@anchor LinearSolversApplicationMainPage

LinearSolversApplication

The LinearSolversApplication is a thin wrapper for the Eigen linear algebra library.

Direct sparse solvers

The application provides the following direct sparse solvers:

*SPD = Symmetric Positive Definite

Example:

{
    "solver_type": "eigen_sparse_lu"
}

Direct dense solvers

The application provides the following direct solvers for dense systems of equations:

*SPD = Symmetric Positive Definite

**Can also be used to solve equation systems with multiple right hand sides.

Generalized eigensystem solvers

The application provides the following generalized eigensystem Ax=λBx solver for sparse matrices.

*SPD = Symmetric Positive Definite **A special version for symmetric matrices can be triggered in the solver settings.

EigensystemSolver and SpectraSymGEigsShiftSolver compute the smallest eigenvalues and corresponding eigenvectors of the system. MKL routines are used automatically if they are available.

SpectraSymGEigsShiftSolver interfaces a solver from the Spectra library, and has a shift mode that can be used to compute the smallest eigenvalues > shift.

Example:

{
    "solver_type": "spectra_sym_g_eigs_shift",
    "number_of_eigenvalues": 3,
    "max_iteration": 1000,
    "echo_level": 1
}

If the application is compiled with MKL, FEAST 4.0 can be used to solve the generalized eigenvalue problem for real and complex systems (symmetric or unsymmetric). The cmake switch USE_EIGEN_FEAST must be set to ON with

-DUSE_EIGEN_FEAST=ON \

Example:

{
    "solver_type": "feast",
    "symmetric": true,
    "number_of_eigenvalues": 3,
    "search_lowest_eigenvalues": true,
    "e_min" : 0.0,
    "e_max" : 0.2
}

Build instructions

1. Set the required definitions for cmake

   As any other app:

   Windows: in configure.bat

   set KRATOS_APPLICATIONS=%KRATOS_APPLICATIONS%%KRATOS_APP_DIR%\LinearSolversApplication;
   

   Linux: in configure.sh

   add_app ${KRATOS_APP_DIR}/LinearSolversApplication
   

2. Build Kratos

3. Setup the ProjectParameters.json

   "linear_solver_settings": {
       "solver_type" : "LinearSolversApplication.sparse_lu"
   }
   

4. Run the simulation

Enable MKL (optional)

In case you have installed MKL (see below), you can also use the Pardiso solvers.

1. Run the MKL setup script before building Kratos:

   Windows:

   call "C:\Program Files (x86)\Intel\oneAPI\mkl\latest\env\vars.bat" intel64 lp64
   

   Linux:

   source /opt/intel/oneapi/setvars.sh intel64
   

2. Add the following flag to CMake to your configure script:

   Windows:

   -DUSE_EIGEN_MKL=ON ^
   

   Linux:

   -DUSE_EIGEN_MKL=ON \
   

3. Build Kratos

4. Usage:

   Windows:

   call "C:\Program Files (x86)\Intel\oneAPI\mkl\latest\env\vars.bat" intel64 lp64
   

   Linux:

   Set the environment before using MKL

   source /opt/intel/oneapi/setvars.sh intel64
   

Install MKL on Ubuntu with apt

Intel MKL can be installed with apt on Ubuntu. A guide can be found in here. For example to install the MKL 2022 version

sudo bash
# <type your user password when prompted.  this will put you in a root shell>
# If they are not installed, you can install using the following command:
sudo apt update
sudo apt -y install cmake pkg-config build-essential
# use wget to fetch the Intel repository public key
wget https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
# add to your apt sources keyring so that archives signed with this key will be trusted.
sudo apt-key add GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
# remove the public key
rm GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
# Configure apt client to use Intel repository
sudo add-apt-repository "deb https://apt.repos.intel.com/oneapi all main"
# Install all MKL related dependencies. You can install full HPC with: sudo apt install intel-hpckit
sudo apt install intel-oneapi-mkl-devel
# Exit
exit

To enable the MKL environment (needs to be done before build/run) use

source /opt/intel/oneapi/setvars.sh intel64

Enable SuiteSparse (optional)

SuiteSparse is a collection of open-source sparse matrix algorithms, including efficient implementations of various factorizations. Kratos currently provides wrappers for CHOLMOD, UMFPACK, and SPQR.

Install the SuiteSparse package on your system, and set the USE_EIGEN_SUITESPARSE flag in your CMake configuration to ON. One way of doing this is appending the list of arguments you pass to CMake in your configure script with:

-DUSE_EIGEN_SUITESPARSE:BOOL=ON

Installing SuiteSparse

• Arch Linux

  • SuiteSparse is available through your package manager.
  • sudo pacman -S suitesparse

• Debian (and derivatives, including Ubuntu)

  • SuiteSparse is available through your package manager.
  • sudo apt install libsuitesparse-dev

• Fedora (and derivatives, including RHEL)

  • SuiteSparse is available through your package manager.
  • sudo dnf install suitesparse-devel

• MacOS

  • SuiteSparse is available on Homebrew for both Intel and Apple Silicon machines.
  • brew install suite-sparse

• Windows

  • SuiteSparse is available through vcpkg and MSYS2 package managers.
  • vcpkg install suitesparse suitesparse-spqr suitesparse-umfpack (remember to add to build script -DCMAKE_TOOLCHAIN_FILE="vcpkg_path\vcpkg\scripts\buildsystems\vcpkg.cmake")
  • pacman -S mingw-w64-x86_64-suitesparse