This package provides tools for numerical analysis under conditions where the
underlying mathematics is scale-invariant. We work on computers with finite
precision, so operations like matrix-multiplication and matrix-division are
expected to have some error. However, naive estimates of the error, based on
quantities like norm(x), may not be scale-invariant.
For example, if H is a diagonal nonnegative (Hessian) matrix (i.e., a rank-2
covariant tensor), with a change-of-scale in the variables all such H are
equivalent to the identity matrix. Therefore we might think that its condition
number should in fact be 1.
This package provides tool to calculate condition number, as well as several
other quantities, under conditions of scale-invariance.
See the documentation for more information.