Sat Mar 24 17:05:35 CET 2012

Invertibility through sparseness

Thinking a bit more about constraint solving.

If all constrains are linear, then GE is the way to go.

What makes local propagation interesting is that it allows solution of
nonlinear equations that remain unique (invertible) through

I.e. there's a difference between an equation like

             xy = 1
             x^2 - y = 0

The former is a bijection while the latter isn't.

The interesting observation is that many practical systems are
nonlinearly constrained but remain invertible, or at least locally
invertable for a wide range around the solution.