1
$\begingroup$

In this question: Lipschitz implies bounded gradient it is shown that if $f: \mathbb{R}^n \to \mathbb{R}^n$ is convex and L-Lipschitz the gradient is bounded by L. I wonder if this holds in Banach spaces? If $f$ is a L-Lipschtiz, convex functional from a Banach space $X$ to $\mathbb{R}$ and $\lambda \in \partial f(x) = \{ x^*: f(x) - f(u) \leq \langle x*,x-u\rangle, \forall u \in X\} $, is it true that $\lVert \lambda \rVert_* \leq L$?

If I consider the inverse duality mapping $J^{-1}(\lambda) = \{ \in X: \langle \lambda,u\rangle = \lVert u \rVert^2= \lVert \lambda \rVert_*^2\}$ and choose $y$ such that $x-y \in J^{-1}(\lambda)$ then I think the proof can be modified:

$$ L \lVert \lambda_t \rVert_* = \lVert x-y \rVert L \geq | \ell_t(x) - \ell_t(y)| \geq | \left(\lambda_t, x-y \right) |= \lVert \lambda_t \rVert_*^2, $$ and then divide both sides by $\lVert \lambda_t \rVert_* $. Is this correct or am I missing something?

$\endgroup$

1 Answer 1

2
$\begingroup$

In the case that $X$ is not reflexive, $J^{-1}(\lambda)$ might be empty.

However, we can argue directly. For arbitrary $h \in X$ we have $$\langle x^*, h\rangle = \langle x^*, (x+h) - x\rangle \le f(x+h) - f(x) \le L \| (x+h) - h)\|_X = L \|h\|_X. $$ Thus, $$ \|x^*\|_{X^*} = \sup_{\|h\|_X\le1}\langle x^*,h\rangle \le L.$$

$\endgroup$

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.