argc, char *argv[])
 #endif
         }

-         // Diffuse the interface
+         //Diffuse the interface
         DiffuseField(disc,2);

-         // Get  partials with respect to x and y of the grid function
+         //Get  partials with respect to x and y of the grid function
         disc.GetDerivative(1,0,d_x);
         disc.GetDerivative(1,1,d_y);

-         // Compute the squared magnitude of the gradient
+         //Compute the squared magnitude of the gradient
         for (int i = 0; i < size.Size(); i++)
         {
            size(i) = std::pow(d_x(i),2)+std::pow(d_y(i),2);
@@ -557,7 +517,7 @@ int main(int argc, char *argv[])
 #endif
         }

-         if (metric_id == 14 || metric_id == 36)
+         if (metric_id == 14)
         {
            ConstantCoefficient ind_coeff(0.1*0.1);
            size.ProjectCoefficient(ind_coeff);
@@ -702,40 +662,6 @@ int main(int argc, char *argv[])
   }
   else { a.AddDomainIntegrator(he_nlf_integ); }

-   // Compute the minimum det(J) of the starting mesh.
-   tauval = infinity();
-   const int NE = mesh->GetNE();
-   for (int i = 0; i < NE; i++)
-   {
-      const IntegrationRule &ir =
-         irules->Get(fespace->GetFE(i)->GetGeomType(), quad_order);
-      ElementTransformation *transf = mesh->GetElementTransformation(i);
-      for (int j = 0; j < ir.GetNPoints(); j++)
-      {
-         transf->SetIntPoint(&ir.IntPoint(j));
-         tauval = min(tauval, transf->Jacobian().Det());
-      }
-   }
-   cout << "Minimum det(J) of the original mesh is " << tauval << endl;
-
-   if (tauval < 0.0 && metric_id != 22 && metric_id != 211 && metric_id != 252
-       && metric_id != 311 && metric_id != 313 && metric_id != 352)
-   {
-      MFEM_ABORT("The input mesh is inverted! Try an untangling metric.");
-   }
-   if (tauval < 0.0)
-   {
-      MFEM_VERIFY(target_t == TargetConstructor::IDEAL_SHAPE_UNIT_SIZE,
-                  "Untangling is supported only for ideal targets.");
-
-      const DenseMatrix &Wideal =
-         Geometries.GetGeomToPerfGeomJac(fespace->GetFE(0)->GetGeomType());
-      tauval /= Wideal.Det();
-
-      // Slightly below minJ0 to avoid div by 0.
-      tauval -= 0.01 * h0.Min();
-   }
-
   const double init_energy = a.GetGridFunctionEnergy(x);

   // Visualize the starting mesh and metric values.
@@ -804,7 +730,7 @@ int main(int argc, char *argv[])

   // 14. As we use the Newton method to solve the resulting nonlinear system,
   //     here we setup the linear solver for the system's Jacobian.
-   Solver *S = NULL, *S_prec = NULL;
+   Solver *S = NULL;
   const double linsol_rtol = 1e-12;
   if (lin_solver == 0)
   {
@@ -825,16 +751,27 @@ int main(int argc, char *argv[])
      minres->SetMaxIter(max_lin_iter);
      minres->SetRelTol(linsol_rtol);
      minres->SetAbsTol(0.0);
-      if (verbosity_level > 2) { minres->SetPrintLevel(1); }
-      minres->SetPrintLevel(verbosity_level == 2 ? 3 : -1);
-      if (lin_solver == 3 || lin_solver == 4)
-      {
-         S_prec = new DSmoother((lin_solver == 3) ? 0 : 1, 1.0, 1);
-         minres->SetPreconditioner(*S_prec);
-      }
+      minres->SetPrintLevel(verbosity_level >= 2 ? 3 : -1);
      S = minres;
   }

+   // Compute the minimum det(J) of the starting mesh.
+   tauval = infinity();
+   const int NE = mesh->GetNE();
+   for (int i = 0; i < NE; i++)
+   {
+      const IntegrationRule &ir =
+         irules->Get(fespace->GetFE(i)->GetGeomType(), quad_order);
+      ElementTransformation *transf = mesh->GetElementTransformation(i);
+      for (int j = 0; j < ir.GetNPoints(); j++)
+      {
+         transf->SetIntPoint(&ir.IntPoint(j));
+         tauval = min(tauval, transf->Jacobian().Det());
+      }
+   }
+   cout << "Minimum det(J) of the original mesh is " << tauval << endl;
+   tauval -= 0.01 * h0.Min(); // Slightly below minJ0 to avoid div by 0.
+
   // Perform the nonlinear optimization.
   const IntegrationRule &ir =
      irules->Get(fespace->GetFE(0)->GetGeomType(), quad_order);
@@ -846,15 +783,9 @@ int main(int argc, char *argv[])
      // Specify linear solver when we use a Newton-based solver.
      solver.SetPreconditioner(*S);
   }
-   // For untangling, the solver will update the min det(T) values.
-   if (tauval < 0.0) { solver.SetMinDetPtr(&tauval); }
   solver.SetMaxIter(solver_iter);
   solver.SetRelTol(solver_rtol);
   solver.SetAbsTol(0.0);
-   if (solver_art_type > 0)
-   {
-      solver.SetAdaptiveLinRtol(solver_art_type, 0.5, 0.9);
-   }
   solver.SetPrintLevel(verbosity_level >= 1 ? 1 : -1);
   solver.SetOperator(a);
   solver.Mult(b, x.GetTrueVector());
@@ -922,7 +853,6 @@ int main(int argc, char *argv[])
   }

