fixed BC issue

fix sign error
fixed invertibility bug
2024-04-14 14:59:23 -04:00 · 2024-04-12 08:12:42 -04:00 · 2024-04-11 22:28:15 -04:00 · 2024-04-11 18:56:53 -04:00 · 2024-04-11 18:05:56 -04:00 · 2024-04-11 17:37:39 -04:00
505 changed files with 12288 additions and 34379 deletions
@@ -0,0 +1,61 @@
+# Configuration for probot-stale - https://github.com/probot/stale
+
+# Number of days of inactivity before an Issue or Pull Request becomes stale
+daysUntilStale: 30
+
+# Number of days of inactivity before an Issue or Pull Request with the stale
+# label is closed.  Set to false to disable. If disabled, issues still need to
+# be closed manually, but will remain marked as stale.
+daysUntilClose: 7
+
+# Only issues or pull requests with all of these labels are check if stale.
+# Defaults to `[]` (disabled)
+onlyLabels: []
+
+# Issues or Pull Requests with these labels will never be considered stale. Set
+# to `[]` to disable
+exemptLabels:
+  - bug
+  - WIP
+  - ready-for-review
+  - in-review
+  - in-next
+
+# Set to true to ignore issues in a project (defaults to false)
+exemptProjects: false
+
+# Set to true to ignore issues in a milestone (defaults to false)
+exemptMilestones: false
+
+# Set to true to ignore issues with an assignee (defaults to false)
+exemptAssignees: false
+
+# Label to use when marking an issue as stale
+staleLabel: stale
+
+# Comment to post when marking an issue as stale. Set to `false` to disable
+markComment: >
+  :warning: This issue or PR has been automatically marked as stale because it has not
+  had any activity in the last month. *If no activity occurs in the next week, it will
+  be automatically closed.* Thank you for your contributions.
+
+# Comment to post when closing a stale issue. Set to `false` to disable
+closeComment: false
+
+# Limit the number of actions per hour, from 1-30. Default is 30
+limitPerRun: 30
+
+# Limit to only `issues` or `pulls`
+# only: issues
+
+# Optionally, specify configuration settings that are specific to just 'issues' or 'pulls':
+# pulls:
+#   daysUntilStale: 30
+#   markComment: >
+#     This pull request has been automatically marked as stale because it has not had
+#     recent activity. It will be closed if no further activity occurs. Thank you
+#     for your contributions.
+
+# issues:
+#   exemptLabels:
+#     - confirmed
@@ -33,7 +33,6 @@ env:
  HYPRE_ARCHIVE: v2.19.0.tar.gz
  HYPRE_TOP_DIR: hypre-2.19.0
  METIS_ARCHIVE: metis-4.0.3.tar.gz
-  METIS_ARCHIVE_MAC: metis-4.0.3-mac.tgz
  METIS_TOP_DIR: metis-4.0.3
  MFEM_TOP_DIR: mfem

@@ -53,7 +52,6 @@ jobs:
        mpi: [seq, par]
        build-system: [make, cmake]
        hypre-target: [int32]
-        precision: [fp64]
        exclude:
          - os: ubuntu-latest
            build-system: cmake
@@ -77,8 +75,6 @@ jobs:
          - os: ubuntu-latest
            target: dbg
            config-opts: 'CPPFLAGS+=-Og'
-          - os: macos-latest
-            codecov: NO
          - os: windows-latest
            codecov: NO
          - os: windows-latest
@@ -91,7 +87,6 @@ jobs:
            mpi: par
            build-system: cmake
            hypre-target: int32
-            precision: fp64
            # This option can be set to pass additional configuration options to
            # the MFEM configuration command.
            # config-opts: '-DCMAKE_VERBOSE_MAKEFILE=ON'
@@ -101,15 +96,7 @@ jobs:
            mpi: par
            build-system: make
            hypre-target: int64
-            precision: fp64
-          - os: ubuntu-latest
-            target: opt
-            codecov: NO
-            mpi: par
-            build-system: make
-            hypre-target: int32
-            precision: fp32
-    name: ${{ matrix.os }}-${{ matrix.build-system }}-${{ matrix.target }}-${{ matrix.mpi }}-${{ matrix.hypre-target }}-${{ matrix.precision }}
+    name: ${{ matrix.os }}-${{ matrix.build-system }}-${{ matrix.target }}-${{ matrix.mpi }}-${{ matrix.hypre-target }}

    runs-on: ${{ matrix.os }}

@@ -139,17 +126,6 @@ jobs:
          # Fetch the complete history for codecov to access commits ID
          fetch-depth: 0

-      - name: Xcode version setup (MacOS)
-        if: matrix.os == 'macos-latest'
-        run: |
-          XCODE_PATH="/Applications/Xcode_15.3.app"
-          echo "> sudo xcode-select -s ${XCODE_PATH}"
-          sudo xcode-select -s ${XCODE_PATH}
-          echo "> g++ -v"
-          g++ -v
-          echo "> clang++ -v"
-          clang++ -v
-
      # Only get MPI if defined for the job.
      # TODO: It would be nice to have only one step, e.g. with a dedicated
      # action, but I (@adrienbernede) don't see how at the moment.
@@ -193,27 +169,25 @@ jobs:
        uses: actions/cache@v4
        with:
          path: ${{ env.HYPRE_TOP_DIR }}
-          key: ${{ runner.os }}-build-${{ env.HYPRE_TOP_DIR }}-${{ matrix.hypre-target }}-${{ matrix.precision }}-v2.5
+          key: ${{ runner.os }}-build-${{ env.HYPRE_TOP_DIR }}-${{ matrix.hypre-target }}-v2.2

      - name: get hypre
        if: matrix.mpi == 'par' && steps.hypre-cache.outputs.cache-hit != 'true' && matrix.os != 'windows-latest'
-        uses: mfem/github-actions/build-hypre@v2.5
+        uses: mfem/github-actions/build-hypre@v2.4
        with:
          archive: ${{ env.HYPRE_ARCHIVE }}
          dir: ${{ env.HYPRE_TOP_DIR }}
          target: ${{ matrix.hypre-target }}
          build-system: make
-          precision: ${{ matrix.precision }}

      - name: get hypre (Windows)
        if: matrix.mpi == 'par' && steps.hypre-cache.outputs.cache-hit != 'true' && matrix.os == 'windows-latest'
-        uses: mfem/github-actions/build-hypre@v2.5
+        uses: mfem/github-actions/build-hypre@v2.4
        with:
          archive: ${{ env.HYPRE_ARCHIVE }}
          dir: ${{ env.HYPRE_TOP_DIR }}
          target: ${{ matrix.hypre-target }}
          build-system: cmake
-          precision: ${{ matrix.precision }}

      # Get Metis through cache, or build it.
      # Install will only run on cache miss.
@@ -223,13 +197,13 @@ jobs:
        uses: actions/cache@v4
        with:
          path: ${{ env.METIS_TOP_DIR }}
-          key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2.5
+          key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2.2

      - name: install metis
        if: matrix.mpi == 'par' && matrix.os != 'windows-latest' && steps.metis-cache.outputs.cache-hit != 'true'
-        uses: mfem/github-actions/build-metis@v2.5
+        uses: mfem/github-actions/build-metis@v2.4
        with:
-          archive: ${{ matrix.os != 'macos-latest' && env.METIS_ARCHIVE || env.METIS_ARCHIVE_MAC }}
+          archive: ${{ env.METIS_ARCHIVE }}
          dir: ${{ env.METIS_TOP_DIR }}

      - name: cache vcpkg (Windows)
@@ -254,7 +228,7 @@ jobs:

      # MFEM build and test
      - name: build
-        uses: mfem/github-actions/build-mfem@v2.5
+        uses: mfem/github-actions/build-mfem@v2.4
        env:
          VCPKG_DEFAULT_BINARY_CACHE: ${{ github.workspace }}/vcpkg_cache
        with:
@@ -266,7 +240,6 @@ jobs:
          hypre-dir: ${{ env.HYPRE_TOP_DIR }}
          metis-dir: ${{ env.METIS_TOP_DIR }}
          mfem-dir: ${{ env.MFEM_TOP_DIR }}
-          precision: ${{ matrix.precision }}
          config-options: ${{ matrix.config-opts }}
          library-only: ${{ matrix.target == 'dbg' && matrix.os != 'ubuntu-latest' }}

@@ -309,7 +282,7 @@ jobs:
      # Code coverage (process and upload reports)
      - name: codecov
        if: matrix.codecov == 'YES'
-        uses: mfem/github-actions/upload-coverage@v2.5
+        uses: mfem/github-actions/upload-coverage@v2.4
        with:
          name: ${{ matrix.os }}-${{ matrix.build-system }}-${{ matrix.target }}-${{ matrix.mpi }}-${{ matrix.hypre-target }}
          project_dir: ${{ env.MFEM_TOP_DIR }}
@@ -53,11 +53,11 @@ jobs:
        uses: actions/cache@v4
        with:
          path: ${{ env.HYPRE_TOP_DIR }}
-          key: ${{ runner.os }}-build-${{ env.HYPRE_TOP_DIR }}-v2.5
+          key: ${{ runner.os }}-build-${{ env.HYPRE_TOP_DIR }}-v2.2

      - name: Get Hypre
        if: steps.hypre-cache.outputs.cache-hit != 'true'
-        uses: mfem/github-actions/build-hypre@v2.5
+        uses: mfem/github-actions/build-hypre@v2.4
        with:
          archive: ${{ env.HYPRE_ARCHIVE }}
          dir: ${{ env.HYPRE_TOP_DIR }}
@@ -68,18 +68,18 @@ jobs:
        uses: actions/cache@v4
        with:
          path: ${{ env.METIS_TOP_DIR }}
-          key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2.5
+          key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2.2

      - name: Install Metis
        if: steps.metis-cache.outputs.cache-hit != 'true'
-        uses: mfem/github-actions/build-metis@v2.5
+        uses: mfem/github-actions/build-metis@v2.4
        with:
          archive: ${{ env.METIS_ARCHIVE }}
          dir: ${{ env.METIS_TOP_DIR }}

      # MFEM build and test
      - name: build-mfem
-        uses: mfem/github-actions/build-mfem@v2.5
+        uses: mfem/github-actions/build-mfem@v2.4
        with:
          os: ${{ runner.os }}
          target: opt
@@ -44,7 +44,7 @@ jobs:
          path: mfem

      - name: MFEM Build
-        uses: mfem/github-actions/build-mfem@v2.5
+        uses: mfem/github-actions/build-mfem@v2.4
        with:
          os: ${{ runner.os }}
          target: opt
@@ -1,31 +0,0 @@
-# This workflow warns and then closes issues and PRs that have had no activity for a specified amount of time.
-# For more information, see: https://github.com/actions/stale
-name: Mark stale issues and pull requests
-
-on:
-  workflow_dispatch:
-  schedule:
-  - cron: '0 0 * * *'
-
-jobs:
-  stale:
-
-    runs-on: ubuntu-latest
-    permissions:
-      issues: write
-      pull-requests: write
-      actions: write
-
-    steps:
-    - uses: actions/stale@v9
-      with:
-        repo-token: ${{ secrets.GITHUB_TOKEN }}
-        stale-issue-message: ':warning: This issue has been automatically marked as stale because it has not had any activity in the last month. *If no activity occurs in the next week, it will be automatically closed.* Thank you for your contributions.'
-        stale-pr-message: ':warning: This PR has been automatically marked as stale because it has not had any activity in the last month. *If no activity occurs in the next week, it will be automatically closed.* Thank you for your contributions.'
-        days-before-stale: 30
-        days-before-close: 7
-        stale-issue-label: 'stale'
-        stale-pr-label: 'stale'
-        operations-per-run: 500
-        exempt-issue-labels: "bug,WIP,ready-for-review,in-review,in-next"
-        exempt-pr-labels: "bug,WIP,ready-for-review,in-review,in-next"
@@ -1,31 +0,0 @@
-# Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
-# at the Lawrence Livermore National Laboratory. All Rights reserved. See files
-# LICENSE and NOTICE for details. LLNL-CODE-806117.
-#
-# This file is part of the MFEM library. For more information and source code
-# availability visit https://mfem.org.
-#
-# MFEM is free software; you can redistribute it and/or modify it under the
-# terms of the BSD-3 license. We welcome feedback and contributions, see file
-# CONTRIBUTING.md for details.
-
-name: "Trigger PyMFEM CI"
-
-on:
-  push:
-    branches:
-      - master
-
-jobs:
-  trigger-pymfem:
-    runs-on: ubuntu-latest
-    steps:
-    - name: Send POST request to trigger PyMFEM CI
-      run: |
-        curl -L \
-             -X POST \
-             -H "Accept: application/vnd.github+json" \
-             -H "Authorization: Bearer ${{ secrets.PYMFEM_CI_TOKEN }}" \
-             -H "X-GitHub-Api-Version: 2022-11-28" \
-             https://api.github.com/repos/mfem/pymfem/actions/workflows/build-and-test-dispatch.yml/dispatches \
-             -d '{"ref":"master", "inputs":{"test_options":"fast"}}'
@@ -15,9 +15,6 @@
 CMakeCache.txt
 CMakeFiles/

-# Clangd server cache
-*.cache*
-
 # Backup files
 *~

@@ -60,8 +57,6 @@ examples/ex2[0-9]
 examples/ex2[0-9]p
 examples/ex3[0-9]
 examples/ex3[0-9]p
-examples/ex4[0-9]
-examples/ex4[0-9]p

 examples/refined.mesh
 examples/displaced.mesh
@@ -237,7 +232,7 @@ miniapps/meshing/mobius-strip.mesh
 miniapps/meshing/klein-bottle.mesh
 miniapps/meshing/toroid-*.mesh
 miniapps/meshing/twist-*.mesh
-miniapps/meshing/mesh-explorer.mesh*
+miniapps/meshing/mesh-explorer.mesh
 miniapps/meshing/partitioning.txt
 miniapps/meshing/mesh-explorer-visit*
 miniapps/meshing/mesh-explorer-paraview/
@@ -275,27 +270,16 @@ miniapps/navier/*_output

 miniapps/nurbs/nurbs_ex1
 miniapps/nurbs/nurbs_ex1p
-miniapps/nurbs/nurbs_ex3
-miniapps/nurbs/nurbs_ex5
 miniapps/nurbs/nurbs_ex11p
-miniapps/nurbs/nurbs_ex24
-miniapps/nurbs/nurbs_solenoidal
 miniapps/nurbs/nurbs_printfunc
 miniapps/nurbs/nurbs_patch_ex1
 miniapps/nurbs/nurbs_curveint
 miniapps/nurbs/refined.mesh
 miniapps/nurbs/mesh.*
-miniapps/nurbs/sol_?.gf
 miniapps/nurbs/sol.*
 miniapps/nurbs/mode_*
 miniapps/nurbs/Example1*
-miniapps/nurbs/Example3*
-miniapps/nurbs/Example5*
-miniapps/nurbs/Solenoidal*
-miniapps/nurbs/ParaView
 miniapps/nurbs/sin-fit.mesh
-miniapps/nurbs/ex5.mesh
-miniapps/nurbs/exsol.mesh
 miniapps/nurbs/CurveInt
 miniapps/nurbs/nurbs_naca_cmesh
 miniapps/nurbs/naca-cmesh.mesh
@@ -385,8 +369,6 @@ miniapps/dpg/ParaView
 miniapps/spde/generate_random_field
 miniapps/spde/ParaView

-miniapps/tribol/contact-patch-test
-
 # Unit test binary and outputs
 tests/unit/output_meshes
 tests/unit/unit_tests
@@ -13,16 +13,13 @@
 # at Lawrence Livermore National Laboratory (LLNL). This entire pipeline is
 # LLNL-specific!

-include:
-  - project: 'lc-templates/id_tokens'
-    file: 'id_tokens.yml'

 # The pipeline is divided into stages. Usually, jobs in a given stage wait for
 # the preceding stages to complete before to start. However, we sometimes use
 # the "needs" keyword and express the DAG of jobs for more efficiency.
 # - We use setup and setup_baseline phases to download content outside of mfem
 #   directory.
-# - Allocate/Release is where ruby resource are allocated/released once for all.
+# - Allocate/Release is where quartz resource are allocated/released once for all.
 # - Build and Test is where we build and MFEM for multiple toolchains.
 # - Baseline_checks gathers baseline-type test suites execution
 # - Baseline_publish, only available on master, allows to update baseline
@@ -53,7 +50,7 @@ variables:
  AUTOTEST_COMMIT: "YES"

 # Trigger subpipelines:
-ruby-build-and-test:
+quartz-build-and-test:
  stage: sub-pipelines
  variables:
    # Explicitly pass down values that we want to be able to set when triggering
@@ -61,10 +58,10 @@ ruby-build-and-test:
    AUTOTEST: "${AUTOTEST}"
    AUTOTEST_COMMIT: "${AUTOTEST_COMMIT}"
  trigger:
-    include: .gitlab/ruby-build-and-test.yml
+    include: .gitlab/quartz-build-and-test.yml
    strategy: depend

-ruby-baseline:
+quartz-baseline:
  stage: sub-pipelines
  variables:
    # Explicitly pass down values that we want to be able to set when triggering
@@ -73,7 +70,7 @@ ruby-baseline:
    AUTOTEST: "${AUTOTEST}"
    AUTOTEST_COMMIT: "${AUTOTEST_COMMIT}"
  trigger:
-    include: .gitlab/ruby-baseline.yml
+    include: .gitlab/quartz-baseline.yml
    strategy: depend

 lassen-build-and-test:
@@ -24,7 +24,7 @@ and `test type`.

 Machines typically include:

-* Ruby: 2nd Gen Intel Xeon (Cascade Lake)
+* Quartz: Intel bi-socket x86
 * Lassen: Power9 + Nvidia GPU
 * Corona: AMD GPU

@@ -76,13 +76,13 @@ with a spack spec of MFEM, within the limits permitted by the MFEM spack
 package.

 In any build-and-test sub-pipeline a job basically consists in defining the
-spack spec to use. Adding a job on ruby for example resumes to:
+spack spec to use. Adding a job on quartz for example resumes to:

 ```yaml
 <job_name>:
  variables:
    SPEC: "<spack_spec>"
-  extends: .build_and_test_on_ruby
+  extends: .build_and_test_on_quartz
 ```

 The remaining and non trivial work is to make sure this spec is working. To
@@ -9,10 +9,6 @@
 # terms of the BSD-3 license. We welcome feedback and contributions, see file
 # CONTRIBUTING.md for details.

-include:
-  - project: 'lc-templates/id_tokens'
-    file: 'id_tokens.yml'
-
 # We define the following GitLab pipeline variables:
 variables:

@@ -24,7 +20,7 @@ variables:
 # TODO: add a clean-up mechanism
  BUILD_ROOT: ${USER_CI_TOP_DIR}/${CI_PROJECT_NAME}-${MACHINE_NAME}-pipeline-${CI_PIPELINE_ID}

-# On LLNL's ruby, there is only one allocation shared among jobs in order to
+# On LLNL's quartz, there is only one allocation shared among jobs in order to
 # save time and resource. This allocation has to be uniquely named so that we
 # are sure to retrieve it.
  ALLOC_NAME: ${CI_PROJECT_NAME}_ci_${CI_PIPELINE_ID}
@@ -35,8 +35,9 @@ variables:
    - when: on_success

 # Lassen uses a different job scheduler (spectrum lsf) that does not allow
-# pre-allocation the same way slurm does.  We use the pci queue on lassen
-# to speed-up the allocation.
+# pre-allocation the same way slurm does.  We use pdebug queue on lassen
+# to speed-up the allocation.  However this would not be scalable to
+# multiple builds.
 .build_and_test_on_lassen:
  extends: [.on_lassen]
  stage: build_and_test
@@ -44,5 +45,5 @@ variables:
    - echo ${MFEM_DATA_DIR}
    - echo ${SPEC}
    # Next script uses 'THREADS': leaving it empty --> it uses 'make all -j'
-    - lalloc 1 -W 45 -q pci --atsdisable tests/gitlab/build_and_test --spec "${SPEC}" --data-dir "${MFEM_DATA_DIR}" --data
+    - lalloc 1 -W 45 -q pdebug --atsdisable tests/gitlab/build_and_test --spec "${SPEC}" --data-dir "${MFEM_DATA_DIR}" --data
  needs: [setup]
@@ -9,17 +9,17 @@
 # terms of the BSD-3 license. We welcome feedback and contributions, see file
 # CONTRIBUTING.md for details.

-# GitLab pipelines configurations for the Ruby machine at LLNL
+# GitLab pipelines configurations for the Quartz machine at LLNL
 variables:
-  MACHINE_NAME: ruby
+  MACHINE_NAME: quartz

-.on_ruby:
+.on_quartz:
  tags:
    - shell
-    - ruby
+    - quartz
  rules:
-    # Don't run ruby jobs if...
-    - if: '$CI_COMMIT_BRANCH =~ /_qnone/ || $ON_RUBY == "OFF"'
+    # Don't run quartz jobs if...
+    - if: '$CI_COMMIT_BRANCH =~ /_qnone/ || $ON_QUARTZ == "OFF"'
      when: never
    # Don't run autotest update if...
    - if: '$CI_JOB_NAME =~ /report/ && $AUTOTEST != "YES"'
@@ -40,16 +40,16 @@ variables:
    - when: on_success

 # Spack helped builds
-# Generic ruby build job, extending build script
-.build_and_test_on_ruby:
-  extends: [.on_ruby]
+# Generic quartz build job, extending build script
+.build_and_test_on_quartz:
+  extends: [.on_quartz]
  stage: build_and_test
  script:
    # THREADS is used by 'tests/gitlab/build_and_test', run below
-    - export THREADS=16
+    - export THREADS=12
    - echo ${ALLOC_NAME}
    - export JOBID=$(squeue -h --name=${ALLOC_NAME} --format=%A)
    - echo ${JOBID}
    - echo ${MFEM_DATA_DIR}
    - echo ${SPEC}
-    - srun $( [[ -n "${JOBID}" ]] && echo "--jobid=${JOBID}" ) --reservation=ci -t 45 -N 1 tests/gitlab/build_and_test --spec "${SPEC}" --data-dir "${MFEM_DATA_DIR}" --data
+    - srun $( [[ -n "${JOBID}" ]] && echo "--jobid=${JOBID}" ) -t 45 -N 1 tests/gitlab/build_and_test --spec "${SPEC}" --data-dir "${MFEM_DATA_DIR}" --data
@@ -18,7 +18,7 @@
 setup_baseline:
  tags:
    - shell
-    - ruby
+    - quartz
  stage: setup
  variables:
    GIT_STRATEGY: none
@@ -16,7 +16,7 @@
 setup:
  tags:
    - shell
-    - ruby
+    - quartz
  stage: setup
  variables:
    GIT_STRATEGY: none
@@ -14,14 +14,14 @@ stages:
  - build_and_test
  - report

-opt_mpi_cuda_gcc:
+opt_mpi_cuda_xl_16_1_1_12:
  variables:
-    SPEC: "%gcc@8.3.1 +mpi +cuda cuda_arch=70"
+    SPEC: "%xl@16.1.1.12 +mpi +cuda cuda_arch=70"
  extends: .build_and_test_on_lassen

-opt_mpi_cuda_hypre_cuda_gcc:
+opt_mpi_cuda_hypre_cuda_xl:
  variables:
-    SPEC: "%gcc@8.3.1 +mpi +cuda cuda_arch=70 ^hypre+cuda~shared cuda_arch=70"
+    SPEC: "%xl@16.1.1.12 +mpi +cuda cuda_arch=70 ^hypre+cuda~shared cuda_arch=70"
  extends: .build_and_test_on_lassen

 # Jobs report
@@ -19,8 +19,8 @@ stages:
  - cleanup
  - baseline_publish

-baselinecheck_mfem_intel_ruby:
-  extends: [.on_ruby]
+baselinecheck_mfem_intel_quartz:
+  extends: [.on_quartz]
  stage: baseline_check
  variables:
    # TPLS_DIR is used in .gitlab/scripts/baseline to provide the tpls location
@@ -32,7 +32,7 @@ baselinecheck_mfem_intel_ruby:
    - echo ${BUILD_ROOT}
    - echo ${TPLS_DIR}
    # Used by the tests in MFEM/tests:
-    - export MFEM_TEST_NP=48
+    - export MFEM_TEST_NP=32
    # The next script uses the following environment variables:
    # * BASELINE_TEST, SYS_TYPE, CI_PROJECT_DIR, ARTIFACTS_DIR,
    # * BUILD_ROOT, TPLS_DIR, MACHINE_NAME
@@ -44,16 +44,18 @@ baselinecheck_mfem_intel_ruby:
  allow_failure: true

 cleanup:
-  extends: .on_ruby
+  extends: .on_quartz
  stage: cleanup
  variables:
    GIT_STRATEGY: none
  script:
    - echo "BUILD_ROOT=${BUILD_ROOT}"
    - rm -rf "${BUILD_ROOT}" || true
+    - echo "CI_PROJECT_DIR=${CI_PROJECT_DIR}"
+    - make -C "${CI_PROJECT_DIR}" distclean

 report_baseline:
-  extends: [.on_ruby]
+  extends: [.on_quartz]
  stage: baseline_report
  script:
    - echo ${MACHINE_NAME}
@@ -113,8 +115,8 @@ report_baseline:
        exit $err
      ) 9> autotest.lock

-baselinepublish_mfem_ruby:
-  extends: [.on_ruby]
+baselinepublish_mfem_quartz:
+  extends: [.on_quartz]
  stage: baseline_publish
  rules:
    # - if: '$CI_COMMIT_BRANCH == "master" || $REBASELINE == "YES"'
@@ -129,5 +131,5 @@ baselinepublish_mfem_ruby:

 include:
  - local: .gitlab/configs/common.yml
-  - local: .gitlab/configs/ruby-config.yml
+  - local: .gitlab/configs/quartz-config.yml
  - local: .gitlab/configs/setup-baseline.yml
@@ -19,54 +19,54 @@ stages:
 allocate_resource:
  variables:
    GIT_STRATEGY: none
-  extends: .on_ruby
+  extends: .on_quartz
  stage: allocate_resource
  script:
    - echo ${ALLOC_NAME}
    - salloc --exclusive --nodes=1 --reservation=ci --time=60 --no-shell --job-name=${ALLOC_NAME}
  timeout: 6h

-# GitLab jobs for the Ruby machine at LLNL
+# GitLab jobs for the Quartz machine at LLNL
 debug_ser_gcc_10:
  variables:
    SPEC: "%gcc@10.3.1 +debug~mpi"
-  extends: .build_and_test_on_ruby
+  extends: .build_and_test_on_quartz

 debug_par_gcc_10:
  variables:
    SPEC: "%gcc@10.3.1 +debug+mpi"
-  extends: .build_and_test_on_ruby
+  extends: .build_and_test_on_quartz

 opt_ser_gcc_10:
  variables:
    SPEC: "%gcc@10.3.1 ~mpi"
-  extends: .build_and_test_on_ruby
+  extends: .build_and_test_on_quartz

 opt_par_gcc_10:
  variables:
    SPEC: "%gcc@10.3.1"
-  extends: .build_and_test_on_ruby
+  extends: .build_and_test_on_quartz

 opt_par_gcc_10_sundials:
  variables:
    SPEC: "%gcc@10.3.1 +sundials"
-  extends: .build_and_test_on_ruby
+  extends: .build_and_test_on_quartz

 opt_par_gcc_10_petsc:
  variables:
    SPEC: "%gcc@10.3.1 +petsc ^petsc+mumps~superlu-dist"
-  extends: .build_and_test_on_ruby
+  extends: .build_and_test_on_quartz

 opt_par_gcc_10_pumi:
  variables:
    SPEC: "%gcc@10.3.1 +pumi"
-  extends: .build_and_test_on_ruby
+  extends: .build_and_test_on_quartz

 # Release
 release_resource:
  variables:
    GIT_STRATEGY: none
-  extends: .on_ruby
+  extends: .on_quartz
  stage: release_resource_and_report
  script:
    - echo ${ALLOC_NAME}
@@ -78,17 +78,17 @@ release_resource:
 report_job_success:
  stage: release_resource_and_report
  extends:
-    - .on_ruby
+    - .on_quartz
    - .report_job_success

 report_job_failure:
  stage: release_resource_and_report
  extends:
-    - .on_ruby
+    - .on_quartz
    - .report_job_failure

 include:
  - local: .gitlab/configs/common.yml
-  - local: .gitlab/configs/ruby-config.yml
+  - local: .gitlab/configs/quartz-config.yml
  - local: .gitlab/configs/setup-build-and-test.yml
  - local: .gitlab/configs/report-build-and-test.yml
@@ -14,7 +14,7 @@
 # locals
 glob_err=${BASELINE_TEST}.err
 base=${BASELINE_TEST}-${SYS_TYPE}
-if [[ "${MACHINE_NAME}" == "ruby" ]]; then
+if [[ "${MACHINE_NAME}" == "quartz" ]]; then
  base="${BASELINE_TEST}-${MACHINE_NAME}"
 fi
 base_diff=${base}.diff
@@ -31,21 +31,21 @@ cd tests
 mkdir _${BASELINE_TEST} && cd _${BASELINE_TEST}

 # run
-if [[ "${MACHINE_NAME}" == "ruby" ]]; then
-  salloc --nodes=1 --exclusive --reservation=ci ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
+if [[ "${MACHINE_NAME}" == "quartz" || "${MACHINE_NAME}" == "ruby" ]]; then
+  salloc --nodes=1 -p pdebug ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
 elif [[ ${MACHINE_NAME} == "corona" ]]; then
  salloc --nodes=1 -t 60 -p pbatch ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
 elif [[ ${MACHINE_NAME} == "lassen" ]]; then
-  lalloc 1 -q pci ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
+  lalloc 1 -q pdebug ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
 else
  echo "Unknown machine: MACHINE_NAME=$MACHINE_NAME"
  exit 1
 fi
-status="$?"

 # post
 mkdir ${artifacts_path}

+status=0
 if [[ -f ${BASELINE_TEST}.out ]]; then
  cp ${BASELINE_TEST}.out ${artifacts_path}
 fi
@@ -11,7 +11,7 @@
 # terms of the BSD-3 license. We welcome feedback and contributions, see file
 # CONTRIBUTING.md for details.

-# There will be collision between corona and ruby baselines.
+# There will be collision between corona and quartz baselines.
 # Once the corresponding files have been generated, we can switch to machine
 # specific ref.
 ARTIFACT_PATH=${CI_PROJECT_DIR}/${ARTIFACTS_DIR}/baseline-${SYS_TYPE}
@@ -21,7 +21,7 @@ PATCH_FILE=${ARTIFACT_PATH}.patch
 FULL_FILE=${ARTIFACT_PATH}.out
 DIFF_FILE=${ARTIFACT_PATH}.diff

-# There will be collision between corona and ruby baselines.
+# There will be collision between corona and quartz baselines.
 # Once the corresponding files have been generated, we can switch to machine
 # specific ref.
 SAVED_NAME=baseline-${SYS_TYPE}.saved
@@ -8,199 +8,74 @@
                               https://mfem.org


-Version 4.7.1 (development)
+Version 4.6.1 (development)
 ===========================
- Refactored ALGOIM cut integration rules. The interface is unified with
-  the interface for moment based cut integration rules.
-
-Discretization improvements
---------------------------
- Added NURBS-based H(div) and H(curl) elements in 2D and 3D. Only on single
-  patch meshes. Only implemented for serial computations.
-
- Added support for boundary constraints to the hybridization class.
-
- Added support for external boundary submeshes with nonconformal mesh adaptation.

 Meshing improvements
 --------------------
- The ExodusII reader now handles pyramid and wedge element types. Mixed meshes
-  are also supported.
-
-New and updated examples and miniapps
-------------------------------------
- Added miniapps to demonstrate the H(div) and H(curl) NURBS elements.
-
- Added an MFEM example for the eikonal equation. This new solver is based on
-  the proximal Galerkin method introduced by Keith and Surowiec.
-
- Added a command line option to all miniapps (`-p` or `--send-port`) for
-  specifying the GLVis server socket port (19916 by default).
-
-GPU computing
-------------
- Added support for GPU-accelerated batched linear algebra (using cuBLAS,
-  hipBLAS, MAGMA, or native MFEM functionality) through the BatchedLinAlg class.
-
- A new GPU kernel dispatch mechanism was introduced. Users can instantiate
-  specialized kernels for specific combinations of (for example) polynomial
-  degree and number of quadrature points using
-  `DiffusionIntegrator::AddSpecialization` and
-  `MassIntegrator::AddSpecialization` (this functionality may be added to more
-  integrators in the future).
-
- Calls to slower fallback kernels can be reported to `mfem::err` by setting
-  the environment variable `MFEM_REPORT_KERNELS` to any value other than `NO`
-  or by explicitly calling `KernelReporter::Enable`. Users can then add
-  specializations for these kernels to achieve higher performance.
-
- Element assembly kernels have been added for low-order refined to
-  high-order transfer operators. New kernels can be offloaded as device
-  kernels. Example usage may be found in lor-transfer.cpp under miniapps/tools.
-
-Miscellaneous
-------------
- Added support for SUNDIALS v7. See the section "API changes" for some small
-  changes related to this new version.
-
- Refactored the `ARKStepSolver` class (ARKODE interface) to use
-  `TimeDependentOperator::Mult` only when the associated ODE operator is
-  expressed in explicit form (i.e., `TimeDependentOperator::isExplicit()`),
-  otherwise `TimeDependentOperator::ExplicitMult` is used. A check has been
-  added to `ARKStepSolver` to verify that the associated ODE operator is not in
-  explicit form when a mass matrix solver is enabled via a call to either the
-  `UseMFEMMassLinearSolver` or `UseSundialsMassLinearSolver` methods. This is
-  because enabling a mass matrix solver assumes that F(u,k,t) = M k in the
-  associated ODE operator.
-
- Added support for custom interpolation procedure in FindPointsGSLIB.
-
-API changes
-----------
- API change: in class GridFunction, 'fec' was renamed to 'fec_owned'.
-
- API change: support for SUNDIALS v7:
-  * the SUNDIALS types `realtype` and `booleantype` are no longer defined by v7
-    and therefore MFEM now uses the new type names `sunrealtype` and
-    `sunbooleantype`, respectively, which MFEM defines when using SUNDIALS < v6
-    where these types were not defined.
-  * The SUNDIALS macro `SUNLS_SUCCESS` and some other `*_SUCCESS` macros were
-    removed and replaced by `SUN_SUCCESS` in v7, so to avoid tedious checks for
-    SUNDIALS versions, MFEM now defines and uses the constant `SUN_SUCCESS` when
-    using SUNDIALS < v7.
-  * The constants `SUN_PREC_*`, introduced by SUNDIALS v6 are now introduced by
-    MFEM when using SUNDIALS < v6 to avoid tedious version checks.
-
-
-Version 4.7, released on May 7, 2024
-====================================
-
- Added support for single precision (with corresponding hypre build). The MFEM
-  floating point type was generalized from `double` to `real_t`. For details see
-  https://github.com/orgs/mfem/discussions/4207.
-
-Meshing improvements
--------------------
- Added the capability to partition (big) serial meshes in serial code, see the
-  new classes MeshPartitioner and MeshPart. This capability is also exposed as a
-  menu option in the mesh-explorer miniapp in miniapps/meshing.
-
- Added named attribute sets and basic supporting methods to the Mesh class as a
-  convenient means of referring to sets of domain or boundary attribute numbers.
-  See the new Example 39/39p and data/compass.mesh.
-
 - Introduced formulas for refinement of patches in NURBS meshes. Refinement by
  arbitrary integer factors is also enabled, e.g. in the mesh-explorer miniapp.
  NURBS coarsening and knot removal are also introduced.

- Added support for internal boundary elements in nonconforming meshes.
-
- Added ExodusII output capability. The writer can handle first-order (Pyramid5,
-  Wedge6, Hex8, Tet4) and second-order FE types (Pyramid14, Wedge18, Hex27, Tet10).
-
- The ReadCubit Genesis mesh importer has been rewritten to improve readability.
-
 Discretization improvements
 ---------------------------
- Added a new nonlinear integrator, `HyperbolicFormIntegrator` that implements
+- Introduced support for higher order non conformal Nedelec elements on
+  simplices in ParMesh.
+- Introduced support for internal boundary elements in nonconformal adapted
+  meshes.
+
+- Added functionality for construction of cut-surface and cut-volume
+  IntegrationRules through a moment-fitting approach. The cut is specified by
+  the zero level set of a Coefficient. See fem/intrules_cut.hpp and Example 38.
+
+- Added a new nonlinear integrator, `HyperbolicFormIntegrator`. This implements 
  both element-wise weak divergence and face-wise numerical flux for a general
-  system of hyperbolic conservation laws. To use the integrator for a specific
+  system of hyperbolic conservation laws. To use this integrator for a specific
  flux function, users can define a derived class of `FluxFunction`. Currently,
-  advection, Burgers, shallow-water and Euler equations (see Example 18/18p) are
+  advection, Burgers', shallow-water, Euler equations (see, Example 18) are
  available.

- Added a capability to construct cut-surface and cut-volume IntegrationRules
-  through a moment-fitting approach. The cut is specified by the zero level set
-  of a Coefficient. See fem/intrules_cut.hpp and the new Example 38.
-
- Introduced support for high-order nonconforming Nedelec elements on simplices.
-
-GPU computing
-------------
- Added partial assembly and GPU support for the DG diffusion integrator.
-
- Efficient GPU-accelerated LOR assembly is now supported on surface meshes.
-
- Added functionality to automatically configure hypre's compute policy to match
-  MFEM's compute policy when hypre is built with GPU support. Requires version
-  hypre-2.31.0 or later.
-
+GPU support
+----------------------------
 - Added support for full assembly on simplices.
-
- Added partial assembly for linear elasticity (no sum factorization for now).
-
- Added functionality for BilinearFormIntegrators to use kernels that work for
-  both tensor and unstructured elements.
-
- The RAJA backend will use `seq_exec` for serial loop execution when RAJA
-  v2023.06.00 and beyond is detected as `loop_exec` is deprecated.
-
- API change: The macro MFEM_HYPRE_FORALL (from hypre.hpp) which was intended
-  for internal use, has been removed and replaced by the function template
-  mfem::hypre_forall in general/forall.hpp.
+- Added functionality for BilinearFormIntegrators to use kernels that work for both
+  tensor and unstructured elements.
+- Added partial assembly for linear elasticity. Does not use sum factorization for now.

 New and updated examples and miniapps
 -------------------------------------
- Added a new miniapp illustrating elastic contact based on the Tribol library,
-  (https://github.com/LLNL/Tribol). See miniapps/tribol.
+- Added a new block solver in miniapp/solvers for the Darcy problem.
+  The new solver is based on a Bramble-Pasciak preconditioning. User can
+  use and implement their own preconditioner for the mass matrix.

- Added a miniapp to demonstrate low order refined (LOR) block preconditioning
-  for linear elasticity on GPUs. See miniapps/solvers/lor_elast.
-
- Added a new block solver in miniapp/solvers for the Darcy problem. The new
-  solver is based on a Bramble-Pasciak preconditioning. User can use and
-  implement their own preconditioner for the mass matrix.
-
- Added a small miniapp for printing the shape functions of a KnotVector. See
-  miniapps/nurbs/nurbs_printfunc.cpp.
-
- Added two new example codes: 38 and 39/39p described above. Substantially
-  updated Example 18/18p.
-
- Added ODE solvers selection routines. This creates a uniformity across examples,
-  miniapps and other executables in regard to ODE(time-integrator) selection.
-
- Added new mechanism for retrieving and setting state vectors in ODE solvers.
-  This is relevant for AB/AM and gen-alpha solvers.
-
- Added ODEsolver/ODEsolver2 unit tests to verify order of convergence and
-  read/write functionality.
+- Added miniapp to demonstrate new elasticity integrator and unstructured element GPU support,
+  and a block diagonal preconditioner using low order refinement. Allows comparison with
+  currently existing legacy mode integrator. See miniapps/solvers/lor_elast.

 Miscellaneous
 -------------
+- Added support for single and double precision, with corresponding hypre build.
+  Generalized the floating point type from `double` to `real_t`. For more
+  details see https://github.com/orgs/mfem/discussions/4207.
+
+- The ReadCubit Genesis mesh importer has been rewritten to improve readability.
+
 - Updated the Doxygen documentation style, which now requires Doxygen version
  1.9.8 or later. See the doc/ directory.

- Improved thread safety for global variables in the library, e.g. for IntRules,
-  RefinedIntRules, GlobGeometryRefiner, and FiniteElement::dof2quad_array.
+- Improved thread safety for global variables in the library, for example
+  IntegrationRules IntRules, RefinedIntRules, GeometryRefiner
+  GlobGeometryRefiner, and FiniteElement::dof2quad_array.

 - PETSc integration now generally requires PETSc version 3.21 or later, though
  depending on the functionality older versions may still work.

- Various other simplifications, extensions, and bugfixes in the code.
-
- Added GSLIB-based gather-scatter operator.
+- RAJA backend will use seq_exec for serial loop execution when RAJA
+  v2023.06.00 and beyond is detected as loop_exec is deprecated.

+- Adding named attribute sets and basic supporting methods to the Mesh class as
+  a convenient means of referring to sets of domain or boundary attribute
+  numbers. Also adding related serial and parallel examples which illustrate.

 Version 4.6, released on September 27, 2023
 ===========================================
@@ -221,6 +96,7 @@ Meshing improvements
  * The edge to knot map for NURBS meshes can be determined automatically. It is
    no longer needed to specify this in the NURBS mesh.
  * Added curve interpolation method for NURBS.
+  * Added new small miniapp for printing of shape functions of a KnotVector
  * See miniapps/nurbs for example meshes and miniapps.

 Discretization improvements
@@ -267,6 +143,8 @@ Linear and nonlinear solvers

 - Added HIP support to the PETSc and SUNDIALS interfaces.

+- Efficient GPU-accelerated LOR assembly now supports surface meshes.
+
 New and updated examples and miniapps
 -------------------------------------
 - Added a new H(div) solver miniapp demonstrating the use of a matrix-free
@@ -58,7 +58,7 @@ project(mfem NONE)
 # Current version of MFEM, see also `makefile`.
 #   mfem_VERSION = (string)
 #   MFEM_VERSION = (int)   [automatically derived from mfem_VERSION]
-set(${PROJECT_NAME}_VERSION 4.7.1)
+set(${PROJECT_NAME}_VERSION 4.6.1)

 # Prohibit in-source build
 if (${PROJECT_SOURCE_DIR} STREQUAL ${PROJECT_BINARY_DIR})
@@ -87,11 +87,10 @@ if (MFEM_USE_STRUMPACK OR MFEM_USE_MUMPS)
  # Just needed to find the MPI_Fortran libraries to link with
  set(XSDK_ENABLE_Fortran ON)
 endif()
-# SUNDIALS, STRUMPACK, Ginkgo, Tribol, RAJA and Umpire require C++14:
+# SUNDIALS, STRUMPACK, Ginkgo, RAJA and Umpire require C++14:
 if ((MFEM_USE_SUNDIALS OR
     MFEM_USE_STRUMPACK OR
     MFEM_USE_GINKGO OR
-     MFEM_USE_TRIBOL OR
     MFEM_USE_RAJA OR
     MFEM_USE_UMPIRE) AND
    ("${CMAKE_CXX_STANDARD}" LESS "14"))
@@ -146,9 +145,7 @@ if (MFEM_USE_CUDA)
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} ${CUDA_FLAGS}")
   find_package(CUDAToolkit REQUIRED)
   set(CUSPARSE_FOUND TRUE)
-   set(CUBLAS_FOUND TRUE)
   get_target_property(CUSPARSE_LIBRARIES CUDA::cusparse LOCATION)
-   get_target_property(CUBLAS_LIBRARIES CUDA::cublas LOCATION)
 endif()

 if (XSDK_ENABLE_C)
@@ -233,7 +230,6 @@ if (MFEM_USE_HIP)
    list(INSERT CMAKE_PREFIX_PATH 0 ${ROCM_PATH})
  endif()
  find_package(HIP REQUIRED)
-  find_package(HIPBLAS REQUIRED)
  find_package(HIPSPARSE REQUIRED)
 endif()

@@ -340,10 +336,7 @@ if (MFEM_USE_SUNDIALS)
  if (MFEM_USE_HIP)
    list(APPEND SUNDIALS_COMPONENTS NVector_Hip)
  endif()
-  # The Core component was added in SUNDIALS v7, so we treat it as optional in
-  # order to support older versions.
-  find_package(SUNDIALS REQUIRED ${SUNDIALS_COMPONENTS}
-    OPTIONAL_COMPONENTS Core)
+  find_package(SUNDIALS REQUIRED ${SUNDIALS_COMPONENTS})
 endif()

 # SuperLU_DIST can only be enabled in parallel
@@ -402,10 +395,6 @@ if (MFEM_USE_AMGX)
  find_package(AMGX REQUIRED)
 endif()

-if (MFEM_USE_MAGMA)
-  find_package(MAGMA REQUIRED)
-endif()
-
 if (MFEM_USE_CONDUIT)
    find_package(Conduit REQUIRED conduit relay blueprint)
 endif()
@@ -514,15 +503,6 @@ if (MFEM_USE_PARELAG)
   find_package(PARELAG REQUIRED)
 endif()

-# Tribol
-if (MFEM_USE_TRIBOL)
-  if (MFEM_USE_MPI)
-    find_package(Tribol REQUIRED tribol redecomp)
-  else()
-    message(FATAL_ERROR " *** Tribol requires that MPI be enabled.")
-  endif()
-endif()
-
 # Enzyme
 if (MFEM_USE_ENZYME)
   find_package(ENZYME REQUIRED)
@@ -567,9 +547,8 @@ find_package(Threads REQUIRED)
 set(MFEM_TPLS OPENMP HYPRE LAPACK BLAS SuperLUDist STRUMPACK METIS SuiteSparse
    SUNDIALS PETSC SLEPC MUMPS AXOM FMS CONDUIT Ginkgo GNUTLS GSLIB
    NETCDF MPFR PUMI HIOP POSIXCLOCKS MFEMBacktrace ZLIB OCCA CEED RAJA UMPIRE
-    ADIOS2 MKL_CPARDISO MKL_PARDISO AMGX MAGMA CUSPARSE CUBLAS CALIPER CODIPACK
-    BENCHMARK PARELAG TRIBOL MPI_CXX HIP HIPBLAS HIPSPARSE MOONOLITH BLITZ
-    ALGOIM ENZYME)
+    ADIOS2 CUSPARSE MKL_CPARDISO MKL_PARDISO AMGX CALIPER CODIPACK
+    BENCHMARK PARELAG MPI_CXX HIP HIPSPARSE MOONOLITH BLITZ ALGOIM ENZYME)

 # Add all *_FOUND libraries in the variable TPL_LIBRARIES.
 set(TPL_LIBRARIES "")
@@ -684,7 +663,7 @@ if (NOT ("${PROJECT_SOURCE_DIR}" STREQUAL "${PROJECT_BINARY_DIR}"))
 #include \"${PROJECT_SOURCE_DIR}/${Header}\"
 ")

-    execute_process(COMMAND ${CMAKE_COMMAND} -E copy_if_different
+    execute_process(COMMAND ${CMAKE_COMMAND} -E copy_if_different 
      "${PROJECT_BINARY_DIR}/${Header}.tmp"
      "${PROJECT_BINARY_DIR}/${Header}"
    )
@@ -698,7 +677,7 @@ if (NOT ("${PROJECT_SOURCE_DIR}" STREQUAL "${PROJECT_BINARY_DIR}"))
 #include \"mfem/${Header}\"
 ")

-    execute_process(COMMAND ${CMAKE_COMMAND} -E copy_if_different
+    execute_process(COMMAND ${CMAKE_COMMAND} -E copy_if_different 
      "${PROJECT_BINARY_DIR}/InstallHeaders/${Header}.tmp"
      "${PROJECT_BINARY_DIR}/InstallHeaders/${Header}"
    )
@@ -151,8 +151,7 @@ The MFEM source code has the following structure:
  │   ├── solvers
  │   ├── spde
  │   ├── tools
-  │   ├── toys
-  │   └── tribol
+  │   └── toys
  └── tests
      ├── benchmarks
      ├── convergence
@@ -75,8 +75,6 @@ and miniapps. See https://glvis.org and https://mfem.org/building.

