User:IssaRice/Chain rule proofs

From Machinelearning
Revision as of 01:25, 28 November 2018 by IssaRice (talk | contribs)

Using Newton's approximation

Since g is differentiable at y0, we know g(y0) is a real number, and we can write

g(y)=g(y0)+g(y0)(yy0)+[g(y)(g(y0)+g(y0)(yy0))]

If we define Eg(Δy):=g(y)(g(y0)+g(y0)(yy0)) we can write

g(y)=g(y0)+g(f(x0))(yy0)+Eg(Δy)

Newton's approximation says that |Eg(Δy)|ϵ|yy0| as long as |yy0|δ.

Since f is differentiable at x0, we know that it must be continuous at x0. This means we can keep |f(x)y0|δ as long as we keep |xx0|δ.

Since f(x)Y and |f(x)y0|δ, this means we can substitute y=f(x) and get

g(f(x))=g(y0)+g(f(x0))(f(x)y0)+Eg(Δf)