   // 19. Free the used memory.
-   delete S_prec;
   delete S;
   delete target_c2;
   delete metric2;
@@ -114,7 +114,7 @@ public:
   {
      Vector pos(3);
      T.Transform(ip, pos);
-      if (metric != 14 && metric != 36 && metric != 85)
+      if (metric != 14 && metric != 85)
      {
         const double xc = pos(0) - 0.5, yc = pos(1) - 0.5;
         const double r = sqrt(xc*xc + yc*yc);
@@ -129,7 +129,7 @@ public:
         K(1, 0) = 0.0;
         K(1, 1) = 1.0;
      }
-      else if (metric == 14 || metric == 36) // Size + Alignment
+      else if (metric == 14) // Size + Alignment
      {
         const double xc = pos(0), yc = pos(1);
         double theta = M_PI * yc * (1.0 - yc) * cos(2 * M_PI * xc);
@@ -33,25 +33,25 @@
 //
 // Sample runs:
 //   Adapted analytic shape:
-//     mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 2 -tid 4 -ni 200 -bnd -qt 1 -qo 8
+//     mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 2 -tid 4 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8
 //   Adapted analytic size+orientation:
-//     mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 14 -tid 4 -ni 200 -bnd -qt 1 -qo 8 -fd
+//     mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 14 -tid 4 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8 -fd
 //   Adapted analytic shape+orientation:
-//     mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 85 -tid 4 -ni 100 -bnd -qt 1 -qo 8 -fd
+//     mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 85 -tid 4 -ni 100 -ls 2 -li 100 -bnd -qt 1 -qo 8 -fd
 //
 //   Adapted discrete size:
-//     mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 80 -tid 5 -ni 50 -qo 4 -nor
+//     mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 7 -tid 5 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8
 //   Adapted discrete size; explicit combo of metrics; mixed tri/quad mesh:
 //     mpirun -np 4 pmesh-optimizer -m ../../data/square-mixed.mesh -o 2 -rs 2 -mid 2 -tid 5 -ni 200 -bnd -qo 6 -cmb 2 -nor
 //   Adapted discrete size+aspect_ratio:
 //     mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 7 -tid 6 -ni 100
 //     mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 7 -tid 6 -ni 100 -qo 6 -ex -st 1 -nor
 //   Adapted discrete size+orientation (requires GSLIB):
-//   * mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 36 -tid 8 -qo 4 -fd -ae 1 -nor
+//   * mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 14 -tid 8 -ni 100  -ls 2 -li 100 -bnd -qt 1 -qo 8 -fd -ae 1
 //   Adapted discrete aspect_ratio+orientation (requires GSLIB):
-//   * mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 85 -tid 8 -ni 10 -bnd -qt 1 -qo 8 -fd -ae 1
+//   * mpirun -np 4 pmesh-optimizer -m square01.mesh -o 2 -rs 2 -mid 85 -tid 8 -ni 10  -ls 2 -li 100 -bnd -qt 1 -qo 8 -fd -ae 1
 //   Adapted discrete aspect ratio (3D):
-//     mpirun -np 4 pmesh-optimizer -m cube.mesh -o 2 -rs 2 -mid 302 -tid 7 -ni 20 -bnd -qt 1 -qo 8
+//     mpirun -np 4 pmesh-optimizer -m cube.mesh -o 2 -rs 2 -mid 302 -tid 7 -ni 20  -ls 2 -li 100 -bnd -qt 1 -qo 8
 //
 //   Adaptive limiting:
 //     mpirun -np 4 pmesh-optimizer -m stretched2D.mesh -o 2 -mid 2 -tid 1 -ni 50 -qo 5 -nor -vl 1 -alc 0.5
@@ -61,33 +61,27 @@
 //   * mpirun -np 4 pmesh-optimizer -m stretched2D.mesh -o 2 -mid 2 -tid 1 -ni 50 -qo 5 -nor -vl 1 -alc 0.5 -fd -ae 1
 //
 //   Blade shape:
-//     mpirun -np 4 pmesh-optimizer -m blade.mesh -o 4 -mid 2 -tid 1 -ni 30 -ls 3 -art 1 -bnd -qt 1 -qo 8
+//     mpirun -np 4 pmesh-optimizer -m blade.mesh -o 4 -rs 0 -mid 2 -tid 1 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8
 //   Blade shape with FD-based solver:
-//     mpirun -np 4 pmesh-optimizer -m blade.mesh -o 4 -mid 2 -tid 1 -ni 30 -ls 4 -bnd -qt 1 -qo 8 -fd
+//     mpirun -np 4 pmesh-optimizer -m blade.mesh -o 4 -rs 0 -mid 2 -tid 1 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8 -fd
 //   Blade limited shape:
-//     mpirun -np 4 pmesh-optimizer -m blade.mesh -o 4 -mid 2 -tid 1 -bnd -qt 1 -qo 8 -lc 5000
+//     mpirun -np 4 pmesh-optimizer -m blade.mesh -o 4 -rs 0 -mid 2 -tid 1 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8 -lc 5000
 //   ICF shape and equal size:
-//     mpirun -np 4 pmesh-optimizer -o 3 -mid 9 -tid 2 -ni 25 -ls 3 -art 2 -qo 5
+//     mpirun -np 4 pmesh-optimizer -o 3 -rs 0 -mid 9 -tid 2 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8
 //   ICF shape and initial size:
-//     mpirun -np 4 pmesh-optimizer -o 3 -mid 9 -tid 3 -ni 30 -ls 3 -bnd -qt 1 -qo 8
+//     mpirun -np 4 pmesh-optimizer -o 3 -rs 0 -mid 9 -tid 3 -ni 100 -ls 2 -li 100 -bnd -qt 1 -qo 8
 //   ICF shape:
-//     mpirun -np 4 pmesh-optimizer -o 3 -mid 1 -tid 1 -ni 100 -bnd -qt 1 -qo 8
+//     mpirun -np 4 pmesh-optimizer -o 3 -rs 0 -mid 1 -tid 1 -ni 100 -ls 2 -li 100 -bnd -qt 1 -qo 8
 //   ICF limited shape:
-//     mpirun -np 4 pmesh-optimizer -o 3 -mid 1 -tid 1 -ni 100 -bnd -qt 1 -qo 8 -lc 10
+//     mpirun -np 4 pmesh-optimizer -o 3 -rs 0 -mid 1 -tid 1 -ni 100 -ls 2 -li 100 -bnd -qt 1 -qo 8 -lc 10
 //   ICF combo shape + size (rings, slow convergence):
-//     mpirun -np 4 pmesh-optimizer -o 3 -mid 1 -tid 1 -ni 1000 -bnd -qt 1 -qo 8 -cmb 1
+//     mpirun -np 4 pmesh-optimizer -o 3 -rs 0 -mid 1 -tid 1 -ni 1000 -ls 2 -li 100 -bnd -qt 1 -qo 8 -cmb 1
 //   Mixed tet / cube / hex mesh with limiting:
 //     mpirun -np 4 pmesh-optimizer -m ../../data/fichera-mixed-p2.mesh -o 4 -rs 1 -mid 301 -tid 1 -fix-bnd -qo 6 -nor -lc 0.25
 //   3D pinched sphere shape (the mesh is in the mfem/data GitHub repository):
-//   * mpirun -np 4 pmesh-optimizer -m ../../../mfem_data/ball-pert.mesh -o 4 -mid 303 -tid 1 -ni 20 -li 500 -fix-bnd
+//   * mpirun -np 4 pmesh-optimizer -m ../../../mfem_data/ball-pert.mesh -o 4 -rs 0 -mid 303 -tid 1 -ni 20 -ls 2 -li 500 -fix-bnd
 //   2D non-conforming shape and equal size:
-//     mpirun -np 4 pmesh-optimizer -m ./amr-quad-q2.mesh -o 2 -rs 1 -mid 9 -tid 2 -ni 200 -bnd -qt 1 -qo 8
-//
-//   2D untangling:
-//     mpirun -np 4 pmesh-optimizer -m jagged.mesh -o 2 -mid 22 -tid 1 -ni 50 -li 50 -qo 4 -fd -vl 1
-//   3D untangling (the mesh is in the mfem/data GitHub repository):
-//   * mpirun -np 4 pmesh-optimizer -m ../../../mfem_data/cube-holes-inv.mesh -o 3 -mid 313 -tid 1 -rtol 1e-5 -li 50 -qo 4 -fd -vl 1
-//
+//     mpirun -np 4 pmesh-optimizer -m ./amr-quad-q2.mesh -o 2 -rs 1 -mid 9 -tid 2 -ni 200 -ls 2 -li 100 -bnd -qt 1 -qo 8