 Quick start with GNU make
 =========================
-See also: https://mfem.org/building
-
 Serial build:
   make serial -j 4

@@ -85,7 +83,6 @@ Parallel build:
   (build METIS 4 in ../metis-4.0 relative to mfem/)
   (build hypre in ../hypre relative to mfem/)
   make parallel -j 4
-(For METIS 5, see https://mfem.org/building/#parallel-build-using-metis-5)

 CUDA build:
   make cuda -j 4
@@ -119,7 +116,6 @@ Parallel build:
   mkdir <mfem-build-dir> ; cd <mfem-build-dir>
   cmake <mfem-source-dir> -DMFEM_USE_MPI=YES
   make -j 4
-(For METIS 5, see https://mfem.org/building/#parallel-build-using-metis-5)

 CUDA build:
   (this build requires CMake 3.8 or newer)
@@ -273,13 +269,7 @@ Installation options:
   PREFIX  - Specify the installation directory. The library (libmfem.a) will be
             installed in $(PREFIX)/lib, the headers in $(PREFIX)/include, and
             the configuration makefile (config.mk) in $(PREFIX)/share/mfem.
-   INSTALL - Specify the install program, default = /usr/bin/install
-   INSTALL_DEF_PERM - Specify the default install permissions. This affects
-                      headers and configuration makefiles, default = 644
-   INSTALL_BIN_PERM - Specify the install permissions for binaries. This only
-                      affects the shared version of the library, default = 755
-   INSTALL_DIR_PERM - Specify the install permissions for directories and,
-                      on macOS/BSD, for symlinks as well, default = 755
+   INSTALL - Specify the install program, e.g /usr/bin/install

 MFEM library features/options (GNU make)
 ----------------------------------------
@@ -394,11 +384,6 @@ MFEM_USE_AMGX = YES/NO
   Allows the user to use SparseMatrices and HypreParMatrices to solve linear
   systems with the routines from the AmgX library.

-MFEM_USE_MAGMA = YES/NO
-   Enable MFEM functionality based on the MAGMA high-performance linear algebra
-   library. The MAGMA library provides a BLAS/LAPACK interface, with
-   implementations that have been optimized for Nvidia and AMD GPUs.
-
 MFEM_USE_GNUTLS = YES/NO
   Enable secure socket support in class socketstream, using the auxiliary
   GnuTLS_* classes, based on the GnuTLS library. This option may be useful in
@@ -502,14 +487,10 @@ MFEM_USE_CODIPACK = YES/NO
 MFEM_USE_ALGOIM = YES/NO
  Enable the usage of Algoim - a collection of high-order accurate numerical
  methods and C++ algorithms for working with implicitly-defined geometry and
-  level set methods, see https://algoim.github.io. MFEM provides interface to
-  Algoim v1. To check out the specific Algoim state use:
-  https://github.com/algoim/algoim
+  level set methods. The Algoim library requires the Blitz++ library. The MFEM
+  provides interface to Algoim v1. Thus, to check out the specific state use:
  git checkout 9c9ca0ef094d8ab0390ed36367a1151b459bbe0a
-  The Algoim library requires the Blitz++ library. To use the latest state of
-  Blitz++ that has been tested with MFEM, use:
-  https://github.com/blitzpp/blitz
-  git checkout f24a250a43dff88c31ad92916da828b7ea9a98b7
+  https://algoim.github.io

 MFEM_USE_ADFORWARD = YES/NO
  Enable forward mode for AD packages. This option is valid
@@ -590,11 +571,6 @@ MFEM_USE_PARELAG = YES/NO
   use ParELAG. In fact, ParELAG is dependent on MFEM. Therefore, this option
   currently only concerns the miniapps.

-MFEM_USE_TRIBOL = YES/NO
-   Enables the miniapps that use the Tribol library. MFEM does not currently
-   use Tribol. In fact, Tribol is dependent on MFEM. Therefore, this option
-   currently only concerns the miniapps.
-
 MFEM_USE_ENZYME = YES/NO
   Enables automatic differentiation support through the LLVM plugin Enzyme.
   This requires the compiler to be set to clang (>=14.0.0). We also advise to
@@ -631,13 +607,9 @@ The specific libraries and their options are:
            HYPRE >= 2.20.0   (HYPRE built with '--enable-mixedint')
            HYPRE >= 2.22.1   (HYPRE built with CUDA)
            HYPRE >= 2.23.0   (HYPRE built with HIP)
-            HYPRE >= 2.31.0   (runtime selectable HYPRE execution on CPU/GPU)

 - METIS, used when MFEM_USE_METIS = YES. If using METIS 5, set
-  MFEM_USE_METIS_5 = YES (default is to use METIS 4). For building instructions,
-  see the following:
-    - METIS 4.0.3: https://mfem.org/building/#parallel-mpi-version-of-mfem
-    - METIS 5.1.0: https://mfem.org/building/#parallel-build-using-metis-5
+  MFEM_USE_METIS_5 = YES (default is to use METIS 4).
  URL: https://github.com/mfem/tpls (MFEM mirror, see above)
  Options: METIS_OPT, METIS_LIB.
  Versions: METIS 4.0.3 or 5.1.0.
@@ -714,11 +686,6 @@ The specific libraries and their options are:
  Options: AMGX_OPT, AMGX_LIB.
  Versions: AmgX >= 2.1, older versions may work too.

- MAGMA (optional), used with MFEM_USE_MAGMA = YES.
-  URL: https://icl.utk.edu/magma/
-  Options: MAGMA_OPT, MAGMA_LIB
-  Versions: MAGMA >= 2.8.0
-
 - GnuTLS (optional), used when MFEM_USE_GNUTLS = YES. On most Linux systems,
  GnuTLS is available as a development package, e.g. gnutls-devel. On Mac OS X,
  one can get the library through the Homebrew package manager (http://brew.sh).
@@ -890,10 +857,6 @@ The specific libraries and their options are:
  URL: https://github.com/LLNL/parelag
  Options: PARELAG_DIR, PARELAG_OPT, PARELAG_LIB.

- Tribol, used when MFEM_USE_TRIBOL = YES.
-  URL: https://github.com/LLNL/Tribol
-  Options: TRIBOL_DIR, TRIBOL_OPT, TRIBOL_LIB.
-
 - Enzyme, used when MFEM_USE_ENZYME = YES. Requires LLVM/Clang >= 14.0.0.
  URL: https://github.com/EnzymeAD/Enzyme
  Options: ENZYME_DIR, ENZYME_OPT, ENZYME_LIB.
@@ -1038,7 +1001,6 @@ MFEM_USE_CALIPER
 MFEM_USE_FMS
 MFEM_USE_BENCHMARK
 MFEM_USE_PARELAG
-MFEM_USE_TRIBOL
 MFEM_USE_ENZYME

 The following options are CMake specific:
@@ -287,7 +287,3 @@ ENDIF()
 IF (DEFINED TPL_ENABLE_PARELAG)
  SET(MFEM_USE_PARELAG ${TPL_ENABLE_PARELAG} CACHE BOOL "Enable ParELAG" FORCE)
 ENDIF()
-
-IF (DEFINED TPL_ENABLE_TRIBOL)
-  SET(MFEM_USE_TRIBOL ${TPL_ENABLE_TRIBOL} CACHE BOOL "Enable Tribol" FORCE)
-ENDIF()
@@ -37,7 +37,6 @@ set(MFEM_USE_MUMPS @MFEM_USE_MUMPS@)
 set(MFEM_USE_STRUMPACK @MFEM_USE_STRUMPACK@)
 set(MFEM_USE_GINKGO @MFEM_USE_GINKGO@)
 set(MFEM_USE_AMGX @MFEM_USE_AMGX@)
-set(MFEM_USE_MAGMA @MFEM_USE_MAGMA@)
 set(MFEM_USE_HIOP @MFEM_USE_HIOP@)
 set(MFEM_USE_GNUTLS @MFEM_USE_GNUTLS@)
 set(MFEM_USE_GSLIB @MFEM_USE_GSLIB@)
@@ -65,7 +64,6 @@ set(MFEM_USE_CALIPER @MFEM_USE_CALIPER@)
 set(MFEM_USE_ALGOIM @MFEM_USE_ALGOIM@)
 set(MFEM_USE_BENCHMARK @MFEM_USE_BENCHMARK@)
 set(MFEM_USE_PARELAG @MFEM_USE_PARELAG@)
-set(MFEM_USE_TRIBOL @MFEM_USE_TRIBOL@)
 set(MFEM_USE_ENZYME @MFEM_USE_ENZYME@)

 set(MFEM_CXX_COMPILER "@CMAKE_CXX_COMPILER@")
@@ -114,9 +114,6 @@
 // Enable MFEM functionality based on the AmgX library.
 #cmakedefine MFEM_USE_AMGX

-// Enable MFEM functionality based on the MAGMA library.
-#cmakedefine MFEM_USE_MAGMA
-
 // Enable secure socket streams based on the GNUTLS library.
 #cmakedefine MFEM_USE_GNUTLS

@@ -18,13 +18,4 @@ include(MfemCmakeUtilities)
 # Note: components are enabled based on the find_package() parameters.
 mfem_find_package(Axom AXOM AXOM_DIR "include" "" "lib" ""
  "Paths to headers required by Axom." "Libraries required by Axom."
-  ADD_COMPONENT core "include" axom/core.hpp "lib" axom_core
-  ADD_COMPONENT inlet "include" axom/inlet.hpp "lib" axom_inlet
-  ADD_COMPONENT klee "include" axom/klee.hpp "lib" axom_klee
-  ADD_COMPONENT lumberjack "include" axom/lumberjack.hpp "lib" axom_lumberjack
-  ADD_COMPONENT mint "include" axom/mint.hpp "lib" axom_mint
-  ADD_COMPONENT multimat "include" axom/multimat.hpp "lib" axom_multimat
-  ADD_COMPONENT quest "include" axom/quest.hpp "lib" axom_quest
-  ADD_COMPONENT sidre "include" axom/sidre.hpp "lib" axom_sidre
-  ADD_COMPONENT slam "include" axom/slam.hpp "lib" axom_slam
-  ADD_COMPONENT slic "include" axom/slic.hpp "lib" axom_slic)
+  ADD_COMPONENT Axom "include" axom/config.hpp "lib" axom)
@@ -36,11 +36,7 @@ include(MfemCmakeUtilities)
 mfem_find_package(Conduit CONDUIT CONDUIT_DIR
  "include;include/conduit" conduit.hpp "lib" conduit
  "Paths to headers required by Conduit." "Libraries required by Conduit."
-  ADD_COMPONENT blueprint
-    "include;include/conduit" conduit_blueprint.hpp "lib" conduit_blueprint
-  ADD_COMPONENT blueprint_mpi
-    "include;include/conduit" conduit_blueprint_mpi.hpp "lib" conduit_blueprint_mpi
  ADD_COMPONENT relay
    "include;include/conduit" conduit_relay.hpp "lib" conduit_relay
-  ADD_COMPONENT relay_mpi
-    "include;include/conduit" conduit_relay_mpi.hpp "lib" conduit_relay_mpi)
+  ADD_COMPONENT blueprint
+    "include;include/conduit" conduit_blueprint.hpp "lib" conduit_blueprint)
@@ -1,37 +0,0 @@
-# Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
-# at the Lawrence Livermore National Laboratory. All Rights reserved. See files
-# LICENSE and NOTICE for details. LLNL-CODE-806117.
-#
-# This file is part of the MFEM library. For more information and source code
-# availability visit https://mfem.org.
-#
-# MFEM is free software; you can redistribute it and/or modify it under the
-# terms of the BSD-3 license. We welcome feedback and contributions, see file
-# CONTRIBUTING.md for details.
-
-# Defines the following variables:
-#   - MAGMA_FOUND
-#   - MAGMA_LIBRARIES
-#   - MAGMA_INCLUDE_DIRS
-
-include(MfemCmakeUtilities)
-mfem_find_package(MAGMA MAGMA MAGMA_DIR "include" "magma.h" "lib" "magma"
-  "Paths to headers required by MAGMA." "Libraries required by MAGMA.")
-
-if (MAGMA_FOUND AND MFEM_USE_CUDA)
-  get_target_property(CUSPARSE_LIBRARIES CUDA::cusparse LOCATION)
-  get_target_property(CUBLAS_LIBRARIES CUDA::cublas LOCATION)
-  list(APPEND MAGMA_LIBRARIES ${CUSPARSE_LIBRARIES} ${CUBLAS_LIBRARIES})
-  set(MAGMA_LIBRARIES ${MAGMA_LIBRARIES} CACHE STRING
-      "MAGMA libraries + dependencies." FORCE)
-  message(STATUS "Updated MAGMA_LIBRARIES: ${MAGMA_LIBRARIES}")
-endif()
-
-if (MAGMA_FOUND AND MFEM_USE_HIP)
-  find_package(HIPBLAS REQUIRED)
-  find_package(HIPSPARSE REQUIRED)
-  list(APPEND MAGMA_LIBRARIES ${HIPBLAS_LIBRARIES} ${HIPSPARSE_LIBRARIES})
-  set(MAGMA_LIBRARIES ${MAGMA_LIBRARIES} CACHE STRING
-      "MAGMA libraries + dependencies." FORCE)
-  message(STATUS "Updated MAGMA_LIBRARIES: ${MAGMA_LIBRARIES}")
-endif()
@@ -16,22 +16,12 @@
 #   - MUMPS_VERSION

 include(MfemCmakeUtilities)
-
-# Toggle which precision of MUMPS to use depending on the precision of MFEM.
-if (MFEM_USE_DOUBLE)
-  set(_mumps_header dmumps_c.h)
-  set(_mumps_lib dmumps)
-elseif(MFEM_USE_SINGLE)
-  set(_mumps_header smumps_c.h)
-  set(_mumps_lib smumps)
-endif()
-
 mfem_find_package(MUMPS MUMPS MUMPS_DIR
-  "include" ${_mumps_header} "lib" ${_mumps_lib}
+  "include" dmumps_c.h "lib" dmumps
  "Paths to headers required by MUMPS."
  "Libraries required by MUMPS."
-  ADD_COMPONENT mumps_common "include" ${_mumps_header} "lib" mumps_common
-  ADD_COMPONENT pord "include" ${_mumps_header} "lib" pord)
+  ADD_COMPONENT mumps_common "include" dmumps_c.h "lib" mumps_common
+  ADD_COMPONENT pord "include" dmumps_c.h "lib" pord)

 if (MUMPS_FOUND AND (NOT MUMPS_VERSION))
  try_run(MUMPS_VERSION_RUN_RESULT MUMPS_VERSION_COMPILE_RESULT
@@ -31,5 +31,4 @@ mfem_find_package(SUNDIALS SUNDIALS SUNDIALS_DIR
  ADD_COMPONENT CVODE "include" cvode/cvode.h "lib" sundials_cvode
  ADD_COMPONENT CVODES "include" cvodes/cvodes.h "lib" sundials_cvodes
  ADD_COMPONENT ARKODE "include" arkode/arkode.h "lib" sundials_arkode
-  ADD_COMPONENT KINSOL "include" kinsol/kinsol.h "lib" sundials_kinsol
-  ADD_COMPONENT Core "include" sundials/sundials_core.h "lib" sundials_core)
+  ADD_COMPONENT KINSOL "include" kinsol/kinsol.h "lib" sundials_kinsol)
@@ -1,22 +0,0 @@
-# Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
-# at the Lawrence Livermore National Laboratory. All Rights reserved. See files
-# LICENSE and NOTICE for details. LLNL-CODE-806117.
-#
-# This file is part of the MFEM library. For more information and source code
-# availability visit https://mfem.org.
-#
-# MFEM is free software; you can redistribute it and/or modify it under the
-# terms of the BSD-3 license. We welcome feedback and contributions, see file
-# CONTRIBUTING.md for details.
-
-# Defines the following variables:
-#   - TRIBOL_FOUND
-#   - TRIBOL_LIBRARIES
-#   - TRIBOL_INCLUDE_DIRS
-
-include(MfemCmakeUtilities)
-# Note: components are enabled based on the find_package() parameters.
-mfem_find_package(Tribol TRIBOL TRIBOL_DIR "include" tribol/config.hpp "lib" tribol
-  "Paths to headers required by Tribol." "Libraries required by Tribol."
-  ADD_COMPONENT redecomp
-    "include" redecomp/redecomp.hpp "lib" redecomp)
@@ -846,14 +846,14 @@ function(mfem_export_mk_files)
      MFEM_USE_ZLIB MFEM_USE_LIBUNWIND MFEM_USE_LAPACK MFEM_THREAD_SAFE
      MFEM_USE_LEGACY_OPENMP MFEM_USE_OPENMP MFEM_USE_MEMALLOC MFEM_USE_SUNDIALS
      MFEM_USE_SUITESPARSE MFEM_USE_SUPERLU MFEM_USE_SUPERLU5 MFEM_USE_MUMPS
-      MFEM_USE_STRUMPACK MFEM_USE_GINKGO MFEM_USE_AMGX MFEM_USE_MAGMA
-      MFEM_USE_GNUTLS MFEM_USE_NETCDF MFEM_USE_PETSC MFEM_USE_SLEPC
-      MFEM_USE_MPFR MFEM_USE_SIDRE MFEM_USE_FMS MFEM_USE_CONDUIT MFEM_USE_PUMI
-      MFEM_USE_HIOP MFEM_USE_GSLIB MFEM_USE_CUDA MFEM_USE_HIP MFEM_USE_RAJA
-      MFEM_USE_OCCA MFEM_USE_CEED MFEM_USE_CALIPER MFEM_USE_UMPIRE MFEM_USE_SIMD
-      MFEM_USE_ADIOS2 MFEM_USE_MKL_CPARDISO MFEM_USE_MKL_PARDISO
-      MFEM_USE_ADFORWARD MFEM_USE_CODIPACK MFEM_USE_BENCHMARK MFEM_USE_PARELAG
-      MFEM_USE_TRIBOL MFEM_USE_MOONOLITH MFEM_USE_ALGOIM MFEM_USE_ENZYME)
+      MFEM_USE_STRUMPACK MFEM_USE_GINKGO MFEM_USE_AMGX MFEM_USE_GNUTLS
+      MFEM_USE_NETCDF MFEM_USE_PETSC MFEM_USE_SLEPC MFEM_USE_MPFR MFEM_USE_SIDRE
+      MFEM_USE_FMS MFEM_USE_CONDUIT MFEM_USE_PUMI MFEM_USE_HIOP MFEM_USE_GSLIB
+      MFEM_USE_CUDA MFEM_USE_HIP MFEM_USE_RAJA MFEM_USE_OCCA MFEM_USE_CEED
+      MFEM_USE_CALIPER MFEM_USE_UMPIRE MFEM_USE_SIMD MFEM_USE_ADIOS2
+      MFEM_USE_MKL_CPARDISO MFEM_USE_MKL_PARDISO MFEM_USE_ADFORWARD
+      MFEM_USE_CODIPACK MFEM_USE_BENCHMARK MFEM_USE_PARELAG MFEM_USE_MOONOLITH
+      MFEM_USE_ALGOIM MFEM_USE_ENZYME)
  foreach(var ${CONFIG_MK_BOOL_VARS})
    if (${var})
      set(${var} YES)
@@ -120,15 +120,6 @@ constexpr real_t operator""_r(unsigned long long v)

 // Check dependencies:

-// Define MFEM_MPI_REAL_T to be the appropriate MPI real type
-#ifdef MFEM_USE_MPI
-#ifdef MFEM_USE_SINGLE
-#define MFEM_MPI_REAL_T MPI_FLOAT
-#elif defined MFEM_USE_DOUBLE
-#define MFEM_MPI_REAL_T MPI_DOUBLE
-#endif
-#endif
-
 // Options that require MPI
 #ifndef MFEM_USE_MPI
 #ifdef MFEM_USE_SUPERLU
@@ -114,9 +114,6 @@
 // Enable MFEM functionality based on the AmgX library.
 // #define MFEM_USE_AMGX

-// Enable MFEM functionality based on the MAGMA library.
-// #define MFEM_USE_MAGMA
-
 // Enable secure socket streams based on the GNUTLS library.
 // #define MFEM_USE_GNUTLS

@@ -38,7 +38,6 @@ MFEM_USE_MUMPS         = @MFEM_USE_MUMPS@
 MFEM_USE_STRUMPACK     = @MFEM_USE_STRUMPACK@
 MFEM_USE_GINKGO        = @MFEM_USE_GINKGO@
 MFEM_USE_AMGX          = @MFEM_USE_AMGX@
-MFEM_USE_MAGMA         = @MFEM_USE_MAGMA@
 MFEM_USE_GNUTLS        = @MFEM_USE_GNUTLS@
 MFEM_USE_NETCDF        = @MFEM_USE_NETCDF@
 MFEM_USE_PETSC         = @MFEM_USE_PETSC@
@@ -66,7 +65,6 @@ MFEM_USE_ADFORWARD     = @MFEM_USE_ADFORWARD@
 MFEM_USE_CODIPACK      = @MFEM_USE_CODIPACK@
 MFEM_USE_BENCHMARK     = @MFEM_USE_BENCHMARK@
 MFEM_USE_PARELAG       = @MFEM_USE_PARELAG@
-MFEM_USE_TRIBOL        = @MFEM_USE_TRIBOL@
 MFEM_USE_ENZYME        = @MFEM_USE_ENZYME@

 # Compiler, compile options, and link options
@@ -40,7 +40,6 @@ option(MFEM_USE_MUMPS "Enable MUMPS usage" OFF)
 option(MFEM_USE_STRUMPACK "Enable STRUMPACK usage" OFF)
 option(MFEM_USE_GINKGO "Enable Ginkgo usage" OFF)
 option(MFEM_USE_AMGX "Enable AmgX usage" OFF)
-option(MFEM_USE_MAGMA "Enable MAGMA usage" OFF)
 option(MFEM_USE_GNUTLS "Enable GNUTLS usage" OFF)
 option(MFEM_USE_GSLIB "Enable GSLIB usage" OFF)
 option(MFEM_USE_NETCDF "Enable NETCDF usage" OFF)
@@ -68,7 +67,6 @@ option(MFEM_USE_ADFORWARD "Enable forward mode for AD" OFF)
 option(MFEM_USE_CODIPACK "Enable automatic differentiation (AD) using CoDiPack" OFF)
 option(MFEM_USE_BENCHMARK "Enable Google Benchmark" OFF)
 option(MFEM_USE_PARELAG "Enable ParELAG" OFF)
-option(MFEM_USE_TRIBOL "Enable Tribol" OFF)
 option(MFEM_USE_ENZYME "Enable Enzyme" OFF)

 # Optional overrides for autodetected MPIEXEC and MPIEXEC_NUMPROC_FLAG
@@ -184,10 +182,6 @@ set(Ginkgo_DIR "${MFEM_DIR}/../ginkgo" CACHE PATH "Path to the Ginkgo library.")

 set(AMGX_DIR "${MFEM_DIR}/../amgx" CACHE PATH "Path to AmgX")

-set(MAGMA_DIR "${MFEM_DIR}/../magma" CACHE PATH "Path to MAGMA")
-set(MAGMA_REQUIRED_PACKAGES "BLAS" "LAPACK" CACHE STRING
-    "Additional packages required by MAGMA.")
-
 set(GNUTLS_DIR "" CACHE PATH "Path to the GnuTLS library.")

 set(GSLIB_DIR "" CACHE PATH "Path to the GSLIB library.")
@@ -218,15 +212,8 @@ set(CONDUIT_DIR "${MFEM_DIR}/../conduit" CACHE PATH

 set(AXOM_DIR "${MFEM_DIR}/../axom" CACHE PATH "Path to the Axom library.")
 # May need to add "Boost" as requirement.
-if (MFEM_USE_SIDRE)
-    if (MFEM_USE_MPI)
-        set(Axom_REQUIRED_PACKAGES "Conduit/blueprint/blueprint_mpi/relay/relay_mpi" CACHE STRING
-            "Additional packages required by Axom.")
-    elseif()
-        set(Axom_REQUIRED_PACKAGES "Conduit/blueprint/relay" CACHE STRING
-            "Additional packages required by Axom.")
-    endif()
-endif()
+set(Axom_REQUIRED_PACKAGES "Conduit/relay/blueprint" CACHE STRING
+    "Additional packages required by Axom.")

 set(PUMI_DIR "${MFEM_DIR}/../pumi-2.1.0" CACHE STRING
    "Directory where PUMI is installed")
@@ -263,10 +250,6 @@ set(PARELAG_INCLUDE_DIRS "${PARELAG_DIR}/src;${PARELAG_DIR}/build/src" CACHE
 set(PARELAG_LIBRARIES "${PARELAG_DIR}/build/src/libParELAG.a" CACHE STRING
    "The ParELAG library.")

-set(TRIBOL_DIR "${MFEM_DIR}/../tribol" CACHE PATH "Path to Tribol")
-set(Tribol_REQUIRED_PACKAGES "Axom/core/mint/slam/slic" CACHE STRING
-    "Additional packages required by Tribol")
-
 set(BLAS_INCLUDE_DIRS "" CACHE STRING "Path to BLAS headers.")
 set(BLAS_LIBRARIES "" CACHE STRING "The BLAS library.")
 set(LAPACK_INCLUDE_DIRS "" CACHE STRING "Path to LAPACK headers.")
@@ -95,10 +95,6 @@ else
   # Silence unused command line argument warnings when generating dependencies
   # with mpicxx and clang
   DEP_FLAGS := -Wno-unused-command-line-argument $(DEP_FLAGS)
-   # Silence "ignoring duplicate libraries" warnings on new (Xcode 15) linker
-   ifneq (,$(findstring PROJECT:dyld,$(shell ld -v 2>&1)))
-      LDFLAGS_INTERNAL = -Xlinker -no_warn_duplicate_libraries
-   endif
 endif

 # Set CXXFLAGS to overwrite the default selection of DEBUG_FLAGS/OPTIM_FLAGS
@@ -143,7 +139,6 @@ MFEM_USE_MUMPS         = NO
 MFEM_USE_STRUMPACK     = NO
 MFEM_USE_GINKGO        = NO
 MFEM_USE_AMGX          = NO
-MFEM_USE_MAGMA         = NO
 MFEM_USE_GNUTLS        = NO
 MFEM_USE_NETCDF        = NO
 MFEM_USE_PETSC         = NO
@@ -172,21 +167,8 @@ MFEM_USE_ADFORWARD     = NO
 MFEM_USE_CODIPACK      = NO
 MFEM_USE_BENCHMARK     = NO
 MFEM_USE_PARELAG       = NO
-MFEM_USE_TRIBOL        = NO
 MFEM_USE_ENZYME        = NO

-# Process MFEM_PRECISION -> MFEM_USE_SINGLE, MFEM_USE_DOUBLE
-ifneq ($(filter double Double DOUBLE,$(MFEM_PRECISION)),)
-   MFEM_USE_DOUBLE = YES
-   MFEM_USE_SINGLE = NO
-else ifneq ($(filter single Single SINGLE,$(MFEM_PRECISION)),)
-   MFEM_USE_DOUBLE = NO
-   MFEM_USE_SINGLE = YES
-else ifeq ($(MAKECMDGOALS),config)
-   $(error Invalid floating-point precision: \
-     MFEM_PRECISION = $(MFEM_PRECISION))
-endif
-
 # MPI library compile and link flags
 # These settings are used only when building MFEM with MPI + HIP
 ifeq ($(MFEM_USE_MPI)$(MFEM_USE_HIP),YESYES)
@@ -289,13 +271,6 @@ endif
 ifeq ($(MFEM_USE_HIP),YES)
   SUNDIALS_LIB += -lsundials_nvechip
 endif
-SUNDIALS_CORE_PAT = $(subst\
- @MFEM_DIR@,$(MFEM_DIR),$(SUNDIALS_DIR))/lib*/libsundials_core.*
-ifeq ($(MFEM_USE_SUNDIALS),YES)
-   ifneq ($(wildcard $(SUNDIALS_CORE_PAT)),)
-      SUNDIALS_LIB += -lsundials_core
-   endif
-endif
 # If SUNDIALS was built with KLU:
 # MFEM_USE_SUITESPARSE = YES

@@ -343,13 +318,13 @@ MPI_FORTRAN_LIB = -lmpifort
 # MUMPS library configuration
 MUMPS_DIR = @MFEM_DIR@/../MUMPS_5.5.0
 MUMPS_OPT = -I$(MUMPS_DIR)/include
-MUMPS_LIB = $(XLINKER)-rpath,$(MUMPS_DIR)/lib -L$(MUMPS_DIR)/lib
+MUMPS_LIB = $(XLINKER)-rpath,$(MUMPS_DIR)/lib -L$(MUMPS_DIR)/lib \
+ -lmumps_common -lpord $(SCALAPACK_LIB) $(LAPACK_LIB) $(MPI_FORTRAN_LIB)
 ifeq ($(MFEM_USE_SINGLE),YES)
   MUMPS_LIB += -lsmumps
 else
   MUMPS_LIB += -ldmumps
 endif
-MUMPS_LIB += -lmumps_common -lpord $(SCALAPACK_LIB) $(LAPACK_LIB) $(MPI_FORTRAN_LIB)

 # STRUMPACK library configuration
 STRUMPACK_DIR = @MFEM_DIR@/../STRUMPACK-build
@@ -400,12 +375,7 @@ GINKGO_LIB = $(XLINKER)-rpath,$(GINKGO_LINK_LIB_DIR) -L$(GINKGO_LINK_LIB_DIR)\
 # AmgX library configuration
 AMGX_DIR = @MFEM_DIR@/../amgx
 AMGX_OPT = -I$(AMGX_DIR)/include
-AMGX_LIB = -L$(AMGX_DIR)/lib -lamgx -lcusparse -lcusolver -lcublas -lnvToolsExt
-
-# MAGMA library configuration
-MAGMA_DIR = @MFEM_DIR@/../magma
-MAGMA_OPT = -I$(MAGMA_DIR)/include
-MAGMA_LIB = -L$(MAGMA_DIR)/lib -l:libmagma.a -lcublas -lcusparse $(LAPACK_LIB)
+AMGX_LIB = -lcusparse -lcusolver -lcublas -lnvToolsExt -L$(AMGX_DIR)/lib -lamgx

 # GnuTLS library configuration
 GNUTLS_OPT =
@@ -514,11 +484,11 @@ GSLIB_LIB = -L$(GSLIB_DIR)/lib -lgs

 # CUDA library configuration
 CUDA_OPT =
-CUDA_LIB = -lcusparse -lcublas
+CUDA_LIB = -lcusparse

 # HIP library configuration
 HIP_OPT =
-HIP_LIB = -L$(HIP_DIR)/lib $(XLINKER)-rpath,$(HIP_DIR)/lib -lhipsparse -lhipblas
+HIP_LIB = -L$(HIP_DIR)/lib $(XLINKER)-rpath,$(HIP_DIR)/lib -lhipsparse

 # OCCA library configuration
 OCCA_DIR = @MFEM_DIR@/../occa
@@ -540,10 +510,8 @@ ifdef GOTCHA_DIR
 endif

 # BLITZ library configuration
-# BLITZ_DIR must be the custom installation folder (-DCMAKE_INSTALL_PREFIX).
-BLITZ_DIR = @MFEM_DIR@/../blitz/install
+BLITZ_DIR = @MFEM_DIR@/../blitz
 BLITZ_OPT = -I$(BLITZ_DIR)/include
-# On intel machines, use /lib64 instead of /lib.
 BLITZ_LIB = $(XLINKER)-rpath,$(BLITZ_DIR)/lib -L$(BLITZ_DIR)/lib -lblitz

 # ALGOIM library configuration
@@ -608,16 +576,6 @@ PARELAG_DIR = @MFEM_DIR@/../parelag
 PARELAG_OPT = -I$(PARELAG_DIR)/src -I$(PARELAG_DIR)/build/src
 PARELAG_LIB = -L$(PARELAG_DIR)/build/src -lParELAG

-# Tribol library configuration
-ifeq ($(MFEM_USE_TRIBOL),YES)
-   BASE_FLAGS = -std=c++14
-endif
-AXOM_DIR = @MFEM_DIR@/../axom
-TRIBOL_DIR = @MFEM_DIR@/../tribol
-TRIBOL_OPT = -I$(TRIBOL_DIR)/include -I$(AXOM_DIR)/include
-TRIBOL_LIB = -L$(TRIBOL_DIR)/lib -ltribol -lredecomp -L$(AXOM_DIR)/lib -laxom_mint\
-   -laxom_slam -laxom_slic -laxom_core
-
 # Enzyme configuration

 # If you want to enable automatic differentiation at compile time, use the
@@ -110,4 +110,4 @@ config-mk:

 clean:
 	rm -f $(CONFIG_HPP) $(CONFIG_MK) sample-runs-build.log
-	rm -f $(GHV) $(GHV).out $(GMV) $(GMV).out *.dSYM
+	rm -f $(GHV) $(GHV).out $(GMV) $(GMV).out
@@ -32,7 +32,7 @@ groups_serial=(
 '"examples"
   "Examples:"
   "examples"
-   "ex{,[1-9]}[0-9].cpp"'
+   "ex{,1,2,3}[0-9].cpp"'
 #   "ex1.cpp"'
 '"sundials"
   "SUNDIALS examples:"
@@ -58,10 +58,6 @@ groups_serial=(
   "HiOp examples:"
   "examples/hiop"
   "ex9.cpp"'
-'"moonolith"
-   "Moonolith examples:"
-   "examples/moonolith"
-   "ex1.cpp"'
 '"pumi"
   "PUMI examples:"
   "examples/pumi"
@@ -70,38 +66,25 @@ groups_serial=(
 '"meshing"
   "Meshing miniapps:"
   "miniapps/meshing"
-   "mobius-strip.cpp klein-bottle.cpp extruder.cpp toroid.cpp mesh-quality.cpp
-    polar-nc.cpp reflector.cpp shaper.cpp trimmer.cpp twist.cpp
+   "mobius-strip.cpp klein-bottle.cpp extruder.cpp toroid.cpp
    mesh-optimizer.cpp minimal-surface.cpp"'
 '"adjoint"
   "Adjoint miniapps:"
   "miniapps/adjoint"
   "cvsRoberts_ASAi_dns.cpp"'
-'"autodiff"
-   "Autodiff miniapps:"
-   "miniapps/autodiff"
-   "seq_example.cpp seq_test.cpp"' # 'seq_test.cpp' has no sample runs
-'"dpg"
-   "DPG miniapps:"
-   "miniapps/dpg"
-   "{acoustics,convection-diffusion,diffusion,maxwell}.cpp"'
 '"gslib"
   "GSLIB miniapps:"
   "miniapps/gslib"
   "field-diff.cpp field-interp.cpp findpts.cpp schwarz_ex1.cpp "'
-# todo: miniapps/mtop
 '"nurbs"
   "NURBS miniapps:"
   "miniapps/nurbs"
   "nurbs_ex1.cpp"'
-# todo: add other nurbs miniapps
-# todo: miniapps/solvers (serial)
 '"tools"
   "Tools miniapps:"
   "miniapps/tools"
   "convert-dc.cpp display-basis.cpp get-values.cpp load-dc.cpp
    lor-transfer.cpp"'
-# todo: add other tools miniapps
 '"toys"
   "Toys miniapps:"
   "miniapps/toys"
@@ -117,7 +100,7 @@ groups_parallel=(
 '"examples"
   "Examples:"
   "examples"
-   "ex{,[1-9]}[0-9]p.cpp"'
+   "ex{,1,2,3}[0-9]p.cpp"'
 #   "ex1p.cpp"'
 '"sundials"
   "SUNDIALS examples:"
@@ -143,10 +126,6 @@ groups_parallel=(
   "HiOp examples:"
   "examples/hiop"
   "ex9p.cpp"'
-'"moonolith"
-   "Moonolith examples:"
-   "examples/moonolith"
-   "ex{1,2}p.cpp"'
 '"pumi"
   "PUMI examples:"
   "examples/pumi"
@@ -159,41 +138,24 @@ groups_parallel=(
 '"meshing"
   "Meshing miniapps:"
   "miniapps/meshing"
-   "pmesh-optimizer.cpp pmesh-fitting.cpp pminimal-surface.cpp
-    fit-node-position.cpp"'
+   "pmesh-optimizer.cpp pmesh-fitting.cpp pminimal-surface.cpp"'
 '"electromagnetics"
   "Electromagnetics miniapps:"
   "miniapps/electromagnetics"
   "joule.cpp"'
-#   "{joule,maxwell,tesla,volta}.cpp"' # todo: multiline sample runs
+#   "{volta,tesla,joule}.cpp"' # todo: multiline sample runs
 '"adjoint"
   "Adjoint miniapps:"
   "miniapps/adjoint"
   "adjoint_advection_diffusion.cpp"'
-'"autodiff"
-   "Autodiff miniapps:"
-   "miniapps/autodiff"
-   "par_example.cpp"'
-'"dpg"
-   "DPG miniapps:"
-   "miniapps/dpg"
-   "p{acoustics,convection-diffusion,diffusion,maxwell}.cpp"'
 '"gslib"
   "GSLIB miniapps:"
   "miniapps/gslib"
   "pfindpts.cpp schwarz_ex1p.cpp"'
-'"hdiv-linear-solver"
-   "H(div) linear solver miniapps:"
-   "miniapps/hdiv-linear-solver"
-   "grad_div.cpp darcy.cpp"'
-# 'miniapps/hooke/hooke.cpp' has no sample runs
-# todo: miniapps/mtop
-# todo: miniapps/multidomain
 '"navier"
   "Navier miniapps:"
   "miniapps/navier"
   "navier_cht.cpp"'
-# todo: add other navier miniapps
 '"nurbs"
   "NURBS miniapps:"
   "miniapps/nurbs"
@@ -202,18 +164,14 @@ groups_parallel=(
   "Shifted miniapps:"
   "miniapps/shifted"
   "distance.cpp"'
-# todo: add other shifted miniapps
 '"solvers"
   "Solvers miniapps:"
   "miniapps/solvers"
   "block-solvers.cpp"'
-# todo: add other solvers miniapps
-# todo: miniapps/spde
 '"tools"
   "Tools miniapps:"
   "miniapps/tools"
-   "convert-dc.cpp get-values.cpp load-dc.cpp"'
-# todo: add other tools miniapps
+   "convert-cd.cpp get-values.cpp load-dc.cpp"'
 '"convergence"
   "Convergence tests:"
   "tests/convergence"
@@ -228,7 +186,7 @@ groups_all=(
 '"examples"
   "Examples:"
   "examples"
-   "ex\"{,[1-9]}[0-9]\"{,p}.cpp"'
+   "ex\"{,1,2,3}[0-9]\"{,p}.cpp"'
 '"sundials"
   "SUNDIALS examples:"
   "examples/sundials"
@@ -257,14 +215,10 @@ groups_all=(
   "HiOp examples:"
   "examples/hiop"
   "ex9.cpp ex9p.cpp"'
-'"moonolith"
-   "Moonolith examples:"
-   "examples/moonolith"
-   "ex1.cpp ex{1,2}p.cpp"'
 '"pumi"
   "PUMI examples:"
   "examples/pumi"
-   "ex1.cpp ex2.cpp ex1p.cpp ex6p.cpp"'
+   "ex1.cpp ex1p.cpp ex2.cpp ex6p.cpp"'
 '"superlu"
   "Superlu examples:"
   "examples/superlu"
@@ -272,67 +226,43 @@ groups_all=(
 '"meshing"
   "Meshing miniapps:"
   "miniapps/meshing"
-   "mobius-strip.cpp klein-bottle.cpp extruder.cpp toroid.cpp mesh-quality.cpp
-    polar-nc.cpp reflector.cpp shaper.cpp trimmer.cpp twist.cpp
-    {,p}mesh-optimizer.cpp pmesh-fitting.cpp {,p}minimal-surface.cpp
-    fit-node-position.cpp"'
+   "mobius-strip.cpp klein-bottle.cpp extruder.cpp toroid.cpp
+    {,p}mesh-optimizer.cpp pmesh-fitting.cpp {,p}minimal-surface.cpp"'
 '"electromagnetics"
   "Electromagnetics miniapps:"
   "miniapps/electromagnetics"
   "joule.cpp"'
-#   "{joule,maxwell,tesla,volta}.cpp"' # todo: multiline sample runs
+#   "{volta,tesla,joule}.cpp"' # todo: multiline sample runs
 '"adjoint"
   "Adjoint miniapps:"
   "miniapps/adjoint"
-   "cvsRoberts_ASAi_dns.cpp adjoint_advection_diffusion.cpp"'
-'"autodiff"
-   "Autodiff miniapps:"
-   "miniapps/autodiff"
-   "seq_example.cpp seq_test.cpp par_example.cpp"'
-# 'seq_test.cpp' has no sample runs
-'"dpg"
-   "DPG miniapps:"
-   "miniapps/dpg"
-   "{,p}{acoustics,convection-diffusion,diffusion,maxwell}.cpp"'
+   "adjoint_advection_diffusion.cpp cvsRoberts_ASAi_dns.cpp"'
 '"gslib"
   "GSLIB miniapps:"
   "miniapps/gslib"
   "field-diff.cpp field-interp.cpp findpts.cpp schwarz_ex1.cpp pfindpts.cpp
    schwarz_ex1p.cpp"'
-'"hdiv-linear-solver"
-   "H(div) linear solver miniapps:"
-   "miniapps/hdiv-linear-solver"
-   "grad_div.cpp darcy.cpp"'
-# 'miniapps/hooke/hooke.cpp' has no sample runs
-# todo: miniapps/mtop
-# todo: miniapps/multidomain
 '"navier"
   "Navier miniapps:"
   "miniapps/navier"
   "navier_cht.cpp"'
-# todo: add other navier miniapps
 '"nurbs"
   "NURBS miniapps:"
   "miniapps/nurbs"
   "nurbs_ex1.cpp nurbs_ex1p.cpp nurbs_ex11p.cpp"'
-# todo: add other nurbs miniapps
 '"shifted"
   "Shifted miniapps:"
   "miniapps/shifted"
   "distance.cpp"'
-# todo: add other shifted miniapps
 '"solvers"
   "Solvers miniapps:"
   "miniapps/solvers"
   "block-solvers.cpp"'
-# todo: add other solvers miniapps
-# todo: miniapps/spde
 '"tools"
   "Tools miniapps:"
   "miniapps/tools"
   "convert-dc.cpp display-basis.cpp get-values.cpp load-dc.cpp
    lor-transfer.cpp"'
-# todo: add other tools miniapps
 '"toys"
   "Toys miniapps:"
   "miniapps/toys"
@@ -456,7 +386,7 @@ function help_message()
      mfem_config [${mfem_config}]
         Set MFEM configuration options
      make [${make}], mpiexec [${mpiexec}], mpiexec_np [${mpiexec_np}]
-         Their values can also be set using the respective uppercase environment
+         Their values can also set using the respective uppercase environment
         variable
      mfem_build_dir [${mfem_build_dir}]
         Same as '-d': set this variable to something different from <mfem_dir>
@@ -92,5 +92,4 @@ vertices
 -0.70710678 -0.70710678
 0 -1
 0.70710678 -0.70710678
-
 mfem_mesh_end
@@ -18,9 +18,9 @@ elements
 boundary
 4
 1 1 0 1
-2 1 2 3
-3 1 3 0
-4 1 1 2
+1 1 2 3
+1 1 3 0
+1 1 1 2

 edges
 4
@@ -48,7 +48,7 @@ PROJECT_NAME           = MFEM
 # could be handy for archiving the generated documentation or if some version
 # control system is used.

