[vtk-developers] [Proposal] Moving old univariate solvers from vtkMath to vtkPolynomialSolversUnivariate

[Philippe P. Pebay]

Hello dear VTK developers,

About a year ago, I moved all univariate polynomial solvers which David Thompson and I had developed, from vtkMath to 
vtkPolynomialSolversUnivariate. This includes in particular the Sturm, Lin-Bairstow, and Ferrari solvers.
Subsequent development, together with Korben Rusek, of the Habicht solver has been done directly in 
vtkPolynomialSolversUnivariate.

However, at the time, we did not move the "historic" solvers that were already present in vtkMath; more precisely, these 
are the quadratic and cubic solvers which use exact (up to numerical accuracy) algebraic formulas.

I would like to suggest that we move those as well to vtkPolynomialSolversUnivariate. This would present several advantages:
1. consistency (all solvers in the same place)
2. efficiency (centralized documentation for all solvers, hence allowing users to learn more easily what options they 
have in terms of univariate polynomial solvers)
3. simplification (vtkMath is now a bloated package of various mathematics algorithm which have little in common)
4. and other ancillary benefits such as having a single test harness for all univariate polynomial solvers.

Needless to say, backwards-compatibility would be maintained for a long while (until a to be determined future release 
number) so every user has a chance to modify his code as needed. I would suggest something of the order of one year. 
This is similar to what we did with the radian to degree, and degree to radian, conversion functions.

Please let me know what you think. Thank you.

P.
-- 
Philippe Pébay
Sandia National Laboratories