Skip to content

Latest commit

 

History

History
121 lines (87 loc) · 4.27 KB

README.md

File metadata and controls

121 lines (87 loc) · 4.27 KB
Raw

XBraid-convergence-est

Operating systems

This code was developed for use with Linux operating systems.

Installation requirements

The code depends on a number of third-party libraries:

  • C++ compiler
  • MPI (recommended: MPICH2)
  • bootstrap
  • armadillo
  • optional: Doxygen
  • optional: matplotlib (recommended: version >= 1.5.2)

To install bootstrap and armadillo, we recommend using SPACK, since SPACK will install all dependencies (blas, lapack, etc.).

Makefile targets

default : builds the default binary
doxygen : generates the Doxygen source code documentation
load-packages : loads the recommended libraries using SPACK
unload-packages : unloads the recommended libraries using SPACK

Examples

Commandline options

A list of commandline options is displayed when running:

    ./bin/xbraid-convergence-est --help

Compute approximate convergence factor of MGRIT V-cycle with FCF-relaxation

Sample complex plane

This example samples spatial eigenvalues in the comlex plane and computes an approximate convergence factor (according to Hessenthaler et al., Approximation 2) of 5-level MGRIT V-cycles with FCF-relaxation on each level:

    ./bin/xbraid-convergence-est                            \
        --number-of-timesteps  1025                         \
        --number-of-time-grids 5                            \
        --coarsening-factors   2 2 2 2                      \
        --runge-kutta-method L_stable_SDIRK1                \
        --sample-complex-plane -10.0 1.0 -4.0 4.0           \
        --complex-plane-sample-size 110 90                  \
        --bound error_l2_sqrt_expression_approximate_rate   \
        --bound-on-level 1                                  \
        --V-cycle                                           \
        --relaxation-scheme FCF_relaxation                  \
        --output-file V-cycle_FCF-relaxation_approx

This will produce the following output:

  • dteta_l0.txt : contains the sampled spatial eigenvalues
  • V-cycle_FCF-relaxation_approx.txt : contains the approximate convergence factor for all samples
  • max_V-cycle_FCF-relaxation_approx.txt : contains the maximum approximate convergence factor

Results can be visualized by running:

    python3 utils/heatmap.py dteta_l0.txt V-cycle_FCF-relaxation_approx.txt NAN NAN NAN NAN 0.0 1.0

This will visualize the approximate convergence factors in the complex plane:

Image: Approximate convergence factor for 5-level V-cycles with FCF-relaxation (SDIRK1)

The visualized data is also exported as heatmap.png.

To find out more about the arguments of the Python script, run:

    python3 utils/heatmap.py

Sample real axis

This example samples the real axis (by using the same complex plane sampling routines) and computes the l2-norm of the error propagation operator (on level 1; according to Hessenthaler et al., Lemma 3.3 and Equation 4.1) of 5-level MGRIT V-cycles with FCF-relaxation on each level in parallel:

    mpirun -np 4 ./bin/xbraid-convergence-est               \
        --number-of-timesteps  1025                         \
        --number-of-time-grids 5                            \
        --coarsening-factors   2 2 2 2                      \
        --runge-kutta-method L_stable_SDIRK1                \
        --sample-complex-plane -0.9 0.0001 0.0 0.0          \
        --complex-plane-sample-size 12 1                    \
        --bound error_l2_upper_bound                        \
        --bound-on-level 1                                  \
        --V-cycle                                           \
        --relaxation-scheme FCF_relaxation                  \
        --output-file V-cycle_FCF-relaxation_l2error

This will produce the following output:

  • dteta_l0.txt : contains the sampled spatial eigenvalues
  • V-cycle_FCF-relaxation_l2error.txt : contains the value of the upper bound for all samples
  • max_V-cycle_FCF-relaxation_l2error.txt : contains the maximum value of the upper bound for all samples

The script prints:

Convergence <= 0.22158731181372346

Source code documentation

Doxygen is used to automatically generate source code documentation. See doc/doxygen/html/index.html