-PROJECT_NUMBER         = v4.7.1
+PROJECT_NUMBER         = v4.6.1

 # Using the PROJECT_BRIEF tag one can provide an optional one line description
 # for a project that appears at the top of each page and should give viewer a
@@ -938,7 +938,6 @@ INPUT                  = @MFEM_SOURCE_DIR@/doc/CodeDocumentation.dox \
                         @MFEM_SOURCE_DIR@/config \
                         @MFEM_SOURCE_DIR@/general \
                         @MFEM_SOURCE_DIR@/linalg \
-                         @MFEM_SOURCE_DIR@/linalg/batched \
                         @MFEM_SOURCE_DIR@/linalg/simd \
                         @MFEM_SOURCE_DIR@/mesh \
                         @MFEM_SOURCE_DIR@/mesh/submesh \
@@ -988,7 +987,6 @@ INPUT                  = @MFEM_SOURCE_DIR@/doc/CodeDocumentation.dox \
                         @MFEM_SOURCE_DIR@/miniapps/solvers \
                         @MFEM_SOURCE_DIR@/miniapps/tools \
                         @MFEM_SOURCE_DIR@/miniapps/toys \
-                         @MFEM_SOURCE_DIR@/miniapps/tribol \
                         @MFEM_SOURCE_DIR@/miniapps/spde \
                         @MFEM_SOURCE_DIR@/miniapps/dpg \
                         @MFEM_SOURCE_DIR@/miniapps/dpg/util
@@ -1050,8 +1048,7 @@ RECURSIVE              = NO
 EXCLUDE                = @MFEM_SOURCE_DIR@/config/_config.hpp \
                         @MFEM_SOURCE_DIR@/config/get_hypre_version.cpp \
                         @MFEM_SOURCE_DIR@/general/tinyxml2.h \
-                         @MFEM_SOURCE_DIR@/general/tinyxml2.cpp \
-                         @MFEM_SOURCE_DIR@/linalg/lapack.hpp
+                         @MFEM_SOURCE_DIR@/general/tinyxml2.cpp

 # The EXCLUDE_SYMLINKS tag can be used to select whether or not files or
 # directories that are symbolic links (a Unix file system feature) are excluded
@@ -110,13 +110,9 @@ namespace mfem {
 * - <a class="el" href="ex35p_8cpp_source.html">Example 35p</a>: parallel multi-domain damped harmonic oscillators
 * - <a class="el" href="ex36_8cpp_source.html">Example 36</a>: Proximal Galerkin FEM for the obstacle problem
 * - <a class="el" href="ex36p_8cpp_source.html">Example 36p</a>: parallel Proximal Galerkin FEM for the obstacle problem
- * - <a class="el" href="ex37_8cpp_source.html">Example 37</a>: topology optimization
+ * - <a class="el" href="ex37_8cpp_source.html">Example 37</a>: Topology optimization
 * - <a class="el" href="ex37p_8cpp_source.html">Example 37p</a>: parallel topology optimization
 * - <a class="el" href="ex38_8cpp_source.html">Example 38</a>: cut-surface and cut-volume integration
- * - <a class="el" href="ex39_8cpp_source.html">Example 39</a>: named mesh attributes
- * - <a class="el" href="ex39p_8cpp_source.html">Example 39p</a>: parallel named mesh attributes
- * - <a class="el" href="ex40_8cpp_source.html">Example 40</a>: eikonal equation
- * - <a class="el" href="ex40p_8cpp_source.html">Example 40p</a>: parallel eikonal equation
 *
 * <H4>AmgX Examples</H4>
 * - Variants of Examples
@@ -182,21 +178,6 @@ namespace mfem {
 *   <a class="el" href="examples_2superlu_2ex1p_8cpp_source.html">1p</a>,
 *   demonstrating the use of MFEM's \link superlu.hpp SuperLU integration\endlink.
 *
- * <H4>NURBS Examples</H4>
- * - Variants of Examples
- *  <a class="el" href="nurbs__ex1_8cpp_source.html">1</a>,
- *  <a class="el" href="nurbs__ex1p_8cpp_source.html">1p</a>,
- *  <a class="el" href="nurbs__ex3_8cpp_source.html">3</a>,
- *  <a class="el" href="nurbs__ex5_8cpp_source.html">5</a>,
- *  <a class="el" href="nurbs__ex11p_8cpp_source.html">11p</a>, and
- *  <a class="el" href="nurbs__ex24_8cpp_source.html">24</a>,
- *   demonstrating howto perform NURBS-based Isogeometric Analysis.
- * - Variant of Example <a class="el" href="nurbs__patch__ex1_8cpp_source.html">1</a>: demonstrates the use of patch integration
- * - <a class="el" href="nurbs__solenoidal_8cpp_source.html">NURBS Divergence-free</a>: solve a solenoidal vector projection with NURBS-based H(div) elements
- * - <a class="el" href="nurbs__curveint_8cpp_source.html">NURBS Interpolation</a>: NURBS interpolation of given geometry
- * - <a class="el" href="nurbs__naca__cmesh_8cpp_source.html">NURBS NACA Mesher</a>: generate NURBS based mesh around a NACA foil
- * - <a class="el" href="nurbs__printfunc_8cpp_source.html">NURBS Printer</a>: print the NURBS-basis
- *
 * <H3>Miniapps</H3>
 * - <a class="el" href="volta_8cpp_source.html">Volta</a>: simple electrostatics simulation code
 * - <a class="el" href="tesla_8cpp_source.html">Tesla</a>: simple magnetostatics simulation code
@@ -233,8 +214,6 @@ namespace mfem {
 * - <a class="el" href="miniapps_2performance_2ex1_8cpp_source.html">HPC Example 1</a>: high-performance nodal H1 FEM for the Laplace problem
 * - <a class="el" href="miniapps_2performance_2ex1p_8cpp_source.html">HPC Example 1p</a>: high-performance parallel nodal H1 FEM for the Laplace problem
 * - <a class="el" href="generate__random__field_8cpp_source.html">SPDE Solvers</a>: SPDE solver random field generation
- * - <a class="el" href="contact-patch-test_8cpp_source.html">Contact</a>: mortar contact patch test for elasticity
- * - <a class="el" href="multidomain_8cpp_source.html">Multidomain miniapp</a>: Multidomain and Submesh demonstration miniapp
 * - <a class="el" href="pdiffusion_8cpp_source.html">DPG Diffusion example</a>: DPG formulation for the diffusion problem
 * - <a class="el" href="pmaxwell_8cpp_source.html">DPG Maxwell example</a>: DPG formulation for the indefinite Maxwell problem
 * - <a class="el" href="lor__elast_8cpp_source.html">LOR Elasticity</a>: solve linear elasticity with LOR preconditioning on GPUs
@@ -46,7 +46,7 @@ class DoxygenAwesomeDarkModeToggle extends HTMLElement {
            DoxygenAwesomeDarkModeToggle.onSystemPreferenceChanged()
        })
        // Update the color scheme when the tab is made visible again.
-        // It is possible that the appearance was changed in another tab
+        // It is possible that the appearance was changed in another tab 
        // while this tab was in the background.
        document.addEventListener("visibilitychange", visibilityState => {
            if (document.visibilityState === 'visible') {
@@ -97,7 +97,7 @@ class DoxygenAwesomeDarkModeToggle extends HTMLElement {
     * @returns `true` for dark-mode, `false` for light-mode user preference
     */
    static get userPreference() {
-        return (!DoxygenAwesomeDarkModeToggle.systemPreference && localStorage.getItem(DoxygenAwesomeDarkModeToggle.prefersDarkModeInLightModeKey)) ||
+        return (!DoxygenAwesomeDarkModeToggle.systemPreference && localStorage.getItem(DoxygenAwesomeDarkModeToggle.prefersDarkModeInLightModeKey)) || 
        (DoxygenAwesomeDarkModeToggle.systemPreference && !localStorage.getItem(DoxygenAwesomeDarkModeToggle.prefersLightModeInDarkModeKey))
    }

@@ -148,10 +148,10 @@ if (MFEM_ENABLE_TESTING)
  # Add CUDA/HIP tests.
  set(DEVICE_EXAMPLES
      # serial examples with device support:
-      ex1 ex3 ex4 ex5 ex6 ex9 ex14 ex22 ex24 ex25 ex26 ex34
+      ex1 ex3 ex4 ex5 ex6 ex9 ex22 ex24 ex25 ex26 ex34
      # parallel examples with device support:
-      ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex9p ex13p ex14p ex22p ex24p ex25p
-      ex26p ex34p ex35p)
+      ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex9p ex13p ex22p ex24p ex25p ex26p
+      ex34p ex35p)
  set(MFEM_TEST_DEVICE)
  if (MFEM_USE_CUDA)
    set(MFEM_TEST_DEVICE "cuda")
@@ -161,11 +161,6 @@ if (MFEM_ENABLE_TESTING)
  if (MFEM_TEST_DEVICE)
    foreach(TEST_NAME ${DEVICE_EXAMPLES})
      set(THIS_TEST_OPTIONS "-no-vis" "-d" "${MFEM_TEST_DEVICE}")
-      if (${TEST_NAME} MATCHES "ex14p")
-        list(APPEND THIS_TEST_OPTIONS "-rs" "2" "-rp" "0" "-pa")
-      elseif (${TEST_NAME} MATCHES "ex14")
-        list(APPEND THIS_TEST_OPTIONS "-r" "2" "-pa")
-      endif()
      if (NOT (${TEST_NAME} MATCHES ".*p$"))
        add_test(NAME ${TEST_NAME}_${MFEM_TEST_DEVICE}_ser
          COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
@@ -12,10 +12,11 @@
 # Use the MFEM build directory
 MFEM_DIR ?= ../..
 MFEM_BUILD_DIR ?= ../..
-MFEM_INSTALL_DIR ?= ../../mfem
 SRC = $(if $(MFEM_DIR:../..=),$(MFEM_DIR)/examples/amgx/,)
-CONFIG_MK = $(or $(wildcard $(MFEM_BUILD_DIR)/config/config.mk),\
-   $(wildcard $(MFEM_INSTALL_DIR)/share/mfem/config.mk))
+CONFIG_MK = $(MFEM_BUILD_DIR)/config/config.mk
+# Use the MFEM install directory
+# MFEM_INSTALL_DIR = ../../mfem
+# CONFIG_MK = $(MFEM_INSTALL_DIR)/share/mfem/config.mk

 MFEM_LIB_FILE = mfem_is_not_built
 -include $(CONFIG_MK)
@@ -12,10 +12,11 @@
 # Use the MFEM build directory
 MFEM_DIR ?= ../..
 MFEM_BUILD_DIR ?= ../..
-MFEM_INSTALL_DIR ?= ../../mfem
 SRC = $(if $(MFEM_DIR:../..=),$(MFEM_DIR)/examples/caliper,)
-CONFIG_MK = $(or $(wildcard $(MFEM_BUILD_DIR)/config/config.mk),\
-   $(wildcard $(MFEM_INSTALL_DIR)/share/mfem/config.mk))
+CONFIG_MK = $(MFEM_BUILD_DIR)/config/config.mk
+# Use the MFEM install directory
+# MFEM_INSTALL_DIR = ../../mfem
+# CONFIG_MK = $(MFEM_INSTALL_DIR)/share/mfem/config.mk

 MFEM_LIB_FILE = mfem_is_not_built
 -include $(CONFIG_MK)
@@ -3,14 +3,14 @@
 // Compile with: make ex10
 //
 // Sample runs:
-//    ex10 -m ../data/beam-quad.mesh -s 23 -r 2 -o 2 -dt 3
-//    ex10 -m ../data/beam-tri.mesh -s 23 -r 2 -o 2 -dt 3
-//    ex10 -m ../data/beam-hex.mesh -s 22 -r 1 -o 2 -dt 3
-//    ex10 -m ../data/beam-tet.mesh -s 22 -r 1 -o 2 -dt 3
-//    ex10 -m ../data/beam-wedge.mesh -s 22 -r 1 -o 2 -dt 3
-//    ex10 -m ../data/beam-quad.mesh -s 4 -r 2 -o 2 -dt 0.03 -vs 20
-//    ex10 -m ../data/beam-hex.mesh -s 4 -r 1 -o 2 -dt 0.05 -vs 20
-//    ex10 -m ../data/beam-quad-amr.mesh -s 23 -r 2 -o 2 -dt 3
+//    ex10 -m ../data/beam-quad.mesh -s 3 -r 2 -o 2 -dt 3
+//    ex10 -m ../data/beam-tri.mesh -s 3 -r 2 -o 2 -dt 3
+//    ex10 -m ../data/beam-hex.mesh -s 2 -r 1 -o 2 -dt 3
+//    ex10 -m ../data/beam-tet.mesh -s 2 -r 1 -o 2 -dt 3
+//    ex10 -m ../data/beam-wedge.mesh -s 2 -r 1 -o 2 -dt 3
+//    ex10 -m ../data/beam-quad.mesh -s 14 -r 2 -o 2 -dt 0.03 -vs 20
+//    ex10 -m ../data/beam-hex.mesh -s 14 -r 1 -o 2 -dt 0.05 -vs 20
+//    ex10 -m ../data/beam-quad-amr.mesh -s 3 -r 2 -o 2 -dt 3
 //
 // Description:  This examples solves a time dependent nonlinear elasticity
 //               problem of the form dv/dt = H(x) + S v, dx/dt = v, where H is a
@@ -87,16 +87,16 @@ public:
                        real_t visc, real_t mu, real_t K);

   /// Compute the right-hand side of the ODE system.
-   void Mult(const Vector &vx, Vector &dvx_dt) const override;
+   virtual void Mult(const Vector &vx, Vector &dvx_dt) const;
   /** Solve the Backward-Euler equation: k = f(x + dt*k, t), for the unknown k.
       This is the only requirement for high-order SDIRK implicit integration.*/
-   void ImplicitSolve(const real_t dt, const Vector &x, Vector &k) override;
+   virtual void ImplicitSolve(const real_t dt, const Vector &x, Vector &k);

   real_t ElasticEnergy(const Vector &x) const;
   real_t KineticEnergy(const Vector &v) const;
   void GetElasticEnergyDensity(const GridFunction &x, GridFunction &w) const;

-   ~HyperelasticOperator() override;
+   virtual ~HyperelasticOperator();
 };

 /** Nonlinear operator of the form:
@@ -120,12 +120,12 @@ public:
   void SetParameters(real_t dt_, const Vector *v_, const Vector *x_);

   /// Compute y = H(x + dt (v + dt k)) + M k + S (v + dt k).
-   void Mult(const Vector &k, Vector &y) const override;
+   virtual void Mult(const Vector &k, Vector &y) const;

   /// Compute J = M + dt S + dt^2 grad_H(x + dt (v + dt k)).
-   Operator &GetGradient(const Vector &k) const override;
+   virtual Operator &GetGradient(const Vector &k) const;

-   ~ReducedSystemOperator() override;
+   virtual ~ReducedSystemOperator();
 };


@@ -141,8 +141,8 @@ private:
 public:
   ElasticEnergyCoefficient(HyperelasticModel &m, const GridFunction &x_)
      : model(m), x(x_) { }
-   real_t Eval(ElementTransformation &T, const IntegrationPoint &ip) override;
-   ~ElasticEnergyCoefficient() override { }
+   virtual real_t Eval(ElementTransformation &T, const IntegrationPoint &ip);
+   virtual ~ElasticEnergyCoefficient() { }
 };

 void InitialDeformation(const Vector &x, Vector &y);
@@ -160,7 +160,7 @@ int main(int argc, char *argv[])
   const char *mesh_file = "../data/beam-quad.mesh";
   int ref_levels = 2;
   int order = 2;
-   int ode_solver_type = 23;
+   int ode_solver_type = 3;
   real_t t_final = 300.0;
   real_t dt = 3.0;
   real_t visc = 1e-2;
@@ -177,7 +177,11 @@ int main(int argc, char *argv[])
   args.AddOption(&order, "-o", "--order",
                  "Order (degree) of the finite elements.");
   args.AddOption(&ode_solver_type, "-s", "--ode-solver",
-                  ODESolver::Types.c_str());
+                  "ODE solver: 1 - Backward Euler, 2 - SDIRK2, 3 - SDIRK3,\n\t"
+                  "            11 - Forward Euler, 12 - RK2,\n\t"
+                  "            13 - RK3 SSP, 14 - RK4."
+                  "            22 - Implicit Midpoint Method,\n\t"
+                  "            23 - SDIRK23 (A-stable), 24 - SDIRK34");
   args.AddOption(&t_final, "-tf", "--t-final",
                  "Final time; start time is 0.");
   args.AddOption(&dt, "-dt", "--time-step",
@@ -209,7 +213,28 @@ int main(int argc, char *argv[])
   // 3. Define the ODE solver used for time integration. Several implicit
   //    singly diagonal implicit Runge-Kutta (SDIRK) methods, as well as
   //    explicit Runge-Kutta methods are available.
-   unique_ptr<ODESolver> ode_solver = ODESolver::Select(ode_solver_type);
+   ODESolver *ode_solver;
+   switch (ode_solver_type)
+   {
+      // Implicit L-stable methods
+      case 1: ode_solver = new BackwardEulerSolver; break;
+      case 2: ode_solver = new SDIRK23Solver(2); break;
+      case 3: ode_solver = new SDIRK33Solver; break;
+      // Explicit methods
+      case 11: ode_solver = new ForwardEulerSolver; break;
+      case 12: ode_solver = new RK2Solver(0.5); break; // midpoint method
+      case 13: ode_solver = new RK3SSPSolver; break;
+      case 14: ode_solver = new RK4Solver; break;
+      case 15: ode_solver = new GeneralizedAlphaSolver(0.5); break;
+      // Implicit A-stable methods (not L-stable)
+      case 22: ode_solver = new ImplicitMidpointSolver; break;
+      case 23: ode_solver = new SDIRK23Solver; break;
+      case 24: ode_solver = new SDIRK34Solver; break;
+      default:
+         cout << "Unknown ODE solver type: " << ode_solver_type << '\n';
+         delete mesh;
+         return 3;
+   }

   // 4. Refine the mesh to increase the resolution. In this example we do
   //    'ref_levels' of uniform refinement, where 'ref_levels' is a
@@ -346,6 +371,7 @@ int main(int argc, char *argv[])
   }

   // 10. Free the used memory.
+   delete ode_solver;
   delete mesh;

   return 0;
@@ -3,14 +3,14 @@
 // Compile with: make ex10p
 //
 // Sample runs:
-//    mpirun -np 4 ex10p -m ../data/beam-quad.mesh -s 23 -rs 2 -dt 3
-//    mpirun -np 4 ex10p -m ../data/beam-tri.mesh -s 23 -rs 2 -dt 3
-//    mpirun -np 4 ex10p -m ../data/beam-hex.mesh -s 22 -rs 1 -dt 3
-//    mpirun -np 4 ex10p -m ../data/beam-tet.mesh -s 22 -rs 1 -dt 3
-//    mpirun -np 4 ex10p -m ../data/beam-wedge.mesh -s 22 -rs 1 -dt 3
-//    mpirun -np 4 ex10p -m ../data/beam-quad.mesh -s 4 -rs 2 -dt 0.03 -vs 20
-//    mpirun -np 4 ex10p -m ../data/beam-hex.mesh -s 4 -rs 1 -dt 0.05 -vs 20
-//    mpirun -np 4 ex10p -m ../data/beam-quad-amr.mesh -s 23 -rs 2 -dt 3
+//    mpirun -np 4 ex10p -m ../data/beam-quad.mesh -s 3 -rs 2 -dt 3
+//    mpirun -np 4 ex10p -m ../data/beam-tri.mesh -s 3 -rs 2 -dt 3
+//    mpirun -np 4 ex10p -m ../data/beam-hex.mesh -s 2 -rs 1 -dt 3
+//    mpirun -np 4 ex10p -m ../data/beam-tet.mesh -s 2 -rs 1 -dt 3
+//    mpirun -np 4 ex10p -m ../data/beam-wedge.mesh -s 2 -rs 1 -dt 3
+//    mpirun -np 4 ex10p -m ../data/beam-quad.mesh -s 14 -rs 2 -dt 0.03 -vs 20
+//    mpirun -np 4 ex10p -m ../data/beam-hex.mesh -s 14 -rs 1 -dt 0.05 -vs 20
+//    mpirun -np 4 ex10p -m ../data/beam-quad-amr.mesh -s 3 -rs 2 -dt 3
 //
 // Description:  This examples solves a time dependent nonlinear elasticity
 //               problem of the form dv/dt = H(x) + S v, dx/dt = v, where H is a
@@ -89,17 +89,17 @@ public:
                        real_t visc, real_t mu, real_t K);

   /// Compute the right-hand side of the ODE system.
-   void Mult(const Vector &vx, Vector &dvx_dt) const override;
+   virtual void Mult(const Vector &vx, Vector &dvx_dt) const;
   /** Solve the Backward-Euler equation: k = f(x + dt*k, t), for the unknown k.
       This is the only requirement for high-order SDIRK implicit integration.*/
-   void ImplicitSolve(const real_t dt, const Vector &x, Vector &k) override;
+   virtual void ImplicitSolve(const real_t dt, const Vector &x, Vector &k);

   real_t ElasticEnergy(const ParGridFunction &x) const;
   real_t KineticEnergy(const ParGridFunction &v) const;
   void GetElasticEnergyDensity(const ParGridFunction &x,
                                ParGridFunction &w) const;

-   ~HyperelasticOperator() override;
+   virtual ~HyperelasticOperator();
 };

 /** Nonlinear operator of the form:
@@ -125,12 +125,12 @@ public:
   void SetParameters(real_t dt_, const Vector *v_, const Vector *x_);

   /// Compute y = H(x + dt (v + dt k)) + M k + S (v + dt k).
-   void Mult(const Vector &k, Vector &y) const override;
+   virtual void Mult(const Vector &k, Vector &y) const;

   /// Compute J = M + dt S + dt^2 grad_H(x + dt (v + dt k)).
-   Operator &GetGradient(const Vector &k) const override;
+   virtual Operator &GetGradient(const Vector &k) const;

-   ~ReducedSystemOperator() override;
+   virtual ~ReducedSystemOperator();
 };


@@ -146,8 +146,8 @@ private:
 public:
   ElasticEnergyCoefficient(HyperelasticModel &m, const ParGridFunction &x_)
      : model(m), x(x_) { }
-   real_t Eval(ElementTransformation &T, const IntegrationPoint &ip) override;
-   ~ElasticEnergyCoefficient() override { }
+   virtual real_t Eval(ElementTransformation &T, const IntegrationPoint &ip);
+   virtual ~ElasticEnergyCoefficient() { }
 };

 void InitialDeformation(const Vector &x, Vector &y);
@@ -172,7 +172,7 @@ int main(int argc, char *argv[])
   int ser_ref_levels = 2;
   int par_ref_levels = 0;
   int order = 2;
-   int ode_solver_type = 23;
+   int ode_solver_type = 3;
   real_t t_final = 300.0;
   real_t dt = 3.0;
   real_t visc = 1e-2;
@@ -192,7 +192,11 @@ int main(int argc, char *argv[])
   args.AddOption(&order, "-o", "--order",
                  "Order (degree) of the finite elements.");
   args.AddOption(&ode_solver_type, "-s", "--ode-solver",
-                  ODESolver::Types.c_str());
+                  "ODE solver: 1 - Backward Euler, 2 - SDIRK2, 3 - SDIRK3,\n\t"
+                  "            11 - Forward Euler, 12 - RK2,\n\t"
+                  "            13 - RK3 SSP, 14 - RK4."
+                  "            22 - Implicit Midpoint Method,\n\t"
+                  "            23 - SDIRK23 (A-stable), 24 - SDIRK34");
   args.AddOption(&t_final, "-tf", "--t-final",
                  "Final time; start time is 0.");
   args.AddOption(&dt, "-dt", "--time-step",
@@ -234,7 +238,31 @@ int main(int argc, char *argv[])
   // 4. Define the ODE solver used for time integration. Several implicit
   //    singly diagonal implicit Runge-Kutta (SDIRK) methods, as well as
   //    explicit Runge-Kutta methods are available.
-   unique_ptr<ODESolver> ode_solver = ODESolver::Select(ode_solver_type);
+   ODESolver *ode_solver;
+   switch (ode_solver_type)
+   {
+      // Implicit L-stable methods
+      case 1:  ode_solver = new BackwardEulerSolver; break;
+      case 2:  ode_solver = new SDIRK23Solver(2); break;
+      case 3:  ode_solver = new SDIRK33Solver; break;
+      // Explicit methods
+      case 11: ode_solver = new ForwardEulerSolver; break;
+      case 12: ode_solver = new RK2Solver(0.5); break; // midpoint method
+      case 13: ode_solver = new RK3SSPSolver; break;
+      case 14: ode_solver = new RK4Solver; break;
+      case 15: ode_solver = new GeneralizedAlphaSolver(0.5); break;
+      // Implicit A-stable methods (not L-stable)
+      case 22: ode_solver = new ImplicitMidpointSolver; break;
+      case 23: ode_solver = new SDIRK23Solver; break;
+      case 24: ode_solver = new SDIRK34Solver; break;
+      default:
+         if (myid == 0)
+         {
+            cout << "Unknown ODE solver type: " << ode_solver_type << '\n';
+         }
+         delete mesh;
+         return 3;
+   }

   // 5. Refine the mesh in serial to increase the resolution. In this example
   //    we do 'ser_ref_levels' of uniform refinement, where 'ser_ref_levels' is
@@ -405,6 +433,7 @@ int main(int argc, char *argv[])
   }

   // 12. Free the used memory.
+   delete ode_solver;
   delete pmesh;

   return 0;
@@ -617,7 +646,10 @@ real_t HyperelasticOperator::ElasticEnergy(const ParGridFunction &x) const

 real_t HyperelasticOperator::KineticEnergy(const ParGridFunction &v) const
 {
-   real_t energy = 0.5*M.ParInnerProduct(v, v);
+   real_t loc_energy = 0.5*M.InnerProduct(v, v);
+   real_t energy;
+   MPI_Allreduce(&loc_energy, &energy, 1, MPITypeMap<real_t>::mpi_type,
+                 MPI_SUM, fespace.GetComm());
   return energy;
 }

@@ -18,12 +18,6 @@
 //               ex14 -m ../data/amr-quad.mesh -r 3
 //               ex14 -m ../data/amr-hex.mesh
 //               ex14 -m ../data/fichera-amr.mesh
-//               ex14 -pa -r 1 -o 3
-//               ex14 -pa -r 1 -o 3 -m ../data/fichera.mesh
-//
-// Device sample runs:
-//               ex14 -pa -r 2 -d cuda -o 3
-//               ex14 -pa -r 2 -d cuda -o 3 -m ../data/fichera.mesh
 //
 // Description:  This example code demonstrates the use of MFEM to define a
 //               discontinuous Galerkin (DG) finite element discretization of
@@ -52,9 +46,7 @@ int main(int argc, char *argv[])
   real_t sigma = -1.0;
   real_t kappa = -1.0;
   real_t eta = 0.0;
-   bool pa = false;
   bool visualization = 1;
-   const char *device_config = "cpu";

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -70,13 +62,9 @@ int main(int argc, char *argv[])
                  "One of the three DG penalty parameters, should be positive."
                  " Negative values are replaced with (order+1)^2.");
   args.AddOption(&eta, "-e", "--eta", "BR2 penalty parameter.");
-   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
-                  "--no-partial-assembly", "Enable Partial Assembly.");
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
-   args.AddOption(&device_config, "-d", "--device",
-                  "Device configuration string, see Device::Configure().");
   args.Parse();
   if (!args.Good())
   {
@@ -89,129 +77,117 @@ int main(int argc, char *argv[])
   }
   args.PrintOptions(cout);

-   // 2. Enable hardware devices such as GPUs, and programming models such as
-   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
-   Device device(device_config);
-   device.Print();
-
-   // 3. Read the mesh from the given mesh file. We can handle triangular,
+   // 2. Read the mesh from the given mesh file. We can handle triangular,
   //    quadrilateral, tetrahedral and hexahedral meshes with the same code.
   //    NURBS meshes are projected to second order meshes.
-   Mesh mesh(mesh_file);
-   const int dim = mesh.Dimension();
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
+   int dim = mesh->Dimension();

-   // 4. Refine the mesh to increase the resolution. In this example we do
+   // 3. Refine the mesh to increase the resolution. In this example we do
   //    'ref_levels' of uniform refinement. By default, or if ref_levels < 0,
   //    we choose it to be the largest number that gives a final mesh with no
   //    more than 50,000 elements.
   {
      if (ref_levels < 0)
      {
-         ref_levels = (int)floor(log(50000./mesh.GetNE())/log(2.)/dim);
+         ref_levels = (int)floor(log(50000./mesh->GetNE())/log(2.)/dim);
      }
      for (int l = 0; l < ref_levels; l++)
      {
-         mesh.UniformRefinement();
+         mesh->UniformRefinement();
      }
   }
-   if (mesh.NURBSext)
+   if (mesh->NURBSext)
   {
-      mesh.SetCurvature(max(order, 1));
+      mesh->SetCurvature(max(order, 1));
   }

-   // 5. Define a finite element space on the mesh. Here we use discontinuous
+   // 4. Define a finite element space on the mesh. Here we use discontinuous
   //    finite elements of the specified order >= 0.
-   const auto bt = pa ? BasisType::GaussLobatto : BasisType::GaussLegendre;
-   DG_FECollection fec(order, dim, bt);
-   FiniteElementSpace fespace(&mesh, &fec);
-   cout << "Number of unknowns: " << fespace.GetVSize() << endl;
+   FiniteElementCollection *fec = new DG_FECollection(order, dim);
+   FiniteElementSpace *fespace = new FiniteElementSpace(mesh, fec);
+   cout << "Number of unknowns: " << fespace->GetVSize() << endl;

-   // 6. Set up the linear form b(.) which corresponds to the right-hand side of
+   // 5. Set up the linear form b(.) which corresponds to the right-hand side of
   //    the FEM linear system.
-   LinearForm b(&fespace);
+   LinearForm *b = new LinearForm(fespace);
   ConstantCoefficient one(1.0);
   ConstantCoefficient zero(0.0);
-   b.AddDomainIntegrator(new DomainLFIntegrator(one));
-   b.AddBdrFaceIntegrator(
+   b->AddDomainIntegrator(new DomainLFIntegrator(one));
+   b->AddBdrFaceIntegrator(
      new DGDirichletLFIntegrator(zero, one, sigma, kappa));
-   b.Assemble();
+   b->Assemble();

-   // 7. Define the solution vector x as a finite element grid function
+   // 6. Define the solution vector x as a finite element grid function
   //    corresponding to fespace. Initialize x with initial guess of zero.
-   GridFunction x(&fespace);
+   GridFunction x(fespace);
   x = 0.0;

-   // 8. Set up the bilinear form a(.,.) on the finite element space
+   // 7. Set up the bilinear form a(.,.) on the finite element space
   //    corresponding to the Laplacian operator -Delta, by adding the Diffusion
   //    domain integrator and the interior and boundary DG face integrators.
   //    Note that boundary conditions are imposed weakly in the form, so there
   //    is no need for dof elimination. After assembly and finalizing we
   //    extract the corresponding sparse matrix A.
-   BilinearForm a(&fespace);
-   a.AddDomainIntegrator(new DiffusionIntegrator(one));
-   a.AddInteriorFaceIntegrator(new DGDiffusionIntegrator(one, sigma, kappa));
-   a.AddBdrFaceIntegrator(new DGDiffusionIntegrator(one, sigma, kappa));
+   BilinearForm *a = new BilinearForm(fespace);
+   a->AddDomainIntegrator(new DiffusionIntegrator(one));
+   a->AddInteriorFaceIntegrator(new DGDiffusionIntegrator(one, sigma, kappa));
+   a->AddBdrFaceIntegrator(new DGDiffusionIntegrator(one, sigma, kappa));
   if (eta > 0)
   {
-      MFEM_VERIFY(!pa, "BR2 not yet compatible with partial assembly.");
-      a.AddInteriorFaceIntegrator(new DGDiffusionBR2Integrator(fespace, eta));
-      a.AddBdrFaceIntegrator(new DGDiffusionBR2Integrator(fespace, eta));
+      a->AddInteriorFaceIntegrator(new DGDiffusionBR2Integrator(*fespace, eta));
+      a->AddBdrFaceIntegrator(new DGDiffusionBR2Integrator(*fespace, eta));
   }
-   if (pa) { a.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
-   a.Assemble();
-   a.Finalize();
+   a->Assemble();
+   a->Finalize();
+   const SparseMatrix &A = a->SpMat();

-   // 9. Define a simple symmetric Gauss-Seidel preconditioner and use it to
+#ifndef MFEM_USE_SUITESPARSE
+   // 8. Define a simple symmetric Gauss-Seidel preconditioner and use it to
   //    solve the system Ax=b with PCG in the symmetric case, and GMRES in the
-   //    non-symmetric one. (Note that tolerances are squared: 1e-12 corresponds
-   //    to a relative tolerance of 1e-6).
-   //
-   //    If MFEM was compiled with SuiteSparse, use UMFPACK to solve the system.
-   if (pa)
+   //    non-symmetric one.
+   GSSmoother M(A);
+   if (sigma == -1.0)
   {
-      MFEM_VERIFY(sigma == -1.0,
-                  "The case of PA with sigma != -1 is not yet supported.");
-      CG(a, b, x, 1, 500, 1e-12, 0.0);
+      PCG(A, M, *b, x, 1, 500, 1e-12, 0.0);
   }
   else
   {
-      const SparseMatrix &A = a.SpMat();
-#ifndef MFEM_USE_SUITESPARSE
-      GSSmoother M(A);
-      if (sigma == -1.0)
-      {
-         PCG(A, M, b, x, 1, 500, 1e-12, 0.0);
-      }
-      else
-      {
-         GMRES(A, M, b, x, 1, 500, 10, 1e-12, 0.0);
-      }
-#else
-      UMFPackSolver umf_solver;
-      umf_solver.Control[UMFPACK_ORDERING] = UMFPACK_ORDERING_METIS;
-      umf_solver.SetOperator(A);
-      umf_solver.Mult(b, x);
-#endif
+      GMRES(A, M, *b, x, 1, 500, 10, 1e-12, 0.0);
   }
+#else
+   // 8. If MFEM was compiled with SuiteSparse, use UMFPACK to solve the system.
+   UMFPackSolver umf_solver;
+   umf_solver.Control[UMFPACK_ORDERING] = UMFPACK_ORDERING_METIS;
+   umf_solver.SetOperator(A);
+   umf_solver.Mult(*b, x);
+#endif

-   // 10. Save the refined mesh and the solution. This output can be viewed
-   //     later using GLVis: "glvis -m refined.mesh -g sol.gf".
+   // 9. Save the refined mesh and the solution. This output can be viewed later
+   //    using GLVis: "glvis -m refined.mesh -g sol.gf".
   ofstream mesh_ofs("refined.mesh");
   mesh_ofs.precision(8);
-   mesh.Print(mesh_ofs);
+   mesh->Print(mesh_ofs);
   ofstream sol_ofs("sol.gf");
   sol_ofs.precision(8);
   x.Save(sol_ofs);

-   // 11. Send the solution by socket to a GLVis server.
+   // 10. Send the solution by socket to a GLVis server.
   if (visualization)
   {
      char vishost[] = "localhost";
      int  visport   = 19916;
      socketstream sol_sock(vishost, visport);
      sol_sock.precision(8);
-      sol_sock << "solution\n" << mesh << x << flush;
+      sol_sock << "solution\n" << *mesh << x << flush;
   }

+   // 11. Free the used memory.
+   delete a;
+   delete b;
+   delete fespace;
+   delete fec;
+   delete mesh;
+
   return 0;
 }
@@ -17,12 +17,6 @@
 //               mpirun -np 4 ex14p -m ../data/inline-segment.mesh -rs 5
 //               mpirun -np 4 ex14p -m ../data/amr-quad.mesh -rs 3
 //               mpirun -np 4 ex14p -m ../data/amr-hex.mesh
-//               mpirun -np 4 ex14p -pa -rs 1 -rp 0 -o 3
-//               mpirun -np 4 ex14p -pa -rs 1 -rp 0 -m ../data/fichera.mesh -o 3
-//
-// Device sample runs:
-//               mpirun -np 4 ex14p -pa -rs 2 -rp 0 -d cuda -o 3
-//               mpirun -np 4 ex14p -pa -rs 2 -rp 0 -d cuda -m ../data/fichera.mesh -o 3
 //
 // Description:  This example code demonstrates the use of MFEM to define a
 //               discontinuous Galerkin (DG) finite element discretization of
@@ -44,39 +38,42 @@ using namespace mfem;

 class CustomSolverMonitor : public IterativeSolverMonitor
 {
-private:
-   const ParMesh &pmesh;
-   ParGridFunction &pgf;
 public:
-   CustomSolverMonitor(const ParMesh &pmesh_,
-                       ParGridFunction &pgf_) :
-      pmesh(pmesh_),
-      pgf(pgf_) {}
+   CustomSolverMonitor(const ParMesh *m,
+                       ParGridFunction *f) :
+      pmesh(m),
+      pgf(f) {}

-   void MonitorSolution(int i, real_t norm, const Vector &x, bool final) override
+   void MonitorSolution(int i, real_t norm, const Vector &x, bool final)
   {
      char vishost[] = "localhost";
      int  visport   = 19916;
      int  num_procs, myid;

-      MPI_Comm_size(pmesh.GetComm(), &num_procs);
-      MPI_Comm_rank(pmesh.GetComm(), &myid);
+      MPI_Comm_size(pmesh->GetComm(),&num_procs);
+      MPI_Comm_rank(pmesh->GetComm(),&myid);

-      pgf.SetFromTrueDofs(x);
+      pgf->SetFromTrueDofs(x);

      socketstream sol_sock(vishost, visport);
      sol_sock << "parallel " << num_procs << " " << myid << "\n";
      sol_sock.precision(8);
-      sol_sock << "solution\n" << pmesh << pgf
+      sol_sock << "solution\n" << *pmesh << *pgf
               << "window_title 'Iteration no " << i << "'"
               << "keys rRjlc\n" << flush;
   }
+
+private:
+   const ParMesh *pmesh;
+   ParGridFunction *pgf;
 };

 int main(int argc, char *argv[])
 {
   // 1. Initialize MPI and HYPRE.
   Mpi::Init(argc, argv);
+   int num_procs = Mpi::WorldSize();
+   int myid = Mpi::WorldRank();
   Hypre::Init();

   // 2. Parse command-line options.
@@ -87,9 +84,7 @@ int main(int argc, char *argv[])
   real_t sigma = -1.0;
   real_t kappa = -1.0;
   real_t eta = 0.0;
-   bool pa = false;
   bool visualization = 1;
-   const char *device_config = "cpu";

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -108,17 +103,13 @@ int main(int argc, char *argv[])
                  "One of the three DG penalty parameters, should be positive."
                  " Negative values are replaced with (order+1)^2.");
   args.AddOption(&eta, "-e", "--eta", "BR2 penalty parameter.");
-   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
-                  "--no-partial-assembly", "Enable Partial Assembly.");
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
-   args.AddOption(&device_config, "-d", "--device",
-                  "Device configuration string, see Device::Configure().");
   args.Parse();
   if (!args.Good())
   {
-      if (Mpi::Root())
+      if (myid == 0)
      {
         args.PrintUsage(cout);
      }
@@ -128,19 +119,16 @@ int main(int argc, char *argv[])
   {
      kappa = (order+1)*(order+1);
   }
-   if (Mpi::Root())
+   if (myid == 0)
   {
      args.PrintOptions(cout);
   }

-   Device device(device_config);
-   if (Mpi::Root()) { device.Print(); }
-
   // 3. Read the (serial) mesh from the given mesh file on all processors. We
   //    can handle triangular, quadrilateral, tetrahedral and hexahedral meshes
   //    with the same code. NURBS meshes are projected to second order meshes.
-   Mesh mesh(mesh_file);
-   int dim = mesh.Dimension();
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
+   int dim = mesh->Dimension();

   // 4. Refine the serial mesh on all processors to increase the resolution. In
   //    this example we do 'ser_ref_levels' of uniform refinement. By default,
@@ -149,54 +137,53 @@ int main(int argc, char *argv[])
   {
      if (ser_ref_levels < 0)
      {
-         ser_ref_levels = (int)floor(log(10000./mesh.GetNE())/log(2.)/dim);
+         ser_ref_levels = (int)floor(log(10000./mesh->GetNE())/log(2.)/dim);
      }
      for (int l = 0; l < ser_ref_levels; l++)
      {
-         mesh.UniformRefinement();
+         mesh->UniformRefinement();
      }
   }
-   if (mesh.NURBSext)
+   if (mesh->NURBSext)
   {
-      mesh.SetCurvature(max(order, 1));
+      mesh->SetCurvature(max(order, 1));
   }

   // 5. 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.
-   ParMesh pmesh(MPI_COMM_WORLD, mesh);
-   mesh.Clear();
+   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
+   delete mesh;
   {
      for (int l = 0; l < par_ref_levels; l++)
      {
-         pmesh.UniformRefinement();
+         pmesh->UniformRefinement();
      }
   }

   // 6. Define a parallel finite element space on the parallel mesh. Here we
   //    use discontinuous finite elements of the specified order >= 0.
-   const auto bt = pa ? BasisType::GaussLobatto : BasisType::GaussLegendre;
-   DG_FECollection fec(order, dim, bt);
-   ParFiniteElementSpace fespace(&pmesh, &fec);
-   HYPRE_BigInt size = fespace.GlobalTrueVSize();
-   if (Mpi::Root())
+   FiniteElementCollection *fec = new DG_FECollection(order, dim);
+   ParFiniteElementSpace *fespace = new ParFiniteElementSpace(pmesh, fec);
+   HYPRE_BigInt size = fespace->GlobalTrueVSize();
+   if (myid == 0)
   {
      cout << "Number of unknowns: " << size << endl;
   }

   // 7. Set up the parallel linear form b(.) which corresponds to the
   //    right-hand side of the FEM linear system.
-   ParLinearForm b(&fespace);
+   ParLinearForm *b = new ParLinearForm(fespace);
   ConstantCoefficient one(1.0);
   ConstantCoefficient zero(0.0);
-   b.AddDomainIntegrator(new DomainLFIntegrator(one));
-   b.AddBdrFaceIntegrator(
+   b->AddDomainIntegrator(new DomainLFIntegrator(one));
+   b->AddBdrFaceIntegrator(
      new DGDirichletLFIntegrator(zero, one, sigma, kappa));
-   b.Assemble();
+   b->Assemble();

   // 8. Define the solution vector x as a parallel finite element grid function
   //    corresponding to fespace. Initialize x with initial guess of zero.
-   ParGridFunction x(&fespace);
+   ParGridFunction x(fespace);
   x = 0.0;

   // 9. Set up the bilinear form a(.,.) on the finite element space
@@ -205,51 +192,42 @@ int main(int argc, char *argv[])
   //    Note that boundary conditions are imposed weakly in the form, so there
   //    is no need for dof elimination. After serial and parallel assembly we
   //    extract the corresponding parallel matrix A.
-   ParBilinearForm a(&fespace);
-   a.AddDomainIntegrator(new DiffusionIntegrator(one));
-   a.AddInteriorFaceIntegrator(new DGDiffusionIntegrator(one, sigma, kappa));
-   a.AddBdrFaceIntegrator(new DGDiffusionIntegrator(one, sigma, kappa));
+   ParBilinearForm *a = new ParBilinearForm(fespace);
+   a->AddDomainIntegrator(new DiffusionIntegrator(one));
+   a->AddInteriorFaceIntegrator(new DGDiffusionIntegrator(one, sigma, kappa));
+   a->AddBdrFaceIntegrator(new DGDiffusionIntegrator(one, sigma, kappa));
   if (eta > 0)
   {
-      MFEM_VERIFY(!pa, "BR2 not yet compatible with partial assembly.");
-      a.AddInteriorFaceIntegrator(new DGDiffusionBR2Integrator(fespace, eta));
-      a.AddBdrFaceIntegrator(new DGDiffusionBR2Integrator(fespace, eta));
+      a->AddInteriorFaceIntegrator(new DGDiffusionBR2Integrator(*fespace, eta));
+      a->AddBdrFaceIntegrator(new DGDiffusionBR2Integrator(*fespace, eta));
   }
-   if (pa) { a.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
-   a.Assemble();
-   a.Finalize();
+   a->Assemble();
+   a->Finalize();