 #include "mfem.hpp"
 #include "../common/mfem-common.hpp"
@@ -119,9 +113,8 @@ int main (int argc, char *argv[])
   int quad_type         = 1;
   int quad_order        = 8;
   int solver_type       = 0;
-   int solver_iter       = 20;
+   int solver_iter       = 10;
   double solver_rtol    = 1e-10;
-   int solver_art_type   = 0;
   int lin_solver        = 2;
   int max_lin_iter      = 100;
   bool move_bnd         = true;
@@ -147,38 +140,25 @@ int main (int argc, char *argv[])
                  "Random perturbation scaling factor.");
   args.AddOption(&metric_id, "-mid", "--metric-id",
                  "Mesh optimization metric:\n\t"
-                  "T-metrics\n\t"
                  "1  : |T|^2                          -- 2D shape\n\t"
                  "2  : 0.5|T|^2/tau-1                 -- 2D shape (condition number)\n\t"
                  "7  : |T-T^-t|^2                     -- 2D shape+size\n\t"
                  "9  : tau*|T-T^-t|^2                 -- 2D shape+size\n\t"
-                  "14 : |T-I|^2                        -- 2D shape+size+orientation\n\t"
                  "22 : 0.5(|T|^2-2*tau)/(tau-tau_0)   -- 2D untangling\n\t"
                  "50 : 0.5|T^tT|^2/tau^2-1            -- 2D shape\n\t"
                  "55 : (tau-1)^2                      -- 2D size\n\t"
                  "56 : 0.5(sqrt(tau)-1/sqrt(tau))^2   -- 2D size\n\t"
                  "58 : |T^tT|^2/(tau^2)-2*|T|^2/tau+2 -- 2D shape\n\t"
                  "77 : 0.5(tau-1/tau)^2               -- 2D size\n\t"
-                  "80 : (1-gamma)mu_2 + gamma mu_77    -- 2D shape+size\n\t"
-                  "85 : |T-|T|/sqrt(2)I|^2             -- 2D shape+orientation\n\t"
-                  "98 : (1/tau)|T-I|^2                 -- 2D shape+size+orientation\n\t"
-                  // "211: (tau-1)^2-tau+sqrt(tau^2+eps)  -- 2D untangling\n\t"
-                  // "252: 0.5(tau-1)^2/(tau-tau_0)       -- 2D untangling\n\t"
+                  "211: (tau-1)^2-tau+sqrt(tau^2)      -- 2D untangling\n\t"
+                  "252: 0.5(tau-1)^2/(tau-tau_0)       -- 2D untangling\n\t"
                  "301: (|T||T^-1|)/3-1              -- 3D shape\n\t"
                  "302: (|T|^2|T^-1|^2)/9-1          -- 3D shape\n\t"
                  "303: (|T|^2)/3*tau^(2/3)-1        -- 3D shape\n\t"
-                  // "311: (tau-1)^2-tau+sqrt(tau^2+eps)-- 3D untangling\n\t"
-                  "313: (|T|^2)(tau-tau0)^(-2/3)/3   -- 3D untangling\n\t"
                  "315: (tau-1)^2                    -- 3D size\n\t"
                  "316: 0.5(sqrt(tau)-1/sqrt(tau))^2 -- 3D size\n\t"
                  "321: |T-T^-t|^2                   -- 3D shape+size\n\t"
-                  // "352: 0.5(tau-1)^2/(tau-tau_0)     -- 3D untangling\n\t"
-                  "A-metrics\n\t"
-                  "11 : (1/4*alpha)|A-(adjA)^T(W^TW)/omega|^2 -- 2D shape\n\t"
-                  "36 : (1/alpha)|A-W|^2                      -- 2D shape+size+orientation\n\t"
-                  "107: (1/2*alpha)|A-|A|/|W|W|^2             -- 2D shape+orientation\n\t"
-                  "126: (1-gamma)nu_11 + gamma*nu_14a         -- 2D shape+size\n\t"
-                 );
+                  "352: 0.5(tau-1)^2/(tau-tau_0)     -- 3D untangling");
   args.AddOption(&target_id, "-tid", "--target-id",
                  "Target (ideal element) type:\n\t"
                  "1: Ideal shape, unit size\n\t"
@@ -202,25 +182,15 @@ 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"
-                  "1: CG\n\t"
-                  "2: MINRES\n\t"
-                  "3: MINRES + Jacobi preconditioner\n\t"
-                  "4: MINRES + l1-Jacobi preconditioner");
+                  "Linear solver: 0 - l1-Jacobi, 1 - CG, 2 - MINRES.");
   args.AddOption(&max_lin_iter, "-li", "--lin-iter",
                  "Maximum number of iterations in the linear solve.");
   args.AddOption(&move_bnd, "-bnd", "--move-boundary", "-fix-bnd",
                  "--fix-boundary",
                  "Enable motion along horizontal and vertical boundaries.");
   args.AddOption(&combomet, "-cmb", "--combo-type",
-                  "Combination of metrics options:\n\t"
+                  "Combination of metrics options:"
                  "0: Use single metric\n\t"
                  "1: Shape + space-dependent size given analytically\n\t"
                  "2: Shape + adapted size given discretely; shared target");
@@ -373,37 +343,27 @@ int main (int argc, char *argv[])
   TMOP_QualityMetric *metric = NULL;
   switch (metric_id)
   {
-      // T-metrics
      case 1: metric = new TMOP_Metric_001; break;
      case 2: metric = new TMOP_Metric_002; break;
      case 7: metric = new TMOP_Metric_007; break;
      case 9: metric = new TMOP_Metric_009; break;
-      case 14: metric = new TMOP_Metric_014; break;
+      case 14: metric = new TMOP_Metric_SSA2D; break;
      case 22: metric = new TMOP_Metric_022(tauval); break;
      case 50: metric = new TMOP_Metric_050; break;
      case 55: metric = new TMOP_Metric_055; break;
      case 56: metric = new TMOP_Metric_056; break;
      case 58: metric = new TMOP_Metric_058; break;
      case 77: metric = new TMOP_Metric_077; break;
-      case 80: metric = new TMOP_Metric_080(0.5); break;
      case 85: metric = new TMOP_Metric_085; break;
-      case 98: metric = new TMOP_Metric_098; break;
-      // case 211: metric = new TMOP_Metric_211; break;
-      // case 252: metric = new TMOP_Metric_252(tauval); break;
+      case 211: metric = new TMOP_Metric_211; break;
+      case 252: metric = new TMOP_Metric_252(tauval); break;
      case 301: metric = new TMOP_Metric_301; break;
      case 302: metric = new TMOP_Metric_302; break;
      case 303: metric = new TMOP_Metric_303; break;
-      // case 311: metric = new TMOP_Metric_311; break;
-      case 313: metric = new TMOP_Metric_313(tauval); break;
      case 315: metric = new TMOP_Metric_315; break;
      case 316: metric = new TMOP_Metric_316; break;
      case 321: metric = new TMOP_Metric_321; break;
-      // case 352: metric = new TMOP_Metric_352(tauval); break;
-      // A-metrics
-      case 11: metric = new TMOP_AMetric_011; break;
-      case 36: metric = new TMOP_AMetric_036; break;
-      case 107: metric = new TMOP_AMetric_107a; break;
-      case 126: metric = new TMOP_AMetric_126(0.9); break;
+      case 352: metric = new TMOP_Metric_352(tauval); break;
      default:
         if (myid == 0) { cout << "Unknown metric_id: " << metric_id << endl; }
         return 3;
@@ -415,7 +375,7 @@ int main (int argc, char *argv[])
   ParFiniteElementSpace ind_fes(pmesh, &ind_fec);
   ParFiniteElementSpace ind_fesv(pmesh, &ind_fec, dim);
   ParGridFunction size(&ind_fes), aspr(&ind_fes), disc(&ind_fes), ori(&ind_fes);
-   ParGridFunction aspr3d(&ind_fesv);
+   ParGridFunction aspr3d(&ind_fesv), size3d(&ind_fesv);

