notes-to-remove.txt

LAPCK is limited to roght looking algroithmm, cannot  really doleft looking by reconstructing each time 

construction of T from Demmel is not stable, can we do a similar construction
in the standard case? What is he doing? Where does this construction comes
from? 

Needs to count these flops in Householder for my claim

What CG? Can CG keep the orthogonality if we use Householder reflectors somewhere? It sounds that reflecting to have a smaller matrix is a better idea than projecting. We have the zero on top that essentially prevent us from losing orthogonality

Think about the pop for MGS, MGS basically runs Householder in the small world
so that it never have to push and pop. Is that it?  Can we use this insight to
do something about it? 


How do get from Q to V in the one vector case? This is pretty easy. And i think this explains LU.
Think about the

Work on this LARFT that improves stability 
what matters is
how ``H * ( I in small world)`` so H is orthogonal.
And then look at representatiity
plot in term of condition number
rpve


#include <Accelerate/Accelerate.h>

/usr/local/bin/gcc -framework Accelerate xxx.c 


Gustavson proposes to do ( A, T ). We propose ( A, T and Q ).


change interface for A,ia,ja, B,ib,jb