   // 10. Define the parallel (hypre) matrix and vectors representing a(.,.),
   //     b(.) and the finite element approximation.
-   OperatorHandle A;
+   HypreParMatrix *A = a->ParallelAssemble();
+   HypreParVector *B = b->ParallelAssemble();
+   HypreParVector *X = x.ParallelProject();

-   std::unique_ptr<HypreBoomerAMG> amg;
-   if (pa)
-   {
-      A.Reset(&a, false);
-   }
-   else
-   {
-      A.SetType(Operator::Hypre_ParCSR);
-      a.ParallelAssemble(A);
-      amg.reset(new HypreBoomerAMG(*A.As<HypreParMatrix>()));
-   }
+   delete a;
+   delete b;

   // 11. Depending on the symmetry of A, define and apply a parallel PCG or
   //     GMRES solver for AX=B using the BoomerAMG preconditioner from hypre.
+   HypreSolver *amg = new HypreBoomerAMG(*A);
   if (sigma == -1.0)
   {
-      CGSolver cg(MPI_COMM_WORLD);
-      cg.SetRelTol(1e-12);
-      cg.SetMaxIter(500);
-      cg.SetPrintLevel(1);
-      cg.SetOperator(*A);
-      if (amg) { cg.SetPreconditioner(*amg); }
-      cg.Mult(b, x);
+      HyprePCG pcg(*A);
+      pcg.SetTol(1e-12);
+      pcg.SetMaxIter(500);
+      pcg.SetPrintLevel(2);
+      pcg.SetPreconditioner(*amg);
+      pcg.Mult(*B, *X);
   }
   else
   {
-      CustomSolverMonitor monitor(pmesh, x);
+      CustomSolverMonitor monitor(pmesh, &x);
      GMRESSolver gmres(MPI_COMM_WORLD);
      gmres.SetAbsTol(0.0);
      gmres.SetRelTol(1e-12);
@@ -257,37 +235,51 @@ int main(int argc, char *argv[])
      gmres.SetKDim(10);
      gmres.SetPrintLevel(1);
      gmres.SetOperator(*A);
-      if (amg) { gmres.SetPreconditioner(*amg); }
+      gmres.SetPreconditioner(*amg);
      gmres.SetMonitor(monitor);
-      gmres.Mult(b, x);
+      gmres.Mult(*B, *X);
   }
+   delete amg;

-   // 12. Save the refined mesh and the solution in parallel. This output can
+   // 12. Extract the parallel grid function corresponding to the finite element
+   //     approximation X. This is the local solution on each processor.
+   x = *X;
+
+   // 13. 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) << Mpi::WorldRank();
-      sol_name << "sol." << setfill('0') << setw(6) << Mpi::WorldRank();
+      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);
+      pmesh->Print(mesh_ofs);

      ofstream sol_ofs(sol_name.str().c_str());
      sol_ofs.precision(8);
      x.Save(sol_ofs);
   }

-   // 13. Send the solution by socket to a GLVis server.
+   // 14. 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 " << Mpi::WorldSize() << " " << Mpi::WorldRank() << "\n";
+      sol_sock << "parallel " << num_procs << " " << myid << "\n";
      sol_sock.precision(8);
-      sol_sock << "solution\n" << pmesh << x << flush;
+      sol_sock << "solution\n" << *pmesh << x << flush;
   }

+   // 15. Free the used memory.
+   delete X;
+   delete B;
+   delete A;
+
+   delete fespace;
+   delete fec;
+   delete pmesh;
+
   return 0;
 }
@@ -5,10 +5,10 @@
 // Sample runs:  ex16
 //               ex16 -m ../data/inline-tri.mesh
 //               ex16 -m ../data/disc-nurbs.mesh -tf 2
-//               ex16 -s 21 -a 0.0 -k 1.0
-//               ex16 -s 22 -a 1.0 -k 0.0
-//               ex16 -s 23 -a 0.5 -k 0.5 -o 4
-//               ex16 -s 4 -dt 1.0e-4 -tf 4.0e-2 -vs 40
+//               ex16 -s 1 -a 0.0 -k 1.0
+//               ex16 -s 2 -a 1.0 -k 0.0
+//               ex16 -s 3 -a 0.5 -k 0.5 -o 4
+//               ex16 -s 14 -dt 1.0e-4 -tf 4.0e-2 -vs 40
 //               ex16 -m ../data/fichera-q2.mesh
 //               ex16 -m ../data/fichera-mixed.mesh
 //               ex16 -m ../data/escher.mesh
@@ -76,15 +76,15 @@ public:
   ConductionOperator(FiniteElementSpace &f, real_t alpha, real_t kappa,
                      const Vector &u);

-   void Mult(const Vector &u, Vector &du_dt) const override;
+   virtual void Mult(const Vector &u, Vector &du_dt) const;
   /** Solve the Backward-Euler equation: k = f(u + dt*k, t), for the unknown k.
       This is the only requirement for high-order SDIRK implicit integration.*/
-   void ImplicitSolve(const real_t dt, const Vector &u, Vector &k) override;
+   virtual void ImplicitSolve(const real_t dt, const Vector &u, Vector &k);

   /// Update the diffusion BilinearForm K using the given true-dof vector `u`.
   void SetParameters(const Vector &u);

-   ~ConductionOperator() override;
+   virtual ~ConductionOperator();
 };

 real_t InitialTemperature(const Vector &x);
@@ -95,13 +95,11 @@ int main(int argc, char *argv[])
   const char *mesh_file = "../data/star.mesh";
   int ref_levels = 2;
   int order = 2;
-
-   int ode_solver_type = 23;  // SDIRK33Solver
+   int ode_solver_type = 3;
   real_t t_final = 0.5;
   real_t dt = 1.0e-2;
   real_t alpha = 1.0e-2;
   real_t kappa = 0.5;
-
   bool visualization = true;
   bool visit = false;
   int vis_steps = 5;
@@ -117,7 +115,8 @@ int main(int argc, char *argv[])
   args.AddOption(&order, "-o", "--order",
                  "Order (degree) of the finite elements.");
   args.AddOption(&ode_solver_type, "-s", "--ode-solver",
-                  ODESolver::Types.c_str());
+                  "ODE solver: 1 - Backward Euler, 2 - SDIRK2, 3 - SDIRK3,\n\t"
+                  "\t   11 - Forward Euler, 12 - RK2, 13 - RK3 SSP, 14 - RK4.");
   args.AddOption(&t_final, "-tf", "--t-final",
                  "Final time; start time is 0.");
   args.AddOption(&dt, "-dt", "--time-step",
@@ -150,7 +149,28 @@ int main(int argc, char *argv[])
   // 3. Define the ODE solver used for time integration. Several implicit
   //    singly diagonal implicit Runge-Kutta (SDIRK) methods, as well as
   //    explicit Runge-Kutta methods are available.
-   unique_ptr<ODESolver> ode_solver = ODESolver::Select(ode_solver_type);
+   ODESolver *ode_solver;
+   switch (ode_solver_type)
+   {
+      // Implicit L-stable methods
+      case 1:  ode_solver = new BackwardEulerSolver; break;
+      case 2:  ode_solver = new SDIRK23Solver(2); break;
+      case 3:  ode_solver = new SDIRK33Solver; break;
+      // Explicit methods
+      case 11: ode_solver = new ForwardEulerSolver; break;
+      case 12: ode_solver = new RK2Solver(0.5); break; // midpoint method
+      case 13: ode_solver = new RK3SSPSolver; break;
+      case 14: ode_solver = new RK4Solver; break;
+      case 15: ode_solver = new GeneralizedAlphaSolver(0.5); break;
+      // Implicit A-stable methods (not L-stable)
+      case 22: ode_solver = new ImplicitMidpointSolver; break;
+      case 23: ode_solver = new SDIRK23Solver; break;
+      case 24: ode_solver = new SDIRK34Solver; break;
+      default:
+         cout << "Unknown ODE solver type: " << ode_solver_type << '\n';
+         delete mesh;
+         return 3;
+   }

   // 4. Refine the mesh to increase the resolution. In this example we do
   //    'ref_levels' of uniform refinement, where 'ref_levels' is a
@@ -267,6 +287,7 @@ int main(int argc, char *argv[])
   }

   // 10. Free the used memory.
+   delete ode_solver;
   delete mesh;

   return 0;
@@ -5,10 +5,10 @@
 // Sample runs:  mpirun -np 4 ex16p
 //               mpirun -np 4 ex16p -m ../data/inline-tri.mesh
 //               mpirun -np 4 ex16p -m ../data/disc-nurbs.mesh -tf 2
-//               mpirun -np 4 ex16p -s 21 -a 0.0 -k 1.0
-//               mpirun -np 4 ex16p -s 22 -a 1.0 -k 0.0
-//               mpirun -np 8 ex16p -s 23 -a 0.5 -k 0.5 -o 4
-//               mpirun -np 4 ex16p -s 4 -dt 1.0e-4 -tf 4.0e-2 -vs 40
+//               mpirun -np 4 ex16p -s 1 -a 0.0 -k 1.0
+//               mpirun -np 4 ex16p -s 2 -a 1.0 -k 0.0
+//               mpirun -np 8 ex16p -s 3 -a 0.5 -k 0.5 -o 4
+//               mpirun -np 4 ex16p -s 14 -dt 1.0e-4 -tf 4.0e-2 -vs 40
 //               mpirun -np 16 ex16p -m ../data/fichera-q2.mesh
 //               mpirun -np 16 ex16p -m ../data/fichera-mixed.mesh
 //               mpirun -np 16 ex16p -m ../data/escher-p2.mesh
@@ -78,15 +78,15 @@ public:
   ConductionOperator(ParFiniteElementSpace &f, real_t alpha, real_t kappa,
                      const Vector &u);

-   void Mult(const Vector &u, Vector &du_dt) const override;
+   virtual void Mult(const Vector &u, Vector &du_dt) const;
   /** Solve the Backward-Euler equation: k = f(u + dt*k, t), for the unknown k.
       This is the only requirement for high-order SDIRK implicit integration.*/
-   void ImplicitSolve(const real_t dt, const Vector &u, Vector &k) override;
+   virtual void ImplicitSolve(const real_t dt, const Vector &u, Vector &k);

   /// Update the diffusion BilinearForm K using the given true-dof vector `u`.
   void SetParameters(const Vector &u);

-   ~ConductionOperator() override;
+   virtual ~ConductionOperator();
 };

 real_t InitialTemperature(const Vector &x);
@@ -104,13 +104,11 @@ int main(int argc, char *argv[])
   int ser_ref_levels = 2;
   int par_ref_levels = 1;
   int order = 2;
-
-   int ode_solver_type = 23;  // SDIRK33Solver
+   int ode_solver_type = 3;
   real_t t_final = 0.5;
   real_t dt = 1.0e-2;
   real_t alpha = 1.0e-2;
   real_t kappa = 0.5;
-
   bool visualization = true;
   bool visit = false;
   int vis_steps = 5;
@@ -129,7 +127,8 @@ int main(int argc, char *argv[])
   args.AddOption(&order, "-o", "--order",
                  "Order (degree) of the finite elements.");
   args.AddOption(&ode_solver_type, "-s", "--ode-solver",
-                  ODESolver::Types.c_str());
+                  "ODE solver: 1 - Backward Euler, 2 - SDIRK2, 3 - SDIRK3,\n\t"
+                  "\t   11 - Forward Euler, 12 - RK2, 13 - RK3 SSP, 14 - RK4.");
   args.AddOption(&t_final, "-tf", "--t-final",
                  "Final time; start time is 0.");
   args.AddOption(&dt, "-dt", "--time-step",
@@ -170,7 +169,28 @@ int main(int argc, char *argv[])
   // 4. Define the ODE solver used for time integration. Several implicit
   //    singly diagonal implicit Runge-Kutta (SDIRK) methods, as well as
   //    explicit Runge-Kutta methods are available.
-   unique_ptr<ODESolver> ode_solver = ODESolver::Select(ode_solver_type);
+   ODESolver *ode_solver;
+   switch (ode_solver_type)
+   {
+      // Implicit L-stable methods
+      case 1:  ode_solver = new BackwardEulerSolver; break;
+      case 2:  ode_solver = new SDIRK23Solver(2); break;
+      case 3:  ode_solver = new SDIRK33Solver; break;
+      // Explicit methods
+      case 11: ode_solver = new ForwardEulerSolver; break;
+      case 12: ode_solver = new RK2Solver(0.5); break; // midpoint method
+      case 13: ode_solver = new RK3SSPSolver; break;
+      case 14: ode_solver = new RK4Solver; break;
+      case 15: ode_solver = new GeneralizedAlphaSolver(0.5); break;
+      // Implicit A-stable methods (not L-stable)
+      case 22: ode_solver = new ImplicitMidpointSolver; break;
+      case 23: ode_solver = new SDIRK23Solver; break;
+      case 24: ode_solver = new SDIRK34Solver; break;
+      default:
+         cout << "Unknown ODE solver type: " << ode_solver_type << '\n';
+         delete mesh;
+         return 3;
+   }

   // 5. Refine the mesh in serial to increase the resolution. In this example
   //    we do 'ser_ref_levels' of uniform refinement, where 'ser_ref_levels' is
@@ -356,6 +376,7 @@ int main(int argc, char *argv[])
   }

   // 12. Free the used memory.
+   delete ode_solver;
   delete pmesh;

   return 0;
@@ -69,7 +69,7 @@ public:
   void SetDisplacement(GridFunction &u_) { u = &u_; }
   void SetComponent(int i, int j) { si = i; sj = j; }

-   real_t Eval(ElementTransformation &T, const IntegrationPoint &ip) override;
+   virtual real_t Eval(ElementTransformation &T, const IntegrationPoint &ip);
 };

 // Simple GLVis visualization manager.
@@ -89,7 +89,7 @@ public:
   void NewWindow();
   void CloseConnection();
   void PositionWindow();
-   ~VisMan() override;
+   virtual ~VisMan();
 };

 // Manipulators for the GLVis visualization manager.
@@ -69,7 +69,7 @@ public:
   void SetDisplacement(GridFunction &u_) { u = &u_; }
   void SetComponent(int i, int j) { si = i; sj = j; }

-   real_t Eval(ElementTransformation &T, const IntegrationPoint &ip) override;
+   virtual real_t Eval(ElementTransformation &T, const IntegrationPoint &ip);
 };

 // Simple GLVis visualization manager.
@@ -89,7 +89,7 @@ public:
   void NewWindow();
   void CloseConnection();
   void PositionWindow();
-   ~VisMan() override;
+   virtual ~VisMan();
 };

 // Manipulators for the GLVis visualization manager.
@@ -90,7 +90,8 @@ int main(int argc, char *argv[])
   args.AddOption(&order, "-o", "--order",
                  "Order (degree) of the finite elements.");
   args.AddOption(&ode_solver_type, "-s", "--ode-solver",
-                  ODESolver::ExplicitTypes.c_str());
+                  "ODE solver: 1 - Forward Euler,\n\t"
+                  "            2 - RK2 SSP, 3 - RK3 SSP, 4 - RK4, 6 - RK6.");
   args.AddOption(&t_final, "-tf", "--t-final", "Final time; start time is 0.");
   args.AddOption(&dt, "-dt", "--time-step",
                  "Time step. Positive number skips CFL timestep calculation.");
@@ -124,7 +125,18 @@ int main(int argc, char *argv[])

   // 3. Define the ODE solver used for time integration. Several explicit
   //    Runge-Kutta methods are available.
-   unique_ptr<ODESolver> ode_solver = ODESolver::SelectExplicit(ode_solver_type);
+   ODESolver *ode_solver = NULL;
+   switch (ode_solver_type)
+   {
+      case 1: ode_solver = new ForwardEulerSolver; break;
+      case 2: ode_solver = new RK2Solver(1.0); break;
+      case 3: ode_solver = new RK3SSPSolver; break;
+      case 4: ode_solver = new RK4Solver; break;
+      case 6: ode_solver = new RK6Solver; break;
+      default:
+         cout << "Unknown ODE solver type: " << ode_solver_type << '\n';
+         return 3;
+   }

   // 4. Define the discontinuous DG finite element space of the given
   //    polynomial order on the refined mesh.
@@ -292,5 +304,8 @@ int main(int argc, char *argv[])
   const real_t error = sol.ComputeLpError(2, u0);
   cout << "Solution error: " << error << endl;

+   // Free the used memory.
+   delete ode_solver;
+
   return 0;
 }
@@ -39,8 +39,8 @@ private:
   // Base Nonlinear Form
   std::unique_ptr<NonlinearForm> nonlinearForm;
   // element-wise inverse mass matrix
-   std::vector<DenseMatrix> invmass; // local scalar inverse mass
-   std::vector<DenseMatrix> weakdiv; // local weak divergence (trial space ByDim)
+   std::vector<DenseMatrix> invmass; // local scalar inverse mass.
+   std::vector<DenseMatrix> weakdiv; // local weakdivergence. Trial space is ByDim.
   // global maximum characteristic speed. Updated by form integrators
   mutable real_t max_char_speed;
   // auxiliary variable used in Mult
@@ -169,9 +169,9 @@ void DGHyperbolicConservationLaws::Mult(const Vector &x, Vector &y) const
 {
   // 0. Reset wavespeed computation before operator application.
   formIntegrator->ResetMaxCharSpeed();
-   // 1. Apply Nonlinear form to obtain an auxiliary result
+   // 1. Apply Nonlinear form to obtain an axiliary result
   //         z = - <F̂(u_h,n), [[v]]>_e
-   //    If weak-divergence is not preassembled, we also have weak-divergence
+   //    If weak-divergencee is not preassembled, we also have weak-divergence
   //         z = - <F̂(u_h,n), [[v]]>_e + (F(u_h), ∇v)
   nonlinearForm->Mult(x, z);
   if (!weakdiv.empty()) // if weak divergence is pre-assembled
@@ -99,7 +99,8 @@ int main(int argc, char *argv[])
   args.AddOption(&order, "-o", "--order",
                  "Order (degree) of the finite elements.");
   args.AddOption(&ode_solver_type, "-s", "--ode-solver",
-                  ODESolver::ExplicitTypes.c_str());
+                  "ODE solver: 1 - Forward Euler,\n\t"
+                  "            2 - RK2 SSP, 3 - RK3 SSP, 4 - RK4, 6 - RK6.");
   args.AddOption(&t_final, "-tf", "--t-final", "Final time; start time is 0.");
   args.AddOption(&dt, "-dt", "--time-step",
                  "Time step. Positive number skips CFL timestep calculation.");
@@ -147,7 +148,18 @@ int main(int argc, char *argv[])

   // 3. Define the ODE solver used for time integration. Several explicit
   //    Runge-Kutta methods are available.
-   unique_ptr<ODESolver> ode_solver = ODESolver::SelectExplicit(ode_solver_type);
+   ODESolver *ode_solver = NULL;
+   switch (ode_solver_type)
+   {
+      case 1: ode_solver = new ForwardEulerSolver; break;
+      case 2: ode_solver = new RK2Solver(1.0); break;
+      case 3: ode_solver = new RK3SSPSolver; break;
+      case 4: ode_solver = new RK4Solver; break;
+      case 6: ode_solver = new RK6Solver; break;
+      default:
+         cout << "Unknown ODE solver type: " << ode_solver_type << '\n';
+         return 3;
+   }

   // 4. Define the discontinuous DG finite element space of the given
   //    polynomial order on the refined mesh.
@@ -348,5 +360,8 @@ int main(int argc, char *argv[])
      cout << "Solution error: " << error << endl;
   }

+   // Free the used memory.
+   delete ode_solver;
+
   return 0;
 }
@@ -48,7 +48,7 @@ public:
      print_level = print_lvl;
   }

-   void MonitorResidual(int it, real_t norm, const Vector &r, bool final) override;
+   virtual void MonitorResidual(int it, real_t norm, const Vector &r, bool final);

 private:
   const std::string prefix;
@@ -116,10 +116,10 @@ public:
   JacobianPreconditioner(Array<FiniteElementSpace *> &fes,
                          SparseMatrix &mass, Array<int> &offsets);

-   void Mult(const Vector &k, Vector &y) const override;
-   void SetOperator(const Operator &op) override;
+   virtual void Mult(const Vector &k, Vector &y) const;
+   virtual void SetOperator(const Operator &op);

-   ~JacobianPreconditioner() override;
+   virtual ~JacobianPreconditioner();
 };

 // After spatial discretization, the rubber model can be written as:
@@ -161,13 +161,13 @@ public:
                  int iter, Coefficient &mu);

   // Required to use the native newton solver
-   Operator &GetGradient(const Vector &xp) const override;
-   void Mult(const Vector &k, Vector &y) const override;
+   virtual Operator &GetGradient(const Vector &xp) const;
+   virtual void Mult(const Vector &k, Vector &y) const;

   // Driver for the newton solver
   void Solve(Vector &xp) const;

-   ~RubberOperator() override;
+   virtual ~RubberOperator();
 };

 // Visualization driver
@@ -62,7 +62,7 @@ public:
 #endif
   }

-   void MonitorResidual(int it, real_t norm, const Vector &r, bool final) override;
+   virtual void MonitorResidual(int it, real_t norm, const Vector &r, bool final);

 private:
   const std::string prefix;
@@ -130,10 +130,10 @@ public:
   JacobianPreconditioner(Array<ParFiniteElementSpace *> &fes,
                          Operator &mass, Array<int> &offsets);

-   void Mult(const Vector &k, Vector &y) const override;
-   void SetOperator(const Operator &op) override;
+   virtual void Mult(const Vector &k, Vector &y) const;
+   virtual void SetOperator(const Operator &op);

-   ~JacobianPreconditioner() override;
+   virtual ~JacobianPreconditioner();
 };

 // After spatial discretization, the rubber model can be written as:
@@ -175,13 +175,13 @@ public:
                  int iter, Coefficient &mu);

   // Required to use the native newton solver
-   Operator &GetGradient(const Vector &xp) const override;
-   void Mult(const Vector &k, Vector &y) const override;
+   virtual Operator &GetGradient(const Vector &xp) const;
+   virtual void Mult(const Vector &k, Vector &y) const;

   // Driver for the newton solver
   void Solve(Vector &xp) const;

-   ~RubberOperator() override;
+   virtual ~RubberOperator();
 };

 // Visualization driver
@@ -79,14 +79,14 @@ class GradT : public Operator
 {
 public:
   GradT() : Operator(1) {}
-   void Mult(const Vector &x, Vector &y) const override { y.Set(1.0/m_, x); }
+   void Mult(const Vector &x, Vector &y) const { y.Set(1.0/m_, x); }
 };

 class NegGradV : public TimeDependentOperator
 {
 public:
   NegGradV() : TimeDependentOperator(1) {}
-   void Mult(const Vector &x, Vector &y) const override;
+   void Mult(const Vector &x, Vector &y) const;
 };

 int main(int argc, char *argv[])
@@ -84,14 +84,14 @@ class GradT : public Operator
 {
 public:
   GradT() : Operator(1) {}
-   void Mult(const Vector &x, Vector &y) const override { y.Set(1.0/m_, x); }
+   void Mult(const Vector &x, Vector &y) const { y.Set(1.0/m_, x); }
 };

 class NegGradV : public TimeDependentOperator
 {
 public:
   NegGradV() : TimeDependentOperator(1) {}
-   void Mult(const Vector &x, Vector &y) const override;
+   void Mult(const Vector &x, Vector &y) const;
 };

 int main(int argc, char *argv[])
@@ -44,7 +44,7 @@ protected:
   BilinearForm *M;
   BilinearForm *K;

-   SparseMatrix Mmat, Kmat;
+   SparseMatrix Mmat, Kmat, Kmat0;
   SparseMatrix *T; // T = M + dt K
   real_t current_dt;

@@ -61,20 +61,20 @@ public:
   WaveOperator(FiniteElementSpace &f, Array<int> &ess_bdr, real_t speed);

   using SecondOrderTimeDependentOperator::Mult;
-   void Mult(const Vector &u, const Vector &du_dt,
-             Vector &d2udt2) const override;
+   virtual void Mult(const Vector &u, const Vector &du_dt,
+                     Vector &d2udt2) const;

   /** Solve the Backward-Euler equation:
       d2udt2 = f(u + fac0*d2udt2,dudt + fac1*d2udt2, t),
       for the unknown d2udt2. */
   using SecondOrderTimeDependentOperator::ImplicitSolve;
-   void ImplicitSolve(const real_t fac0, const real_t fac1,
-                      const Vector &u, const Vector &dudt, Vector &d2udt2) override;
+   virtual void ImplicitSolve(const real_t fac0, const real_t fac1,
+                              const Vector &u, const Vector &dudt, Vector &d2udt2);

   ///
   void SetParameters(const Vector &u);

-   ~WaveOperator() override;
+   virtual ~WaveOperator();
 };


@@ -83,24 +83,25 @@ WaveOperator::WaveOperator(FiniteElementSpace &f,
   : SecondOrderTimeDependentOperator(f.GetTrueVSize(), (real_t) 0.0),
     fespace(f), M(NULL), K(NULL), T(NULL), current_dt(0.0), z(height)
 {
-   // Assemble Laplace matrix
+   const real_t rel_tol = 1e-8;
+
+   fespace.GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+
   c2 = new ConstantCoefficient(speed*speed);
+
   K = new BilinearForm(&fespace);
   K->AddDomainIntegrator(new DiffusionIntegrator(*c2));
   K->Assemble();

-   // Assemble Mass matrix
+   Array<int> dummy;
+   K->FormSystemMatrix(dummy, Kmat0);
+   K->FormSystemMatrix(ess_tdof_list, Kmat);
+
   M = new BilinearForm(&fespace);
   M->AddDomainIntegrator(new MassIntegrator());
   M->Assemble();
-
-   // Apply Bcs
-   fespace.GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
-   K->FormSystemMatrix(ess_tdof_list, Kmat);
   M->FormSystemMatrix(ess_tdof_list, Mmat);

-   // Configure preconditioner
-   const real_t rel_tol = 1e-8;
   M_solver.iterative_mode = false;
   M_solver.SetRelTol(rel_tol);
   M_solver.SetAbsTol(0.0);
@@ -109,13 +110,14 @@ WaveOperator::WaveOperator(FiniteElementSpace &f,
   M_solver.SetPreconditioner(M_prec);
   M_solver.SetOperator(Mmat);

-   // Configure solver
   T_solver.iterative_mode = false;
   T_solver.SetRelTol(rel_tol);
   T_solver.SetAbsTol(0.0);
   T_solver.SetMaxIter(100);
   T_solver.SetPrintLevel(0);
   T_solver.SetPreconditioner(T_prec);
+
+   T = NULL;
 }

 void WaveOperator::Mult(const Vector &u, const Vector &du_dt,
@@ -124,11 +126,9 @@ void WaveOperator::Mult(const Vector &u, const Vector &du_dt,
   // Compute:
   //    d2udt2 = M^{-1}*-K(u)
   // for d2udt2
-   K->FullMult(u, z);
+   Kmat.Mult(u, z);
   z.Neg(); // z = -z
-   z.SetSubVector(ess_tdof_list, 0.0);
   M_solver.Mult(z, d2udt2);
-   d2udt2.SetSubVector(ess_tdof_list, 0.0);
 }

 void WaveOperator::ImplicitSolve(const real_t fac0, const real_t fac1,
@@ -142,11 +142,14 @@ void WaveOperator::ImplicitSolve(const real_t fac0, const real_t fac1,
      T = Add(1.0, Mmat, fac0, Kmat);
      T_solver.SetOperator(*T);
   }
-   K->FullMult(u, z);
+   Kmat0.Mult(u, z);
   z.Neg();
-   z.SetSubVector(ess_tdof_list, 0.0);
+
+   for (int i = 0; i < ess_tdof_list.Size(); i++)
+   {
+      z[ess_tdof_list[i]] = 0.0;
+   }
   T_solver.Mult(z, d2udt2);
-   d2udt2.SetSubVector(ess_tdof_list, 0.0);
 }

 void WaveOperator::SetParameters(const Vector &u)
@@ -201,7 +204,9 @@ int main(int argc, char *argv[])
   args.AddOption(&order, "-o", "--order",
                  "Order (degree) of the finite elements.");
   args.AddOption(&ode_solver_type, "-s", "--ode-solver",
-                  SecondOrderODESolver::Types.c_str());
+                  "ODE solver: [0--10] - GeneralizedAlpha(0.1 * s),\n\t"
+                  "\t   11 - Average Acceleration, 12 - Linear Acceleration\n"
+                  "\t   13 - CentralDifference, 14 - FoxGoodwin");
   args.AddOption(&t_final, "-tf", "--t-final",
                  "Final time; start time is 0.");
   args.AddOption(&dt, "-dt", "--time-step",
@@ -236,7 +241,32 @@ int main(int argc, char *argv[])

   // 3. Define the ODE solver used for time integration. Several second order
   //    time integrators are available.
-   SecondOrderODESolver *ode_solver= SecondOrderODESolver::Select(ode_solver_type);
+   SecondOrderODESolver *ode_solver;
+   switch (ode_solver_type)
+   {
+      // Implicit methods
+      case 0: ode_solver = new GeneralizedAlpha2Solver(0.0); break;
+      case 1: ode_solver = new GeneralizedAlpha2Solver(0.1); break;
+      case 2: ode_solver = new GeneralizedAlpha2Solver(0.2); break;
+      case 3: ode_solver = new GeneralizedAlpha2Solver(0.3); break;
+      case 4: ode_solver = new GeneralizedAlpha2Solver(0.4); break;
+      case 5: ode_solver = new GeneralizedAlpha2Solver(0.5); break;
+      case 6: ode_solver = new GeneralizedAlpha2Solver(0.6); break;
+      case 7: ode_solver = new GeneralizedAlpha2Solver(0.7); break;
+      case 8: ode_solver = new GeneralizedAlpha2Solver(0.8); break;
+      case 9: ode_solver = new GeneralizedAlpha2Solver(0.9); break;
+      case 10: ode_solver = new GeneralizedAlpha2Solver(1.0); break;
+
+      case 11: ode_solver = new AverageAccelerationSolver(); break;
+      case 12: ode_solver = new LinearAccelerationSolver(); break;
+      case 13: ode_solver = new CentralDifferenceSolver(); break;
+      case 14: ode_solver = new FoxGoodwinSolver(); break;
+
+      default:
+         cout << "Unknown ODE solver type: " << ode_solver_type << '\n';
+         delete mesh;
+         return 3;
+   }

   // 4. Refine the mesh to increase the resolution. In this example we do
   //    'ref_levels' of uniform refinement, where 'ref_levels' is a
@@ -284,6 +314,7 @@ int main(int argc, char *argv[])
         ess_bdr = 0;
      }
   }
+
   WaveOperator oper(fespace, ess_bdr, speed);

   u_gf.SetFromTrueDofs(u);
@@ -103,8 +103,8 @@ public:

   using VectorCoefficient::Eval;

-   void Eval(Vector &K, ElementTransformation &T,
-             const IntegrationPoint &ip) override
+   virtual void Eval(Vector &K, ElementTransformation &T,
+                     const IntegrationPoint &ip)
   {
      real_t x[3];
      Vector transip(x, 3);
@@ -102,8 +102,8 @@ public:

   using VectorCoefficient::Eval;

-   void Eval(Vector &K, ElementTransformation &T,
-             const IntegrationPoint &ip) override
+   virtual void Eval(Vector &K, ElementTransformation &T,
+                     const IntegrationPoint &ip)
   {
      real_t x[3];
      Vector transip(x, 3);
@@ -58,7 +58,7 @@ public:
      }
   }

-   ~DiffusionMultigrid() override
+   virtual ~DiffusionMultigrid()
   {
      delete amg;
   }
@@ -19,11 +19,8 @@
 //    ex33 -m ../data/amr-quad.mesh -ver -alpha 2.6 -o 2 -r 2
 //    ex33 -m ../data/inline-hex.mesh -ver -alpha 0.3 -o 2 -r 1
 //
-//  Note: The manufactured solution used in this problem is
-//
-//            u = ∏_{i=0}^{dim-1} sin(π x_i) ,
-//
-//        regardless of the value of alpha.
+//  Note: the analytic solution to this problem is u = ∏_{i=0}^{dim-1} sin(π x_i)
+//        for all alpha.
 //
 // Description:
 //
@@ -117,8 +114,7 @@ int main(int argc, char *argv[])
                  "Enable or disable GLVis visualization.");
   args.AddOption(&verification, "-ver", "--verification", "-no-ver",
                  "--no-verification",
-                  "Use sinusoidal function (f) for manufactured "
-                  "solution test.");
+                  "Use sinusoidal function (f) for analytic comparison.");
   args.Parse();
   if (!args.Good())
   {
@@ -167,7 +163,7 @@ int main(int argc, char *argv[])
   // 5. Define a finite element space on the mesh.
   H1_FECollection fec(order, dim);
   FiniteElementSpace fespace(&mesh, &fec);
-   cout << "Number of degrees of freedom: "
+   cout << "Number of finite element unknowns: "
        << fespace.GetTrueVSize() << endl;

   // 6. Determine the list of true (i.e. conforming) essential boundary dofs.
@@ -383,29 +379,29 @@ int main(int argc, char *argv[])
      FunctionCoefficient sol(solution);
      real_t l2_error = u.ComputeL2Error(sol);

-      string manufactured_solution,expected_mesh;
+      string analytic_solution,expected_mesh;
      switch (dim)
      {
         case 1:
-            manufactured_solution = "sin(π x)";
+            analytic_solution = "sin(π x)";
            expected_mesh = "inline_segment.mesh";
            break;
         case 2:
-            manufactured_solution = "sin(π x) sin(π y)";
+            analytic_solution = "sin(π x) sin(π y)";
            expected_mesh = "inline_quad.mesh";
            break;
         default:
-            manufactured_solution = "sin(π x) sin(π y) sin(π z)";
+            analytic_solution = "sin(π x) sin(π y) sin(π z)";
            expected_mesh = "inline_hex.mesh";
            break;
      }

      mfem::out << "\n" << string(80,'=')
                << "\n\nSolution Verification in "<< dim << "D \n\n"
-                << "Manufactured solution : " << manufactured_solution << "\n"
-                << "Expected mesh         : " << expected_mesh <<"\n"
-                << "Your mesh             : " << mesh_file << "\n"
-                << "L2 error              : " << l2_error << "\n\n"
+                << "Analytic solution : " << analytic_solution << "\n"
+                << "Expected mesh     : " << expected_mesh <<"\n"
+                << "Your mesh         : " << mesh_file << "\n"
+                << "L2 error          : " << l2_error << "\n\n"
                << string(80,'=') << endl;
   }

@@ -131,7 +131,7 @@ void RationalApproximation_AAA(const Vector &val, const Vector &pt,
      }

 #ifdef MFEM_USE_LAPACK
-      DenseMatrixSVD svd(Am,'N','A');
+      DenseMatrixSVD svd(Am,false,true);
      svd.Eval(Am);
      DenseMatrix &v = svd.RightSingularvectors();
      v.GetRow(k,w);
@@ -346,7 +346,7 @@ void ComputePartialFractionApproximation(real_t & alpha,
   }
   else
   {
-      if (abs(alpha - 0.5) > eps)
+      if (abs(alpha - 0.5) > eps && print_warning)
      {
         alpha = 0.5;
      }
@@ -368,8 +368,6 @@ void ComputePartialFractionApproximation(real_t & alpha,


   return;
-#else
-   MFEM_CONTRACT_VAR(print_warning);
 #endif

   Vector x(npoints);
@@ -19,11 +19,8 @@
 //    mpirun -np 4 ex33p -m ../data/amr-quad.mesh -ver -alpha 2.6 -o 2 -r 2
 //    mpirun -np 4 ex33p -m ../data/inline-hex.mesh -ver -alpha 0.3 -o 2 -r 1

-//  Note: The manufactured solution used in this problem is
-//
-//            u = ∏_{i=0}^{dim-1} sin(π x_i) ,
-//
-//        regardless of the value of alpha.
+//  Note: the analytic solution to this problem is u = ∏_{i=0}^{dim-1} sin(π x_i)
+//        for all alpha.
 //
 // Description:
 //
@@ -123,8 +120,7 @@ int main(int argc, char *argv[])
                  "Enable or disable GLVis visualization.");
   args.AddOption(&verification, "-ver", "--verification", "-no-ver",
                  "--no-verification",
-                  "Use sinusoidal function (f) for manufactured "
-                  "solution test.");
+                  "Use sinusoidal function (f) for analytic comparison.");
   args.Parse();
   if (!args.Good())
   {
@@ -184,11 +180,10 @@ int main(int argc, char *argv[])
   // 5. Define a finite element space on the mesh.
   H1_FECollection fec(order, dim);
   ParFiniteElementSpace fespace(&pmesh, &fec);
-   HYPRE_BigInt size = fespace.GlobalTrueVSize();
   if (Mpi::Root())
   {
-      cout << "Number of degrees of freedom: "
-           << size << endl;
+      cout << "Number of finite element unknowns: "
+           << fespace.GetTrueVSize() << endl;
   }

   // 6. Determine the list of true (i.e. conforming) essential boundary dofs.
@@ -228,7 +223,7 @@ int main(int argc, char *argv[])
   if (verification)
   {
      // This statement is only relevant for the verification of the code. It
-      // uses a different f such that an manufactured solution is known and easy
+      // uses a different f such that an analytic solution is known and easy
      // to compare with the numerical one. The FPDE becomes:
      // (-Δ)^α u = (2\pi ^2)^α sin(\pi x) sin(\pi y) on [0,1]^2
      // -> u(x,y) = sin(\pi x) sin(\pi y)
@@ -420,29 +415,29 @@ int main(int argc, char *argv[])

      if (Mpi::Root())
      {
-         string manufactured_solution,expected_mesh;
+         string analytic_solution,expected_mesh;
         switch (dim)
         {
            case 1:
-               manufactured_solution = "sin(π x)";
+               analytic_solution = "sin(π x)";
               expected_mesh = "inline_segment.mesh";
               break;
            case 2:
-               manufactured_solution = "sin(π x) sin(π y)";
+               analytic_solution = "sin(π x) sin(π y)";
               expected_mesh = "inline_quad.mesh";
               break;
            default:
-               manufactured_solution = "sin(π x) sin(π y) sin(π z)";
+               analytic_solution = "sin(π x) sin(π y) sin(π z)";
               expected_mesh = "inline_hex.mesh";
               break;
         }

         mfem::out << "\n" << string(80,'=')
                   << "\n\nSolution Verification in "<< dim << "D \n\n"
-                   << "Manufactured solution : " << manufactured_solution << "\n"
-                   << "Expected mesh         : " << expected_mesh <<"\n"
-                   << "Your mesh             : " << mesh_file << "\n"
-                   << "L2 error              : " << l2_error << "\n\n"
+                   << "Analytic solution : " << analytic_solution << "\n"
+                   << "Expected mesh     : " << expected_mesh <<"\n"
+                   << "Your mesh         : " << mesh_file << "\n"
+                   << "L2 error          : " << l2_error << "\n\n"
                   << string(80,'=') << endl;
      }
   }
@@ -53,7 +53,7 @@ public:
                                    real_t min_val_=-36)
      : u(&u_), obstacle(&obst_), min_val(min_val_) { }

-   real_t Eval(ElementTransformation &T, const IntegrationPoint &ip) override;
+   virtual real_t Eval(ElementTransformation &T, const IntegrationPoint &ip);
 };

 class ExponentialGridFunctionCoefficient : public Coefficient
@@ -69,7 +69,7 @@ public:
                                      real_t min_val_=0.0, real_t max_val_=1e6)
      : u(&u_), obstacle(&obst_), min_val(min_val_), max_val(max_val_) { }

-   real_t Eval(ElementTransformation &T, const IntegrationPoint &ip) override;
+   virtual real_t Eval(ElementTransformation &T, const IntegrationPoint &ip);
 };

 int main(int argc, char *argv[])
@@ -258,7 +258,7 @@ int main(int argc, char *argv[])
         MixedBilinearForm a10(&H1fes,&L2fes);
         a10.AddDomainIntegrator(new MixedScalarMassIntegrator());
         a10.Assemble();
-         a10.EliminateTrialEssentialBC(ess_bdr, x.GetBlock(0), rhs.GetBlock(1));
+         a10.EliminateTrialDofs(ess_bdr, x.GetBlock(0), rhs.GetBlock(1));
         a10.Finalize();
         SparseMatrix &A10 = a10.SpMat();

@@ -53,7 +53,7 @@ public:
                                    real_t min_val_=-36)
      : u(&u_), obstacle(&obst_), min_val(min_val_) { }

-   real_t Eval(ElementTransformation &T, const IntegrationPoint &ip) override;
+   virtual real_t Eval(ElementTransformation &T, const IntegrationPoint &ip);
 };

 class ExponentialGridFunctionCoefficient : public Coefficient
@@ -69,7 +69,7 @@ public:
                                      real_t min_val_=0.0, real_t max_val_=1e6)
      : u(&u_), obstacle(&obst_), min_val(min_val_), max_val(max_val_) { }

-   real_t Eval(ElementTransformation &T, const IntegrationPoint &ip) override;
+   virtual real_t Eval(ElementTransformation &T, const IntegrationPoint &ip);
 };

 int main(int argc, char *argv[])
@@ -52,8 +52,8 @@ public:
       fun(fun_) {}


-   real_t Eval(ElementTransformation &T,
-               const IntegrationPoint &ip) override
+   virtual real_t Eval(ElementTransformation &T,
+                       const IntegrationPoint &ip)
   {
      return fun(GridFunctionCoefficient::Eval(T, ip));
   }
@@ -83,8 +83,8 @@ public:
       OtherGridF_cf(OtherGridF),
       fun(fun_) {}

-   real_t Eval(ElementTransformation &T,
-               const IntegrationPoint &ip) override
+   virtual real_t Eval(ElementTransformation &T,
+                       const IntegrationPoint &ip)
   {
      const real_t value1 = fun(GridFunctionCoefficient::Eval(T, ip));
      const real_t value2 = fun(OtherGridF_cf.Eval(T, ip));
@@ -108,7 +108,7 @@ public:
      : rho_filter(rho_filter_), min_val(min_val_), max_val(max_val_),
        exponent(exponent_) { }

-   real_t Eval(ElementTransformation &T, const IntegrationPoint &ip) override
+   virtual real_t Eval(ElementTransformation &T, const IntegrationPoint &ip)
   {
      real_t val = rho_filter->GetValue(T, ip);
      real_t coeff = min_val + pow(val,exponent)*(max_val-min_val);
@@ -142,7 +142,7 @@ public:
      MFEM_ASSERT(rho_filter, "density field is not set");
   }

-   real_t Eval(ElementTransformation &T, const IntegrationPoint &ip) override
+   virtual real_t Eval(ElementTransformation &T, const IntegrationPoint &ip)
   {
      real_t L = lambda->Eval(T, ip);
      real_t M = mu->Eval(T, ip);
@@ -176,8 +176,8 @@ public:

   using VectorCoefficient::Eval;

-   void Eval(Vector &V, ElementTransformation &T,
-             const IntegrationPoint &ip) override
+   virtual void Eval(Vector &V, ElementTransformation &T,
+                     const IntegrationPoint &ip)
   {
      Vector xx; xx.SetSize(T.GetDimension());
      T.Transform(ip,xx);
@@ -3,18 +3,18 @@
 // Compile with: make ex38
 //
 // Sample runs:
-// (since all sample runs require LAPACK or ALGOIM, the * symbol is used to
-//  exclude them from the automatically generated internal MFEM tests).
+// (since all sample runs require LAPACK, the * symbol is used to exclude them
+// from the automatically generated internal MFEM tests).
 //              * ex38
 //              * ex38 -i volumetric1d
 //              * ex38 -i surface2d
-//              * ex38 -i surface2d -o 4 -r 5 -m 1
+//              * ex38 -i surface2d -o 4 -r 5
 //              * ex38 -i volumetric2d
-//              * ex38 -i volumetric2d -o 4 -r 5 -m 1
+//              * ex38 -i volumetric2d -o 4 -r 5
 //              * ex38 -i surface3d
-//              * ex38 -i surface3d -o 3 -r 4 -m 1
+//              * ex38 -i surface3d -o 4 -r 5
 //              * ex38 -i volumetric3d
-//              * ex38 -i volumetric3d -o 3 -r 4 -m 1
+//              * ex38 -i volumetric3d -o 4 -r 5
 //
 // Description: This example code demonstrates the use of MFEM to integrate
 //              functions over implicit interfaces and subdomains bounded by
@@ -71,7 +71,7 @@ real_t integrand(const Vector& X)
   switch (itype)
   {
      case IntegrationType::Volumetric1D:
-         return pow(X(0), 2.);
+         return 1.;
      case IntegrationType::Surface2D:
         return 3. * pow(X(0), 2.) - pow(X(1), 2.);
      case IntegrationType::Volumetric2D:
@@ -91,7 +91,7 @@ real_t Surface()
   switch (itype)
   {
      case IntegrationType::Volumetric1D:
-         return .3025;
+         return 1.;
      case IntegrationType::Surface2D:
         return 2. * M_PI;
      case IntegrationType::Volumetric2D:
@@ -111,7 +111,7 @@ real_t Volume()
   switch (itype)
   {
      case IntegrationType::Volumetric1D:
-         return pow(.55, 3.) / 3.;
+         return .55;
      case IntegrationType::Surface2D:
         return NAN;
      case IntegrationType::Volumetric2D:
@@ -125,6 +125,7 @@ real_t Volume()
   }
 }