   switch (target_id)
   {
@@ -453,7 +413,7 @@ int main (int argc, char *argv[])
         target_c = tc;
         break;
      }
-      case 6: // material indicator 2D
+      case 6: //material indicator 2D
      {
         ParGridFunction d_x(&ind_fes), d_y(&ind_fes);

@@ -473,14 +433,14 @@ int main (int argc, char *argv[])
            MFEM_ABORT("MFEM is not built with GSLIB.");
 #endif
         }
-         // Diffuse the interface
+         //Diffuse the interface
         DiffuseField(disc,2);

-         // Get  partials with respect to x and y of the grid function
+         //Get  partials with respect to x and y of the grid function
         disc.GetDerivative(1,0,d_x);
         disc.GetDerivative(1,1,d_y);

-         // Compute the squared magnitude of the gradient
+         //Compute the squared magnitude of the gradient
         for (int i = 0; i < size.Size(); i++)
         {
            size(i) = std::pow(d_x(i),2)+std::pow(d_y(i),2);
@@ -593,7 +553,7 @@ int main (int argc, char *argv[])
 #endif
         }

-         if (metric_id == 14 || metric_id == 36)
+         if (metric_id == 14)
         {
            ConstantCoefficient ind_coeff(0.1*0.1);
            size.ProjectCoefficient(ind_coeff);
@@ -743,46 +703,6 @@ int main (int argc, char *argv[])
   }
   else { a.AddDomainIntegrator(he_nlf_integ); }

-   // Compute the minimum det(J) of the starting mesh.
-   tauval = infinity();
-   const int NE = pmesh->GetNE();
-   for (int i = 0; i < NE; i++)
-   {
-      const IntegrationRule &ir =
-         irules->Get(pfespace->GetFE(i)->GetGeomType(), quad_order);
-      ElementTransformation *transf = pmesh->GetElementTransformation(i);
-      for (int j = 0; j < ir.GetNPoints(); j++)
-      {
-         transf->SetIntPoint(&ir.IntPoint(j));
-         tauval = min(tauval, transf->Jacobian().Det());
-      }
-   }
-   double minJ0;
-   MPI_Allreduce(&tauval, &minJ0, 1, MPI_DOUBLE, MPI_MIN, MPI_COMM_WORLD);
-   tauval = minJ0;
-   if (myid == 0)
-   { cout << "Minimum det(J) of the original mesh is " << tauval << endl; }
-
-   if (tauval < 0.0 && metric_id != 22 && metric_id != 211 && metric_id != 252
-       && metric_id != 311 && metric_id != 313 && metric_id != 352)
-   {
-      MFEM_ABORT("The input mesh is inverted! Try an untangling metric.");
-   }
-   if (tauval < 0.0)
-   {
-      MFEM_VERIFY(target_t == TargetConstructor::IDEAL_SHAPE_UNIT_SIZE,
-                  "Untangling is supported only for ideal targets.");
-
-      const DenseMatrix &Wideal =
-         Geometries.GetGeomToPerfGeomJac(pfespace->GetFE(0)->GetGeomType());
-      tauval /= Wideal.Det();
-
-      double h0min = h0.Min(), h0min_all;
-      MPI_Allreduce(&h0min, &h0min_all, 1, MPI_DOUBLE, MPI_MIN, MPI_COMM_WORLD);
-      // Slightly below minJ0 to avoid div by 0.
-      tauval -= 0.01 * h0min_all;
-   }
-
   const double init_energy = a.GetParGridFunctionEnergy(x);

   // Visualize the starting mesh and metric values.
@@ -851,7 +771,7 @@ int main (int argc, char *argv[])

   // 15. As we use the Newton method to solve the resulting nonlinear system,
   //     here we setup the linear solver for the system's Jacobian.
-   Solver *S = NULL, *S_prec = NULL;
+   Solver *S = NULL;
   const double linsol_rtol = 1e-12;
   if (lin_solver == 0)
   {
@@ -872,19 +792,33 @@ int main (int argc, char *argv[])
      minres->SetMaxIter(max_lin_iter);
      minres->SetRelTol(linsol_rtol);
      minres->SetAbsTol(0.0);
-      if (verbosity_level > 2) { minres->SetPrintLevel(1); }
-      else { minres->SetPrintLevel(verbosity_level == 2 ? 3 : -1); }
-      if (lin_solver == 3 || lin_solver == 4)
-      {
-         HypreSmoother *hs = new HypreSmoother;
-         hs->SetType((lin_solver == 3) ? HypreSmoother::Jacobi
-                     : HypreSmoother::l1Jacobi, 1);
-         S_prec = hs;
-         minres->SetPreconditioner(*S_prec);
-      }
+      minres->SetPrintLevel(verbosity_level >= 2 ? 3 : -1);
      S = minres;
   }

