Is there a connection between condition number of a problem (a problem is a function from one normed vector space X of data to another normed vector space Y of solutions ) and radius of curvature of the function? In other words, can we infer whether a problem is ill- or well conditioned near a point x in X just by calculating the radius of curvature of the function near that point x?

For example, lets consider f(x) = tan (x). We want to determine whether this problem is ill- or well conditioned near x=pi/2. If we calculate the condition number with the help of the Jacobian, we can readily infer that the condition number is very very large which makes the problem ill-conditioned. Now let's take on a different angle of reasoning.

In the small left neighborhood of x=pi/2, a small change in x will result in a large change of f(x), which makes the problem ill-conditioned near pi/2. This massive change in f(x) near pi/2 results from the radius of curvature of the function near pi/2. Thus from the order of magnitude of the radius of curvature gives the information of the conditioning of any problem.

Any suggestions or feedback would be highly appreciated.

More Avishek Mukherjee's questions See All
Similar questions and discussions