+#ifdef MFEM_USE_LAPACK
 /**
 @brief Class for surface IntegrationRule

@@ -134,14 +135,11 @@ real_t Volume()
 class SIntegrationRule : public IntegrationRule
 {
 protected:
-   /// method 0 is moments-based, 1 is Algoim.
-   int method, ir_order, ls_order;
-   Coefficient &level_set;
-   /// Space Dimension of the IntegrationRule
+   /// @brief Space Dimension of the IntegrationRule
   int dim;
-   /// Column-wise matrix of the quadtrature weights
+   /// @brief Column-wise matrix of the quadtrature weights
   DenseMatrix Weights;
-   /// Column-wise matrix of the transformation weights of the normal
+   /// @brief Column-wise matrix of the transformation weights of the normal
   DenseMatrix SurfaceWeights;

 public:
@@ -155,21 +153,15 @@ public:
    @param [in] lsOrder Polynomial degree for approx of level-set function
    @param [in] mesh Pointer to the mesh that is used
   */
-   SIntegrationRule(int method_, int Order,
-                    Coefficient& LvlSet, int lsOrder, Mesh* mesh)
-      : method(method_), ir_order(Order), ls_order(lsOrder),
-        level_set(LvlSet), dim(mesh->Dimension())
+   SIntegrationRule(int Order, Coefficient& LvlSet, int lsOrder, Mesh* mesh)
   {
-      // Nothing gets pre-computed for Algoim.
-      if (method == 1) { return; }
-
-#ifdef MFEM_USE_LAPACK
-      MomentFittingIntRules mf_ir(ir_order, level_set, ls_order);
+      dim = mesh->Dimension();

      IsoparametricTransformation Tr;
+      MomentFittingIntRules MFIRs(Order, LvlSet, lsOrder);
      mesh->GetElementTransformation(0, &Tr);
      IntegrationRule ir;
-      mf_ir.GetSurfaceIntegrationRule(Tr, ir);
+      MFIRs.GetSurfaceIntegrationRule(Tr, ir);
      if (dim >1)
      {
         Weights.SetSize(ir.GetNPoints(), mesh->GetNE());
@@ -180,7 +172,7 @@ public:
      }
      SurfaceWeights.SetSize(ir.GetNPoints(), mesh->GetNE());
      Vector w;
-      mf_ir.GetSurfaceWeights(Tr, ir, w);
+      MFIRs.GetSurfaceWeights(Tr, ir, w);
      SurfaceWeights.SetCol(0, w);
      SetSize(ir.GetNPoints());

@@ -206,8 +198,9 @@ public:
      for (int elem = 1; elem < mesh->GetNE(); elem++)
      {
         mesh->GetElementTransformation(elem, &Tr);
-         mf_ir.GetSurfaceIntegrationRule(Tr, ir);
-         mf_ir.GetSurfaceWeights(Tr, ir, w);
+         MFIRs.GetSurfaceIntegrationRule(Tr, ir);
+         Vector w;
+         MFIRs.GetSurfaceWeights(Tr, ir, w);
         SurfaceWeights.SetCol(elem, w);

         for (int ip = 0; ip < GetNPoints(); ip++)
@@ -223,48 +216,48 @@ public:
            }
         }
      }
-#else
-      MFEM_ABORT("Moment-fitting requires MFEM to be built with LAPACK!");
-#endif
   }

   /**
    @brief Set the weights for the given element and multiply them with the
    transformation of the interface
    */
-   void SetElementAndSurfaceWeight(ElementTransformation &Tr)
+   void SetElementinclSurfaceWeight(int Element)
   {
-      if (method == 1)
-      {
-#ifdef MFEM_USE_ALGOIM
-         AlgoimIntegrationRules a_ir(ir_order, level_set, ls_order);
-         a_ir.GetSurfaceIntegrationRule(Tr, *this);
-         Vector w;
-         a_ir.GetSurfaceWeights(Tr, *this, w);
-         for (int ip = 0; ip < GetNPoints(); ip++)
-         {
-            IntPoint(ip).weight *= w(ip);
-         }
-         return;
-#else
-         MFEM_ABORT("MFEM is not built with Algoim support!");
-#endif
-      }
-
      if (dim == 1)
      {
-         IntPoint(0).x = Weights(0, Tr.ElementNo);
-         IntPoint(0).weight = Weights(1, Tr.ElementNo);
+         IntegrationPoint &intp = IntPoint(0);
+         intp.x = Weights(0, Element);
+         intp.weight = Weights(1, Element);
+         cout << intp.x << " " << Element << endl;
      }
      else
-      {
         for (int ip = 0; ip < GetNPoints(); ip++)
         {
-            IntPoint(ip).weight = Weights(ip, Tr.ElementNo) *
-                                  SurfaceWeights(ip, Tr.ElementNo);
+            IntegrationPoint &intp = IntPoint(ip);
+            intp.weight = Weights(ip, Element) * SurfaceWeights(ip, Element);
         }
-      }
   }
+
+   /// @brief Set the weights for the given element
+   void SetElement(int Element)
+   {
+      if (dim == 1)
+      {
+         IntegrationPoint &intp = IntPoint(0);
+         intp.x = Weights(0, Element);
+         intp.weight = Weights(1, Element);
+      }
+      else
+         for (int ip = 0; ip < GetNPoints(); ip++)
+         {
+            IntegrationPoint &intp = IntPoint(ip);
+            intp.weight = Weights(ip, Element);
+         }
+   }
+
+   /// @brief Destructor of SIntegrationRule
+   ~SIntegrationRule() {}
 };

 /**
@@ -276,12 +269,9 @@ public:
 class CIntegrationRule : public IntegrationRule
 {
 protected:
-   /// method 0 is moments-based, 1 is Algoim.
-   int method, ir_order, ls_order;
-   Coefficient &level_set;
-   /// Space Dimension of the IntegrationRule
+   /// @brief Space Dimension of the IntegrationRule
   int dim;
-   /// Column-wise matrix of the quadtrature positions and weights.
+   /// @brief Column-wise matrix of the quadtrature weights
   DenseMatrix Weights;

 public:
@@ -295,21 +285,15 @@ public:
    @param [in] lsOrder Polynomial degree for approx of level-set function
    @param [in] mesh Pointer to the mesh that is used
   */
-   CIntegrationRule(int method_, int Order,
-                    Coefficient &LvlSet, int lsOrder, Mesh *mesh)
-      : method(method_), ir_order(Order), ls_order(lsOrder),
-        level_set(LvlSet), dim(mesh->Dimension())
+   CIntegrationRule(int Order, Coefficient& LvlSet, int lsOrder, Mesh* mesh)
   {
-      // Nothing gets pre-computed for Algoim.
-      if (method == 1) { return; }
-
-#ifdef MFEM_USE_LAPACK
-      MomentFittingIntRules mf_ir(ir_order, level_set, ls_order);
+      dim = mesh->Dimension();

      IsoparametricTransformation Tr;
+      MomentFittingIntRules MFIRs(Order, LvlSet, lsOrder);
      mesh->GetElementTransformation(0, &Tr);
      IntegrationRule ir;
-      mf_ir.GetVolumeIntegrationRule(Tr, ir);
+      MFIRs.GetVolumeIntegrationRule(Tr, ir);
      if (dim > 1)
      {
         Weights.SetSize(ir.GetNPoints(), mesh->GetNE());
@@ -341,9 +325,9 @@ public:
      for (int elem = 1; elem < mesh->GetNE(); elem++)
      {
         mesh->GetElementTransformation(elem, &Tr);
-         mf_ir.GetVolumeIntegrationRule(Tr, ir);
+         MFIRs.GetVolumeIntegrationRule(Tr, ir);

-         for (int ip = 0; ip < ir.GetNPoints(); ip++)
+         for (int ip = 0; ip < GetNPoints(); ip++)
         {
            if (dim > 1)
            {
@@ -356,39 +340,29 @@ public:
            }
         }
      }
-#else
-      MFEM_ABORT("Moment-fitting requires MFEM to be built with LAPACK!");
-#endif
   }

   /// @brief Set the weights for the given element
-   void SetElement(ElementTransformation &Tr)
+   void SetElement(int Element)
   {
-      if (method == 1)
-      {
-#ifdef MFEM_USE_ALGOIM
-         AlgoimIntegrationRules a_ir(ir_order, level_set, ls_order);
-         a_ir.GetVolumeIntegrationRule(Tr, *this);
-         return;
-#else
-         MFEM_ABORT("MFEM is not built with Algoim support!");
-#endif
-      }
-
-      for (int ip = 0; ip < GetNPoints(); ip++)
-      {
-         IntegrationPoint &intp = IntPoint(ip);
-         if (dim == 1)
+      if (dim == 1)
+         for (int ip = 0; ip < GetNPoints(); ip++)
         {
-            intp.x = Weights(2 * ip, Tr.ElementNo);
-            intp.weight = Weights(2 * ip + 1, Tr.ElementNo);
+            IntegrationPoint &intp = IntPoint(ip);
+            intp.x = Weights(2 * ip, Element);
+            intp.weight = Weights(2 * ip + 1, Element);
+         }
+      else
+         for (int ip = 0; ip < GetNPoints(); ip++)
+         {
+            IntegrationPoint &intp = IntPoint(ip);
+            intp.weight = Weights(ip, Element);
         }
-         else { intp.weight = Weights(ip, Tr.ElementNo); }
-      }
   }
+
+   /// @brief Destructor of CIntegrationRule
+   ~CIntegrationRule() {}
 };
-
-
 /**
 @brief Class for surface linearform integrator

@@ -435,9 +409,9 @@ public:
    @param [in] Tr transformation of finite element
    @param [out] elvect vector containing the
   */
-   void AssembleRHSElementVect(const FiniteElement &el,
-                               ElementTransformation &Tr,
-                               Vector &elvect) override
+   virtual void AssembleRHSElementVect(const FiniteElement &el,
+                                       ElementTransformation &Tr,
+                                       Vector &elvect) override
   {
      int dof = el.GetDof();
      shape.SetSize(dof);
@@ -445,7 +419,7 @@ public:
      elvect = 0.;

      // Update the surface integration rule for the current element
-      SIntRule->SetElementAndSurfaceWeight(Tr);
+      SIntRule->SetElementinclSurfaceWeight(Tr.ElementNo);

      for (int ip = 0; ip < SIntRule->GetNPoints(); ip++)
      {
@@ -455,8 +429,6 @@ public:
         add(elvect, SIntRule->IntPoint(ip).weight * val, shape, elvect);
      }
   }
-
-   using LinearFormIntegrator::AssembleRHSElementVect;
 };

 /**
@@ -505,9 +477,9 @@ public:
    @param [in] Tr transformation of finite element
    @param [out] elvect vector containing the
   */
-   void AssembleRHSElementVect(const FiniteElement &el,
-                               ElementTransformation &Tr,
-                               Vector &elvect) override
+   virtual void AssembleRHSElementVect(const FiniteElement &el,
+                                       ElementTransformation &Tr,
+                                       Vector &elvect) override
   {
      int dof = el.GetDof();
      shape.SetSize(dof);
@@ -515,7 +487,7 @@ public:
      elvect = 0.;

      // Update the subdomain integration rule
-      CIntRule->SetElement(Tr);
+      CIntRule->SetElement(Tr.ElementNo);

      for (int ip = 0; ip < CIntRule->GetNPoints(); ip++)
      {
@@ -526,17 +498,18 @@ public:
         add(elvect, CIntRule->IntPoint(ip).weight * val, shape, elvect);
      }
   }
-
-   using LinearFormIntegrator::AssembleRHSElementVect;
 };
+#endif // MFEM_USE_LAPACK

 int main(int argc, char *argv[])
 {
-#if defined(MFEM_USE_LAPACK) || defined(MFEM_USE_ALGOIM)
+#ifndef MFEM_USE_LAPACK
+   cout << "MFEM must be built with LAPACK for this example." << endl;
+   return MFEM_SKIP_RETURN_VALUE;
+#else
   // 1. Parse he command-line options.
   int ref_levels = 3;
   int order = 2;
-   int method = 0;
   const char *inttype = "surface2d";
   bool visualization = true;
   itype = IntegrationType::Surface2D;
@@ -544,8 +517,6 @@ int main(int argc, char *argv[])
   OptionsParser args(argc, argv);
   args.AddOption(&order, "-o", "--order", "Order of quadrature rule");
   args.AddOption(&ref_levels, "-r", "--refine", "Number of meh refinements");
-   args.AddOption(&method, "-m", "--method",
-                  "Cut integration method: 0 for moments-based, 1 for Algoim.");
   args.AddOption(&inttype, "-i", "--integrationtype",
                  "IntegrationType to demonstrate");
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
@@ -580,7 +551,7 @@ int main(int argc, char *argv[])
   }

   // 2. Construct and refine the mesh.
-   Mesh *mesh = nullptr;
+   Mesh *mesh;
   if (itype == IntegrationType::Volumetric1D)
   {
      mesh = new Mesh("../data/inline-segment.mesh");
@@ -628,14 +599,13 @@ int main(int argc, char *argv[])
   // 5. Define the necessary Integration rules on element 0.
   IsoparametricTransformation Tr;
   mesh->GetElementTransformation(0, &Tr);
-   SIntegrationRule* sir = new SIntegrationRule(method, order,
-                                                levelset, 2, mesh);
+   SIntegrationRule* sir = new SIntegrationRule(order, levelset, 2, mesh);
   CIntegrationRule* cir = NULL;
   if (itype == IntegrationType::Volumetric1D
       || itype == IntegrationType::Volumetric2D
       || itype == IntegrationType::Volumetric3D)
   {
-      cir = new CIntegrationRule(method, order, levelset, 2, mesh);
+      cir = new CIntegrationRule(order, levelset, 2, mesh);
   }

   // 6. Define and assemble the linear forms on the finite element space.
@@ -678,11 +648,11 @@ int main(int argc, char *argv[])
      cout << "Number of div free basis functions: " << nbasis << endl;
      cout << "Number of quadrature points:        " << ir.GetNPoints() << endl;
   }
-   cout << scientific << setprecision(10);
+   cout << scientific << setprecision(2);
   cout << "============================================" << endl;
   cout << "Computed value of surface integral: " << surface.Sum() << endl;
   cout << "True value of surface integral:     " << Surface() << endl;
-   cout << "Absolute Error (Surface):           ";
+   cout << "Absolute Error (Surface):            ";
   cout << abs(surface.Sum() - Surface()) << endl;
   cout << "Relative Error (Surface):           ";
   cout << abs(surface.Sum() - Surface()) / Surface() << endl;
@@ -693,7 +663,7 @@ int main(int argc, char *argv[])
      cout << "--------------------------------------------" << endl;
      cout << "Computed value of volume integral:  " << volume.Sum() << endl;
      cout << "True value of volume integral:      " << Volume() << endl;
-      cout << "Absolute Error (Volume):            ";
+      cout << "Absolute Error (Volume):             ";
      cout << abs(volume.Sum() - Volume()) << endl;
      cout << "Relative Error (Volume):            ";
      cout << abs(volume.Sum() - Volume()) / Volume() << endl;
@@ -722,8 +692,5 @@ int main(int argc, char *argv[])
   delete fespace;
   delete mesh;
   return EXIT_SUCCESS;
-#else
-   cout << "MFEM must be built with LAPACK or ALGOIM for this example." << endl;
-   return MFEM_SKIP_RETURN_VALUE;
-#endif // MFEM_USE_LAPACK
+#endif //MFEM_USE_LAPACK
 }
@@ -2,53 +2,39 @@
 //
 // Compile with: make ex40
 //
-// Sample runs: ex40 -step 10 -gr 2.0
-//              ex40 -step 10 -gr 2.0 -o 3 -r 1
-//              ex40 -step 10 -gr 2.0 -r 4 -m ../data/l-shape.mesh
-//              ex40 -step 10 -gr 2.0 -r 2 -m ../data/fichera.mesh
+// Sample runs: ex40 -o 2
+//              ex40 -o 2 -r 4
 //
-// Description: This example code demonstrates how to use MFEM to solve the
-//              eikonal equation,
+// Description: This example code demonstrates to how to use MFEM to solve
+//              the Monge–Ampère equation
 //
-//                      |∇𝑢| = 1 in Ω,  𝑢 = g on ∂Ω.
+//                      det(∇²u) = f in Ω,   u = 0 on ∂Ω.
 //
-//              The solution of this problem coincides with the unique optimum of
-//              the nonlinear program
+//              This example highlights the ExponentialMatrixCoefficient
+//              class, which is used in Newton's method to solve the
+//              variational formulation
 //
-//                   maximize ∫_Ω 𝑢 d𝑥 subject to |∇𝑢| ≤ 1,   𝑢 = g on Ω,      (⋆)
+//               Find M ∈ H₀(div,Ω)ⁿ and u ∈ H₀¹(Ω) such that
+//                 (exp(M), N) + (∇u, ∇⋅N) = 0           ∀ N ∈ H₀(div,Ω)ⁿ
+//                 (tr(M), v)              = (ln f, v)   ∀ v ∈ H₀¹(Ω)
 //
-//              which is the foundation for method implemented below.
+//              where n is the spatial dimension of the domain Ω.
 //
-//              Following the proximal Galerkin methodology [1] (see also Example
-//              36), we construct a Legendre function for the unit ball
-//              𝐵₁ := {𝑥 ∈ Rⁿ | |𝑥| < 1}. Our choice is the Hellinger entropy,
 //
-//                    h(𝑥) = −( 1 − |𝑥|² )^{1/2},
+//              The linearized subproblem is
 //
-//              although other choices are possible, each leading to a slightly
-//              different algorithm. We then adaptively regularize the optimization
-//              problem (⋆) with the Bregman divergence of the Hellinger entropy,
+//               Find δM ∈ H₀(div,Ω)ⁿ and u ∈ H₀¹(Ω) such that
+//                 (exp(M) δM, N) + (∇u, ∇⋅N) = -(exp(M), N)        ∀ N ∈ H₀(div,Ω)ⁿ
+//                 (tr(δM), v)                = (ln f - tr(M), v)   ∀ v ∈ H₀¹(Ω)
 //
-//                 maximize  ∫_Ω 𝑢 d𝑥 - αₖ⁻¹ Dₕ(∇𝑢,∇𝑢ₖ₋₁)  subject to  𝑢 = g on Ω.
 //
-//              This results in a sequence of functions ( 𝜓ₖ , 𝑢ₖ ),
+//              (exp(M) δM, N)    :::   VectorFEMassIntegrator
+//              (∇u, ∇⋅N)         :::   MixedGradDivIntegrator
+//              (tr(δM), v)       :::   MixedDotProductIntegrator
+//              (exp(M), N)       :::   VectorFEDomainLFIntegrator
+//              (ln f - tr(M), v) :::   DomainLFIntegrator
 //
-//                      𝑢ₖ → 𝑢,    𝜓ₖ/|𝜓ₖ| → ∇𝑢    as k → \infty,
 //
-//              defined by the nonlinear saddle-point problems
-//
-//               Find 𝜓ₖ ∈ H(div,Ω) and 𝑢ₖ ∈ L²(Ω) such that
-//               ( Zₖ(𝜓ₖ) , τ ) + ( 𝑢ₖ , ∇⋅τ ) = ⟨ g , τ⋅n ⟩         ∀ τ ∈ H(div,Ω)
-//               ( ∇⋅𝜓ₖ , v )                 = ( ∇⋅𝜓ₖ₋₁ - 1 , v )   ∀ v ∈ L²(Ω)
-//
-//              where Zₖ(𝜓) := ∇h⁻¹(αₖ 𝜓) = 𝜓 / ( αₖ⁻² + |𝜓|² )^{1/2} and step size
-//              αₖ > 0. These saddle-point problems are solved using a damped Newton's
-//              method. This example assumes that g = 0 and allows the step size to
-//              grow geometrically, αₖ = α₀rᵏ, where r ≥ 1 is the growth rate.
-//
-//              [1] Keith, B. and Surowiec, T. (2023) Proximal Galerkin: A structure-
-//                  preserving finite element method for pointwise bound constraints.
-//                  arXiv:2307.12444 [math.NA]

 #include "mfem.hpp"
 #include <fstream>
@@ -57,55 +43,23 @@
 using namespace std;
 using namespace mfem;

-class ZCoefficient : public VectorCoefficient
-{
-protected:
-   GridFunction *psi;
-   real_t alpha;
-
-public:
-   ZCoefficient(int vdim, GridFunction &psi_, real_t alpha_ = 1.0)
-      : VectorCoefficient(vdim), psi(&psi_), alpha(alpha_) { }
-
-   using VectorCoefficient::Eval;
-
-   void Eval(Vector &V, ElementTransformation &T,
-             const IntegrationPoint &ip) override;
-   void SetAlpha(real_t alpha_) { alpha = alpha_; }
-};
-
-class DZCoefficient : public MatrixCoefficient
-{
-protected:
-   GridFunction *psi;
-   real_t alpha;
-
-public:
-   DZCoefficient(int height, GridFunction &psi_, real_t alpha_ = 1.0)
-      : MatrixCoefficient(height),  psi(&psi_), alpha(alpha_) { }
-
-   void Eval(DenseMatrix &K, ElementTransformation &T,
-             const IntegrationPoint &ip) override;
-   void SetAlpha(real_t alpha_) { alpha = alpha_; }
-};
+real_t exact_solution(const Vector &pt);
+void exact_solution_gradient(const Vector &pt, Vector &grad);

 int main(int argc, char *argv[])
 {
   // 1. Parse command-line options.
-   const char *mesh_file = "../data/star.mesh";
-   int order = 1;
-   int max_it = 5;
-   int ref_levels = 3;
-   real_t alpha = 1.0;
-   real_t growth_rate = 1.0;
-   real_t newton_scaling = 0.9;
-   real_t tichonov = 1e-1;
-   real_t tol = 1e-4;
+   const char *mesh_file = "../data/disc-nurbs.mesh";
+   // const char *mesh_file = "../data/star.mesh";
+   int order = 2;
+   int max_it = 10;
+   int ref_levels = 1;
+   real_t tol = 1e-5;
   bool visualization = true;

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
-                  "Mesh file to use.");
+                  "Mesh file.");
   args.AddOption(&order, "-o", "--order",
                  "Finite element order (polynomial degree).");
   args.AddOption(&ref_levels, "-r", "--refs",
@@ -115,10 +69,6 @@ int main(int argc, char *argv[])
   args.AddOption(&tol, "-tol", "--tol",
                  "Stopping criteria based on the difference between"
                  "successive solution updates");
-   args.AddOption(&alpha, "-step", "--step",
-                  "Initial size alpha");
-   args.AddOption(&growth_rate, "-gr", "--growth-rate",
-                  "Growth rate of the step size alpha");
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
@@ -133,12 +83,11 @@ int main(int argc, char *argv[])
   // 2. Read the mesh from the mesh file.
   Mesh mesh(mesh_file, 1, 1);
   int dim = mesh.Dimension();
-   int sdim = mesh.SpaceDimension();

-   MFEM_ASSERT(mesh.bdr_attributes.Size(),
-               "This example does not currently support meshes"
-               " without boundary attributes."
-              )
+   if (dim != 2)
+   {
+      MFEM_ABORT("Example 40 currently only supports 2D problems")
+   }

   // 3. Postprocess the mesh.
   // 3A. Refine the mesh to increase the resolution.
@@ -153,45 +102,73 @@ int main(int argc, char *argv[])
   mesh.SetCurvature(curvature_order);

   // 4. Define the necessary finite element spaces on the mesh.
-   RT_FECollection RTfec(order, dim);
+   H1_FECollection H1fec(order, dim);
+   FiniteElementSpace H1fes(&mesh, &H1fec);
+
+   RT_FECollection RTfec(order-1, dim);
   FiniteElementSpace RTfes(&mesh, &RTfec);

-   L2_FECollection L2fec(order, dim);
-   FiniteElementSpace L2fes(&mesh, &L2fec);
+   cout << "Number of H¹ degrees of freedom: "
+        << H1fes.GetTrueVSize() << endl;
+   cout << "Number of H(div) degrees of freedom: "
+        << RTfes.GetTrueVSize() * dim << endl;

-   cout << "Number of H(div) dofs: "
-        << RTfes.GetTrueVSize() << endl;
-   cout << "Number of L² dofs: "
-        << L2fes.GetTrueVSize() << endl;
-
-   // 5. Define the offsets for the block matrices
-   Array<int> offsets(3);
+   Array<int> offsets(4);
   offsets[0] = 0;
   offsets[1] = RTfes.GetVSize();
-   offsets[2] = L2fes.GetVSize();
+   offsets[2] = RTfes.GetVSize();
+   offsets[3] = H1fes.GetVSize();
   offsets.PartialSum();

   BlockVector x(offsets), rhs(offsets);
   x = 0.0; rhs = 0.0;

-   // 6. Define the solution vectors as a finite element grid functions
+   // 5. Determine the list of true (i.e., conforming) essential boundary dofs.
+   Array<int> ess_bdr;
+   if (mesh.bdr_attributes.Size())
+   {
+      ess_bdr.SetSize(mesh.bdr_attributes.Max());
+      ess_bdr = 1;
+   }
+
+   // 6. Define constants to be used later.
+   ConstantCoefficient one(1.0);
+   ConstantCoefficient neg_one(-1.0);
+   ConstantCoefficient zero(0.0);
+   Vector V1(2), V2(2);
+   V1(0) = 1.0; V1(1) = 0.0;
+   V2(0) = 0.0; V2(1) = 1.0;
+   VectorConstantCoefficient onezero(V1);
+   VectorConstantCoefficient zeroone(V2);
+   ScalarVectorProductCoefficient neg_onezero(-1.0, onezero);
+   ScalarVectorProductCoefficient neg_zeroone(-1.0, zeroone);
+
+   // 7. Define the solution vectors as finite element grid functions
   //    corresponding to the fespaces.
-   GridFunction u_gf, delta_psi_gf;
-   delta_psi_gf.MakeRef(&RTfes,x,offsets[0]);
-   u_gf.MakeRef(&L2fes,x,offsets[1]);
+   GridFunction delta_M1_gf, delta_M2_gf, delta_u_gf;

-   GridFunction psi_old_gf(&RTfes);
-   GridFunction psi_gf(&RTfes);
-   GridFunction u_old_gf(&L2fes);
+   delta_M1_gf.MakeRef(&RTfes,x,offsets[0]);
+   delta_M2_gf.MakeRef(&RTfes,x,offsets[1]);
+   delta_u_gf.MakeRef(&H1fes,x,offsets[2]);

-   // 7. Define initial guesses for the solution variables.
-   delta_psi_gf = 0.0;
-   psi_gf = 0.0;
-   u_gf = 0.0;
-   psi_old_gf = psi_gf;
-   u_old_gf = u_gf;
+   GridFunction M1_gf(&RTfes);
+   GridFunction M2_gf(&RTfes);
+   GridFunction u_gf(&H1fes);
+
+   // 8. Define the function coefficients for the solution and use them to
+   //    initialize the initial guess
+   FunctionCoefficient exact_coef(exact_solution);
+   VectorFunctionCoefficient exact_grad_coef(dim,exact_solution_gradient);
+   ConstantCoefficient ln_rhs_coef(0.0);
+   u_gf.ProjectCoefficient(exact_coef);
+   // u_gf.ProjectCoefficient(zero);
+   M1_gf = 0.0;
+   M2_gf = 0.0;
+
+   delta_M1_gf = 0.0;
+   delta_M2_gf = 0.0;
+   delta_u_gf = 0.0;

-   // 8. Prepare for glvis output.
   char vishost[] = "localhost";
   int  visport   = 19916;
   socketstream sol_sock;
@@ -201,176 +178,234 @@ int main(int argc, char *argv[])
      sol_sock.precision(8);
   }

-   // 9. Coefficients to be used later.
-   ConstantCoefficient neg_one(-1.0);
-   ConstantCoefficient zero(0.0);
-   ConstantCoefficient tichonov_cf(tichonov);
-   ConstantCoefficient neg_tichonov_cf(-1.0*tichonov);
-   ZCoefficient Z(sdim, psi_gf, alpha);
-   DZCoefficient DZ(sdim, psi_gf, alpha);
-   ScalarVectorProductCoefficient neg_Z(-1.0, Z);
-   DivergenceGridFunctionCoefficient div_psi_cf(&psi_gf);
-   DivergenceGridFunctionCoefficient div_psi_old_cf(&psi_old_gf);
-   SumCoefficient psi_old_minus_psi(div_psi_old_cf, div_psi_cf, 1.0, -1.0);
-
-   // 10. Assemble constant matrices/vectors to avoid reassembly in the loop.
-   LinearForm b0, b1;
-   b0.MakeRef(&RTfes,rhs.GetBlock(0),0);
-   b1.MakeRef(&L2fes,rhs.GetBlock(1),0);
-
-   b0.AddDomainIntegrator(new VectorFEDomainLFIntegrator(neg_Z));
-   b1.AddDomainIntegrator(new DomainLFIntegrator(neg_one));
-   b1.AddDomainIntegrator(new DomainLFIntegrator(psi_old_minus_psi));
-
-   BilinearForm a00(&RTfes);
-   a00.AddDomainIntegrator(new VectorFEMassIntegrator(DZ));
-   a00.AddDomainIntegrator(new VectorFEMassIntegrator(tichonov_cf));
-
-   MixedBilinearForm a10(&RTfes,&L2fes);
-   a10.AddDomainIntegrator(new VectorFEDivergenceIntegrator());
-   a10.Assemble();
-   a10.Finalize();
-   SparseMatrix &A10 = a10.SpMat();
-   SparseMatrix *A01 = Transpose(A10);
-
-   BilinearForm a11(&L2fes);
-   a11.AddDomainIntegrator(new MassIntegrator(neg_tichonov_cf));
-   a11.Assemble();
-   a11.Finalize();
-   SparseMatrix &A11 = a11.SpMat();
-
-   // 11. Iterate.
+   // 10. Iterate
   int k;
-   int total_iterations = 0;
-   real_t increment_u = 0.1;
-   GridFunction u_tmp(&L2fes);
   for (k = 0; k < max_it; k++)
   {
-      u_tmp = u_old_gf;
-      Z.SetAlpha(alpha);
-      DZ.SetAlpha(alpha);
+      mfem::out << "\nITERATION " << k+1 << endl;

-      mfem::out << "\nOUTER ITERATION " << k+1 << endl;
+      LinearForm b0,b1,b2;
+      b0.Update(&RTfes,rhs.GetBlock(0),0);
+      b1.Update(&RTfes,rhs.GetBlock(1),0);
+      b2.Update(&H1fes,rhs.GetBlock(2),0);

-      int j;
-      for ( j = 0; j < 5; j++)
+      VectorGridFunctionCoefficient M1(&M1_gf);
+      VectorGridFunctionCoefficient M2(&M2_gf);
+
+      MatrixArrayVectorCoefficient M(dim);
+      M.Set(0, &M1, false);
+      M.Set(1, &M2, false);
+      ExponentialMatrixCoefficient exp_M(M);
+
+      MatrixVectorProductCoefficient exp_M1(exp_M, onezero);
+      MatrixVectorProductCoefficient exp_M2(exp_M, zeroone);
+      InnerProductCoefficient exp_M11(exp_M1, onezero);
+      InnerProductCoefficient exp_M12(exp_M1, zeroone);
+      InnerProductCoefficient exp_M21(exp_M2, onezero);
+      InnerProductCoefficient exp_M22(exp_M2, zeroone);
+
+      GradientGridFunctionCoefficient grad_u(&u_gf);
+      InnerProductCoefficient neg_dudx(neg_onezero, grad_u);
+      ScalarVectorProductCoefficient neg_exp_M1(-1.0, exp_M1);
+      b0.AddDomainIntegrator(new VectorFEDomainLFDivIntegrator(neg_dudx));
+      b0.AddDomainIntegrator(new VectorFEDomainLFIntegrator(neg_exp_M1));
+      b0.Assemble();
+
+      InnerProductCoefficient neg_dudy(neg_zeroone, grad_u);
+      b1.AddDomainIntegrator(new VectorFEDomainLFDivIntegrator(neg_dudy));
+      ScalarVectorProductCoefficient neg_exp_M2(-1.0, exp_M2);
+      b1.AddDomainIntegrator(new VectorFEDomainLFIntegrator(neg_exp_M2));
+      b1.Assemble();
+
+      InnerProductCoefficient M11(M1, onezero);
+      InnerProductCoefficient M22(M2, zeroone);
+      SumCoefficient trace_M(M11, M22);
+      SumCoefficient rhs2(ln_rhs_coef, trace_M, 1.0, -1.0);
+      b2.AddDomainIntegrator(new DomainLFIntegrator(rhs2));
+      b2.Assemble();
+
+      cout << "b0.Norml2() = " << b0.Norml2() << endl;
+      cout << "b1.Norml2() = " << b1.Norml2() << endl;
+      cout << "b2.Norml2() = " << b2.Norml2() << endl;
+
+      BilinearForm a00(&RTfes);
+      a00.AddDomainIntegrator(new VectorFEMassIntegrator());
+      // a00.AddDomainIntegrator(new VectorFEMassIntegrator(exp_M11));
+      a00.Assemble();
+      a00.EliminateEssentialBC(ess_bdr,x.GetBlock(0),rhs.GetBlock(0),mfem::Operator::DIAG_ONE);
+      a00.Finalize();
+      SparseMatrix &A00 = a00.SpMat();
+
+      BilinearForm a01(&RTfes);
+      a01.AddDomainIntegrator(new VectorFEMassIntegrator(zero));
+      // a01.AddDomainIntegrator(new VectorFEMassIntegrator(exp_M12));
+      a01.Assemble();
+      a01.EliminateEssentialBC(ess_bdr,mfem::Operator::DIAG_ZERO);
+      a01.Finalize();
+      SparseMatrix &A01 = a01.SpMat();
+
+      MixedBilinearForm a02(&H1fes,&RTfes);
+      a02.AddDomainIntegrator(new MixedGradDivIntegrator(neg_onezero));
+      a02.Assemble(false);
+      a02.EliminateTrialDofs(ess_bdr,x.GetBlock(2),rhs.GetBlock(0));
+      a02.EliminateTestDofs(ess_bdr);
+      a02.Finalize();
+      SparseMatrix &A02 = a02.SpMat();
+
+      BilinearForm a10(&RTfes);
+      a10.AddDomainIntegrator(new VectorFEMassIntegrator(zero));
+      // a10.AddDomainIntegrator(new VectorFEMassIntegrator(exp_M21));
+      a10.Assemble();
+      a10.EliminateEssentialBC(ess_bdr,mfem::Operator::DIAG_ZERO);
+      a10.Finalize();
+      SparseMatrix &A10 = a10.SpMat();
+
+      BilinearForm a11(&RTfes);
+      a11.AddDomainIntegrator(new VectorFEMassIntegrator());
+      // a11.AddDomainIntegrator(new VectorFEMassIntegrator(exp_M22));
+      a11.Assemble();
+      a11.EliminateEssentialBC(ess_bdr,x.GetBlock(1),rhs.GetBlock(1),mfem::Operator::DIAG_ONE);
+      a11.Finalize();
+      SparseMatrix &A11 = a11.SpMat();
+
+      MixedBilinearForm a12(&H1fes,&RTfes);
+      a12.AddDomainIntegrator(new MixedGradDivIntegrator(neg_zeroone));
+      a12.Assemble(false);
+      a12.EliminateTrialDofs(ess_bdr,x.GetBlock(2),rhs.GetBlock(1));
+      a12.EliminateTestDofs(ess_bdr);
+      a12.Finalize();
+      SparseMatrix &A12 = a12.SpMat();
+
+      MixedBilinearForm a20(&RTfes,&H1fes);
+      a20.AddDomainIntegrator(new MixedDotProductIntegrator(onezero));
+      a20.Assemble();
+      a20.EliminateTrialDofs(ess_bdr,x.GetBlock(0),rhs.GetBlock(2));
+      a20.EliminateTestDofs(ess_bdr);
+      a20.Finalize();
+      SparseMatrix &A20 = a20.SpMat();
+
+      MixedBilinearForm a21(&RTfes,&H1fes);
+      a21.AddDomainIntegrator(new MixedDotProductIntegrator(zeroone));
+      a21.Assemble();
+      a21.EliminateTrialDofs(ess_bdr,x.GetBlock(1),rhs.GetBlock(2));
+      a21.EliminateTestDofs(ess_bdr);
+      a21.Finalize();
+      SparseMatrix &A21 = a21.SpMat();
+
+      BilinearForm a22(&H1fes);
+      // a22.AddDomainIntegrator(new MassIntegrator(neg_one));
+      a22.AddDomainIntegrator(new MassIntegrator(zero));
+      a22.Assemble(false);
+      a22.EliminateEssentialBC(ess_bdr,x.GetBlock(2),rhs.GetBlock(2),mfem::Operator::DIAG_ONE);
+      a22.Finalize();
+      SparseMatrix &A22 = a22.SpMat();
+
+      cout << "b0.Norml2() = " << b0.Norml2() << endl;
+      cout << "b1.Norml2() = " << b1.Norml2() << endl;
+      cout << "b2.Norml2() = " << b2.Norml2() << endl;
+
+      // BlockOperator A(offsets);
+      // A.SetBlock(0,0,&A00);
+      // A.SetBlock(0,1,&A01);
+      // A.SetBlock(0,2,&A02);
+      // A.SetBlock(1,0,&A10);
+      // A.SetBlock(1,1,&A11);
+      // A.SetBlock(1,2,&A12);
+      // A.SetBlock(2,0,&A20);
+      // A.SetBlock(2,1,&A21);
+      // A.SetBlock(2,2,&A22);
+
+      // BlockDiagonalPreconditioner prec(offsets);
+      // prec.SetDiagonalBlock(0,new GSSmoother(A00));
+      // prec.SetDiagonalBlock(1,new GSSmoother(A11));
+      // prec.SetDiagonalBlock(1,new GSSmoother(A22));
+      // prec.owns_blocks = 1;
+
+      // GMRES(A,prec,rhs,x,1,10000,500,1e-12,0.0);
+
+      BlockMatrix A(offsets);
+      A.SetBlock(0,0,&A00);
+      A.SetBlock(0,1,&A01);
+      A.SetBlock(0,2,&A02);
+      A.SetBlock(1,0,&A10);
+      A.SetBlock(1,1,&A11);
+      A.SetBlock(1,2,&A12);
+      A.SetBlock(2,0,&A20);
+      A.SetBlock(2,1,&A21);
+      A.SetBlock(2,2,&A22);
+
+      SparseMatrix * A_mono = A.CreateMonolithic();
+      UMFPackSolver umf(*A_mono);
+      umf.Mult(rhs,x);
+
+      delta_M1_gf.MakeRef(&RTfes, x.GetBlock(0), 0);
+      delta_M2_gf.MakeRef(&RTfes, x.GetBlock(1), 0);
+      delta_u_gf.MakeRef(&H1fes, x.GetBlock(2), 0);
+
+      real_t Newton_update_size = delta_u_gf.ComputeL2Error(zero);
+
+      real_t gamma = 0.3;
+      delta_M1_gf *= gamma;
+      delta_M2_gf *= gamma;
+      delta_u_gf *= gamma;
+      M1_gf += delta_M1_gf;
+      M2_gf += delta_M2_gf;
+      u_gf += delta_u_gf;
+
+      if (visualization)
      {
-         total_iterations++;
-
-         b0.Assemble();
-         b1.Assemble();
-
-         a00.Assemble(false);
-         a00.Finalize(false);
-         SparseMatrix &A00 = a00.SpMat();
-
-         // Construct Schur-complement preconditioner
-         Vector A00_diag(a00.Height());
-         A00.GetDiag(A00_diag);
-         A00_diag.Reciprocal();
-         SparseMatrix *S = Mult_AtDA(*A01, A00_diag);
-
-         BlockDiagonalPreconditioner prec(offsets);
-         prec.SetDiagonalBlock(0,new DSmoother(A00));
-#ifndef MFEM_USE_SUITESPARSE
-         prec.SetDiagonalBlock(1,new GSSmoother(*S));
-#else
-         prec.SetDiagonalBlock(1,new UMFPackSolver(*S));
-#endif
-         prec.owns_blocks = 1;
-
-         BlockOperator A(offsets);
-         A.SetBlock(0,0,&A00);
-         A.SetBlock(1,0,&A10);
-         A.SetBlock(0,1,A01);
-         A.SetBlock(1,1,&A11);
-
-         GMRES(A,prec,rhs,x,0,2000,500,1e-12,0.0);
-         delete S;
-
-         u_tmp -= u_gf;
-         real_t Newton_update_size = u_tmp.ComputeL2Error(zero);
-         u_tmp = u_gf;
-
-         // Damped Newton update
-         psi_gf.Add(newton_scaling, delta_psi_gf);
-         a00.Update();
-
-         if (visualization)
-         {
-            sol_sock << "solution\n" << mesh << u_gf << "window_title 'Discrete solution'"
-                     << flush;
-         }
-
-         mfem::out << "Newton_update_size = " << Newton_update_size << endl;
-
-         if (Newton_update_size < increment_u)
-         {
-            break;
-         }
+         // sol_sock << "solution\n" << mesh << delta_M1_gf << "window_title 'Discrete solution'"
+         sol_sock << "solution\n" << mesh << u_gf << "window_title 'Discrete solution'"
+                  << flush;
+         mfem::out << "Increment (|| uₕ - uₕ_prvs||) = " << Newton_update_size <<
+                   endl;
      }

-      u_tmp = u_gf;
-      u_tmp -= u_old_gf;
-      increment_u = u_tmp.ComputeL2Error(zero);
+      // if (Newton_update_size < tol || k == max_it-1)
+      // {
+      //    break;
+      // }

-      mfem::out << "Number of Newton iterations = " << j+1 << endl;
-      mfem::out << "Increment (|| uₕ - uₕ_prvs||) = " << increment_u << endl;
+      real_t H1_error = u_gf.ComputeH1Error(&exact_coef,&exact_grad_coef);
+      real_t L2_error = u_gf.ComputeL2Error(exact_coef);
+      mfem::out << "L2-error  (|| u - uₕᵏ||)       = " << L2_error << endl;
+      // mfem::out << "H1-error  (|| u - uₕᵏ||)       = " << H1_error << endl;

-      u_old_gf = u_gf;
-      psi_old_gf = psi_gf;
-
-      if (increment_u < tol || k == max_it-1)
-      {
-         break;
-      }
-
-      alpha *= max(growth_rate, 1_r);
+      cin.get();

   }

-   mfem::out << "\n Outer iterations: " << k+1
-             << "\n Total iterations: " << total_iterations
-             << "\n Total dofs:       " << RTfes.GetTrueVSize() + L2fes.GetTrueVSize()
+   mfem::out << "\n Total iterations: " << k+1
+             << "\n Total dofs:       " << RTfes.GetTrueVSize() * 2 + H1fes.GetTrueVSize()
             << endl;

-   delete A01;
+   // 11. Exact solution.
+   // if (visualization)
+   // {
+   //    socketstream err_sock(vishost, visport);
+   //    err_sock.precision(8);
+
+   //    GridFunction error_gf(&H1fes);
+   //    error_gf.ProjectCoefficient(exact_coef);
+   //    error_gf -= u_gf;
+
+   //    err_sock << "solution\n" << mesh << error_gf << "window_title 'Error'"  <<
+   //             flush;
+   // }
+
   return 0;
 }

-void ZCoefficient::Eval(Vector &V, ElementTransformation &T,
-                        const IntegrationPoint &ip)
+real_t exact_solution(const Vector &pt)
 {
-   MFEM_ASSERT(psi != NULL, "grid function is not set");
-   MFEM_ASSERT(alpha > 0, "alpha is not positive");
-
-   Vector psi_vals(vdim);
-   psi->GetVectorValue(T, ip, psi_vals);
-   real_t norm = psi_vals.Norml2();
-   real_t phi = 1.0 / sqrt(1.0/(alpha*alpha) + norm*norm);
-
-   V = psi_vals;
-   V *= phi;
+   real_t x = pt(0), y = pt(1);
+   return (x*x + y*y) / 2.0 - 4.0;
 }

-void DZCoefficient::Eval(DenseMatrix &K, ElementTransformation &T,
-                         const IntegrationPoint &ip)
+void exact_solution_gradient(const Vector &pt, Vector &grad)
 {
-   MFEM_ASSERT(psi != NULL, "grid function is not set");
-   MFEM_ASSERT(alpha > 0, "alpha is not positive");
+   real_t x = pt(0), y = pt(1);