+   // Compute the minimum det(J) of the starting mesh.
+   tauval = infinity();
+   const int NE = pmesh->GetNE();
+   for (int i = 0; i < NE; i++)
+   {
+      const IntegrationRule &ir =
+         irules->Get(pfespace->GetFE(i)->GetGeomType(), quad_order);
+      ElementTransformation *transf = pmesh->GetElementTransformation(i);
+      for (int j = 0; j < ir.GetNPoints(); j++)
+      {
+         transf->SetIntPoint(&ir.IntPoint(j));
+         tauval = min(tauval, transf->Jacobian().Det());
+      }
+   }
+   double minJ0;
+   MPI_Allreduce(&tauval, &minJ0, 1, MPI_DOUBLE, MPI_MIN, MPI_COMM_WORLD);
+   tauval = minJ0;
+   if (myid == 0)
+   { cout << "Minimum det(J) of the original mesh is " << tauval << endl; }
+   double h0min = h0.Min(), h0min_all;
+   MPI_Allreduce(&h0min, &h0min_all, 1, MPI_DOUBLE, MPI_MIN, MPI_COMM_WORLD);
+   tauval -= 0.01 * h0min_all; // Slightly below minJ0 to avoid div by 0.
+
   // Perform the nonlinear optimization.
   const IntegrationRule &ir =
      irules->Get(pfespace->GetFE(0)->GetGeomType(), quad_order);
@@ -896,15 +830,9 @@ int main (int argc, char *argv[])
      // Specify linear solver when we use a Newton-based solver.
      solver.SetPreconditioner(*S);
   }
-   // For untangling, the solver will update the min det(T) values.
-   if (tauval < 0.0) { solver.SetMinDetPtr(&tauval); }
   solver.SetMaxIter(solver_iter);
   solver.SetRelTol(solver_rtol);
   solver.SetAbsTol(0.0);
-   if (solver_art_type > 0)
-   {
-      solver.SetAdaptiveLinRtol(solver_art_type, 0.5, 0.9);
-   }
   solver.SetPrintLevel(verbosity_level >= 1 ? 1 : -1);
   solver.SetOperator(a);
   solver.Mult(b, x.GetTrueVector());
@@ -983,7 +911,6 @@ int main (int argc, char *argv[])
   }

   // 20. Free the used memory.
-   delete S_prec;
   delete S;
   delete target_c2;
   delete metric2;
@@ -34,11 +34,6 @@ if (MFEM_USE_MPI)
    ${NAVIER_COMMON_FILES}
    LIBRARIES mfem)

-  add_mfem_miniapp(navier_kovasznay_vs
-    MAIN navier_kovasznay_vs.cpp
-    ${NAVIER_COMMON_FILES}
-    LIBRARIES mfem)
-
  add_mfem_miniapp(navier_tgv
    MAIN navier_tgv.cpp
    ${NAVIER_COMMON_FILES}
@@ -24,7 +24,7 @@ MFEM_LIB_FILE = mfem_is_not_built
 NAVIER_COMMON_SRC = navier_solver.cpp ortho_solver.cpp
 NAVIER_COMMON_OBJ = $(NAVIER_COMMON_SRC:.cpp=.o)

-PAR_MINIAPPS = navier_mms navier_kovasznay navier_kovasznay_vs navier_tgv navier_shear navier_3dfoc
+PAR_MINIAPPS = navier_mms navier_kovasznay navier_tgv navier_shear navier_3dfoc

