Data of Newton method of "A unifying framework for ADI-like methods for linear matrix equations and beneficial consequences"

This deposit contains the datasets generated by the Newton method for the paper:

• J. Schulze, J. Saak: "A unifying framework for ADI-like methods for linear matrix equations and beneficial consequences".

Download the individual files and store them inside the directory data/newton-adi/.

The file names are structured as follows.

• Rail5177: Steel Profile benchmark problem of dimension 5177
• adi_initprev=true|false: whether the initial ADI iterate was set to the solution of the previous Newton step (or zero)
• adi_kwargs=...: keyword arguments passed to ADI method
  • maxiters=1000: maximum number of iterations
  • shifts=...: shift strategy
• newton_kwargs=...: keyword arguments passed to Newton method
  • inexact=true|false: whether to use inexact Newton method
  • inexact_hybrid=true|false: whether to switch back to classical Newton method in later iterations (only present if inexact=true)
  • linesearch=true|false: whether to employ line search
  • reltol=1e-10: relative tolerance to reason about convergence
• β=1000: scaling of the quadratic term in the ARE
• .jld2: file suffix. All file have been generated with JLD2.jl version 0.4.38

Load the dataset via using JLD2 and file = load(FILENAME). This will yield a dictionary having the following entries:

• file["newton_metrics"]: data frame containing execution metrics of Newton iterations
• file["adi_metrics"]: data frame containing execution metrocs of ADI iterations of all Newton iterations
• file["timer"]: isolated runtime metrics generated with TimerOutputs.jl version 0.5.23
• file["timer_metrics"]: runtime metrics of seperate run with additional data observers enabled
• file["config"]: internal configuration object that led to this dataset (information also embedded in file name)
• file["failed"]: Boolean on whether configuration has failed

All data frames were generated with DataFrames.jl version 1.6.1 and have their columns documented using metadata.

Generating this dataset took ~3h and consumed ~0.25kWh of electricity.