-   Vector psi_vals(height);
-   psi->GetVectorValue(T, ip, psi_vals);
-   real_t norm = psi_vals.Norml2();
-   real_t phi = 1.0 / sqrt(1.0/(alpha*alpha) + norm*norm);
-
-   K = 0.0;
-   for (int i = 0; i < height; i++)
-   {
-      K(i,i) = phi;
-      for (int j = 0; j < height; j++)
-      {
-         K(i,j) -= psi_vals(i) * psi_vals(j) * pow(phi, 3);
-      }
-   }
+   grad(0) = x;
+   grad(1) = y;
 }
@@ -1,438 +0,0 @@
-//                       MFEM Example 40 - Parallel Version
-//
-// Compile with: make ex40p
-//
-// Sample runs: mpirun -np 4 ex40p -step 10 -gr 2.0
-//              mpirun -np 4 ex40p -step 10 -gr 2.0 -o 3 -r 1
-//              mpirun -np 4 ex40p -step 10 -gr 2.0 -r 4 -m ../data/l-shape.mesh
-//              mpirun -np 4 ex40p -step 10 -gr 2.0 -r 2 -m ../data/fichera.mesh
-//
-// Description: This example code demonstrates how to use MFEM to solve the
-//              eikonal equation,
-//
-//                      |∇𝑢| = 1 in Ω,  𝑢 = g on ∂Ω.
-//
-//              The solution of this problem coincides with the unique optimum of
-//              the nonlinear program
-//
-//                   maximize ∫_Ω 𝑢 d𝑥 subject to |∇𝑢| ≤ 1,   𝑢 = g on Ω,      (⋆)
-//
-//              which is the foundation for method implemented below.
-//
-//              Following the proximal Galerkin methodology [1] (see also Example
-//              36), we construct a Legendre function for the unit ball
-//              𝐵₁ := {𝑥 ∈ Rⁿ | |𝑥| < 1}. Our choice is the Hellinger entropy,
-//
-//                    h(𝑥) = −( 1 − |𝑥|² )^{1/2},
-//
-//              although other choices are possible, each leading to a slightly
-//              different algorithm. We then adaptively regularize the optimization
-//              problem (⋆) with the Bregman divergence of the Hellinger entropy,
-//
-//                 maximize  ∫_Ω 𝑢 d𝑥 - αₖ⁻¹ Dₕ(∇𝑢,∇𝑢ₖ₋₁)  subject to  𝑢 = g on Ω.
-//
-//              This results in a sequence of functions ( 𝜓ₖ , 𝑢ₖ ),
-//
-//                      𝑢ₖ → 𝑢,    𝜓ₖ/|𝜓ₖ| → ∇𝑢    as k → \infty,
-//
-//              defined by the nonlinear saddle-point problems
-//
-//               Find 𝜓ₖ ∈ H(div,Ω) and 𝑢ₖ ∈ L²(Ω) such that
-//               ( Zₖ(𝜓ₖ) , τ ) + ( 𝑢ₖ , ∇⋅τ ) = ⟨ g , τ⋅n ⟩         ∀ τ ∈ H(div,Ω)
-//               ( ∇⋅𝜓ₖ , v )                 = ( ∇⋅𝜓ₖ₋₁ - 1 , v )   ∀ v ∈ L²(Ω)
-//
-//              where Zₖ(𝜓) := ∇h⁻¹(αₖ 𝜓) = 𝜓 / ( αₖ⁻² + |𝜓|² )^{1/2} and step size
-//              αₖ > 0. These saddle-point problems are solved using a damped Newton's
-//              method. This example assumes that g = 0 and allows the step size to
-//              grow geometrically, αₖ = α₀rᵏ, where r ≥ 1 is the growth rate.
-//
-//              [1] Keith, B. and Surowiec, T. (2023) Proximal Galerkin: A structure-
-//                  preserving finite element method for pointwise bound constraints.
-//                  arXiv:2307.12444 [math.NA]
-
-#include "mfem.hpp"
-#include <fstream>
-#include <iostream>
-
-using namespace std;
-using namespace mfem;
-
-class ZCoefficient : public VectorCoefficient
-{
-protected:
-   ParGridFunction *psi;
-   real_t alpha;
-
-public:
-   ZCoefficient(int vdim, ParGridFunction &psi_, real_t alpha_ = 1.0)
-      : VectorCoefficient(vdim), psi(&psi_), alpha(alpha_) { }
-
-   using VectorCoefficient::Eval;
-
-   void Eval(Vector &V, ElementTransformation &T,
-             const IntegrationPoint &ip) override;
-   void SetAlpha(real_t alpha_) { alpha = alpha_; }
-};
-
-class DZCoefficient : public MatrixCoefficient
-{
-protected:
-   ParGridFunction *psi;
-   real_t alpha;
-
-public:
-   DZCoefficient(int height, ParGridFunction &psi_, real_t alpha_ = 1.0)
-      : MatrixCoefficient(height),  psi(&psi_), alpha(alpha_) { }
-
-   void Eval(DenseMatrix &K, ElementTransformation &T,
-             const IntegrationPoint &ip) override;
-   void SetAlpha(real_t alpha_) { alpha = alpha_; }
-};
-
-int main(int argc, char *argv[])
-{
-   // 0. Initialize MPI and HYPRE.
-   Mpi::Init();
-   int num_procs = Mpi::WorldSize();
-   int myid = Mpi::WorldRank();
-   Hypre::Init();
-
-   // 1. Parse command-line options.
-   const char *mesh_file = "../data/star.mesh";
-   int order = 1;
-   int max_it = 5;
-   int ref_levels = 3;
-   real_t alpha = 1.0;
-   real_t growth_rate = 1.0;
-   real_t newton_scaling = 0.9;
-   real_t tichonov = 1e-1;
-   real_t tol = 1e-4;
-   bool visualization = true;
-
-   OptionsParser args(argc, argv);
-   args.AddOption(&mesh_file, "-m", "--mesh",
-                  "Mesh file to use.");
-   args.AddOption(&order, "-o", "--order",
-                  "Finite element order (polynomial degree).");
-   args.AddOption(&ref_levels, "-r", "--refs",
-                  "Number of h-refinements.");
-   args.AddOption(&max_it, "-mi", "--max-it",
-                  "Maximum number of iterations");
-   args.AddOption(&tol, "-tol", "--tol",
-                  "Stopping criteria based on the difference between"
-                  "successive solution updates");
-   args.AddOption(&alpha, "-step", "--step",
-                  "Initial size alpha");
-   args.AddOption(&growth_rate, "-gr", "--growth-rate",
-                  "Growth rate of the step size alpha");
-   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
-                  "--no-visualization",
-                  "Enable or disable GLVis visualization.");
-   args.Parse();
-   if (!args.Good())
-   {
-      if (myid == 0)
-      {
-         args.PrintUsage(cout);
-      }
-      return 1;
-   }
-   if (myid == 0)
-   {
-      args.PrintOptions(cout);
-   }
-
-   // 2. Read the mesh from the mesh file.
-   Mesh mesh(mesh_file, 1, 1);
-   int dim = mesh.Dimension();
-   int sdim = mesh.SpaceDimension();
-
-   MFEM_ASSERT(mesh.bdr_attributes.Size(),
-               "This example does not currently support meshes"
-               " without boundary attributes."
-              )
-
-   // 3. Postprocess the mesh.
-   // 3A. Refine the mesh to increase the resolution.
-   for (int l = 0; l < ref_levels; l++)
-   {
-      mesh.UniformRefinement();
-   }
-
-   // 3B. Interpolate the geometry after refinement to control geometry error.
-   // NOTE: Minimum second-order interpolation is used to improve the accuracy.
-   int curvature_order = max(order,2);
-   mesh.SetCurvature(curvature_order);
-
-   ParMesh pmesh(MPI_COMM_WORLD, mesh);
-   mesh.Clear();
-
-   // 4. Define the necessary finite element spaces on the mesh.
-   RT_FECollection RTfec(order, dim);
-   ParFiniteElementSpace RTfes(&pmesh, &RTfec);
-
-   L2_FECollection L2fec(order, dim);
-   ParFiniteElementSpace L2fes(&pmesh, &L2fec);
-
-   int num_dofs_RT = RTfes.GlobalTrueVSize();
-   int num_dofs_L2 = L2fes.GlobalTrueVSize();
-   if (myid == 0)
-   {
-      cout << "Number of H(div) dofs: "
-           << num_dofs_RT << endl;
-      cout << "Number of L² dofs: "
-           << num_dofs_L2 << endl;
-   }
-
-   // 5. Define the offsets for the block matrices
-   Array<int> offsets(3);
-   offsets[0] = 0;
-   offsets[1] = RTfes.GetVSize();
-   offsets[2] = L2fes.GetVSize();
-   offsets.PartialSum();
-
-   Array<int> toffsets(3);
-   toffsets[0] = 0;
-   toffsets[1] = RTfes.GetTrueVSize();
-   toffsets[2] = L2fes.GetTrueVSize();
-   toffsets.PartialSum();
-
-   BlockVector x(offsets), rhs(offsets);
-   x = 0.0; rhs = 0.0;
-
-   BlockVector tx(toffsets), trhs(toffsets);
-   tx = 0.0; trhs = 0.0;
-
-   // 6. Define the solution vectors as a finite element grid functions
-   //    corresponding to the fespaces.
-   ParGridFunction u_gf, delta_psi_gf;
-   delta_psi_gf.MakeRef(&RTfes,x,offsets[0]);
-   u_gf.MakeRef(&L2fes,x,offsets[1]);
-
-   ParGridFunction psi_old_gf(&RTfes);
-   ParGridFunction psi_gf(&RTfes);
-   ParGridFunction u_old_gf(&L2fes);
-
-   // 7. Define initial guesses for the solution variables.
-   delta_psi_gf = 0.0;
-   psi_gf = 0.0;
-   u_gf = 0.0;
-   psi_old_gf = psi_gf;
-   u_old_gf = u_gf;
-
-   // 8. Prepare for glvis output.
-   char vishost[] = "localhost";
-   int  visport   = 19916;
-   socketstream sol_sock;
-   if (visualization)
-   {
-      sol_sock.open(vishost,visport);
-      sol_sock.precision(8);
-   }
-
-   // 9. Coefficients to be used later.
-   ConstantCoefficient neg_one(-1.0);
-   ConstantCoefficient zero(0.0);
-   ConstantCoefficient tichonov_cf(tichonov);
-   ConstantCoefficient neg_tichonov_cf(-1.0*tichonov);
-   ZCoefficient Z(sdim, psi_gf, alpha);
-   DZCoefficient DZ(sdim, psi_gf, alpha);
-   ScalarVectorProductCoefficient neg_Z(-1.0, Z);
-   DivergenceGridFunctionCoefficient div_psi_cf(&psi_gf);
-   DivergenceGridFunctionCoefficient div_psi_old_cf(&psi_old_gf);
-   SumCoefficient psi_old_minus_psi(div_psi_old_cf, div_psi_cf, 1.0, -1.0);
-
-   // 10. Assemble constant matrices/vectors to avoid reassembly in the loop.
-   ParLinearForm b0, b1;
-   b0.MakeRef(&RTfes,rhs.GetBlock(0),0);
-   b1.MakeRef(&L2fes,rhs.GetBlock(1),0);
-
-   b0.AddDomainIntegrator(new VectorFEDomainLFIntegrator(neg_Z));
-   b1.AddDomainIntegrator(new DomainLFIntegrator(neg_one));
-   b1.AddDomainIntegrator(new DomainLFIntegrator(psi_old_minus_psi));
-
-   ParBilinearForm a00(&RTfes);
-   a00.AddDomainIntegrator(new VectorFEMassIntegrator(DZ));
-   a00.AddDomainIntegrator(new VectorFEMassIntegrator(tichonov_cf));
-
-   ParMixedBilinearForm a10(&RTfes,&L2fes);
-   a10.AddDomainIntegrator(new VectorFEDivergenceIntegrator());
-   a10.Assemble();
-   a10.Finalize();
-   HypreParMatrix *A10 = a10.ParallelAssemble();
-
-   HypreParMatrix *A01 = A10->Transpose();
-
-   ParBilinearForm a11(&L2fes);
-   a11.AddDomainIntegrator(new MassIntegrator(neg_tichonov_cf));
-   a11.Assemble();
-   a11.Finalize();
-   HypreParMatrix *A11 = a11.ParallelAssemble();
-
-   // 11. Iterate.
-   int k;
-   int total_iterations = 0;
-   real_t increment_u = 0.1;
-   ParGridFunction u_tmp(&L2fes);
-   for (k = 0; k < max_it; k++)
-   {
-      u_tmp = u_old_gf;
-      Z.SetAlpha(alpha);
-      DZ.SetAlpha(alpha);
-
-      if (myid == 0)
-      {
-         mfem::out << "\nOUTER ITERATION " << k+1 << endl;
-      }
-
-      int j;
-      for ( j = 0; j < 5; j++)
-      {
-         total_iterations++;
-
-         b0.Assemble();
-         b0.ParallelAssemble(trhs.GetBlock(0));
-
-         b1.Assemble();
-         b1.ParallelAssemble(trhs.GetBlock(1));
-
-         a00.Assemble(false);
-         a00.Finalize(false);
-         HypreParMatrix *A00 = a00.ParallelAssemble();
-
-         // Construct Schur-complement preconditioner
-         HypreParVector A00_diag(MPI_COMM_WORLD, A00->GetGlobalNumRows(),
-                                 A00->GetRowStarts());
-         A00->GetDiag(A00_diag);
-         HypreParMatrix S_tmp(*A01);
-         S_tmp.InvScaleRows(A00_diag);
-         HypreParMatrix *S = ParMult(A10, &S_tmp, true);
-
-         BlockDiagonalPreconditioner prec(toffsets);
-         HypreBoomerAMG P00(*A00);
-         P00.SetPrintLevel(0);
-         HypreBoomerAMG P11(*S);
-         P11.SetPrintLevel(0);
-         prec.SetDiagonalBlock(0,&P00);
-         prec.SetDiagonalBlock(1,&P11);
-
-         BlockOperator A(toffsets);
-         A.SetBlock(0,0,A00);
-         A.SetBlock(1,0,A10);
-         A.SetBlock(0,1,A01);
-         A.SetBlock(1,1,A11);
-
-         GMRESSolver gmres(MPI_COMM_WORLD);
-         gmres.SetPrintLevel(-1);
-         gmres.SetRelTol(1e-8);
-         gmres.SetMaxIter(2000);
-         gmres.SetKDim(500);
-         gmres.SetOperator(A);
-         gmres.SetPreconditioner(prec);
-         gmres.Mult(trhs,tx);
-         delete S;
-         delete A00;
-
-         delta_psi_gf.SetFromTrueDofs(tx.GetBlock(0));
-         u_gf.SetFromTrueDofs(tx.GetBlock(1));
-
-         u_tmp -= u_gf;
-         real_t Newton_update_size = u_tmp.ComputeL2Error(zero);
-         u_tmp = u_gf;
-
-         // Damped Newton update
-         psi_gf.Add(newton_scaling, delta_psi_gf);
-         a00.Update();
-
-         if (visualization)
-         {
-            sol_sock << "parallel " << num_procs << " " << myid << "\n";
-            sol_sock << "solution\n" << pmesh << u_gf << "window_title 'Discrete solution'"
-                     << flush;
-         }
-
-         if (myid == 0)
-         {
-            mfem::out << "Newton_update_size = " << Newton_update_size << endl;
-         }
-
-         if (Newton_update_size < increment_u)
-         {
-            break;
-         }
-      }
-
-      u_tmp = u_gf;
-      u_tmp -= u_old_gf;
-      increment_u = u_tmp.ComputeL2Error(zero);
-
-      if (myid == 0)
-      {
-         mfem::out << "Number of Newton iterations = " << j+1 << endl;
-         mfem::out << "Increment (|| uₕ - uₕ_prvs||) = " << increment_u << endl;
-      }
-
-      u_old_gf = u_gf;
-      psi_old_gf = psi_gf;
-
-      if (increment_u < tol || k == max_it-1)
-      {
-         break;
-      }
-
-      alpha *= max(growth_rate, 1_r);
-
-   }
-
-   // 12. Print stats.
-   if (myid == 0)
-   {
-      mfem::out << "\n Outer iterations: " << k+1
-                << "\n Total iterations: " << total_iterations
-                << "\n Total dofs:       " << RTfes.GetTrueVSize() + L2fes.GetTrueVSize()
-                << endl;
-   }
-
-   // 13. Free the used memory.
-   delete A01;
-   delete A10;
-   delete A11;
-   return 0;
-}
-
-void ZCoefficient::Eval(Vector &V, ElementTransformation &T,
-                        const IntegrationPoint &ip)
-{
-   MFEM_ASSERT(psi != NULL, "grid function is not set");
-   MFEM_ASSERT(alpha > 0, "alpha is not positive");
-
-   Vector psi_vals(vdim);
-   psi->GetVectorValue(T, ip, psi_vals);
-   real_t norm = psi_vals.Norml2();
-   real_t phi = 1.0 / sqrt(1.0/(alpha*alpha) + norm*norm);
-
-   V = psi_vals;
-   V *= phi;
-}
-
-void DZCoefficient::Eval(DenseMatrix &K, ElementTransformation &T,
-                         const IntegrationPoint &ip)
-{
-   MFEM_ASSERT(psi != NULL, "grid function is not set");
-   MFEM_ASSERT(alpha > 0, "alpha is not positive");
-
-   Vector psi_vals(height);
-   psi->GetVectorValue(T, ip, psi_vals);
-   real_t norm = psi_vals.Norml2();
-   real_t phi = 1.0 / sqrt(1.0/(alpha*alpha) + norm*norm);
-
-   K = 0.0;
-   for (int i = 0; i < height; i++)
-   {
-      K(i,i) = phi;
-      for (int j = 0; j < height; j++)
-      {
-         K(i,j) -= psi_vals(i) * psi_vals(j) * pow(phi, 3);
-      }
-   }
-}
@@ -157,7 +157,7 @@ int main(int argc, char *argv[])
   MixedBilinearForm *B0 = new MixedBilinearForm(x0_space,test_space);
   B0->AddDomainIntegrator(new DiffusionIntegrator(one));
   B0->Assemble();
-   B0->EliminateTrialEssentialBC(ess_bdr, x.GetBlock(x0_var), F);
+   B0->EliminateTrialDofs(ess_bdr, x.GetBlock(x0_var), F);
   B0->Finalize();

   MixedBilinearForm *Bhat = new MixedBilinearForm(xhat_space,test_space);
@@ -9,7 +9,7 @@
 //    ex9 -m ../data/periodic-square.mesh -p 1 -r 2 -dt 0.005 -tf 9
 //    ex9 -m ../data/periodic-hexagon.mesh -p 1 -r 2 -dt 0.005 -tf 9
 //    ex9 -m ../data/amr-quad.mesh -p 1 -r 2 -dt 0.002 -tf 9
-//    ex9 -m ../data/amr-quad.mesh -p 1 -r 2 -dt 0.02 -s 23 -tf 9
+//    ex9 -m ../data/amr-quad.mesh -p 1 -r 2 -dt 0.02 -s 13 -tf 9
 //    ex9 -m ../data/star-q3.mesh -p 1 -r 2 -dt 0.005 -tf 9
 //    ex9 -m ../data/star-mixed.mesh -p 1 -r 2 -dt 0.005 -tf 9
 //    ex9 -m ../data/disc-nurbs.mesh -p 1 -r 3 -dt 0.005 -tf 9
@@ -104,12 +104,12 @@ public:
      }
   }

-   void SetOperator(const Operator &op) override
+   void SetOperator(const Operator &op)
   {
      linear_solver.SetOperator(op);
   }

-   void Mult(const Vector &x, Vector &y) const override
+   virtual void Mult(const Vector &x, Vector &y) const
   {
      linear_solver.Mult(x, y);
   }
@@ -134,10 +134,10 @@ private:
 public:
   FE_Evolution(BilinearForm &M_, BilinearForm &K_, const Vector &b_);

-   void Mult(const Vector &x, Vector &y) const override;
-   void ImplicitSolve(const real_t dt, const Vector &x, Vector &k) override;
+   virtual void Mult(const Vector &x, Vector &y) const;
+   virtual void ImplicitSolve(const real_t dt, const Vector &x, Vector &k);

-   ~FE_Evolution() override;
+   virtual ~FE_Evolution();
 };


@@ -182,7 +182,12 @@ int main(int argc, char *argv[])
   args.AddOption(&device_config, "-d", "--device",
                  "Device configuration string, see Device::Configure().");
   args.AddOption(&ode_solver_type, "-s", "--ode-solver",
-                  ODESolver::Types.c_str());
+                  "ODE solver: 1 - Forward Euler,\n\t"
+                  "            2 - RK2 SSP, 3 - RK3 SSP, 4 - RK4, 6 - RK6,\n\t"
+                  "            11 - Backward Euler,\n\t"
+                  "            12 - SDIRK23 (L-stable), 13 - SDIRK33,\n\t"
+                  "            22 - Implicit Midpoint Method,\n\t"
+                  "            23 - SDIRK23 (A-stable), 24 - SDIRK34");
   args.AddOption(&t_final, "-tf", "--t-final",
                  "Final time; start time is 0.");
   args.AddOption(&dt, "-dt", "--time-step",
@@ -219,7 +224,28 @@ int main(int argc, char *argv[])

   // 3. Define the ODE solver used for time integration. Several explicit
   //    Runge-Kutta methods are available.
-   unique_ptr<ODESolver> ode_solver = ODESolver::Select(ode_solver_type);
+   ODESolver *ode_solver = NULL;
+   switch (ode_solver_type)
+   {
+      // Explicit methods
+      case 1: ode_solver = new ForwardEulerSolver; break;
+      case 2: ode_solver = new RK2Solver(1.0); break;
+      case 3: ode_solver = new RK3SSPSolver; break;
+      case 4: ode_solver = new RK4Solver; break;
+      case 6: ode_solver = new RK6Solver; break;
+      // Implicit (L-stable) methods
+      case 11: ode_solver = new BackwardEulerSolver; break;
+      case 12: ode_solver = new SDIRK23Solver(2); break;
+      case 13: ode_solver = new SDIRK33Solver; break;
+      // Implicit A-stable methods (not L-stable)
+      case 22: ode_solver = new ImplicitMidpointSolver; break;
+      case 23: ode_solver = new SDIRK23Solver; break;
+      case 24: ode_solver = new SDIRK34Solver; break;
+
+      default:
+         cout << "Unknown ODE solver type: " << ode_solver_type << '\n';
+         return 3;
+   }

   // 4. Refine the mesh to increase the resolution. In this example we do
   //    'ref_levels' of uniform refinement, where 'ref_levels' is a
@@ -414,6 +440,7 @@ int main(int argc, char *argv[])
   }

   // 10. Free the used memory.
+   delete ode_solver;
   delete pd;
   delete dc;

@@ -9,7 +9,7 @@
 //    mpirun -np 4 ex9p -m ../data/periodic-square.mesh -p 1 -dt 0.005 -tf 9
 //    mpirun -np 4 ex9p -m ../data/periodic-hexagon.mesh -p 1 -dt 0.005 -tf 9
 //    mpirun -np 4 ex9p -m ../data/amr-quad.mesh -p 1 -rp 1 -dt 0.002 -tf 9
-//    mpirun -np 4 ex9p -m ../data/amr-quad.mesh -p 1 -rp 1 -dt 0.02 -s 23 -tf 9
+//    mpirun -np 4 ex9p -m ../data/amr-quad.mesh -p 1 -rp 1 -dt 0.02 -s 13 -tf 9
 //    mpirun -np 4 ex9p -m ../data/star-q3.mesh -p 1 -rp 1 -dt 0.004 -tf 9
 //    mpirun -np 4 ex9p -m ../data/star-mixed.mesh -p 1 -rp 1 -dt 0.004 -tf 9
 //    mpirun -np 4 ex9p -m ../data/disc-nurbs.mesh -p 1 -rp 1 -dt 0.005 -tf 9
@@ -92,7 +92,7 @@ private:
 public:
   AIR_prec(int blocksize_) : AIR_solver(NULL), blocksize(blocksize_) { }

-   void SetOperator(const Operator &op) override
+   void SetOperator(const Operator &op)
   {
      width = op.Width();
      height = op.Height();
@@ -110,7 +110,7 @@ public:
      AIR_solver->SetMaxLevels(50);
   }

-   void Mult(const Vector &x, Vector &y) const override
+   virtual void Mult(const Vector &x, Vector &y) const
   {
      // Scale the rhs by block inverse and solve system
      HypreParVector z_s;
@@ -119,7 +119,7 @@ public:
      AIR_solver->Mult(z_s, y);
   }

-   ~AIR_prec() override
+   ~AIR_prec()
   {
      delete AIR_solver;
   }
@@ -185,17 +185,17 @@ public:
      }
   }

-   void SetOperator(const Operator &op) override
+   void SetOperator(const Operator &op)
   {
      linear_solver.SetOperator(op);
   }

-   void Mult(const Vector &x, Vector &y) const override
+   virtual void Mult(const Vector &x, Vector &y) const
   {
      linear_solver.Mult(x, y);
   }

-   ~DG_Solver() override
+   ~DG_Solver()
   {
      delete prec;
      delete A;
@@ -223,10 +223,10 @@ public:
   FE_Evolution(ParBilinearForm &M_, ParBilinearForm &K_, const Vector &b_,
                PrecType prec_type);

-   void Mult(const Vector &x, Vector &y) const override;
-   void ImplicitSolve(const real_t dt, const Vector &x, Vector &k) override;
+   virtual void Mult(const Vector &x, Vector &y) const;
+   virtual void ImplicitSolve(const real_t dt, const Vector &x, Vector &k);

-   ~FE_Evolution() override;
+   virtual ~FE_Evolution();
 };


@@ -285,7 +285,12 @@ int main(int argc, char *argv[])
   args.AddOption(&device_config, "-d", "--device",
                  "Device configuration string, see Device::Configure().");
   args.AddOption(&ode_solver_type, "-s", "--ode-solver",
-                  ODESolver::Types.c_str());
+                  "ODE solver: 1 - Forward Euler,\n\t"
+                  "            2 - RK2 SSP, 3 - RK3 SSP, 4 - RK4, 6 - RK6,\n\t"
+                  "            11 - Backward Euler,\n\t"
+                  "            12 - SDIRK23 (L-stable), 13 - SDIRK33,\n\t"
+                  "            22 - Implicit Midpoint Method,\n\t"
+                  "            23 - SDIRK23 (A-stable), 24 - SDIRK34");
   args.AddOption(&t_final, "-tf", "--t-final",
                  "Final time; start time is 0.");
   args.AddOption(&dt, "-dt", "--time-step",
@@ -333,7 +338,31 @@ int main(int argc, char *argv[])

   // 4. Define the ODE solver used for time integration. Several explicit
   //    Runge-Kutta methods are available.
-   unique_ptr<ODESolver> ode_solver = ODESolver::Select(ode_solver_type);
+   ODESolver *ode_solver = NULL;
+   switch (ode_solver_type)
+   {
+      // Explicit methods
+      case 1: ode_solver = new ForwardEulerSolver; break;
+      case 2: ode_solver = new RK2Solver(1.0); break;
+      case 3: ode_solver = new RK3SSPSolver; break;
+      case 4: ode_solver = new RK4Solver; break;
+      case 6: ode_solver = new RK6Solver; break;
+      // Implicit (L-stable) methods
+      case 11: ode_solver = new BackwardEulerSolver; break;
+      case 12: ode_solver = new SDIRK23Solver(2); break;
+      case 13: ode_solver = new SDIRK33Solver; break;
+      // Implicit A-stable methods (not L-stable)
+      case 22: ode_solver = new ImplicitMidpointSolver; break;
+      case 23: ode_solver = new SDIRK23Solver; break;
+      case 24: ode_solver = new SDIRK34Solver; break;
+      default:
+         if (Mpi::Root())
+         {
+            cout << "Unknown ODE solver type: " << ode_solver_type << '\n';
+         }
+         delete mesh;
+         return 3;
+   }

   // 5. Refine the mesh in serial to increase the resolution. In this example
   //    we do 'ser_ref_levels' of uniform refinement, where 'ser_ref_levels' is
@@ -613,6 +642,7 @@ int main(int argc, char *argv[])
   delete m;
   delete fes;
   delete pmesh;
+   delete ode_solver;
   delete pd;
 #ifdef MFEM_USE_ADIOS2
   if (adios2)
@@ -12,10 +12,11 @@
 # Use the MFEM build directory
 MFEM_DIR ?= ../..
 MFEM_BUILD_DIR ?= ../..
-MFEM_INSTALL_DIR ?= ../../mfem
 SRC = $(if $(MFEM_DIR:../..=),$(MFEM_DIR)/examples/ginkgo/,)
-CONFIG_MK = $(or $(wildcard $(MFEM_BUILD_DIR)/config/config.mk),\
-   $(wildcard $(MFEM_INSTALL_DIR)/share/mfem/config.mk))
+CONFIG_MK = $(MFEM_BUILD_DIR)/config/config.mk
+# Use the MFEM install directory
+# MFEM_INSTALL_DIR = ../../mfem
+# CONFIG_MK = $(MFEM_INSTALL_DIR)/share/mfem/config.mk

 MFEM_LIB_FILE = mfem_is_not_built
 -include $(CONFIG_MK)
@@ -96,7 +96,6 @@ public:
   {
      Vector w_glob(width);
      pfes.Dof_TrueDof_Matrix()->MultTranspose(w, w_glob);
-      w_glob.HostReadWrite(); // read+write -> can use w_glob(i) (non-const)
      for (int i = 0; i < width; i++) { grad(0, i) = w_glob(i); }
   }

@@ -12,10 +12,11 @@
 # Use the MFEM build directory
 MFEM_DIR ?= ../..
 MFEM_BUILD_DIR ?= ../..
-MFEM_INSTALL_DIR ?= ../../mfem
 SRC = $(if $(MFEM_DIR:../..=),$(MFEM_DIR)/examples/hiop/,)
-CONFIG_MK = $(or $(wildcard $(MFEM_BUILD_DIR)/config/config.mk),\
-   $(wildcard $(MFEM_INSTALL_DIR)/share/mfem/config.mk))
+CONFIG_MK = $(MFEM_BUILD_DIR)/config/config.mk
+# Use the MFEM install directory
+# MFEM_INSTALL_DIR = ../../mfem
+# CONFIG_MK = $(MFEM_INSTALL_DIR)/share/mfem/config.mk

 MFEM_LIB_FILE = mfem_is_not_built
 -include $(CONFIG_MK)
@@ -12,10 +12,11 @@
 # Use the MFEM build directory
 MFEM_DIR ?= ..
 MFEM_BUILD_DIR ?= ..
-MFEM_INSTALL_DIR ?= ../mfem
 SRC = $(if $(MFEM_DIR:..=),$(MFEM_DIR)/examples/,)
-CONFIG_MK = $(or $(wildcard $(MFEM_BUILD_DIR)/config/config.mk),\
-   $(wildcard $(MFEM_INSTALL_DIR)/share/mfem/config.mk))
+CONFIG_MK = $(MFEM_BUILD_DIR)/config/config.mk
+# Use the MFEM install directory
+# MFEM_INSTALL_DIR = ../mfem
+# CONFIG_MK = $(MFEM_INSTALL_DIR)/share/mfem/config.mk

 MFEM_LIB_FILE = mfem_is_not_built
 -include $(CONFIG_MK)
@@ -27,9 +28,9 @@ PAR_EXAMPLES = ex0p ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex8p ex9p ex10p ex11p \
   ex12p ex13p ex14p ex15p ex16p ex17p ex18p ex19p ex20p ex21p ex22p ex24p \
   ex25p ex26p ex27p ex28p ex29p ex30p ex31p ex32p ex33p ex34p ex35p ex36p \
   ex37p ex39p ex40p
-SEQ_DEVICE_EXAMPLES = ex1 ex3 ex4 ex5 ex6 ex9 ex14 ex22 ex24 ex25 ex26 ex34
-PAR_DEVICE_EXAMPLES = ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex9p ex13p ex14p \
-   ex22p ex24p ex25p ex26p ex34p ex35p
+SEQ_DEVICE_EXAMPLES = ex1 ex3 ex4 ex5 ex6 ex9 ex22 ex24 ex25 ex26 ex34
+PAR_DEVICE_EXAMPLES = ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex9p ex13p ex22p \
+   ex24p ex25p ex26p ex34p ex35p

 ifeq ($(MFEM_USE_LAPACK),YES)
   SEQ_EXAMPLES += ex38
@@ -137,14 +138,6 @@ ex10-test-seq: ex10
 	@$(call mfem-test,$<,, Serial example,-tf 5)
 ex10p-test-par: ex10p
 	@$(call mfem-test,$<, $(RUN_MPI), Parallel example,-tf 5)
-ex14-test-seq-cuda: ex14
-	@$(call mfem-test,$<,, Serial CUDA example,-r 2 -pa -d cuda)
-ex14p-test-par-cuda: ex14p
-	@$(call mfem-test,$<, $(RUN_MPI), Parallel CUDA example,-rs 2 -rp 0 -pa -d cuda)
-ex14-test-seq-hip: ex14
-	@$(call mfem-test,$<,, Serial HIP example,-r 2 -pa -d hip)
-ex14p-test-par-hip: ex14p
-	@$(call mfem-test,$<, $(RUN_MPI), Parallel HIP example,-rs 2 -rp 0 -pa -d hip)
 ex15-test-seq: ex15
 	@$(call mfem-test,$<,, Serial example,-e 1)
 ex15p-test-par: ex15p
@@ -1,14 +1,3 @@
-// Copyright (c) 2010-2024, Lawrence Livermore National Security, LLC. Produced
-// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
-// LICENSE and NOTICE for details. LLNL-CODE-806117.
-//
-// This file is part of the MFEM library. For more information and source code
-// availability visit https://mfem.org.
-//
-// MFEM is free software; you can redistribute it and/or modify it under the
-// terms of the BSD-3 license. We welcome feedback and contributions, see file
-// CONTRIBUTING.md for details.
-
 #include <algorithm>
 #include <assert.h>
 #include <cstdlib>
@@ -12,10 +12,11 @@
 # Use the MFEM build directory
 MFEM_DIR ?= ../..
 MFEM_BUILD_DIR ?= ../..
-MFEM_INSTALL_DIR ?= ../../mfem
 SRC = $(if $(MFEM_DIR:../..=),$(MFEM_DIR)/examples/moonolith/,)
-CONFIG_MK = $(or $(wildcard $(MFEM_BUILD_DIR)/config/config.mk),\
-   $(wildcard $(MFEM_INSTALL_DIR)/share/mfem/config.mk))
+CONFIG_MK = $(MFEM_BUILD_DIR)/config/config.mk
+# Use the MFEM install directory
+# MFEM_INSTALL_DIR = ../../mfem
+# CONFIG_MK = $(MFEM_INSTALL_DIR)/share/mfem/config.mk

 MFEM_LIB_FILE = mfem_is_not_built
 -include $(CONFIG_MK)
@@ -206,7 +206,6 @@ int main(int argc, char *argv[])
   bool use_petsc = true;
   const char *petscrc_file = "";
   bool petsc_use_jfnk = false;
-   const char *device_config = "cpu";

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -244,8 +243,6 @@ int main(int argc, char *argv[])
   args.AddOption(&petsc_use_jfnk, "-jfnk", "--jfnk", "-no-jfnk",
                  "--no-jfnk",
                  "Use JFNK with user-defined preconditioner factory.");
-   args.AddOption(&device_config, "-d", "--device",
-                  "Device configuration string, see Device::Configure().");
   args.Parse();
   if (!args.Good())
   {
@@ -260,12 +257,7 @@ int main(int argc, char *argv[])
      args.PrintOptions(cout);
   }

-   // 2b. 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(); }
-
-   // 2c. We initialize PETSc
+   // 2b. We initialize PETSc
   if (use_petsc)
   {
      MFEMInitializePetsc(NULL,NULL,petscrc_file,NULL);
@@ -717,7 +709,10 @@ real_t HyperelasticOperator::ElasticEnergy(const ParGridFunction &x) const

 real_t HyperelasticOperator::KineticEnergy(const ParGridFunction &v) const
 {
-   real_t energy = 0.5*M.ParInnerProduct(v, v);
+   real_t loc_energy = 0.5*M.InnerProduct(v, v);
+   real_t energy;
+   MPI_Allreduce(&loc_energy, &energy, 1, MPITypeMap<real_t>::mpi_type, MPI_SUM,
+                 fespace.GetComm());
   return energy;
 }

@@ -67,7 +67,6 @@ int main(int argc, char *argv[])
   bool use_petsc = true;
   const char *petscrc_file = "";
   bool use_nonoverlapping = false;
-   const char *device_config = "cpu";

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -96,8 +95,6 @@ int main(int argc, char *argv[])
                  "-no-nonoverlapping", "--no-nonoverlapping",
                  "Use or not the block diagonal PETSc's matrix format "
                  "for non-overlapping domain decomposition.");
-   args.AddOption(&device_config, "-d", "--device",
-                  "Device configuration string, see Device::Configure().");
   args.Parse();
   if (!args.Good())
   {
@@ -112,12 +109,7 @@ int main(int argc, char *argv[])
      args.PrintOptions(cout);
   }

-   // 2b. 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(); }
-
-   // 2c. We initialize PETSc
+   // 2b. We initialize PETSc
   if (use_petsc) { MFEMInitializePetsc(NULL,NULL,petscrc_file,NULL); }

   // 3. Read the (serial) mesh from the given mesh file on all processors.  We
@@ -61,7 +61,6 @@ int main(int argc, char *argv[])
   bool use_petsc = true;
   const char *petscrc_file = "";
   bool use_nonoverlapping = false;
-   const char *device_config = "cpu";

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -88,8 +87,6 @@ int main(int argc, char *argv[])
                  "-no-nonoverlapping", "--no-nonoverlapping",
                  "Use or not the block diagonal PETSc's matrix format "
                  "for non-overlapping domain decomposition.");
-   args.AddOption(&device_config, "-d", "--device",
-                  "Device configuration string, see Device::Configure().");
   args.Parse();
   if (!args.Good())
   {
@@ -103,15 +100,9 @@ int main(int argc, char *argv[])
   {
      args.PrintOptions(cout);
   }
-   kappa = freq * M_PI;
-
-   // 2b. 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(); }
-
-   // 2c. We initialize PETSc
+   // 2b. We initialize PETSc
   if (use_petsc) { MFEMInitializePetsc(NULL,NULL,petscrc_file,NULL); }
+   kappa = freq * M_PI;

   // 3. Read the (serial) mesh from the given mesh file on all processors.  We
   //    can handle triangular, quadrilateral, tetrahedral, hexahedral, surface
@@ -58,7 +58,6 @@ int main(int argc, char *argv[])
   bool use_petsc = true;
   const char *petscrc_file = "";
   bool use_nonoverlapping = false;
-   const char *device_config = "cpu";

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -89,8 +88,6 @@ int main(int argc, char *argv[])
                  "-no-nonoverlapping", "--no-nonoverlapping",
                  "Use or not the block diagonal PETSc's matrix format "
                  "for non-overlapping domain decomposition.");
-   args.AddOption(&device_config, "-d", "--device",
-                  "Device configuration string, see Device::Configure().");
   args.Parse();
   if (!args.Good())
   {
@@ -104,15 +101,9 @@ int main(int argc, char *argv[])
   {
      args.PrintOptions(cout);
   }
-   kappa = freq * M_PI;
-
-   // 2b. 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(); }
-
-   // 2c. We initialize PETSc
+   // 2b. We initialize PETSc
   if (use_petsc) { MFEMInitializePetsc(NULL,NULL,petscrc_file,NULL); }
+   kappa = freq * M_PI;

   // 3. Read the (serial) mesh from the given mesh file on all processors.  We
   //    can handle triangular, quadrilateral, tetrahedral, hexahedral, surface
@@ -59,8 +59,6 @@ int main(int argc, char *argv[])

   // 2. Parse command-line options.
   const char *mesh_file = "../../data/star.mesh";
-   int ser_ref_levels = -1;
-   int par_ref_levels = 2;
   int order = 1;
   bool par_format = false;
   bool visualization = 1;
@@ -68,22 +66,15 @@ int main(int argc, char *argv[])
   bool use_nonoverlapping = false;
   bool local_bdr_spec = false;
   const char *petscrc_file = "";
-   const char *device_config = "cpu";

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
                  "Mesh file to use.");
-   args.AddOption(&ser_ref_levels, "-rs", "--refine-serial",
-                  "Number of times to refine the mesh uniformly in serial.");
-   args.AddOption(&par_ref_levels, "-rp", "--refine-parallel",
-                  "Number of times to refine the mesh uniformly in parallel.");
   args.AddOption(&order, "-o", "--order",
                  "Finite element order (polynomial degree).");
   args.AddOption(&par_format, "-pf", "--parallel-format", "-sf",
                  "--serial-format",
                  "Format to use when saving the results for VisIt.");
-   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.");
@@ -112,13 +103,7 @@ int main(int argc, char *argv[])
   {
      args.PrintOptions(cout);
   }
-
-   // 2b. 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(); }
-
-   // 2c. We initialize PETSc
+   // 2b. We initialize PETSc
   if (use_petsc) { MFEMInitializePetsc(NULL,NULL,petscrc_file,NULL); }

   // 3. Read the (serial) mesh from the given mesh file on all processors.  We
@@ -132,11 +117,9 @@ int main(int argc, char *argv[])
   //    'ref_levels' to be the largest number that gives a final mesh with no
   //    more than 10,000 elements.
   {
-      if (ser_ref_levels < 0)
-      {
-         ser_ref_levels = (int)floor(log(10000./mesh->GetNE())/log(2.)/dim);
-      }
-      for (int l = 0; l < ser_ref_levels; l++)
+      int ref_levels =
+         (int)floor(log(10000./mesh->GetNE())/log(2.)/dim);
+      for (int l = 0; l < ref_levels; l++)
      {
         mesh->UniformRefinement();
      }
@@ -148,6 +131,7 @@ int main(int argc, char *argv[])
   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
   delete mesh;
   {
+      int par_ref_levels = 2;
      for (int l = 0; l < par_ref_levels; l++)
      {
         pmesh->UniformRefinement();
@@ -203,26 +187,21 @@ int main(int argc, char *argv[])

   // 9. Define the parallel grid function and parallel linear forms, solution
   //    vector and rhs.
-   MemoryType mt = device.GetMemoryType();
-   BlockVector x(block_offsets, mt), rhs(block_offsets, mt);
-   BlockVector trueX(block_trueOffsets, mt), trueRhs(block_trueOffsets, mt);
+   BlockVector x(block_offsets), rhs(block_offsets);
+   BlockVector trueX(block_trueOffsets), trueRhs(block_trueOffsets);

   ParLinearForm *fform(new ParLinearForm);
   fform->Update(R_space, rhs.GetBlock(0), 0);
   fform->AddDomainIntegrator(new VectorFEDomainLFIntegrator(fcoeff));
   fform->AddBoundaryIntegrator(new VectorFEBoundaryFluxLFIntegrator(fnatcoeff));
   fform->Assemble();
-   fform->SyncAliasMemory(rhs);
   fform->ParallelAssemble(trueRhs.GetBlock(0));
-   trueRhs.GetBlock(0).SyncAliasMemory(trueRhs);

   ParLinearForm *gform(new ParLinearForm);
   gform->Update(W_space, rhs.GetBlock(1), 0);
   gform->AddDomainIntegrator(new DomainLFIntegrator(gcoeff));
   gform->Assemble();
-   gform->SyncAliasMemory(rhs);
   gform->ParallelAssemble(trueRhs.GetBlock(1));
-   trueRhs.GetBlock(1).SyncAliasMemory(trueRhs);

   // 10. Assemble the finite element matrices for the Darcy operator
   //
@@ -53,7 +53,6 @@ int main(int argc, char *argv[])
   bool use_petsc = true;
   const char *petscrc_file = "";
   bool use_nonoverlapping = false;
-   const char *device_config = "cpu";

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -74,8 +73,6 @@ int main(int argc, char *argv[])
                  "-no-nonoverlapping", "--no-nonoverlapping",
                  "Use or not the block diagonal PETSc's matrix format "
                  "for non-overlapping domain decomposition.");
-   args.AddOption(&device_config, "-d", "--device",
-                  "Device configuration string, see Device::Configure().");
   args.Parse();
   if (!args.Good())
   {
@@ -89,13 +86,7 @@ int main(int argc, char *argv[])
   {
      args.PrintOptions(cout);
   }
-
-   // 2b. 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(); }
-
-   // 2c. We initialize PETSc
+   // 2b. We initialize PETSc
   if (use_petsc) { MFEMInitializePetsc(NULL,NULL,petscrc_file,NULL); }

   // 3. Read the (serial) mesh from the given mesh file on all processors.  We
@@ -12,10 +12,11 @@
 # Use the MFEM build directory
 MFEM_DIR ?= ../..
 MFEM_BUILD_DIR ?= ../..
-MFEM_INSTALL_DIR ?= ../../mfem
 SRC = $(if $(MFEM_DIR:../..=),$(MFEM_DIR)/examples/petsc/,)
-CONFIG_MK = $(or $(wildcard $(MFEM_BUILD_DIR)/config/config.mk),\
-   $(wildcard $(MFEM_INSTALL_DIR)/share/mfem/config.mk))
+CONFIG_MK = $(MFEM_BUILD_DIR)/config/config.mk
+# Use the MFEM install directory
+# MFEM_INSTALL_DIR = ../../mfem
+# CONFIG_MK = $(MFEM_INSTALL_DIR)/share/mfem/config.mk

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

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

   // 2. Parse command-line options.
@@ -12,10 +12,11 @@
 # Use the MFEM build directory
 MFEM_DIR ?= ../..
 MFEM_BUILD_DIR ?= ../..
-MFEM_INSTALL_DIR ?= ../../mfem
 SRC = $(if $(MFEM_DIR:../..=),$(MFEM_DIR)/examples/pumi/,)
-CONFIG_MK = $(or $(wildcard $(MFEM_BUILD_DIR)/config/config.mk),\
-   $(wildcard $(MFEM_INSTALL_DIR)/share/mfem/config.mk))
+CONFIG_MK = $(MFEM_BUILD_DIR)/config/config.mk
+# Use the MFEM install directory
+# MFEM_INSTALL_DIR = ../../mfem
+# CONFIG_MK = $(MFEM_INSTALL_DIR)/share/mfem/config.mk