 ifeq ($(MFEM_USE_MPI),NO)
   MINIAPPS =
@@ -1,305 +0,0 @@
-// Copyright (c) 2010-2020, 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.
-//
-// Navier Kovasznay example with variable time step
-//
-// Solve for the steady Kovasznay flow at Re = 40 defined by
-//
-// u = [1 - exp(L * x) * cos(2 * pi * y),
-//      L / (2 * pi) * exp(L * x) * sin(2 * pi * y)],
-//
-// p = 1/2 * (1 - exp(2 * L * x)),
-//
-// with L = Re/2 - sqrt(Re^2/4 + 4 * pi^2).
-//
-// The problem domain is set up like this
-//
-//            +-------------+
-//            |             |
-//            |             |
-//            |             |
-//            |             |
-//  Inflow -> |             | -> Outflow
-//            |             |
-//            |             |
-//            |             |
-//            |             |
-//            |             |
-//            +-------------+
-//
-// and Dirichlet boundary conditions are applied for the velocity on every
-// boundary. The problem, although steady state, is time integrated up to the
-// final time and the solution is compared with the known exact solution.
-//
-// Additionally, this example shows the usage of variable time steps with
-// Navier. A basic sample algorithm for determining the next time step is
-// provided based on a CFL restriction on the velocity.
-
-#include "navier_solver.hpp"
-#include <fstream>
-
-using namespace mfem;
-using namespace navier;
-
-struct s_NavierContext
-{
-   int ser_ref_levels = 1;
-   int order = 6;
-   double kinvis = 1.0 / 40.0;
-   double t_final = 10 * 0.001;
-   double dt = 0.001;
-   double reference_pressure = 0.0;
-   double reynolds = 1.0 / kinvis;
-   double lam = 0.5 * reynolds
-                - sqrt(0.25 * reynolds * reynolds + 4.0 * M_PI * M_PI);
-   bool pa = true;
-   bool ni = false;
-   bool visualization = false;
-   bool checkres = false;
-} ctx;
-
-void vel_kovasznay(const Vector &x, double t, Vector &u)
-{
-   double xi = x(0);
-   double yi = x(1);
-
-   u(0) = 1.0 - exp(ctx.lam * xi) * cos(2.0 * M_PI * yi);
-   u(1) = ctx.lam / (2.0 * M_PI) * exp(ctx.lam * xi) * sin(2.0 * M_PI * yi);
-}
-
-double pres_kovasznay(const Vector &x, double t)
-{
-   double xi = x(0);
-
-   return 0.5 * (1.0 - exp(2.0 * ctx.lam * xi)) + ctx.reference_pressure;
-}
-
-int main(int argc, char *argv[])
-{
-   MPI_Session mpi(argc, argv);
-
-   OptionsParser args(argc, argv);
-   args.AddOption(&ctx.ser_ref_levels,
-                  "-rs",
-                  "--refine-serial",
-                  "Number of times to refine the mesh uniformly in serial.");
-   args.AddOption(&ctx.order,
-                  "-o",
-                  "--order",
-                  "Order (degree) of the finite elements.");
-   args.AddOption(&ctx.dt, "-dt", "--time-step", "Time step.");
-   args.AddOption(&ctx.t_final, "-tf", "--final-time", "Final time.");
-   args.AddOption(&ctx.pa,
-                  "-pa",
-                  "--enable-pa",
-                  "-no-pa",
-                  "--disable-pa",
-                  "Enable partial assembly.");
-   args.AddOption(&ctx.ni,
-                  "-ni",
-                  "--enable-ni",
-                  "-no-ni",
-                  "--disable-ni",
-                  "Enable numerical integration rules.");
-   args.AddOption(&ctx.visualization,
-                  "-vis",
-                  "--visualization",
-                  "-no-vis",
-                  "--no-visualization",
-                  "Enable or disable GLVis visualization.");
-   args.AddOption(
-      &ctx.checkres,
-      "-cr",
-      "--checkresult",
-      "-no-cr",
-      "--no-checkresult",
-      "Enable or disable checking of the result. Returns -1 on failure.");
-   args.Parse();
-   if (!args.Good())
-   {
-      if (mpi.Root())
-      {
-         args.PrintUsage(mfem::out);
-      }
-      return 1;
-   }
-   if (mpi.Root())
-   {
-      args.PrintOptions(mfem::out);
-   }
-
-   Mesh *mesh = new Mesh(2, 4, Element::QUADRILATERAL, false, 1.5, 2.0);
-
-   mesh->EnsureNodes();
-   GridFunction *nodes = mesh->GetNodes();
-   *nodes -= 0.5;
-
-   for (int i = 0; i < ctx.ser_ref_levels; ++i)
-   {
-      mesh->UniformRefinement();
-   }
-
-   if (mpi.Root())
-   {
-      std::cout << "Number of elements: " << mesh->GetNE() << std::endl;
-   }
-
-   auto *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
-   delete mesh;
-
-   // Create the flow solver.
-   NavierSolver flowsolver(pmesh, ctx.order, ctx.kinvis);
-   flowsolver.EnablePA(ctx.pa);
-   flowsolver.EnableNI(ctx.ni);
-
-   // Set the initial condition.
-   ParGridFunction *u_ic = flowsolver.GetCurrentVelocity();
-   VectorFunctionCoefficient u_excoeff(pmesh->Dimension(), vel_kovasznay);
-   u_ic->ProjectCoefficient(u_excoeff);
-
-   FunctionCoefficient p_excoeff(pres_kovasznay);
-
-   // Add Dirichlet boundary conditions to velocity space restricted to
-   // selected attributes on the mesh.
-   Array<int> attr(pmesh->bdr_attributes.Max());
-   attr = 1;
-   flowsolver.AddVelDirichletBC(vel_kovasznay, attr);
-
-   double t = 0.0;
-   double dt = ctx.dt;
-   double t_final = ctx.t_final;
-   bool last_step = false;
-
-   flowsolver.Setup(dt);
-
-   double err_u = 0.0;
-   double err_p = 0.0;
-   ParGridFunction *u_next_gf = nullptr;
-   ParGridFunction *u_gf = nullptr;
-   ParGridFunction *p_gf = nullptr;
-
-   ParGridFunction p_ex_gf(flowsolver.GetCurrentPressure()->ParFESpace());
-   GridFunctionCoefficient p_ex_gf_coeff(&p_ex_gf);
-
-   double cfl_max = 0.8;
-   double cfl_tol = 1e-4;
-
-   for (int step = 0; !last_step; ++step)
-   {
-      if (t + dt >= t_final - dt / 2)
-      {
-         last_step = true;
-      }
-
-      // Take a provisional step
-      flowsolver.Step(t, dt, step, true);
-
-      // Retrieve the computed provisional velocity
-      u_next_gf = flowsolver.GetProvisionalVelocity();
-
-      // Compute the CFL based on the provisional velocity
-      double cfl = flowsolver.ComputeCFL(*u_next_gf, dt);
-
-      double error_est = cfl / (cfl_max + cfl_tol);
-      if (error_est >= 1.0)
-      {
-         // Reject the time step
-         if (mpi.Root())
-         {
-            std::cout
-                  << "Step reached maximum CFL, retrying with smaller step size..."
-                  << std::endl;
-         }
-         dt *= 0.5;
-         step -= 1;
-      }
-      else
-      {
-         // Accept the time step
-         t += dt;
-
-         // Predict new step size
-         double fac_safety = 2.0;
-         double eta = pow(1.0 / (fac_safety * error_est), 1.0 / (1.0 + 3.0));
-         double fac_min = 0.1;
-         double fac_max = 1.4;
-         dt = dt * std::min(fac_max, std::max(fac_min, eta));
-
-         // Queue new time step in the history array
-         flowsolver.UpdateTimestepHistory(dt);
-      }
-
-      u_gf = flowsolver.GetCurrentVelocity();
-      p_gf = flowsolver.GetCurrentPressure();
-
-      u_excoeff.SetTime(t);
-      p_excoeff.SetTime(t);
-
-      // Remove mean value from exact pressure solution.
-      p_ex_gf.ProjectCoefficient(p_excoeff);
-      flowsolver.MeanZero(p_ex_gf);
-
-      err_u = u_gf->ComputeL2Error(u_excoeff);
-      err_p = p_gf->ComputeL2Error(p_ex_gf_coeff);
-
-      if (mpi.Root())
-      {
-         printf("%5s %8s %8s %8s %11s %11s\n",
-                "Order",
-                "CFL",
-                "Time",
-                "dt",
-                "err_u",
-                "err_p");
-         printf("%5.2d %8.2E %.2E %.2E %.5E %.5E err\n",
-                ctx.order,
-                cfl,
-                t,
-                dt,
-                err_u,
-                err_p);
-         fflush(stdout);
-      }
-   }
-
-   if (ctx.visualization)
-   {
-      char vishost[] = "localhost";
-      int visport = 19916;
-      socketstream sol_sock(vishost, visport);
-      sol_sock.precision(8);
-      sol_sock << "parallel " << mpi.WorldSize() << " " << mpi.WorldRank()
-               << "\n";
-      sol_sock << "solution\n" << *pmesh << *u_ic << std::flush;
-   }
-
-   flowsolver.PrintTimingData();
-
-   // Test if the result for the test run is as expected.
-   if (ctx.checkres)
-   {
-      double tol_u = 1e-6;
-      double tol_p = 1e-5;
-      if (err_u > tol_u || err_p > tol_p)
-      {
-         if (mpi.Root())
-         {
-            mfem::out << "Result has a larger error than expected."
-                      << std::endl;
-         }
-         return -1;
-      }
-   }
-
-   delete pmesh;
-
-   return 0;
-}
@@ -49,8 +49,6 @@ NavierSolver::NavierSolver(ParMesh *mesh, int order, double kin_vis)

   un.SetSize(vfes_truevsize);
   un = 0.0;
-   un_next.SetSize(vfes_truevsize);
-   un_next = 0.0;
   unm1.SetSize(vfes_truevsize);
   unm1 = 0.0;
   unm2.SetSize(vfes_truevsize);
@@ -78,8 +76,6 @@ NavierSolver::NavierSolver(ParMesh *mesh, int order, double kin_vis)

   un_gf.SetSpace(vfes);
   un_gf = 0.0;
-   un_next_gf.SetSpace(vfes);
-   un_next_gf = 0.0;

   Lext_gf.SetSpace(vfes);
   curlu_gf.SetSpace(vfes);
@@ -292,14 +288,7 @@ void NavierSolver::Setup(double dt)
   HInv->SetRelTol(rtol_hsolve);
   HInv->SetMaxIter(200);

