I have a function $g$ which is simply the norm: $g(x)=||F(x)|| $. $F(x)$ is continuously differentiable. In fact it an actually be any norm and these results should still hold since all norms are convex. I have the following two results, and I am trying to figure out how to connect them to a third result:

1) I know that all norms are convex. To remind us, convexity means $\theta f(x)+(1-\theta) f(y) \ge f(\theta x+(1-\theta )y)$ so for the norm that means $\theta ||x||+(1-\theta) ||y|| \ge ||(\theta x+(1-\theta )y||$ where $\theta \in \{0,1\}$.

Convexity also means that the first order approximation is always an underestimator: $f(y) \ge f(x) + \nabla f^T*(y-x)$ which in our case by subustituting what $g(x)$ is, means $g'(x;d) \le ||F(x+hd)|| - ||F(x)||$.

2) I have the following result for the directional derivative at $g$ (the derivative of $g(x)$ in the direction $d$ : $g'(x;d)=\lim_{h\rightarrow0} \frac{||F(x)+hF'(x)d||-||F(x)||}{h} $

And I need to be able to obtain the result:

$\,\,\,\,\,\,g'(x;d) \le ||F(x)+F'(x)d||-||F(x)||$.

what i tried

I was able to use the fact that $g$ is convex, and $g'(x;d) \le ||F(x+hd)|| - ||F(x)||$, and substitute that in to show

$\lim_{h\downarrow0} \frac{||F(x)+hF'(x)d||-||F(x)||}{h} \le ||F(x+hd)||-||F(x)||$ But I cannot seem to obtain the result I need. Can someone help me obtain the result that I need?


For fixed $x$ and $d$, let $$\phi(t) = \left\|F(x) + tF'(x)d\right\|$$ By convexity of the norm $\left\|\cdot\right\|$, $\phi$ is also convex and so for $0\le t \le 1$, $$\phi(t) = \phi((1-t)\cdot 0 + t\cdot 1) \le t\phi(1) + (1-t)\phi(0)$$ Implies $$\frac{\phi(t) - \phi(0)}{t} \le \phi(1) - \phi(0)$$ $t\rightarrow 0$ and you find what you are looking for.

  • $\begingroup$ I see thank you @Youem $\endgroup$ – nundo Apr 29 '18 at 17:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.