 MFEM_LIB_FILE = mfem_is_not_built
 -include $(CONFIG_MK)
@@ -31,21 +31,11 @@ include_directories(BEFORE ${PROJECT_BINARY_DIR})
 add_custom_target(test_sundials
  ${CMAKE_CTEST_COMMAND} -R sundials USES_TERMINAL)

-# Add one executable per cpp file, adding "sundials_" as prefix so the CMake
-# target is unique from those in the non-SUNDIALS examples. Also sets
-# "test_sundials" as a target that depends on the given SUNDIALS examples.
+# Add one executable per cpp file, adding "sundials_" as prefix. Sets
+# "test_sundials" as a target that depends on the given examples.
 set(PFX sundials_)
 add_mfem_examples(SUNDIALS_EXAMPLES_SRCS ${PFX} "" test_sundials)

-# Remove "sundials_" prefix from exectuable name for consistency with GNU build
-# system.
-foreach(SRC_FILE ${SUNDIALS_EXAMPLES_SRCS})
-  get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
-  string(REPLACE ".cpp" "" TARGET_NAME "${PFX}${SRC_FILENAME}")
-  string(REPLACE ${PFX} "" EXE_NAME ${TARGET_NAME})
-  set_target_properties(${TARGET_NAME} PROPERTIES OUTPUT_NAME ${EXE_NAME})
-endforeach()
-
 # Testing.
 # The SUNDIALS tests can be run separately using the target "test_sundials"
 # which builds the examples and runs:
@@ -61,10 +51,7 @@ if (MFEM_ENABLE_TESTING)
  set(EX10_COMMON_OPTS -m ../../data/beam-quad.mesh -o 2 -s 5 -dt 0.15 -tf 6 -vs 10)
  set(EX10_TEST_OPTS ${EX10_COMMON_OPTS}  -r  2)
  set(EX10P_TEST_OPTS ${EX10_COMMON_OPTS} -rp 1)
-  # Example 16: test ARKODE with implicit time stepping using mass form
-  set(EX16_COMMON_OPTS -s 15)
-  set(EX16_TEST_OPTS ${EX16_COMMON_OPTS})
-  set(EX16P_TEST_OPTS ${EX16_COMMON_OPTS})
+  # Example 16: use the default options

  # Add the tests: one test per source file.
  foreach(SRC_FILE ${SUNDIALS_EXAMPLES_SRCS})
@@ -1,17 +1,15 @@
 //                                MFEM Example 10
 //                             SUNDIALS Modification
 //
-// Compile with:
-//    make ex10             (GNU make)
-//    make sundials_ex10    (CMake)
+// Compile with: make ex10
 //
 // Sample runs:
 //    ex10 -m ../../data/beam-quad.mesh -r 2 -o 2 -s 12 -dt 0.15 -vs 10
 //    ex10 -m ../../data/beam-tri.mesh  -r 2 -o 2 -s 16 -dt 0.3  -vs 5
 //    ex10 -m ../../data/beam-hex.mesh  -r 1 -o 2 -s 12 -dt 0.2  -vs 5
-//    ex10 -m ../../data/beam-tri.mesh  -r 2 -o 2 -s  2 -dt 3 -nls 1
-//    ex10 -m ../../data/beam-quad.mesh -r 2 -o 2 -s  2 -dt 3 -nls 2
-//    ex10 -m ../../data/beam-hex.mesh  -r 1 -o 2 -s  2 -dt 3 -nls 4
+//    ex10 -m ../../data/beam-tri.mesh  -r 2 -o 2 -s  2 -dt 3 -nls kinsol
+//    ex10 -m ../../data/beam-quad.mesh -r 2 -o 2 -s  2 -dt 3 -nls kinsol
+//    ex10 -m ../../data/beam-hex.mesh  -r 1 -o 2 -s  2 -dt 3 -nls kinsol
 //    ex10 -m ../../data/beam-quad.mesh -r 2 -o 2 -s 14 -dt 0.15 -vs 10
 //    ex10 -m ../../data/beam-tri.mesh  -r 2 -o 2 -s 17 -dt 0.01 -vs 30
 //    ex10 -m ../../data/beam-hex.mesh  -r 1 -o 2 -s 14 -dt 0.15 -vs 10
@@ -99,11 +97,16 @@ protected:
   double saved_gamma; // saved gamma value from implicit setup

 public:
+   /// Solver type to use in the ImplicitSolve() method, used by SDIRK methods.
+   enum NonlinearSolverType
+   {
+      NEWTON = 0, ///< Use MFEM's plain NewtonSolver
+      KINSOL = 1  ///< Use SUNDIALS' KINSOL (through MFEM's class KINSolver)
+   };

   HyperelasticOperator(FiniteElementSpace &f, Array<int> &ess_bdr,
                        double visc, double mu, double K,
-                        int kinsol_nls_type = -1, double kinsol_damping = 0.0,
-                        int kinsol_aa_n = 0);
+                        NonlinearSolverType nls_type);

   /// Compute the right-hand side of the ODE system.
   virtual void Mult(const Vector &vx, Vector &dvx_dt) const;
@@ -221,10 +224,8 @@ int main(int argc, char *argv[])
   double mu = 0.25;
   double K = 5.0;
   bool visualization = true;
-   int nonlinear_solver_type = 0;
+   const char *nls = "newton";
   int vis_steps = 1;
-   double kinsol_damping = 0.0;
-   int kinsol_aa_n = -1;

   // Relative and absolute tolerances for CVODE and ARKODE.
   const double reltol = 1e-1, abstol = 1e-1;
@@ -261,18 +262,9 @@ int main(int argc, char *argv[])
                  "15 - ARKODE implicit, approximate Jacobian,\n\t"
                  "16 - ARKODE implicit, specified Jacobian,\n\t"
                  "17 - ARKODE explicit, 4th order.");
-   args.AddOption(&nonlinear_solver_type, "-nls", "--nonlinear-solver",
-                  "Nonlinear system solver:\n\t"
-                  "0  - MFEM Newton method,\n\t"
-                  "1  - KINSOL Newton method,\n\t"
-                  "2  - KINSOL Newton method with globalization,\n\t"
-                  "3  - KINSOL fixed-point method (with or without AA),\n\t"
-                  "4  - KINSOL Picard method (with or without AA).");
-   args.AddOption(&kinsol_damping, "-damp", "--kinsol-damping",
-                  "Picard or Fixed-Point damping parameter (only valid with KINSOL): "
-                  "0 < d <= 1.0");
-   args.AddOption(&kinsol_aa_n, "-aan", "--anderson-subspace",
-                  "Anderson Acceleration subspace size (only valid with KINSOL)");
+   args.AddOption(&nls, "-nls", "--nonlinear-solver",
+                  "Nonlinear systems solver: "
+                  "\"newton\" (plain Newton) or \"kinsol\" (KINSOL).");
   args.AddOption(&t_final, "-tf", "--t-final",
                  "Final time; start time is 0.");
   args.AddOption(&dt, "-dt", "--time-step",
@@ -303,32 +295,22 @@ int main(int argc, char *argv[])
      return 1;
   }

-   // check for valid nonlinear solver options
-   if (nonlinear_solver_type < 0 || nonlinear_solver_type > 4)
-   {
-      cout << "Unknown nonlinear solver type: " << nonlinear_solver_type << "\n";
-      return 1;
-   }
-   if (kinsol_damping > 0.0 &&
-       !(nonlinear_solver_type == 3 || nonlinear_solver_type == 4))
-   {
-      cout << "Only KINSOL fixed-point and Picard methods can use damping\n";
-      return 1;
-   }
-   if (kinsol_aa_n > 0 &&
-       !(nonlinear_solver_type == 3 || nonlinear_solver_type == 4))
-   {
-      cout << "Only KINSOL fixed-point and Picard methods can use AA\n";
-      return 1;
-   }
-
-
   // 2. Read the mesh from the given mesh file. We can handle triangular,
   //    quadrilateral, tetrahedral and hexahedral meshes with the same code.
   Mesh *mesh = new Mesh(mesh_file, 1, 1);
   int dim = mesh->Dimension();

-   // 3. Refine the mesh to increase the resolution. In this example we do
+   // 3. Setup the nonlinear solver
+   map<string,HyperelasticOperator::NonlinearSolverType> nls_map;
+   nls_map["newton"] = HyperelasticOperator::NEWTON;
+   nls_map["kinsol"] = HyperelasticOperator::KINSOL;
+   if (nls_map.find(nls) == nls_map.end())
+   {
+      cout << "Unknown type of nonlinear solver: " << nls << endl;
+      return 4;
+   }
+
+   // 4. Refine the mesh to increase the resolution. In this example we do
   //    'ref_levels' of uniform refinement, where 'ref_levels' is a
   //    command-line parameter.
   for (int lev = 0; lev < ref_levels; lev++)
@@ -336,7 +318,7 @@ int main(int argc, char *argv[])
      mesh->UniformRefinement();
   }

-   // 4. Define the vector finite element spaces representing the mesh
+   // 5. Define the vector finite element spaces representing the mesh
   //    deformation x, the velocity v, and the initial configuration, x_ref.
   //    Define also the elastic energy density, w, which is in a discontinuous
   //    higher-order space. Since x and v are integrated in time as a system,
@@ -364,7 +346,7 @@ int main(int argc, char *argv[])
   FiniteElementSpace w_fespace(mesh, &w_fec);
   GridFunction w(&w_fespace);

-   // 5. Set the initial conditions for v and x, and the boundary conditions on
+   // 6. Set the initial conditions for v and x, and the boundary conditions on
   //    a beam-like mesh (see description above).
   VectorFunctionCoefficient velo(dim, InitialVelocity);
   v.ProjectCoefficient(velo);
@@ -377,34 +359,9 @@ int main(int argc, char *argv[])
   ess_bdr = 0;
   ess_bdr[0] = 1; // boundary attribute 1 (index 0) is fixed

-   // 6. Initialize the hyperelastic operator, the GLVis visualization and print
+   // 7. Initialize the hyperelastic operator, the GLVis visualization and print
   //    the initial energies.
-   std::unique_ptr<HyperelasticOperator> oper;
-   if (nonlinear_solver_type == 0)
-      oper = std::make_unique<HyperelasticOperator>(fespace, ess_bdr, visc, mu,
-                                                    K);
-   else
-   {
-      switch (nonlinear_solver_type)
-      {
-         case 1:
-            oper = std::make_unique<HyperelasticOperator>(fespace, ess_bdr,
-                                                          visc, mu, K, KIN_NONE);
-            break;
-         case 2:
-            oper = std::make_unique<HyperelasticOperator>(fespace, ess_bdr,
-                                                          visc, mu, K, KIN_LINESEARCH);
-            break;
-         case 3:
-            oper = std::make_unique<HyperelasticOperator>(fespace, ess_bdr,
-                                                          visc, mu, K, KIN_FP, kinsol_damping, kinsol_aa_n);
-            break;
-         case 4:
-            oper = std::make_unique<HyperelasticOperator>(fespace, ess_bdr,
-                                                          visc, mu, K, KIN_PICARD, kinsol_damping, kinsol_aa_n);
-            break;
-      }
-   }
+   HyperelasticOperator oper(fespace, ess_bdr, visc, mu, K, nls_map[nls]);

   socketstream vis_v, vis_w;
   if (visualization)
@@ -418,23 +375,23 @@ int main(int argc, char *argv[])
      vis_w.open(vishost, visport);
      if (vis_w)
      {
-         oper->GetElasticEnergyDensity(x, w);
+         oper.GetElasticEnergyDensity(x, w);
         vis_w.precision(8);
         visualize(vis_w, mesh, &x, &w, "Elastic energy density", true);
      }
   }

-   double ee0 = oper->ElasticEnergy(x.GetTrueVector());
-   double ke0 = oper->KineticEnergy(v.GetTrueVector());
+   double ee0 = oper.ElasticEnergy(x.GetTrueVector());
+   double ke0 = oper.KineticEnergy(v.GetTrueVector());
   cout << "initial elastic energy (EE) = " << ee0 << endl;
   cout << "initial kinetic energy (KE) = " << ke0 << endl;
   cout << "initial   total energy (TE) = " << (ee0 + ke0) << endl;

-   // 7. Define the ODE solver used for time integration. Several implicit
+   // 8. Define the ODE solver used for time integration. Several implicit
   //    singly diagonal implicit Runge-Kutta (SDIRK) methods, as well as
   //    explicit Runge-Kutta methods are available.
   double t = 0.0;
-   oper->SetTime(t);
+   oper.SetTime(t);

   ODESolver *ode_solver = NULL;
   CVODESolver *cvode = NULL;
@@ -458,7 +415,7 @@ int main(int argc, char *argv[])
      case 11:
      case 12:
         cvode = new CVODESolver(CV_BDF);
-         cvode->Init(*oper);
+         cvode->Init(oper);
         cvode->SetSStolerances(reltol, abstol);
         CVodeSetEpsLin(cvode->GetMem(), cvode_eps_lin);
         cvode->SetMaxStep(dt);
@@ -471,7 +428,7 @@ int main(int argc, char *argv[])
      case 13:
      case 14:
         cvode = new CVODESolver(CV_ADAMS);
-         cvode->Init(*oper);
+         cvode->Init(oper);
         cvode->SetSStolerances(reltol, abstol);
         CVodeSetEpsLin(cvode->GetMem(), cvode_eps_lin);
         cvode->SetMaxStep(dt);
@@ -484,13 +441,9 @@ int main(int argc, char *argv[])
      case 15:
      case 16:
         arkode = new ARKStepSolver(ARKStepSolver::IMPLICIT);
-         arkode->Init(*oper);
+         arkode->Init(oper);
         arkode->SetSStolerances(reltol, abstol);
-#if MFEM_SUNDIALS_VERSION < 70100
         ARKStepSetNonlinConvCoef(arkode->GetMem(), arkode_eps_nonlin);
-#else
-         ARKodeSetNonlinConvCoef(arkode->GetMem(), arkode_eps_nonlin);
-#endif
         arkode->SetMaxStep(dt);
         if (ode_solver_type == 15)
         {
@@ -500,16 +453,16 @@ int main(int argc, char *argv[])
      // ARKStep Explicit methods
      case 17:
         arkode = new ARKStepSolver(ARKStepSolver::EXPLICIT);
-         arkode->Init(*oper);
+         arkode->Init(oper);
         arkode->SetSStolerances(reltol, abstol);
         arkode->SetMaxStep(dt);
         ode_solver = arkode; break;
   }

   // Initialize MFEM integrators, SUNDIALS integrators are initialized above
-   if (ode_solver_type < 11) { ode_solver->Init(*oper); }
+   if (ode_solver_type < 11) { ode_solver->Init(oper); }

-   // 8. Perform time-integration (looping over the time iterations, ti, with a
+   // 9. Perform time-integration (looping over the time iterations, ti, with a
   //    time-step dt).
   bool last_step = false;
   for (int ti = 1; !last_step; ti++)
@@ -522,8 +475,8 @@ int main(int argc, char *argv[])

      if (last_step || (ti % vis_steps) == 0)
      {
-         double ee = oper->ElasticEnergy(x.GetTrueVector());
-         double ke = oper->KineticEnergy(v.GetTrueVector());
+         double ee = oper.ElasticEnergy(x.GetTrueVector());
+         double ke = oper.KineticEnergy(v.GetTrueVector());

         cout << "step " << ti << ", t = " << t << ", EE = " << ee << ", KE = "
              << ke << ", ΔTE = " << (ee+ke)-(ee0+ke0) << endl;
@@ -537,14 +490,14 @@ int main(int argc, char *argv[])
            visualize(vis_v, mesh, &x, &v);
            if (vis_w)
            {
-               oper->GetElasticEnergyDensity(x, w);
+               oper.GetElasticEnergyDensity(x, w);
               visualize(vis_w, mesh, &x, &w);
            }
         }
      }
   }

-   // 9. Save the displaced mesh, the velocity and elastic energy.
+   // 10. Save the displaced mesh, the velocity and elastic energy.
   {
      v.SetFromTrueVector(); x.SetFromTrueVector();
      GridFunction *nodes = &x;
@@ -559,11 +512,11 @@ int main(int argc, char *argv[])
      v.Save(velo_ofs);
      ofstream ee_ofs("elastic_energy.sol");
      ee_ofs.precision(8);
-      oper->GetElasticEnergyDensity(x, w);
+      oper.GetElasticEnergyDensity(x, w);
      w.Save(ee_ofs);
   }

-   // 10. Free the used memory.
+   // 11. Free the used memory.
   delete ode_solver;
   delete mesh;

@@ -647,9 +600,7 @@ ReducedSystemOperator::~ReducedSystemOperator()
 HyperelasticOperator::HyperelasticOperator(FiniteElementSpace &f,
                                           Array<int> &ess_bdr, double visc,
                                           double mu, double K,
-                                           int kinsol_nls_type,
-                                           double kinsol_damping,
-                                           int kinsol_aa_n)
+                                           NonlinearSolverType nls_type)
   : TimeDependentOperator(2*f.GetTrueVSize(), 0.0), fespace(f),
     M(&fespace), S(&fespace), H(&fespace),
     viscosity(visc), z(height/2),
@@ -700,28 +651,15 @@ HyperelasticOperator::HyperelasticOperator(FiniteElementSpace &f,
   J_prec = NULL;
 #endif

-   if (kinsol_nls_type > 0)
+   if (nls_type == KINSOL)
   {
-      KINSolver *kinsolver = new KINSolver(kinsol_nls_type, true);
-      if (kinsol_nls_type != KIN_PICARD)
-      {
-         kinsolver->SetJFNK(true);
-         kinsolver->SetLSMaxIter(100);
-      }
-      if (kinsol_aa_n > 0)
-      {
-         kinsolver->EnableAndersonAcc(kinsol_aa_n);
-      }
+      KINSolver *kinsolver = new KINSolver(KIN_NONE, true);
      newton_solver = kinsolver;
      newton_solver->SetOperator(*reduced_oper);
      newton_solver->SetMaxIter(200);
      newton_solver->SetRelTol(rel_tol);
      newton_solver->SetPrintLevel(0);
      kinsolver->SetMaxSetupCalls(4);
-      if (kinsol_damping > 0.0)
-      {
-         kinsolver->SetDamping(kinsol_damping);
-      }
   }
   else
   {
@@ -1,17 +1,15 @@
 //                       MFEM Example 10 - Parallel Version
 //                             SUNDIALS Modification
 //
-// Compile with:
-//    make ex10p            (GNU make)
-//    make sundials_ex10p   (CMake)
+// Compile with: make ex10p
 //
 // Sample runs:
 //    mpirun -np 4 ex10p -m ../../data/beam-quad.mesh -rp 1 -o 2 -s 12 -dt 0.15 -vs 10
 //    mpirun -np 4 ex10p -m ../../data/beam-tri.mesh  -rp 1 -o 2 -s 16 -dt 0.25 -vs 10
 //    mpirun -np 4 ex10p -m ../../data/beam-hex.mesh  -rp 0 -o 2 -s 12 -dt 0.15 -vs 10
-//    mpirun -np 4 ex10p -m ../../data/beam-tri.mesh  -rp 1 -o 2 -s  2 -dt 3 -nls 1
-//    mpirun -np 4 ex10p -m ../../data/beam-quad.mesh -rp 1 -o 2 -s  2 -dt 3 -nls 2
-//    mpirun -np 4 ex10p -m ../../data/beam-hex.mesh  -rs 1 -o 2 -s  2 -dt 3 -nls 4
+//    mpirun -np 4 ex10p -m ../../data/beam-tri.mesh  -rp 1 -o 2 -s  2 -dt 3 -nls kinsol
+//    mpirun -np 4 ex10p -m ../../data/beam-quad.mesh -rp 1 -o 2 -s  2 -dt 3 -nls kinsol
+//    mpirun -np 4 ex10p -m ../../data/beam-hex.mesh  -rs 1 -o 2 -s  2 -dt 3 -nls kinsol
 //    mpirun -np 4 ex10p -m ../../data/beam-quad.mesh -rp 1 -o 2 -s 14 -dt 0.15 -vs 10
 //    mpirun -np 4 ex10p -m ../../data/beam-tri.mesh  -rp 1 -o 2 -s 17 -dt 5e-3 -vs 60
 //    mpirun -np 4 ex10p -m ../../data/beam-hex.mesh  -rp 0 -o 2 -s 14 -dt 0.15 -vs 10
@@ -101,11 +99,16 @@ protected:
   double saved_gamma; // saved gamma value from implicit setup

 public:
+   /// Solver type to use in the ImplicitSolve() method, used by SDIRK methods.
+   enum NonlinearSolverType
+   {
+      NEWTON = 0, ///< Use MFEM's plain NewtonSolver
+      KINSOL = 1  ///< Use SUNDIALS' KINSOL (through MFEM's class KINSolver)
+   };

   HyperelasticOperator(ParFiniteElementSpace &f, Array<int> &ess_bdr,
                        double visc, double mu, double K,
-                        int kinsol_nls_type = -1, double kinsol_damping = 0.0,
-                        int kinsol_aa_n = 0);
+                        NonlinearSolverType nls_type);

   /// Compute the right-hand side of the ODE system.
   virtual void Mult(const Vector &vx, Vector &dvx_dt) const;
@@ -230,10 +233,8 @@ int main(int argc, char *argv[])
   double mu = 0.25;
   double K = 5.0;
   bool visualization = true;
-   int nonlinear_solver_type = 0;
+   const char *nls = "newton";
   int vis_steps = 1;
-   double kinsol_damping = 0.0;
-   int kinsol_aa_n = -1;

   // Relative and absolute tolerances for CVODE and ARKODE.
   const double reltol = 1e-1, abstol = 1e-1;
@@ -272,18 +273,9 @@ int main(int argc, char *argv[])
                  "15 - ARKODE implicit, approximate Jacobian,\n\t"
                  "16 - ARKODE implicit, specified Jacobian,\n\t"
                  "17 - ARKODE explicit, 4th order.");
-   args.AddOption(&nonlinear_solver_type, "-nls", "--nonlinear-solver",
-                  "Nonlinear system solver:\n\t"
-                  "0  - MFEM Newton method,\n\t"
-                  "1  - KINSOL Newton method,\n\t"
-                  "2  - KINSOL Newton method with globalization,\n\t"
-                  "3  - KINSOL fixed-point method (with or without AA),\n\t"
-                  "4  - KINSOL Picard method (with or without AA).");
-   args.AddOption(&kinsol_damping, "-damp", "--kinsol-damping",
-                  "Picard or Fixed-Point damping parameter (only valid with KINSOL): "
-                  "0 < d <= 1.0");
-   args.AddOption(&kinsol_aa_n, "-aan", "--anderson-subspace",
-                  "Anderson Acceleration subspace size (only valid with KINSOL)");
+   args.AddOption(&nls, "-nls", "--nonlinear-solver",
+                  "Nonlinear systems solver: "
+                  "\"newton\" (plain Newton) or \"kinsol\" (KINSOL).");
   args.AddOption(&t_final, "-tf", "--t-final",
                  "Final time; start time is 0.");
   args.AddOption(&dt, "-dt", "--time-step",
@@ -323,42 +315,27 @@ int main(int argc, char *argv[])
      return 1;
   }

-   // check for valid nonlinear solver options
-   if (nonlinear_solver_type < 0 || nonlinear_solver_type > 4)
-   {
-      if (myid == 0)
-      {
-         cout << "Unknown nonlinear solver type: " << nonlinear_solver_type
-              << "\n";
-      }
-      return 1;
-   }
-   if (kinsol_damping > 0.0 &&
-       !(nonlinear_solver_type == 3 || nonlinear_solver_type == 4))
-   {
-      if (myid == 0)
-      {
-         cout << "Only KINSOL fixed-point and Picard methods can use damping\n";
-      }
-      return 1;
-   }
-   if (kinsol_aa_n > 0 &&
-       !(nonlinear_solver_type == 3 || nonlinear_solver_type == 4))
-   {
-      if (myid == 0)
-      {
-         cout << "Only KINSOL fixed-point and Picard methods can use AA\n";
-      }
-      return 1;
-   }
-
   // 3. Read the serial mesh from the given mesh file on all processors. We can
   //    handle triangular, quadrilateral, tetrahedral and hexahedral meshes
   //    with the same code.
   Mesh *mesh = new Mesh(mesh_file, 1, 1);
   int dim = mesh->Dimension();

-   // 4. Refine the mesh in serial to increase the resolution. In this example
+   // 4. Nonlinear solver
+   map<string,HyperelasticOperator::NonlinearSolverType> nls_map;
+   nls_map["newton"] = HyperelasticOperator::NEWTON;
+   nls_map["kinsol"] = HyperelasticOperator::KINSOL;
+   if (nls_map.find(nls) == nls_map.end())
+   {
+      if (myid == 0)
+      {
+         cout << "Unknown type of nonlinear solver: " << nls << endl;
+      }
+      delete mesh;
+      return 4;
+   }
+
+   // 5. Refine the mesh in serial to increase the resolution. In this example
   //    we do 'ser_ref_levels' of uniform refinement, where 'ser_ref_levels' is
   //    a command-line parameter.
   for (int lev = 0; lev < ser_ref_levels; lev++)
@@ -366,7 +343,7 @@ int main(int argc, char *argv[])
      mesh->UniformRefinement();
   }

-   // 5. Define a parallel mesh by a partitioning of the serial mesh. Refine
+   // 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.
   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
@@ -376,7 +353,7 @@ int main(int argc, char *argv[])
      pmesh->UniformRefinement();
   }

-   // 6. Define the parallel vector finite element spaces representing the mesh
+   // 7. Define the parallel vector finite element spaces representing the mesh
   //    deformation x_gf, the velocity v_gf, and the initial configuration,
   //    x_ref. Define also the elastic energy density, w_gf, which is in a
   //    discontinuous higher-order space. Since x and v are integrated in time
@@ -408,7 +385,7 @@ int main(int argc, char *argv[])
   ParFiniteElementSpace w_fespace(pmesh, &w_fec);
   ParGridFunction w_gf(&w_fespace);

-   // 7. Set the initial conditions for v_gf, x_gf and vx, and define the
+   // 8. Set the initial conditions for v_gf, x_gf and vx, and define the
   //    boundary conditions on a beam-like mesh (see description above).
   VectorFunctionCoefficient velo(dim, InitialVelocity);
   v_gf.ProjectCoefficient(velo);
@@ -423,38 +400,9 @@ int main(int argc, char *argv[])
   ess_bdr = 0;
   ess_bdr[0] = 1; // boundary attribute 1 (index 0) is fixed

-   // 8. Initialize the hyperelastic operator, the GLVis visualization and print
+   // 9. Initialize the hyperelastic operator, the GLVis visualization and print
   //    the initial energies.
-   std::unique_ptr<HyperelasticOperator> oper;
-   if (nonlinear_solver_type == 0)
-      oper = std::make_unique<HyperelasticOperator>(fespace, ess_bdr, visc, mu,
-                                                    K);
-   else
-   {
-      switch (nonlinear_solver_type)
-      {
-         case 1:
-            oper = std::make_unique<HyperelasticOperator>(fespace, ess_bdr,
-                                                          visc, mu, K, KIN_NONE);
-            break;
-         case 2:
-            oper = std::make_unique<HyperelasticOperator>(fespace, ess_bdr,
-                                                          visc, mu, K, KIN_LINESEARCH);
-            break;
-         case 3:
-            oper = std::make_unique<HyperelasticOperator>(fespace, ess_bdr,
-                                                          visc, mu, K, KIN_FP, kinsol_damping, kinsol_aa_n);
-            break;
-         case 4:
-            oper = std::make_unique<HyperelasticOperator>(fespace, ess_bdr,
-                                                          visc, mu, K, KIN_PICARD, kinsol_damping, kinsol_aa_n);
-            break;
-         default:
-            cout << "Unknown type of nonlinear solver: "
-                 << nonlinear_solver_type << endl;
-            return 4;
-      }
-   }
+   HyperelasticOperator oper(fespace, ess_bdr, visc, mu, K, nls_map[nls]);

   socketstream vis_v, vis_w;
   if (visualization)
@@ -470,14 +418,14 @@ int main(int argc, char *argv[])
      vis_w.open(vishost, visport);
      if (vis_w)
      {
-         oper->GetElasticEnergyDensity(x_gf, w_gf);
+         oper.GetElasticEnergyDensity(x_gf, w_gf);
         vis_w.precision(8);
         visualize(vis_w, pmesh, &x_gf, &w_gf, "Elastic energy density", true);
      }
   }

-   double ee0 = oper->ElasticEnergy(x_gf);
-   double ke0 = oper->KineticEnergy(v_gf);
+   double ee0 = oper.ElasticEnergy(x_gf);
+   double ke0 = oper.KineticEnergy(v_gf);
   if (myid == 0)
   {
      cout << "initial elastic energy (EE) = " << ee0 << endl;
@@ -485,11 +433,11 @@ int main(int argc, char *argv[])
      cout << "initial   total energy (TE) = " << (ee0 + ke0) << endl;
   }

-   // 9. Define the ODE solver used for time integration. Several implicit
-   //    singly diagonal implicit Runge-Kutta (SDIRK) methods, as well as
-   //    explicit Runge-Kutta methods are available.
+   // 10. Define the ODE solver used for time integration. Several implicit
+   //     singly diagonal implicit Runge-Kutta (SDIRK) methods, as well as
+   //     explicit Runge-Kutta methods are available.
   double t = 0.0;
-   oper->SetTime(t);
+   oper.SetTime(t);

   ODESolver *ode_solver = NULL;
   CVODESolver *cvode = NULL;
@@ -513,7 +461,7 @@ int main(int argc, char *argv[])
      case 11:
      case 12:
         cvode = new CVODESolver(MPI_COMM_WORLD, CV_BDF);
-         cvode->Init(*oper);
+         cvode->Init(oper);
         cvode->SetSStolerances(reltol, abstol);
         CVodeSetEpsLin(cvode->GetMem(), cvode_eps_lin);
         cvode->SetMaxStep(dt);
@@ -526,7 +474,7 @@ int main(int argc, char *argv[])
      case 13:
      case 14:
         cvode = new CVODESolver(MPI_COMM_WORLD, CV_ADAMS);
-         cvode->Init(*oper);
+         cvode->Init(oper);
         cvode->SetSStolerances(reltol, abstol);
         CVodeSetEpsLin(cvode->GetMem(), cvode_eps_lin);
         cvode->SetMaxStep(dt);
@@ -539,13 +487,9 @@ int main(int argc, char *argv[])
      case 15:
      case 16:
         arkode = new ARKStepSolver(MPI_COMM_WORLD, ARKStepSolver::IMPLICIT);
-         arkode->Init(*oper);
+         arkode->Init(oper);
         arkode->SetSStolerances(reltol, abstol);
-#if MFEM_SUNDIALS_VERSION < 70100
         ARKStepSetNonlinConvCoef(arkode->GetMem(), arkode_eps_nonlin);
-#else
-         ARKodeSetNonlinConvCoef(arkode->GetMem(), arkode_eps_nonlin);
-#endif
         arkode->SetMaxStep(dt);
         if (ode_solver_type == 15)
         {
@@ -555,16 +499,16 @@ int main(int argc, char *argv[])
      // ARKStep Explicit methods
      case 17:
         arkode = new ARKStepSolver(MPI_COMM_WORLD, ARKStepSolver::EXPLICIT);
-         arkode->Init(*oper);
+         arkode->Init(oper);
         arkode->SetSStolerances(reltol, abstol);
         arkode->SetMaxStep(dt);
         ode_solver = arkode; break;
   }

   // Initialize MFEM integrators, SUNDIALS integrators are initialized above
-   if (ode_solver_type < 11) { ode_solver->Init(*oper); }
+   if (ode_solver_type < 11) { ode_solver->Init(oper); }

-   // 10. Perform time-integration
+   // 11. Perform time-integration
   //     (looping over the time iterations, ti, with a time-step dt).
   bool last_step = false;
   for (int ti = 1; !last_step; ti++)
@@ -579,8 +523,8 @@ int main(int argc, char *argv[])
      {
         v_gf.SetFromTrueVector(); x_gf.SetFromTrueVector();

-         double ee = oper->ElasticEnergy(x_gf);
-         double ke = oper->KineticEnergy(v_gf);
+         double ee = oper.ElasticEnergy(x_gf);
+         double ke = oper.KineticEnergy(v_gf);

         if (myid == 0)
         {
@@ -596,14 +540,14 @@ int main(int argc, char *argv[])
            visualize(vis_v, pmesh, &x_gf, &v_gf);
            if (vis_w)
            {
-               oper->GetElasticEnergyDensity(x_gf, w_gf);
+               oper.GetElasticEnergyDensity(x_gf, w_gf);
               visualize(vis_w, pmesh, &x_gf, &w_gf);
            }
         }
      }
   }

-   // 11. Save the displaced mesh, the velocity and elastic energy.
+   // 12. Save the displaced mesh, the velocity and elastic energy.
   {
      v_gf.SetFromTrueVector(); x_gf.SetFromTrueVector();
      GridFunction *nodes = &x_gf;
@@ -624,11 +568,11 @@ int main(int argc, char *argv[])
      v_gf.Save(velo_ofs);
      ofstream ee_ofs(ee_name.str().c_str());
      ee_ofs.precision(8);
-      oper->GetElasticEnergyDensity(x_gf, w_gf);
+      oper.GetElasticEnergyDensity(x_gf, w_gf);
      w_gf.Save(ee_ofs);
   }

-   // 12. Free the used memory.
+   // 13. Free the used memory.
   delete ode_solver;
   delete pmesh;

@@ -718,10 +662,7 @@ ReducedSystemOperator::~ReducedSystemOperator()
 HyperelasticOperator::HyperelasticOperator(ParFiniteElementSpace &f,
                                           Array<int> &ess_bdr, double visc,
                                           double mu, double K,
-                                           int kinsol_nls_type,
-                                           double kinsol_damping,
-                                           int kinsol_aa_n)
-
+                                           NonlinearSolverType nls_type)
   : TimeDependentOperator(2*f.TrueVSize(), 0.0), fespace(f),
     M(&fespace), S(&fespace), H(&fespace),
     viscosity(visc), M_solver(f.GetComm()), z(height/2),
@@ -773,28 +714,17 @@ HyperelasticOperator::HyperelasticOperator(ParFiniteElementSpace &f,
   J_minres->SetPreconditioner(*J_prec);
   J_solver = J_minres;

-   if (kinsol_nls_type > 0)
+   if (nls_type == KINSOL)
   {
-      KINSolver *kinsolver = new KINSolver(f.GetComm(), kinsol_nls_type, true);
-      if (kinsol_nls_type != KIN_PICARD)
-      {
-         kinsolver->SetJFNK(true);
-         kinsolver->SetLSMaxIter(100);
-      }
-      if (kinsol_aa_n > 0)
-      {
-         kinsolver->EnableAndersonAcc(kinsol_aa_n);
-      }
+      KINSolver *kinsolver = new KINSolver(f.GetComm(), KIN_LINESEARCH, true);
+      kinsolver->SetJFNK(true);
+      kinsolver->SetLSMaxIter(100);
      newton_solver = kinsolver;
      newton_solver->SetOperator(*reduced_oper);
      newton_solver->SetMaxIter(200);
      newton_solver->SetRelTol(rel_tol);
-      newton_solver->SetPrintLevel(0);
+      newton_solver->SetPrintLevel(1);
      kinsolver->SetMaxSetupCalls(4);
-      if (kinsol_damping > 0.0)
-      {
-         kinsolver->SetDamping(kinsol_damping);
-      }
   }
   else
   {
@@ -926,7 +856,10 @@ double HyperelasticOperator::ElasticEnergy(const ParGridFunction &x) const

 double HyperelasticOperator::KineticEnergy(const ParGridFunction &v) const
 {
-   double energy = 0.5*M.ParInnerProduct(v, v);
+   double loc_energy = 0.5*M.InnerProduct(v, v);
+   double energy;
+   MPI_Allreduce(&loc_energy, &energy, 1, MPI_DOUBLE, MPI_SUM,
+                 fespace.GetComm());
   return energy;
 }

@@ -1,21 +1,15 @@
 //                                MFEM Example 16
 //                             SUNDIALS Modification
 //
-// Compile with:
-//    make ex16             (GNU make)
-//    make sundials_ex16    (CMake)
+// Compile with: make ex16
 //
 // Sample runs:  ex16
 //               ex16 -m ../../data/inline-tri.mesh
 //               ex16 -m ../../data/disc-nurbs.mesh -tf 2
 //               ex16 -s 12 -a 0.0 -k 1.0
-//               ex16 -s 15 -a 0.0 -k 1.0
 //               ex16 -s 8 -a 1.0 -k 0.0 -dt 1e-4 -tf 5e-2 -vs 25
-//               ex16 -s 11 -a 1.0 -k 0.0 -dt 1e-4 -tf 5e-2 -vs 25
 //               ex16 -s 9 -a 0.5 -k 0.5 -o 4 -dt 1e-4 -tf 2e-2 -vs 25
-//               ex16 -s 12 -a 0.5 -k 0.5 -o 4 -dt 1e-4 -tf 2e-2 -vs 25
 //               ex16 -s 10 -dt 1.0e-4 -tf 4.0e-2 -vs 40
-//               ex16 -s 13 -dt 1.0e-4 -tf 4.0e-2 -vs 40
 //               ex16 -m ../../data/fichera-q2.mesh
 //               ex16 -m ../../data/escher.mesh
 //               ex16 -m ../../data/beam-tet.mesh -tf 10 -dt 0.1
@@ -43,102 +37,75 @@
 using namespace std;
 using namespace mfem;

-/** After spatial discretization, the conduction model is expressed as
+/** After spatial discretization, the conduction model can be written as:
 *
- *   M du/dt = - K(u) u
+ *     du/dt = M^{-1}(-Ku)
 *
 *  where u is the vector representing the temperature, M is the mass matrix,
- *  and K(u) is the diffusion operator with diffusivity depending on u:
+ *  and K is the diffusion operator with diffusivity depending on u:
 *  (\kappa + \alpha u).
 *
- *  Class ConductionOperatorOperator represents the above ODE operator in the
- *  general form F(u, k, t) = G(u, t) where
- *
- *    1. F(u, du/dt, t) = du/dt            (ODE is expressed in EXPLICIT form)
- *       G(u, t) = - inv(M) K(u) u
- *    2. F(u, du/dt, t) = M du/dt          (ODE is expressed in IMPLICIT form)
- *       G(u, t) = - K(u) u
+ *  Class ConductionOperator represents the right-hand side of the above ODE.
 */
 class ConductionOperator : public TimeDependentOperator
 {
+protected:
   FiniteElementSpace &fespace;
   Array<int> ess_tdof_list; // this list remains empty for pure Neumann b.c.