-   // If the initial condition was set, it has to be aligned with dependent
-   // Vectors and GridFunctions
   un_gf.GetTrueDofs(un);
-   un_next = un;
-   un_next_gf.SetFromTrueDofs(un_next);
-
-   // Set initial time step in the history array
-   dthist[0] = dt;

   if (filter_alpha != 0.0)
   {
@@ -319,64 +308,52 @@ void NavierSolver::Setup(double dt)
   sw_setup.Stop();
 }

-void NavierSolver::UpdateTimestepHistory(double dt)
-{
-   // Rotate values in time step history
-   dthist[2] = dthist[1];
-   dthist[1] = dthist[0];
-   dthist[0] = dt;
-
-   // Rotate values in nonlinear extrapolation history
-   Nunm2 = Nunm1;
-   Nunm1 = Nun;
-
-   // Rotate values in solution history
-   unm2 = unm1;
-   unm1 = un;
-
-   // Update the current solution and corresponding GridFunction
-   un_next_gf.GetTrueDofs(un_next);
-   un = un_next;
-   un_gf.SetFromTrueDofs(un);
-}
-
-void NavierSolver::Step(double &time, double dt, int cur_step, bool provisional)
+void NavierSolver::Step(double &time, double dt, int cur_step)
 {
   sw_step.Start();

-   SetTimeIntegrationCoefficients(cur_step);
+   time += dt;

-   // Set current time for velocity Dirichlet boundary conditions.
+   // Set current time for velocity dirichlet boundary conditions.
   for (auto &vel_dbc : vel_dbcs)
   {
-      vel_dbc.coeff->SetTime(time + dt);
+      vel_dbc.coeff->SetTime(time);
   }

-   // Set current time for pressure Dirichlet boundary conditions.
+   // Set current time for pressure dirichlet boundary conditions.
   for (auto &pres_dbc : pres_dbcs)
   {
-      pres_dbc.coeff->SetTime(time + dt);
+      pres_dbc.coeff->SetTime(time);
   }

-   H_bdfcoeff.constant = bd0 / dt;
-   H_form->Update();
-   H_form->Assemble();
-   H_form->FormSystemMatrix(vel_ess_tdof, H);
+   SetTimeIntegrationCoefficients(cur_step);

-   HInv->SetOperator(*H);
-   if (partial_assembly)
+   if (cur_step <= 2)
   {
-      delete HInvPC;
-      Vector diag_pa(vfes->GetTrueVSize());
-      H_form->AssembleDiagonal(diag_pa);
-      HInvPC = new OperatorJacobiSmoother(diag_pa, vel_ess_tdof);
-      HInv->SetPreconditioner(*HInvPC);
+      H_bdfcoeff.constant = bd0 / dt;
+      H_form->Update();
+      H_form->Assemble();
+      H_form->FormSystemMatrix(vel_ess_tdof, H);
+
+      if (partial_assembly)
+      {
+         HInv->SetOperator(*H);
+         delete HInvPC;
+         Vector diag_pa(vfes->GetTrueVSize());
+         H_form->AssembleDiagonal(diag_pa);
+         HInvPC = new OperatorJacobiSmoother(diag_pa, vel_ess_tdof);
+         HInv->SetPreconditioner(*HInvPC);
+      }
+      else
+      {
+         HInv->SetOperator(*H);
+      }
   }

   // Extrapolated f^{n+1}.
   for (auto &accel_term : accel_terms)
   {
-      accel_term.coeff->SetTime(time + dt);
+      accel_term.coeff->SetTime(time);
   }

   f_form->Assemble();
@@ -402,6 +379,10 @@ void NavierSolver::Step(double &time, double dt, int cur_step, bool provisional)
                              ab3_ * d_Nunm2[i];);
   }

+   // Rotate the solutions from previous time steps.
+   Nunm2 = Nunm1;
+   Nunm1 = Nun;
+
   // Fext = M^{-1} (F(u^{n}) + f^{n+1})
   MvInv->Mult(Fext, tmp1);
   iter_mvsolve = MvInv->GetNumIterations();
@@ -522,41 +503,35 @@ void NavierSolver::Step(double &time, double dt, int cur_step, bool provisional)
   Mv->Mult(Fext, tmp1);
   resu.Add(1.0, tmp1);

-   // un_next_gf = un_gf;
-
   for (auto &vel_dbc : vel_dbcs)
   {
-      un_next_gf.ProjectBdrCoefficient(*vel_dbc.coeff, vel_dbc.attr);
+      un_gf.ProjectBdrCoefficient(*vel_dbc.coeff, vel_dbc.attr);
   }

   vfes->GetRestrictionMatrix()->MultTranspose(resu, resu_gf);

+   // Rotate solutions from previous time steps.
+   unm2 = unm1;
+   unm1 = un;
+
   Vector X2, B2;
   if (partial_assembly)
   {
      auto *HC = H.As<ConstrainedOperator>();
-      EliminateRHS(*H_form, *HC, vel_ess_tdof, un_next_gf, resu_gf, X2, B2, 1);
+      EliminateRHS(*H_form, *HC, vel_ess_tdof, un_gf, resu_gf, X2, B2, 1);
   }
   else
   {
-      H_form->FormLinearSystem(vel_ess_tdof, un_next_gf, resu_gf, H, X2, B2, 1);
+      H_form->FormLinearSystem(vel_ess_tdof, un_gf, resu_gf, H, X2, B2, 1);
   }
   sw_hsolve.Start();
   HInv->Mult(B2, X2);
   sw_hsolve.Stop();
   iter_hsolve = HInv->GetNumIterations();
   res_hsolve = HInv->GetFinalNorm();
-   H_form->RecoverFEMSolution(X2, resu_gf, un_next_gf);
+   H_form->RecoverFEMSolution(X2, resu_gf, un_gf);

-   un_next_gf.GetTrueDofs(un_next);
-
-   // If the current time step is not provisional, accept the computed solution
-   // and update the time step history by default.
-   if (!provisional)
-   {
-      UpdateTimestepHistory(dt);
-      time += dt;
-   }
+   un_gf.GetTrueDofs(un);

   if (filter_alpha != 0.0)
   {
@@ -1017,24 +992,7 @@ void NavierSolver::AddAccelTerm(VecFuncT *f, Array<int> &attr)

