My question is a general one, whose answer can probably be found in any decent convex analysis book. I unfortunately don't have any at hand right now, so here it is:

Let's consider a "reasonable" Banach space $X$ (say at least reflexive and separable as usual), a convex subset $C\subset X$ and a function $\Phi:C\to\mathbb{R}$ which is continuous for the strong $X$ topology and convex.

What do I need to assume for $\Phi$ to be weakly lower-semicontinuous? Clearly differentiability works with the classical trick $x_n\rightharpoonup x$ and $\Phi(x_k)\geq \Phi(x)+D\Phi(x).(x_k-x)$, but I guess there must be less restrictive conditions than differentiability?

Thank you!


A function is lower semicontinuous (in some topology on $X$) if and only if its epigraph $\{(x,y)\in X\times \mathbb R:y\ge \Phi(x)\}$ is a closed set in the product topology on $X\times \mathbb R$.

In particular, if we use the weak topology on $X$, the product topology on $X\times \mathbb R$ is the weak topology of the product $X\times \mathbb R$, considered as a Banach space of its own.

It is a well-known consequence of Hahn-Banach theorem that a strongly closed convex set is weakly closed, in any normed space.

Combining the above, we conclude: every lower semicontinuous convex function is weakly lower semicontinuous. No further assumptions are needed.

  • $\begingroup$ Very elegant answer, thank you. This clearly works if $\Phi$ is defined on the whole space $X$. However, I'm a little concerned with the case when $\Phi$ is only defined on a convex subset $C\subset X$: if I'm not mistaken your cute argument would require $C$ to be strongly closed, which I'm not assuming here... What do you think? $\endgroup$ – leo monsaingeon Jul 9 '13 at 21:36
  • $\begingroup$ @leomonsaingeon Since the epigraph $\{(x,y):y\ge \Phi(x),x\in C\}$ is convex, so is its closure (in the norm topology of $X\times \mathbb R$). This closure is the epigraph of a convex lsc function defined on $X$ (which takes on the value $+\infty$ outside of the closure of $C$). $\endgroup$ – 40 votes Jul 9 '13 at 21:55
  • $\begingroup$ Ah, yes indeed... thank you! $\endgroup$ – leo monsaingeon Jul 10 '13 at 12:05

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.