-   BilinearForm M;
-   SparseMatrix Mmat;
+   BilinearForm *M;
+   BilinearForm *K;

-   const real_t alpha, kappa;
-   std::unique_ptr<BilinearForm> K;
-   SparseMatrix Kmat;
-
-   std::unique_ptr<SparseMatrix> T; // T = M + gam K(u)
+   SparseMatrix Mmat, Kmat;
+   SparseMatrix *T; // T = M + dt K

   CGSolver M_solver; // Krylov solver for inverting the mass matrix M
   DSmoother M_prec;  // Preconditioner for the mass matrix M

-   CGSolver T_solver; // Implicit solver for T = M + gam K(u)
+   CGSolver T_solver; // Implicit solver for T = M + dt K
   DSmoother T_prec;  // Preconditioner for the implicit solver

+   double alpha, kappa;
+
   mutable Vector z; // auxiliary vector

 public:
+   ConductionOperator(FiniteElementSpace &f, double alpha, double kappa,
+                      const Vector &u);

-   ConductionOperator(FiniteElementSpace &f, const real_t alpha,
-                      const real_t kappa, const Vector &u,
-                      const Type &ode_expression_type);
+   virtual void Mult(const Vector &u, Vector &du_dt) const;

-   // Compute K(u_n) for use as an approximation in - K(u) u
-   void SetConductionTensor(const Vector &u);
+   /** Solve the Backward-Euler equation: k = f(u + dt*k, t), for the unknown k.
+       This is the only requirement for high-order SDIRK implicit integration.*/
+   virtual void ImplicitSolve(const double dt, const Vector &u, Vector &k);

-   /** Compute G(u, t) as defined in the IMPLICIT expression form of the ODE
-    operator, i.e., @a v = - K(u_n) @a u. Note that K(u_n) is an
-      approximation to K(u). */
-   void ExplicitMult(const Vector &u, Vector &v) const override;
+   /// Custom Jacobian system solver for the SUNDIALS time integrators.
+   /** For the ODE system represented by ConductionOperator

-   /** Solve for k in F(u, k, t) = G(u, t) for either EXPLICIT or IMPLICIT
-       expression forms of the ODE operator, i.e., @a k = - inv(M) K(u_n) @a u.
-       Note that K(u_n) is an approximation to K(u). */
-   void Mult(const Vector &u, Vector &k) const override;
+           M du/dt = -K(u),

-   /** Solve for k in F(u + gam*k, k, t) = G(u + gam*k, t) for either EXPLICIT
-       or IMPLICIT expression forms of the ODE operator, i.e.,
-       [ M + @a gam K(u_n) ] @a k = - K(u_n) @a u . Note that K(u_n) is an
-       approximation to K(u). */
-   void ImplicitSolve(const real_t gam, const Vector &u, Vector &k) override;
+       this class facilitates the solution of linear systems of the form

-   /** Setup to solve for dk in [dF/dk + gam*dF/du - gam*dG/du] dk = G - F for
-       either EXPLICIT or IMPLICIT expression forms of the ODE operator, i.e.,
-       [M - @a gam Jf(u)] dk = G - F, where Jf(u) is an approximation of the
-       Jacobian of -K(u) u. The approximation chosen here is Jf(u) = -K(u_n). */
-   int SUNImplicitSetup(const Vector &u, const Vector &fu, int jok, int *jcur,
-                        real_t gam) override;
+           (M + γK) y = M b,

-   /** Solve for @a dk in the system in SUNImplicitSetup to the given tolerance,
-       with the residual @a r providing either
-        1. @a r = G - F = inv(M) f(u) - k       (EXPLICIT expression form)
-        1. @a r = G - F = f(u) - M k            (IMPLICIT expression form)
-       */
-   int SUNImplicitSolve(const Vector &r, Vector &dk, real_t tol) override;
+       for given b, u (not used), and γ = GetTimeStep(). */

-   int SUNMassSetup() override;
+   /** Setup the system (M + dt K) x = M b. This method is used by the implicit
+       SUNDIALS solvers. */
+   virtual int SUNImplicitSetup(const Vector &x, const Vector &fx,
+                                int jok, int *jcur, double gamma);

-   int SUNMassSolve(const Vector &b, Vector &x, real_t tol) override;
+   /** Solve the system (M + dt K) x = M b. This method is used by the implicit
+       SUNDIALS solvers. */
+   virtual int SUNImplicitSolve(const Vector &b, Vector &x, double tol);

-   int SUNMassMult(const Vector &x, Vector &v) override;
+   /// Update the diffusion BilinearForm K using the given true-dof vector `u`.
+   void SetParameters(const Vector &u);
+
+   virtual ~ConductionOperator();
 };

-real_t InitialTemperature(const Vector &x)
-{
-   if (x.Norml2() < 0.5)
-   {
-      return 2.0;
-   }
-   else
-   {
-      return 1.0;
-   }
-}
-
+double InitialTemperature(const Vector &x);

 int main(int argc, char *argv[])
 {
@@ -150,16 +117,16 @@ int main(int argc, char *argv[])
   int ref_levels = 2;
   int order = 2;
   int ode_solver_type = 9; // CVODE implicit BDF
-   real_t t_final = 0.5;
-   real_t dt = 1.0e-2;
-   real_t alpha = 1.0e-2;
-   real_t kappa = 0.5;
+   double t_final = 0.5;
+   double dt = 1.0e-2;
+   double alpha = 1.0e-2;
+   double kappa = 0.5;
   bool visualization = true;
   bool visit = false;
   int vis_steps = 5;

   // Relative and absolute tolerances for CVODE and ARKODE.
-   const real_t reltol = 1e-4, abstol = 1e-4;
+   const double reltol = 1e-4, abstol = 1e-4;

   int precision = 8;
   cout.precision(precision);
@@ -184,10 +151,7 @@ int main(int argc, char *argv[])
                  "9  - CVODE (implicit BDF),\n\t"
                  "10 - ARKODE (default explicit),\n\t"
                  "11 - ARKODE (explicit Fehlberg-6-4-5),\n\t"
-                  "12 - ARKODE (default implicit),\n\t"
-                  "13 - ARKODE (default explicit with MFEM mass solve),\n\t"
-                  "14 - ARKODE (explicit Fehlberg-6-4-5 with MFEM mass solve),\n\t"
-                  "15 - ARKODE (default implicit with MFEM mass solve).");
+                  "12 - ARKODE (default impicit).");
   args.AddOption(&t_final, "-tf", "--t-final",
                  "Final time; start time is 0.");
   args.AddOption(&dt, "-dt", "--time-step",
@@ -210,13 +174,16 @@ int main(int argc, char *argv[])
      args.PrintUsage(cout);
      return 1;
   }
+   if (ode_solver_type < 1 || ode_solver_type > 12)
+   {
+      cout << "Unknown ODE solver type: " << ode_solver_type << '\n';
+      return 3;
+   }
   args.PrintOptions(cout);

-   bool use_mass_solver = ode_solver_type >= 13;
-
   // 2. Read the mesh from the given mesh file. We can handle triangular,
   //    quadrilateral, tetrahedral and hexahedral meshes with the same code.
-   std::unique_ptr<Mesh> mesh(new Mesh(mesh_file, 1, 1));
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
   int dim = mesh->Dimension();

   // 3. Refine the mesh to increase the resolution. In this example we do
@@ -230,7 +197,7 @@ int main(int argc, char *argv[])
   // 4. Define the vector finite element space representing the current and the
   //    initial temperature, u_ref.
   H1_FECollection fe_coll(order, dim);
-   FiniteElementSpace fespace(mesh.get(), &fe_coll);
+   FiniteElementSpace fespace(mesh, &fe_coll);

   int fe_size = fespace.GetTrueVSize();
   cout << "Number of temperature unknowns: " << fe_size << endl;
@@ -244,17 +211,8 @@ int main(int argc, char *argv[])
   Vector u;
   u_gf.GetTrueDofs(u);

-   // 6. Initialize the conduction ODE operator and the visualization.
-   ConductionOperator::Type ode_expression_type;
-   if (use_mass_solver)
-   {
-      ode_expression_type = ConductionOperator::Type::IMPLICIT;
-   }
-   else
-   {
-      ode_expression_type = ConductionOperator::Type::EXPLICIT;
-   }
-   ConductionOperator oper(fespace, alpha, kappa, u, ode_expression_type);
+   // 6. Initialize the conduction operator and the visualization.
+   ConductionOperator oper(fespace, alpha, kappa, u);

   u_gf.SetFromTrueDofs(u);
   {
@@ -266,7 +224,7 @@ int main(int argc, char *argv[])
      u_gf.Save(osol);
   }

-   VisItDataCollection visit_dc("Example16", mesh.get());
+   VisItDataCollection visit_dc("Example16", mesh);
   visit_dc.RegisterField("temperature", &u_gf);
   if (visit)
   {
@@ -300,75 +258,52 @@ int main(int argc, char *argv[])
   }

   // 7. Define the ODE solver used for time integration.
-   real_t t = 0.0;
-   std::unique_ptr<ODESolver> ode_solver;
+   double t = 0.0;
+   ODESolver *ode_solver = NULL;
+   CVODESolver *cvode = NULL;
+   ARKStepSolver *arkode = NULL;
   switch (ode_solver_type)
   {
      // MFEM explicit methods
-      case 1: ode_solver = std::make_unique<ForwardEulerSolver>(); break;
-      case 2: ode_solver = std::make_unique<RK2Solver>(0.5); break; // midpoint method
-      case 3: ode_solver = std::make_unique<RK3SSPSolver>(); break;
-      case 4: ode_solver = std::make_unique<RK4Solver>(); break;
+      case 1: ode_solver = new ForwardEulerSolver; break;
+      case 2: ode_solver = new RK2Solver(0.5); break; // midpoint method
+      case 3: ode_solver = new RK3SSPSolver; break;
+      case 4: ode_solver = new RK4Solver; break;
      // MFEM implicit L-stable methods
-      case 5: ode_solver = std::make_unique<BackwardEulerSolver>(); break;
-      case 6: ode_solver = std::make_unique<SDIRK23Solver>(2); break;
-      case 7: ode_solver = std::make_unique<SDIRK33Solver>(); break;
+      case 5: ode_solver = new BackwardEulerSolver; break;
+      case 6: ode_solver = new SDIRK23Solver(2); break;
+      case 7: ode_solver = new SDIRK33Solver; break;
      // CVODE
      case 8:
-      case 9:
-      {
-         int cvode_solver_type;
-         if (ode_solver_type == 8)
-         {
-            cvode_solver_type = CV_ADAMS;
-         }
-         else
-         {
-            cvode_solver_type = CV_BDF;
-         }
-         std::unique_ptr<CVODESolver> cvode(new CVODESolver(cvode_solver_type));
+         cvode = new CVODESolver(CV_ADAMS);
         cvode->Init(oper);
         cvode->SetSStolerances(reltol, abstol);
         cvode->SetMaxStep(dt);
-         ode_solver = std::move(cvode);
-         break;
-      }
+         ode_solver = cvode; break;
+      case 9:
+         cvode = new CVODESolver(CV_BDF);
+         cvode->Init(oper);
+         cvode->SetSStolerances(reltol, abstol);
+         cvode->SetMaxStep(dt);
+         ode_solver = cvode; break;
      // ARKODE
      case 10:
      case 11:
-      case 12:
-      case 13:
-      case 14:
-      case 15:
-      {
-         ARKStepSolver::Type arkode_solver_type;
-         if (ode_solver_type == 12 || ode_solver_type == 15)
-         {
-            arkode_solver_type = ARKStepSolver::IMPLICIT;
-         }
-         else
-         {
-            arkode_solver_type = ARKStepSolver::EXPLICIT;
-         }
-         std::unique_ptr<ARKStepSolver> arkode(
-            new ARKStepSolver(arkode_solver_type));
+         arkode = new ARKStepSolver(ARKStepSolver::EXPLICIT);
         arkode->Init(oper);
         arkode->SetSStolerances(reltol, abstol);
         arkode->SetMaxStep(dt);
-         if (ode_solver_type == 11 || ode_solver_type == 14)
+         if (ode_solver_type == 11)
         {
            arkode->SetERKTableNum(ARKODE_FEHLBERG_13_7_8);
         }
-         if (use_mass_solver)
-         {
-            arkode->UseMFEMMassLinearSolver(SUNFALSE);
-         }
-         ode_solver = std::move(arkode);
-         break;
-      }
-      default:
-         cout << "Unknown ODE solver type: " << ode_solver_type << '\n';
-         return 3;
+         ode_solver = arkode; break;
+      case 12:
+         arkode = new ARKStepSolver(ARKStepSolver::IMPLICIT);
+         arkode->Init(oper);
+         arkode->SetSStolerances(reltol, abstol);
+         arkode->SetMaxStep(dt);
+         ode_solver = arkode; break;
   }

   // Initialize MFEM integrators, SUNDIALS integrators are initialized above
@@ -376,14 +311,8 @@ int main(int argc, char *argv[])

   // Since we want to update the diffusion coefficient after every time step,
   // we need to use the "one-step" mode of the SUNDIALS solvers.
-   if (CVODESolver* cvode = dynamic_cast<CVODESolver*>(ode_solver.get()))
-   {
-      cvode->SetStepMode(CV_ONE_STEP);
-   }
-   else if (ARKStepSolver* arkode = dynamic_cast<ARKStepSolver*>(ode_solver.get()))
-   {
-      arkode->SetStepMode(ARK_ONE_STEP);
-   }
+   if (cvode) { cvode->SetStepMode(CV_ONE_STEP); }
+   if (arkode) { arkode->SetStepMode(ARK_ONE_STEP); }

   // 8. Perform time-integration (looping over the time iterations, ti, with a
   //    time-step dt).
@@ -394,7 +323,7 @@ int main(int argc, char *argv[])
   bool last_step = false;
   for (int ti = 1; !last_step; ti++)
   {
-      real_t dt_real = min(dt, t_final - t);
+      double dt_real = min(dt, t_final - t);

      // Note that since we are using the "one-step" mode of the SUNDIALS
      // solvers, they will, generally, step over the final time and will not
@@ -408,14 +337,8 @@ int main(int argc, char *argv[])
      if (last_step || (ti % vis_steps) == 0)
      {
         cout << "step " << ti << ", t = " << t << endl;
-         if (CVODESolver* cvode = dynamic_cast<CVODESolver*>(ode_solver.get()))
-         {
-            cvode->PrintInfo();
-         }
-         else if (ARKStepSolver* arkode = dynamic_cast<ARKStepSolver*>(ode_solver.get()))
-         {
-            arkode->PrintInfo();
-         }
+         if (cvode) { cvode->PrintInfo(); }
+         if (arkode) { arkode->PrintInfo(); }

         u_gf.SetFromTrueDofs(u);
         if (visualization)
@@ -430,153 +353,137 @@ int main(int argc, char *argv[])
            visit_dc.Save();
         }
      }
-      oper.SetConductionTensor(u);
+      oper.SetParameters(u);
   }
   tic_toc.Stop();
   cout << "Done, " << tic_toc.RealTime() << "s." << endl;

   // 9. Save the final solution. This output can be viewed later using GLVis:
   //    "glvis -m ex16.mesh -g ex16-final.gf".
-   u_gf.Save("ex16-final.gf", precision);
+   {
+      ofstream osol("ex16-final.gf");
+      osol.precision(precision);
+      u_gf.Save(osol);
+   }
+
+   // 10. Free the used memory.
+   delete ode_solver;
+   delete mesh;

   return 0;
 }

-ConductionOperator::ConductionOperator(FiniteElementSpace &fes,
-                                       const real_t alpha, const real_t kappa,
-                                       const Vector &u,
-                                       const Type &ode_expression_type)
-   : TimeDependentOperator(fes.GetTrueVSize(), 0.0, ode_expression_type),
-     fespace(fes), M(&fespace), alpha(alpha), kappa(kappa), z(height)
+ConductionOperator::ConductionOperator(FiniteElementSpace &f, double al,
+                                       double kap, const Vector &u)
+   : TimeDependentOperator(f.GetTrueVSize(), 0.0), fespace(f), M(NULL), K(NULL),
+     T(NULL), z(height)
 {
-   // specify a relative tolerance for all solves with MFEM integrators
-   const real_t rel_tol = 1e-8;
+   const double rel_tol = 1e-8;

-   M.AddDomainIntegrator(new MassIntegrator());
-   M.Assemble();
-   M.FormSystemMatrix(ess_tdof_list, Mmat);
+   M = new BilinearForm(&fespace);
+   M->AddDomainIntegrator(new MassIntegrator());
+   M->Assemble();
+   M->FormSystemMatrix(ess_tdof_list, Mmat);

   M_solver.iterative_mode = false;
-   M_solver.SetRelTol(rel_tol); // will be overwritten with SUNDIALS integrators
+   M_solver.SetRelTol(rel_tol);
   M_solver.SetAbsTol(0.0);
   M_solver.SetMaxIter(50);
   M_solver.SetPrintLevel(0);
   M_solver.SetPreconditioner(M_prec);
   M_solver.SetOperator(Mmat);

+   alpha = al;
+   kappa = kap;
+
   T_solver.iterative_mode = false;
-   T_solver.SetRelTol(rel_tol); // will be overwritten with SUNDIALS integrators
+   T_solver.SetRelTol(rel_tol);
   T_solver.SetAbsTol(0.0);
   T_solver.SetMaxIter(100);
   T_solver.SetPrintLevel(0);
   T_solver.SetPreconditioner(T_prec);

-   SetConductionTensor(u);
+   SetParameters(u);
 }

-void ConductionOperator::SetConductionTensor(const Vector &u)
+void ConductionOperator::Mult(const Vector &u, Vector &du_dt) const
+{
+   // Compute:
+   //    du_dt = M^{-1}*-K(u)
+   // for du_dt
+   Kmat.Mult(u, z);
+   z.Neg(); // z = -z
+   M_solver.Mult(z, du_dt);
+}
+
+void ConductionOperator::ImplicitSolve(const double dt,
+                                       const Vector &u, Vector &du_dt)
+{
+   // Solve the equation:
+   //    du_dt = M^{-1}*[-K(u + dt*du_dt)]
+   // for du_dt
+   if (T) { delete T; }
+   T = Add(1.0, Mmat, dt, Kmat);
+   T_solver.SetOperator(*T);
+   Kmat.Mult(u, z);
+   z.Neg();
+   T_solver.Mult(z, du_dt);
+}
+
+void ConductionOperator::SetParameters(const Vector &u)
 {
-   // Compute K(u_n).
   GridFunction u_alpha_gf(&fespace);
   u_alpha_gf.SetFromTrueDofs(u);
   for (int i = 0; i < u_alpha_gf.Size(); i++)
   {
      u_alpha_gf(i) = kappa + alpha*u_alpha_gf(i);
   }
+
+   delete K;
+   K = new BilinearForm(&fespace);
+
   GridFunctionCoefficient u_coeff(&u_alpha_gf);

-   K = std::make_unique<BilinearForm>(&fespace);
   K->AddDomainIntegrator(new DiffusionIntegrator(u_coeff));
   K->Assemble();
   K->FormSystemMatrix(ess_tdof_list, Kmat);
 }

-void ConductionOperator::ExplicitMult(const Vector &u, Vector &v) const
+int ConductionOperator::SUNImplicitSetup(const Vector &x,
+                                         const Vector &fx, int jok, int *jcur,
+                                         double gamma)
 {
-   // Compute - K(u_n) u.
-   Kmat.Mult(u, v);
-   v.Neg();
-}
-
-void ConductionOperator::Mult(const Vector &u, Vector &k) const
-{
-   // Compute - inv(M) K(u_n) u.
-   ExplicitMult(u, z);
-   M_solver.Mult(z, k);
-}
-
-void ConductionOperator::ImplicitSolve(const real_t gam, const Vector &u,
-                                       Vector &k)
-{
-   // Solve for k in M k = - K(u_n) [u + gam*k].
-   ExplicitMult(u, z);
-   T = std::unique_ptr<SparseMatrix>(Add(1.0, Mmat, gam, Kmat));
+   // Setup the ODE Jacobian T = M + gamma K.
+   if (T) { delete T; }
+   T = Add(1.0, Mmat, gamma, Kmat);
   T_solver.SetOperator(*T);
-   T_solver.Mult(z, k);
+   *jcur = 1;
+   return (0);
 }

-int ConductionOperator::SUNImplicitSetup(const Vector &u, const Vector &fu,
-                                         int jok, int *jcur, real_t gam)
+int ConductionOperator::SUNImplicitSolve(const Vector &b, Vector &x, double tol)
 {
-   // Compute T = M + gamma K(u_n).
-   T = std::unique_ptr<SparseMatrix>(Add(1.0, Mmat, gam, Kmat));
-   T_solver.SetOperator(*T);
-   *jcur = SUNTRUE; // this should eventually only be set true if K(u) is used
-   return SUN_SUCCESS;
+   // Solve the system A x = z => (M - gamma K) x = M b.
+   Mmat.Mult(b, z);
+   T_solver.Mult(z, x);
+   return (0);
 }

-int ConductionOperator::SUNImplicitSolve(const Vector &r, Vector &dk,
-                                         real_t tol)
+ConductionOperator::~ConductionOperator()
 {
-   // Solve the system [M + gamma K(u_n)] dk = - K(u_n) u - M k.
-   // What value r is providing depends on the ODE expression form:
-   //   EXPLICIT form: r = -inv(M) K(u_n) u - k
-   //   IMPLICIT form: r = -K(u_n) u - M k
-   T_solver.SetRelTol(tol);
-   if (isExplicit())
+   delete T;
+   delete M;
+   delete K;
+}
+
+double InitialTemperature(const Vector &x)
+{
+   if (x.Norml2() < 0.5)
   {
-      Mmat.Mult(r, z);
-      T_solver.Mult(z, dk);
+      return 2.0;
   }
   else
   {
-      T_solver.Mult(r, dk);
-   }
-   if (T_solver.GetConverged())
-   {
-      return SUN_SUCCESS;
-   }
-   else
-   {
-      return SUNLS_CONV_FAIL;
+      return 1.0;
   }
 }
-
-int ConductionOperator::SUNMassSetup()
-{
-   // Do nothing b/c mass solver was setup in constructor.
-   return SUN_SUCCESS;
-}
-
-int ConductionOperator::SUNMassSolve(const Vector &b, Vector &x, real_t tol)
-{
-   // Solve the system M x = b.
-   M_solver.SetRelTol(tol);
-   M_solver.Mult(b, x);
-   if (M_solver.GetConverged())
-   {
-      return SUN_SUCCESS;
-   }
-   else
-   {
-      return SUNLS_CONV_FAIL;
-   }
-}
-
-int ConductionOperator::SUNMassMult(const Vector &x, Vector &v)
-{
-   // Compute M x.
-   Mmat.Mult(x, v);
-   return SUN_SUCCESS;
-}
-
@@ -1,22 +1,16 @@
 //                         MFEM Example 16 - Parallel Version
 //                              SUNDIALS Modification
 //
-// Compile with:
-//    make ex16p            (GNU make)
-//    make sundials_ex16p   (CMake)
+// Compile with: make ex16p
 //
 // Sample runs:
 //     mpirun -np 4 ex16p
 //     mpirun -np 4 ex16p -m ../../data/inline-tri.mesh
 //     mpirun -np 4 ex16p -m ../../data/disc-nurbs.mesh -tf 2
 //     mpirun -np 4 ex16p -s 12 -a 0.0 -k 1.0
-//     mpirun -np 4 ex16p -s 15 -a 0.0 -k 1.0
 //     mpirun -np 4 ex16p -s 8 -a 1.0 -k 0.0 -dt 4e-6 -tf 2e-2 -vs 50
-//     mpirun -np 4 ex16p -s 11 -a 1.0 -k 0.0 -dt 4e-6 -tf 2e-2 -vs 50
 //     mpirun -np 8 ex16p -s 9 -a 0.5 -k 0.5 -o 4 -dt 8e-6 -tf 2e-2 -vs 50
-//     mpirun -np 8 ex16p -s 12 -a 0.5 -k 0.5 -o 4 -dt 8e-6 -tf 2e-2 -vs 50
 //     mpirun -np 4 ex16p -s 10 -dt 2.0e-4 -tf 4.0e-2
-//     mpirun -np 4 ex16p -s 13 -dt 2.0e-4 -tf 4.0e-2
 //     mpirun -np 16 ex16p -m ../../data/fichera-q2.mesh
 //     mpirun -np 16 ex16p -m ../../data/escher-p2.mesh
 //     mpirun -np 8 ex16p -m ../../data/beam-tet.mesh -tf 10 -dt 0.1
@@ -44,102 +38,66 @@
 using namespace std;
 using namespace mfem;

-/** After spatial discretization, the conduction model is expressed as
+/** After spatial discretization, the conduction model can be written as:
 *
- *   M du/dt = - K(u) u
+ *     du/dt = M^{-1}(-Ku)
 *
 *  where u is the vector representing the temperature, M is the mass matrix,
- *  and K(u) is the diffusion operator with diffusivity depending on u:
+ *  and K is the diffusion operator with diffusivity depending on u:
 *  (\kappa + \alpha u).
 *
- *  Class ConductionOperatorOperator represents the above ODE operator in the
- *  general form F(u, k, t) = G(u, t) where either
- *
- *    1. F(u, du/dt, t) = du/dt            (ODE is expressed in EXPLICIT form)
- *       G(u, t) = - inv(M) K(u) u
- *    2. F(u, du/dt, t) = M du/dt          (ODE is expressed in IMPLICIT form)
- *       G(u, t) = - K(u) u
+ *  Class ConductionOperator represents the right-hand side of the above ODE.
 */
 class ConductionOperator : public TimeDependentOperator
 {
+protected:
   ParFiniteElementSpace &fespace;
   Array<int> ess_tdof_list; // this list remains empty for pure Neumann b.c.

-   ParBilinearForm M;
+   ParBilinearForm *M;
+   ParBilinearForm *K;
+
   HypreParMatrix Mmat;
-
-   const real_t alpha, kappa;
-   std::unique_ptr<BilinearForm> K;
   HypreParMatrix Kmat;
+   HypreParMatrix *T; // T = M + dt K
+   double current_dt;

-   std::unique_ptr<HypreParMatrix> T; // T = M + gam K(u)
+   CGSolver M_solver;    // Krylov solver for inverting the mass matrix M
+   HypreSmoother M_prec; // Preconditioner for the mass matrix M

-   CGSolver M_solver; // Krylov solver for inverting the mass matrix M
-   HypreSmoother M_prec;  // Preconditioner for the mass matrix M
+   CGSolver T_solver;    // Implicit solver for T = M + dt K
+   HypreSmoother T_prec; // Preconditioner for the implicit solver

-   CGSolver T_solver; // Implicit solver for T = M + gam K(u)
-   HypreSmoother T_prec;  // Preconditioner for the implicit solver
+   double alpha, kappa;

   mutable Vector z; // auxiliary vector

 public:
+   ConductionOperator(ParFiniteElementSpace &f, double alpha, double kappa,
+                      const Vector &u);

-   ConductionOperator(ParFiniteElementSpace &f, const real_t alpha,
-                      const real_t kappa, const Vector &u,
-                      const Type &ode_expression_type);
+   virtual void Mult(const Vector &u, Vector &du_dt) const;

-   // Compute K(u_n) for use as an approximation in - K(u) u
-   void SetConductionTensor(const Vector &u);
+   /** Solve the Backward-Euler equation: k = f(u + dt*k, t), for the unknown k.
+       This is the only requirement for high-order SDIRK implicit integration.*/
+   virtual void ImplicitSolve(const double dt, const Vector &u, Vector &k);

-   /** Compute G(u, t) as defined in the IMPLICIT expression form of the ODE
-       operator, i.e., @a v = - K(u_n) @a u. Note that K(u_n) is an
-       approximation to K(u). */
-   void ExplicitMult(const Vector &u, Vector &v) const override;
+   /** Setup the system (M + dt K) x = M b. This method is used by the implicit
+       SUNDIALS solvers. */
+   virtual int SUNImplicitSetup(const Vector &x, const Vector &fx,
+                                int jok, int *jcur, double gamma);

-   /** Solve for k in F(u, k, t) = G(u, t) for either EXPLICIT or IMPLICIT
-       expression forms of the ODE operator, i.e., @a k = - inv(M) K(u_n) @a u.
-       Note that K(u_n) is an approximation to K(u). */
-   void Mult(const Vector &u, Vector &k) const override;
+   /** Solve the system (M + dt K) x = M b. This method is used by the implicit
+       SUNDIALS solvers. */
+   virtual int SUNImplicitSolve(const Vector &b, Vector &x, double tol);

-   /** Solve for k in F(u + gam*k, k, t) = G(u + gam*k, t) for either EXPLICIT
-       or IMPLICIT expression forms of the ODE operator, i.e.,
-       [ M + @a gam K(u_n) ] @a k = - K(u_n) @a u . Note that K(u_n) is an
-       approximation to K(u). */
-   void ImplicitSolve(const real_t gam, const Vector &u, Vector &k) override;
+   /// Update the diffusion BilinearForm K using the given true-dof vector `u`.
+   void SetParameters(const Vector &u);

-   /** Setup to solve for dk in [dF/dk + gam*dF/du - gam*dG/du] dk = G - F for
-       either EXPLICIT or IMPLICIT expression forms of the ODE operator, i.e.,
-       [M - @a gam Jf(u)] dk = G - F, where Jf(u) is an approximation of the
-       Jacobian of -K(u) u. The approximation chosen here is Jf(u) = -K(u_n). */
-   int SUNImplicitSetup(const Vector &u, const Vector &fu, int jok, int *jcur,
-                        real_t gam) override;
-
-   /** Solve for @a dk in the system in SUNImplicitSetup to the given tolerance,
-       with the residual @a r providing either
-        1. @a r = G - F = inv(M) f(u) - k       (EXPLICIT expression form)
-        1. @a r = G - F = f(u) - M k            (IMPLICIT expression form)
-       */
-   int SUNImplicitSolve(const Vector &r, Vector &dk, real_t tol) override;
-
-   int SUNMassSetup() override;
-
-   int SUNMassSolve(const Vector &b, Vector &x, real_t tol) override;
-
-   int SUNMassMult(const Vector &x, Vector &v) override;
+   virtual ~ConductionOperator();
 };

-real_t InitialTemperature(const Vector &x)
-{
-   if (x.Norml2() < 0.5)
-   {
-      return 2.0;
-   }
-   else
-   {
-      return 1.0;
-   }
-}
-
+double InitialTemperature(const Vector &x);

 int main(int argc, char *argv[])
 {
@@ -156,16 +114,16 @@ int main(int argc, char *argv[])
   int par_ref_levels = 1;
   int order = 2;
   int ode_solver_type = 9; // CVODE implicit BDF
-   real_t t_final = 0.5;
-   real_t dt = 1.0e-2;
-   real_t alpha = 1.0e-2;
-   real_t kappa = 0.5;
+   double t_final = 0.5;
+   double dt = 1.0e-2;
+   double alpha = 1.0e-2;
+   double kappa = 0.5;
   bool visualization = true;
   bool visit = false;
   int vis_steps = 5;

   // Relative and absolute tolerances for CVODE and ARKODE.
-   const real_t reltol = 1e-4, abstol = 1e-4;
+   const double reltol = 1e-4, abstol = 1e-4;

   int precision = 8;
   cout.precision(precision);
@@ -192,10 +150,7 @@ int main(int argc, char *argv[])
                  "9  - CVODE (implicit BDF),\n\t"
                  "10 - ARKODE (default explicit),\n\t"
                  "11 - ARKODE (explicit Fehlberg-6-4-5),\n\t"
-                  "12 - ARKODE (default implicit),\n\t"
-                  "13 - ARKODE (default explicit with MFEM mass solve),\n\t"
-                  "14 - ARKODE (explicit Fehlberg-6-4-5 with MFEM mass solve),\n\t"
-                  "15 - ARKODE (default implicit with MFEM mass solve).");
+                  "12 - ARKODE (default impicit).");
   args.AddOption(&t_final, "-tf", "--t-final",
                  "Final time; start time is 0.");
   args.AddOption(&dt, "-dt", "--time-step",
@@ -219,33 +174,40 @@ int main(int argc, char *argv[])
      return 1;
   }

-   if (Mpi::Root())
+   if (myid == 0)
   {
      args.PrintOptions(cout);
   }

-   bool use_mass_solver = ode_solver_type >= 13;
+   // check for valid ODE solver option
+   if (ode_solver_type < 1 || ode_solver_type > 12)
+   {
+      if (myid == 0)
+      {
+         cout << "Unknown ODE solver type: " << ode_solver_type << '\n';
+      }
+      return 1;
+   }

-   // 3. Define a parallel mesh by a partitioning of a serial mesh. Read the
-   //    serial mesh from the given mesh file on all processors. We can
+   // 3. Read the serial mesh from the given mesh file on all processors. We can
   //    handle triangular, quadrilateral, tetrahedral and hexahedral meshes
   //    with the same code.
-   std::unique_ptr<ParMesh> pmesh;
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
+   int dim = mesh->Dimension();
+
+   // 4. Refine the mesh in serial to increase the resolution. In this example
+   //    we do 'ser_ref_levels' of uniform refinement, where 'ser_ref_levels' is
+   //    a command-line parameter.
+   for (int lev = 0; lev < ser_ref_levels; lev++)
   {
-      std::unique_ptr<Mesh> mesh(new Mesh(mesh_file, 1, 1));
-
-      // 4. Refine the mesh in serial to increase the resolution. In this example
-      //    we do 'ser_ref_levels' of uniform refinement, where 'ser_ref_levels' is
-      //    a command-line parameter.
-      for (int lev = 0; lev < ser_ref_levels; lev++)
-      {
-         mesh->UniformRefinement();
-      }
-
-      // 5. Refine this mesh further in parallel to increase the resolution.
-      //    Once the parallel mesh is defined, the serial mesh can be deleted.
-      pmesh = std::make_unique<ParMesh>(MPI_COMM_WORLD, *mesh);
+      mesh->UniformRefinement();
   }
+
+   // 5. 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.
+   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
+   delete mesh;
   for (int lev = 0; lev < par_ref_levels; lev++)
   {
      pmesh->UniformRefinement();
@@ -253,9 +215,8 @@ int main(int argc, char *argv[])

   // 6. Define the vector finite element space representing the current and the
   //    initial temperature, u_ref.
-   int dim = pmesh->Dimension();
   H1_FECollection fe_coll(order, dim);
-   ParFiniteElementSpace fespace(pmesh.get(), &fe_coll);
+   ParFiniteElementSpace fespace(pmesh, &fe_coll);

   int fe_size = fespace.GlobalTrueVSize();
   if (myid == 0)
@@ -272,17 +233,8 @@ int main(int argc, char *argv[])
   Vector u;
   u_gf.GetTrueDofs(u);

-   // 8. Initialize the conduction ODE operator and the visualization.
-   ConductionOperator::Type ode_expression_type;
-   if (use_mass_solver)
-   {
-      ode_expression_type = ConductionOperator::Type::IMPLICIT;
-   }
-   else
-   {
-      ode_expression_type = ConductionOperator::Type::EXPLICIT;
-   }
-   ConductionOperator oper(fespace, alpha, kappa, u, ode_expression_type);
+   // 8. Initialize the conduction operator and the VisIt visualization.
+   ConductionOperator oper(fespace, alpha, kappa, u);

   u_gf.SetFromTrueDofs(u);
   {
@@ -297,7 +249,7 @@ int main(int argc, char *argv[])
      u_gf.Save(osol);
   }

-   VisItDataCollection visit_dc("Example16-Parallel", pmesh.get());
+   VisItDataCollection visit_dc("Example16-Parallel", pmesh);
   visit_dc.RegisterField("temperature", &u_gf);
   if (visit)
   {
@@ -341,76 +293,52 @@ int main(int argc, char *argv[])
   }

   // 9. Define the ODE solver used for time integration.
-   real_t t = 0.0;
-   std::unique_ptr<ODESolver> ode_solver;
+   double t = 0.0;
+   ODESolver *ode_solver = NULL;
+   CVODESolver *cvode = NULL;
+   ARKStepSolver *arkode = NULL;
   switch (ode_solver_type)
   {
      // MFEM explicit methods
-      case 1: ode_solver = std::make_unique<ForwardEulerSolver>(); break;
-      case 2: ode_solver = std::make_unique<RK2Solver>(0.5); break; // midpoint method
-      case 3: ode_solver = std::make_unique<RK3SSPSolver>(); break;
-      case 4: ode_solver = std::make_unique<RK4Solver>(); break;
+      case 1: ode_solver = new ForwardEulerSolver; break;
+      case 2: ode_solver = new RK2Solver(0.5); break; // midpoint method
+      case 3: ode_solver = new RK3SSPSolver; break;
+      case 4: ode_solver = new RK4Solver; break;
      // MFEM implicit L-stable methods
-      case 5: ode_solver = std::make_unique<BackwardEulerSolver>(); break;
-      case 6: ode_solver = std::make_unique<SDIRK23Solver>(2); break;
-      case 7: ode_solver = std::make_unique<SDIRK33Solver>(); break;
+      case 5: ode_solver = new BackwardEulerSolver; break;
+      case 6: ode_solver = new SDIRK23Solver(2); break;
+      case 7: ode_solver = new SDIRK33Solver; break;
      // CVODE
      case 8:
-      case 9:
-      {
-         int cvode_solver_type;
-         if (ode_solver_type == 8)
-         {
-            cvode_solver_type = CV_ADAMS;
-         }
-         else
-         {
-            cvode_solver_type = CV_BDF;
-         }
-         std::unique_ptr<CVODESolver> cvode(
-            new CVODESolver(MPI_COMM_WORLD, cvode_solver_type));
+         cvode = new CVODESolver(MPI_COMM_WORLD, CV_ADAMS);
         cvode->Init(oper);
         cvode->SetSStolerances(reltol, abstol);
         cvode->SetMaxStep(dt);
-         ode_solver = std::move(cvode);
-         break;
-      }
+         ode_solver = cvode; break;
+      case 9:
+         cvode = new CVODESolver(MPI_COMM_WORLD, CV_BDF);
+         cvode->Init(oper);
+         cvode->SetSStolerances(reltol, abstol);
+         cvode->SetMaxStep(dt);
+         ode_solver = cvode; break;
      // ARKODE
      case 10:
      case 11:
-      case 12:
-      case 13:
-      case 14:
-      case 15:
-      {
-         ARKStepSolver::Type arkode_solver_type;
-         if (ode_solver_type == 12 || ode_solver_type == 15)
-         {
-            arkode_solver_type = ARKStepSolver::IMPLICIT;
-         }
-         else
-         {
-            arkode_solver_type = ARKStepSolver::EXPLICIT;
-         }
-         std::unique_ptr<ARKStepSolver> arkode(
-            new ARKStepSolver(MPI_COMM_WORLD, arkode_solver_type));
+         arkode = new ARKStepSolver(MPI_COMM_WORLD, ARKStepSolver::EXPLICIT);
         arkode->Init(oper);
         arkode->SetSStolerances(reltol, abstol);
         arkode->SetMaxStep(dt);
-         if (ode_solver_type == 11 || ode_solver_type == 14)
+         if (ode_solver_type == 11)
         {
            arkode->SetERKTableNum(ARKODE_FEHLBERG_13_7_8);
         }
-         if (use_mass_solver)
-         {
-            arkode->UseMFEMMassLinearSolver(SUNFALSE);
-         }
-         ode_solver = std::move(arkode);
-         break;
-      }
-      default:
-         cout << "Unknown ODE solver type: " << ode_solver_type << '\n';
-         return 3;
+         ode_solver = arkode; break;
+      case 12:
+         arkode = new ARKStepSolver(MPI_COMM_WORLD, ARKStepSolver::IMPLICIT);
+         arkode->Init(oper);
+         arkode->SetSStolerances(reltol, abstol);
+         arkode->SetMaxStep(dt);
+         ode_solver = arkode; break;
   }

   // Initialize MFEM integrators, SUNDIALS integrators are initialized above
@@ -418,18 +346,12 @@ int main(int argc, char *argv[])

   // Since we want to update the diffusion coefficient after every time step,
   // we need to use the "one-step" mode of the SUNDIALS solvers.
-   if (CVODESolver* cvode = dynamic_cast<CVODESolver*>(ode_solver.get()))
-   {
-      cvode->SetStepMode(CV_ONE_STEP);
-   }
-   else if (ARKStepSolver* arkode = dynamic_cast<ARKStepSolver*>(ode_solver.get()))
-   {
-      arkode->SetStepMode(ARK_ONE_STEP);
-   }
+   if (cvode) { cvode->SetStepMode(CV_ONE_STEP); }
+   if (arkode) { arkode->SetStepMode(ARK_ONE_STEP); }

   // 10. Perform time-integration (looping over the time iterations, ti, with a
   //     time-step dt).
-   if (Mpi::Root())
+   if (myid == 0)
   {
      cout << "Integrating the ODE ..." << endl;
   }
@@ -439,7 +361,7 @@ int main(int argc, char *argv[])
   bool last_step = false;
   for (int ti = 1; !last_step; ti++)
   {
-      real_t dt_real = min(dt, t_final - t);
+      double dt_real = min(dt, t_final - t);

      // Note that since we are using the "one-step" mode of the SUNDIALS
      // solvers, they will, generally, step over the final time and will not
@@ -455,14 +377,8 @@ int main(int argc, char *argv[])
         if (myid == 0)
         {
            cout << "step " << ti << ", t = " << t << endl;
-            if (CVODESolver* cvode = dynamic_cast<CVODESolver*>(ode_solver.get()))
-            {
-               cvode->PrintInfo();
-            }
-            else if (ARKStepSolver* arkode = dynamic_cast<ARKStepSolver*>(ode_solver.get()))
-            {
-               arkode->PrintInfo();
-            }
+            if (cvode) { cvode->PrintInfo(); }
+            if (arkode) { arkode->PrintInfo(); }
         }

         u_gf.SetFromTrueDofs(u);
@@ -479,38 +395,46 @@ int main(int argc, char *argv[])
            visit_dc.Save();
         }
      }
-      oper.SetConductionTensor(u);
+      oper.SetParameters(u);
   }
   tic_toc.Stop();
-   if (Mpi::Root())
+   if (myid == 0)
   {
      cout << "Done, " << tic_toc.RealTime() << "s." << endl;
   }

   // 11. Save the final solution in parallel. This output can be viewed later
   //     using GLVis: "glvis -np <np> -m ex16-mesh -g ex16-final".
-   u_gf.Save("ex16-final", precision);
+   {
+      ostringstream sol_name;
+      sol_name << "ex16-final." << setfill('0') << setw(6) << myid;
+      ofstream osol(sol_name.str().c_str());
+      osol.precision(precision);
+      u_gf.Save(osol);
+   }
+
+   // 12. Free the used memory.
+   delete ode_solver;
+   delete pmesh;

   return 0;
 }

-ConductionOperator::ConductionOperator(ParFiniteElementSpace &fes,
-                                       const real_t alpha, const real_t kappa,
-                                       const Vector &u,
-                                       const Type &ode_expression_type)
-   : TimeDependentOperator(fes.GetTrueVSize(), 0.0, ode_expression_type),
-     fespace(fes), M(&fespace), alpha(alpha), kappa(kappa),
-     M_solver(fes.GetComm()), T_solver(fes.GetComm()), z(height)
+ConductionOperator::ConductionOperator(ParFiniteElementSpace &f, double al,
+                                       double kap, const Vector &u)
+   : TimeDependentOperator(f.GetTrueVSize(), 0.0), fespace(f), M(NULL), K(NULL),
+     T(NULL),
+     M_solver(f.GetComm()), T_solver(f.GetComm()), z(height)
 {
-   // specify a relative tolerance for all solves with MFEM integrators
-   const real_t rel_tol = 1e-8;
+   const double rel_tol = 1e-8;

-   M.AddDomainIntegrator(new MassIntegrator());
-   M.Assemble(0); // keep zeros to keep sparsity pattern of M and K the same
-   M.FormSystemMatrix(ess_tdof_list, Mmat);
+   M = new ParBilinearForm(&fespace);
+   M->AddDomainIntegrator(new MassIntegrator());
+   M->Assemble(0); // keep sparsity pattern of M and K the same
+   M->FormSystemMatrix(ess_tdof_list, Mmat);

   M_solver.iterative_mode = false;
-   M_solver.SetRelTol(rel_tol); // will be overwritten with SUNDIALS integrators
+   M_solver.SetRelTol(rel_tol);
   M_solver.SetAbsTol(0.0);
   M_solver.SetMaxIter(100);
   M_solver.SetPrintLevel(0);
@@ -518,118 +442,97 @@ ConductionOperator::ConductionOperator(ParFiniteElementSpace &fes,
   M_solver.SetPreconditioner(M_prec);
   M_solver.SetOperator(Mmat);

+   alpha = al;
+   kappa = kap;
+
   T_solver.iterative_mode = false;
-   T_solver.SetRelTol(rel_tol); // will be overwritten with SUNDIALS integrators
+   T_solver.SetRelTol(rel_tol);
   T_solver.SetAbsTol(0.0);
   T_solver.SetMaxIter(100);
   T_solver.SetPrintLevel(0);
   T_solver.SetPreconditioner(T_prec);

-   SetConductionTensor(u);
+   SetParameters(u);
 }

-void ConductionOperator::SetConductionTensor(const Vector &u)
+void ConductionOperator::Mult(const Vector &u, Vector &du_dt) const
+{
+   // Compute:
+   //    du_dt = M^{-1}*-K(u)
+   // for du_dt
+   Kmat.Mult(u, z);
+   z.Neg(); // z = -z
+   M_solver.Mult(z, du_dt);
+}
+
+void ConductionOperator::ImplicitSolve(const double dt,
+                                       const Vector &u, Vector &du_dt)
+{
+   // Solve the equation:
+   //    du_dt = M^{-1}*[-K(u + dt*du_dt)]
+   // for du_dt
+   if (T) { delete T; }
+   T = Add(1.0, Mmat, dt, Kmat);
+   T_solver.SetOperator(*T);
+   Kmat.Mult(u, z);
+   z.Neg();
+   T_solver.Mult(z, du_dt);
+}
+
+int ConductionOperator::SUNImplicitSetup(const Vector &x,
+                                         const Vector &fx, int jok, int *jcur,
+                                         double gamma)
+{
+   // Setup the ODE Jacobian T = M + gamma K.
+   if (T) { delete T; }
+   T = Add(1.0, Mmat, gamma, Kmat);
+   T_solver.SetOperator(*T);
+   *jcur = 1;
+   return (0);
+}
+
+int ConductionOperator::SUNImplicitSolve(const Vector &b, Vector &x, double tol)
+{
+   // Solve the system A x = z => (M - gamma K) x = M b.
+   Mmat.Mult(b, z);
+   T_solver.Mult(z, x);
+   return (0);
+}
+
+void ConductionOperator::SetParameters(const Vector &u)
 {
-   // Compute K(u_n).
   ParGridFunction u_alpha_gf(&fespace);
   u_alpha_gf.SetFromTrueDofs(u);
   for (int i = 0; i < u_alpha_gf.Size(); i++)
   {
      u_alpha_gf(i) = kappa + alpha*u_alpha_gf(i);
   }
+
+   delete K;
+   K = new ParBilinearForm(&fespace);
+
   GridFunctionCoefficient u_coeff(&u_alpha_gf);

-   K = std::make_unique<ParBilinearForm>(&fespace);
   K->AddDomainIntegrator(new DiffusionIntegrator(u_coeff));
-   K->Assemble(0); // keep zeros to keep sparsity pattern of M and K the same
+   K->Assemble(0); // keep sparsity pattern of M and K the same
   K->FormSystemMatrix(ess_tdof_list, Kmat);
 }

-void ConductionOperator::ExplicitMult(const Vector &u, Vector &v) const
+ConductionOperator::~ConductionOperator()
 {
-   // Compute - K(u_n) u.
-   Kmat.Mult(u, v);
-   v.Neg();
+   delete T;
+   delete M;
+   delete K;
 }

-void ConductionOperator::Mult(const Vector &u, Vector &k) const
+double InitialTemperature(const Vector &x)
 {
-   // Compute - inv(M) K(u_n) u.
-   ExplicitMult(u, z);
-   M_solver.Mult(z, k);
-}
-
-void ConductionOperator::ImplicitSolve(const real_t gam, const Vector &u,
-                                       Vector &k)
-{
-   // Solve for k in M k = - K(u_n) [u + gam*k].
-   ExplicitMult(u, z);
-   T = std::unique_ptr<HypreParMatrix>(Add(1.0, Mmat, gam, Kmat));
-   T_solver.SetOperator(*T);
-   T_solver.Mult(z, k);
-}
-
-int ConductionOperator::SUNImplicitSetup(const Vector &u, const Vector &fu,
-                                         int jok, int *jcur, real_t gam)
-{
-   // Compute T = M + gamma K(u_n).
-   T = std::unique_ptr<HypreParMatrix>(Add(1.0, Mmat, gam, Kmat));
-   T_solver.SetOperator(*T);
-   *jcur = SUNTRUE; // this should eventually only be set true if K(u) is used
-   return SUN_SUCCESS;
-}
-
-int ConductionOperator::SUNImplicitSolve(const Vector &r, Vector &dk,
-                                         real_t tol)
-{
-   // Solve the system [M + gamma K(u_n)] dk = - K(u_n) u - M k.
-   // What value r is providing depends on the ODE expression form:
-   //   EXPLICIT form: r = -inv(M) K(u_n) u - k
-   //   IMPLICIT form: r = -K(u_n) u - M k
-   T_solver.SetRelTol(tol);
-   if (isExplicit())
+   if (x.Norml2() < 0.5)
   {
-      Mmat.Mult(r, z);
-      T_solver.Mult(z, dk);
+      return 2.0;
   }
   else
   {
-      T_solver.Mult(r, dk);
-   }
-   if (T_solver.GetConverged())
-   {
-      return SUN_SUCCESS;
-   }
-   else
-   {
-      return SUNLS_CONV_FAIL;
+      return 1.0;
   }
 }
-
-int ConductionOperator::SUNMassSetup()
-{
-   // Do nothing b/c mass solver was setup in constructor.
-   return SUN_SUCCESS;
-}
-
-int ConductionOperator::SUNMassSolve(const Vector &b, Vector &x, real_t tol)
-{
-   // Solve the system M x = b.
-   M_solver.SetRelTol(tol);
-   M_solver.Mult(b, x);
-   if (M_solver.GetConverged())
-   {
-      return SUN_SUCCESS;
-   }
-   else
-   {
-      return SUNLS_CONV_FAIL;
-   }
-}
-
-int ConductionOperator::SUNMassMult(const Vector &x, Vector &v)
-{
-   // Compute M x.
-   Mmat.Mult(x, v);
-   return SUN_SUCCESS;
-}
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Brendan Keith	e7e0ed79e6	fixed BC issue	2024-04-14 14:59:23 -04:00
Brendan Keith	7359466ce4	fix sign error	2024-04-12 08:12:42 -04:00
Brendan Keith	d5814b9d8e	fixed invertibility bug	2024-04-11 22:28:15 -04:00
Brendan Keith	7dd2312ec4	minor	2024-04-11 18:56:53 -04:00
Brendan Keith	aa25a12086	changing to direct solver	2024-04-11 18:05:56 -04:00
Brendan Keith	3341feb1d3	linearized problem solved. memory leak	2024-04-11 17:37:39 -04:00
Brendan Keith	a6f7baeede	fixed the runtime issues with integrators	2024-04-10 11:32:58 -04:00
Brendan Keith	bc2ee7f3a9	layout of ex40.cpp done	2024-04-10 10:17:49 -04:00
Brendan Keith	2bcae01f71	outline of MA code	2024-04-09 21:17:45 -04:00
Brendan Keith	cde13ad145	Merge branch '2x2_matrix_exponential' into MA	2024-04-09 15:26:27 -04:00
Brendan Keith	f728fadcf7	more work on ex40	2024-04-07 22:11:21 -04:00
Brendan Keith	3db9688894	Merge branch '2x2_matrix_exponential' into MA Adding coefficients from 2x2_matrix_exponential	2024-04-06 15:33:30 -04:00
Brendan Keith	d9dc18c32b	starting ex40	2024-04-05 12:36:35 -04:00