Majorization in a Neighborhood
Consider the cubic . It is an increasing function on the real line, with a root at , and a saddle point at .
We know that cubics do not have quadratic majorizers, but we can try to find a quadratic such that for all and such that if .
The condition for all is which is in our case We have for
If for all and then This can be extended easily to higher order partials
Minimize on , where , and thus if and The majorizer is minimized at . Thus the majorization algorithm is which converges monotonically to zero if started with . Because convergence is sublinear.