[tuhopuu-devel] LSCM sls_solver.cpp line 310

[Bjornmose]

hi blendix and all
as far as i can read it :{
the "least_sqaure" flag modifies the original matrix A by adding some
matrix M "on the fly"  consisting of "square" elements a[i]*a[j] so A
becomes soemthing like A+M.
And the components of the target vector b become scaled "on the fly".
b -> b'
So it seems to be solving (A+M)x= b' which is again a linear system to
solve say (A'')x =b';
I am still struggeling to evaluate this /* bruno help us , i know that
you know! */

so what i know(ref1)
LET y(x):= x^2+b*x+c; y(x) has a extrama, derive dy/dx=2*x+b; so slove
dy/dx =0 to get x in the minimum,maximum (and it is unique prooven!).
expanding this to multiple dimensions is not so hard to do.

i think (ref1) this must do the trick:
Knowing "least squares" in a general sense in our case reduces to solve
a "true square" problem. /* no higher order terms in taylor's series
occur */
So knowing the derivates this becomes a linear problem in derivates
space ( leaving one "unknown" for every dimension, (defined by
pinning) )
/* i am thinking "math" and feel so clumpsy to express my intention /
sorry for that / bruno?? */
}
N.B. I could not resist:
WHEE this is "clean and well documented" code! /* more clean than
documented, i think :-)*/
/* this would be a usesfull comment if it would exist */
/* i know i am one of the coders using "sparse" comments */
( is it just me, that i can't see any comment at all ?)