pymgrit 0.1.2

Python implementation of the MGRIT algorithm

Introduction

PyMGRIT is a package for the Multigrid-Reduction-in-Time (MGRIT) algorithm in Python.

PyMGRIT is currently developed by Jens Hahne and Stephanie Friedhoff.

What is MGRIT?

The MGRIT algorithm is a reduction-based time-multigrid method for solving time-dependent problems. A reduction-based method attempts to reduce the solving of one problem to equivalently solving two smaller problems. Reduction-based multigrid methods are iterative solvers that consist of two parts: relaxation and coarse-grid correction, which are, in the spirit of reduction, designed to be complementary in reducing error associated with different degrees of freedom. Applying this idea in the time domain, MGRIT combines local time stepping on the discretized temporal domain, the fine grid, for a relaxation scheme, with time stepping on a coarse temporal mesh (or a hierarchy of coarse temporal meshes) that uses a larger time step for the coarse-grid correction.

PyMGRIT Features

PyMGRIT features:

• Classical Multigrid-Reduction-in-Time (MGRIT) for solving evolutionary systems of equations

  • Non-intrusive approach

  • Optimal time-multigrid algorithm

  • A variety of cycling strategies, relaxation schemes, and coarsening strategies

• Time parallelism

• Specific to space-time problems

  • Space & time parallelism

  • Additional coarsening in space

Citing

@MISC{PyMGRIT,
  author = "Hahne, J. and Friedhoff, S.",
  title = "{PyMGRIT}: Multigrid-Reduction-in-Time in {Python} v1.0",
  year = "2020",
  url = "https://github.com/pymgrit/pymgrit",
  note = "Release 1.0"
  }

Installation

PyMGRIT requires mpicc (from openmpi or mpich)

>>> pip3 install pymgrit

or

>>> pip3 install .

Example Usage

PyMGRIT is easy to use! The following code generates a discrete Dahlquist test problem and solves the resulting linear system using a two-level MGRIT algorithm.:

# Import PyMGRIT
from pymgrit import *

# Create Dahlquist's test problem with 101 time steps in the interval [0, 5]
dahlquist = Dahlquist(t_start=0, t_stop=5, nt=101)

# Construct a two-level multigrid hierarchy for the test problem using a coarsening factor of 2
dahlquist_multilevel_structure = simple_setup_problem(problem=dahlquist, level=2, coarsening=2)

# Set up the MGRIT solver for the test problem and set the solver tolerance to 1e-10
mgrit = Mgrit(problem=dahlquist_multilevel_structure, tol=1e-10)

# Solve the test problem
info = mgrit.solve()

Program output:

INFO - 21-02-20 16:18:43 - Start setup
INFO - 21-02-20 16:18:43 - Setup took 0.009232759475708008 s
INFO - 21-02-20 16:18:43 - Start solve
INFO - 21-02-20 16:18:43 - iter 1  | conv: 7.186185937031941e-05  | conv factor: -                       | runtime: 0.013237237930297852 s
INFO - 21-02-20 16:18:43 - iter 2  | conv: 1.2461067076355103e-06 | conv factor: 0.017340307063501627    | runtime: 0.010195493698120117 s
INFO - 21-02-20 16:18:43 - iter 3  | conv: 2.1015566145245807e-08 | conv factor: 0.016864981158092696    | runtime: 0.008922338485717773 s
INFO - 21-02-20 16:18:43 - iter 4  | conv: 3.144127445017594e-10  | conv factor: 0.014960945726074891    | runtime: 0.0062139034271240234 s
INFO - 21-02-20 16:18:43 - iter 5  | conv: 3.975214076032893e-12  | conv factor: 0.01264329816633959     | runtime: 0.006150722503662109 s
INFO - 21-02-20 16:18:43 - Solve took 0.05394101142883301 s
INFO - 21-02-20 16:18:43 - Run parameter overview
  time interval             : [0.0, 5.0]
  number of time points     : 101
  max dt                    : 0.05000000000000071
  number of levels          : 2
  coarsening factors        : [2]
  cf_iter                   : 1
  nested iteration          : True
  cycle type                : V
  stopping tolerance        : 1e-10
  time communicator size    : 1
  space communicator size   : -99

Getting Help

For documentation see https://pymgrit.github.io/pymgrit/

Create an issue.

Look at the Quickstart, Tutorial or the Examples.