Re: [Bug-gsl] [bug #34361] gsl_bspline_knots_greville needs inequality constrained linear least squares

[Rhys Ulerich]

Hi Brian,

> [I]s this algorithm for generating points by solving
> that constrained system known to work? i.e. is there a published reference 
> for it?

The algorithm for generating the breakpoints via the constrained system is not
published (to my knowledge) though it's not the sort of thing anyone
would publish.
The system of constraints is natural for the problem setting.  The algorithm for
solving the constraint problem once it is in matrix form is classical
and the difficulty
purely plumbing.

I don't consider it research but you're (of course) welcome to tell me
to yank the
feature and associated code.

>> I intend to perform the same f2c-then-cleanup task starting directly from the
>> Netlib sources to ensure the license pedigree remains firmly public domain.

> One comment: I've never got on well with f2c code, it's difficult to
> maintain.  Writing an equivalent routine from scratch is usually not
> much more difficult.

The Fortran 77 kernels from Netlib run to about 550 lines.  The
published algorithm
from Lawson and Hanson contains the bare bones details but only hints at
some improvements and defers to their source for full details.  The f2c route
(along with refactoring and portability fixes) was more straightforward.
I have reimplemented functionality above those kernels in C using regular GSL
APIs.

- Rhys