Pointwise convergence of $c$-concave envelopes as the sequence of cost functions converges pointwise

262 Views Asked by At

On Remark 1.12, page 33 of Villani's Topics in optimal transport we find the following problem. It should be easy but it's turning me mad.

Let $c:\mathbb{R}^n\times\mathbb{R}^n\rightarrow \mathbb{R}_+\cup\{+\infty\}$ be a measurable, bounded, lower semicontinuous function, then there exists $c_l$ an increasing family of bounded uniformly continuous functions converging pointwise to $c$. Let $\varphi:\mathbb{R}^n\rightarrow\mathbb{R}$ be a bounded function.

Define $$\varphi^c(y)=\inf_{x\in X}[c(x,y)-\varphi(x)],$$ and $$\psi_l(y)=\inf_{x\in X}[c_l(x,y)-\varphi(x)].$$ Then, it is easily seen that each $\psi_l$ is uniformly continuous, but how do we prove that $\psi_l$ converges pointwise to $\varphi^c$?

1

There are 1 best solutions below

1
On BEST ANSWER

This is not true as stated. Define $$\varphi(x) = \begin{cases} 0 \quad &\text{ if } x=0 \\ 1 \quad &\text{ if } x\ne 0 \end{cases}$$ and $c(x,y) = \varphi(x)$. Clearly, $\varphi^c(y)=0$ for all $y$.

Suppose that $c_l$ is a continuous function such that $c_l\le c$. Take a sequence of nonzero numbers $x_k$ converging to $0$: then $$\lim_{k\to\infty}(c_l(x_k,y)-\varphi(x_k)) = c_l(0,y)-1 \le -1$$ and therefore $\psi_l(y)\le -1$ for all $y$.


The author needed this "exercise" to justify the measurability of $\varphi^c$... If I were in your place, I would trust the claim that it's measurable and keep on reading.