 void NavierSolver::SetTimeIntegrationCoefficients(int step)
 {
-   // Maxmium BDF order to use at current time step
-   // step + 1 <= order <= max_bdf_order
-   int bdf_order = std::min(step + 1, max_bdf_order);
-
-   // Ratio of time step history at dt(t_{n}) - dt(t_{n-1})
-   double rho1 = 0.0;
-
-   // Ratio of time step history at dt(t_{n-1}) - dt(t_{n-2})
-   double rho2 = 0.0;
-
-   rho1 = dthist[0] / dthist[1];
-
-   if (bdf_order == 3)
-   {
-      rho2 = dthist[1] / dthist[2];
-   }
-
-   if (step == 0 && bdf_order == 1)
+   if (step == 0)
   {
      bd0 = 1.0;
      bd1 = -1.0;
@@ -1044,27 +1002,25 @@ void NavierSolver::SetTimeIntegrationCoefficients(int step)
      ab2 = 0.0;
      ab3 = 0.0;
   }
-   else if (step >= 1 && bdf_order == 2)
+   else if (step == 1)
   {
-      bd0 = (1.0 + 2.0 * rho1) / (1.0 + rho1);
-      bd1 = -(1.0 + rho1);
-      bd2 = pow(rho1, 2.0) / (1.0 + rho1);
+      bd0 = 3.0 / 2.0;
+      bd1 = -4.0 / 2.0;
+      bd2 = 1.0 / 2.0;
      bd3 = 0.0;
-      ab1 = 1.0 + rho1;
-      ab2 = -rho1;
+      ab1 = 2.0;
+      ab2 = -1.0;
      ab3 = 0.0;
   }
-   else if (step >= 2 && bdf_order == 3)
+   else if (step == 2)
   {
-      bd0 = 1.0 + rho1 / (1.0 + rho1)
-            + (rho2 * rho1) / (1.0 + rho2 * (1 + rho1));
-      bd1 = -1.0 - rho1 - (rho2 * rho1 * (1.0 + rho1)) / (1.0 + rho2);
-      bd2 = pow(rho1, 2.0) * (rho2 + 1.0 / (1.0 + rho1));
-      bd3 = -(pow(rho2, 3.0) * pow(rho1, 2.0) * (1.0 + rho1))
-            / ((1.0 + rho2) * (1.0 + rho2 + rho2 * rho1));
-      ab1 = ((1.0 + rho1) * (1.0 + rho2 * (1.0 + rho1))) / (1.0 + rho2);
-      ab2 = -rho1 * (1.0 + rho2 * (1.0 + rho1));
-      ab3 = (pow(rho2, 2.0) * rho1 * (1.0 + rho1)) / (1.0 + rho2);
+      bd0 = 11.0 / 6.0;
+      bd1 = -18.0 / 6.0;
+      bd2 = 9.0 / 6.0;
+      bd3 = -2.0 / 6.0;
+      ab1 = 3.0;
+      ab2 = -3.0;
+      ab3 = 1.0;
   }
 }

@@ -100,7 +100,7 @@ public:
 /**
 * This implementation of a transient incompressible Navier Stokes solver uses
 * the non-dimensionalized formulation. The coupled momentum and
- * incompressibility equations are decoupled using the split scheme described in
+ * incompressibilty equations are decoupled using the split scheme described in
 * [1]. This leads to three solving steps.
 *
 * 1. An extrapolation step for all nonlinear terms which are treated
@@ -160,31 +160,7 @@ public:
   void Setup(double dt);

   /// Compute solution at the next time step t+dt.
-   /**
-    * This method can be called with the default value @a provisional which
-    * always accepts the computed time step by automatically calling
-    * UpdateTimestepHistory.
-    *
-    * If @a provisional is set to true, the solution at t+dt is not accepted
-    * automatically and the application code has to call UpdateTimestepHistory
-    * and update the @a time variable accordingly.
-    *
-    * The application code can check the provisional step by retrieving the
-    * GridFunction with the method GetProvisionalVelocity. If the check fails,
-    * it is possible to retry the step with a different time step by not
-    * calling UpdateTimestepHistory and calling this method with the previous
-    * @a time and @a cur_step.
-    *
-    * The method and parameter choices are based on [1].
-    *
-    * [1] D. Wang, S.J. Ruuth (2008) Variable step-size implicit-explicit
-    * linear multistep methods for time-dependent partial differential
-    * equations
-    */
-   void Step(double &time, double dt, int cur_step, bool provisional = false);
-
-   /// Return a pointer to the provisional velocity ParGridFunction.
-   ParGridFunction *GetProvisionalVelocity() { return &un_next_gf; }
+   void Step(double &time, double dt, int cur_step);

   /// Return a pointer to the current velocity ParGridFunction.
   ParGridFunction *GetCurrentVelocity() { return &un_gf; }
@@ -262,12 +238,6 @@ public:
    */
   void MeanZero(ParGridFunction &v);

-   /// Rotate entries in the time step and solution history arrays.
-   void UpdateTimestepHistory(double dt);
-
-   /// Set the maximum order to use for the BDF method.
-   void SetMaxBDFOrder(int maxbdforder) { max_bdf_order = maxbdforder; };
-
   /// Compute CFL
   double ComputeCFL(ParGridFunction &u, double dt);

@@ -285,7 +255,7 @@ public:
   void SetFilterAlpha(double a) { filter_alpha = a; }

 protected:
-   /// Print information about the Navier version.
+   /// Print informations about the Navier version.
   void PrintInfo();

   /// Update the EXTk/BDF time integration coefficient.
@@ -293,8 +263,7 @@ protected:
    * Depending on which time step the computation is in, the EXTk/BDF time
    * integration coefficients have to be set accordingly. This allows
    * bootstrapping with a BDF scheme of order 1 and increasing the order each
-    * following time step, up to order 3 (or whichever order is set in
-    * SetMaxBDFOrder).
+    * following time step, up to order 3.
    */
   void SetTimeIntegrationCoefficients(int step);

@@ -387,14 +356,12 @@ protected:
   Solver *HInvPC = nullptr;
   CGSolver *HInv = nullptr;

-   Vector fn, un, un_next, unm1, unm2, Nun, Nunm1, Nunm2, Fext, FText, Lext,
-          resu;
+   Vector fn, un, unm1, unm2, Nun, Nunm1, Nunm2, Fext, FText, Lext, resu;
   Vector tmp1;

   Vector pn, resp, FText_bdr, g_bdr;

-   ParGridFunction un_gf, un_next_gf, curlu_gf, curlcurlu_gf, Lext_gf, FText_gf,
-                   resu_gf;
+   ParGridFunction un_gf, curlu_gf, curlcurlu_gf, Lext_gf, FText_gf, resu_gf;

   ParGridFunction pn_gf, resp_gf;

@@ -415,9 +382,7 @@ protected:
   // Bookkeeping for acceleration (forcing) terms.
   std::vector<AccelTerm_T> accel_terms;

-   int max_bdf_order = 3;
   int cur_step = 0;
-   std::vector<double> dthist = {0.0, 0.0, 0.0};

   // BDFk/EXTk coefficients.
   double bd0 = 0.0;
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Tucker Babcock	cf8b8bc74e	update removeinternalboundariesnotadjacentto	2020-11-21 17:29:02 -05:00
Tucker Babcock	c0fbd3a2e3	added method to remove internal boundaries except for those adjacent to specified regions	2020-11-19 14:22:00 -05:00
Tucker Babcock	31a7a4a71f	added new method to remove mesh internal boundaries except for those in a array of boundary attributes to keep	2020-11-19 13:37:58 -05:00