Consider the conditional expectation of $x$ given $y$, $$ \mathbb{E}(x | y) $$ where $x \in X$ and $y \in Y$ where $X, Y$ are Hilbert spaces (possibly infinite dimensional).

Question : I am looking for conditions under which $$ \mathbb{E}(x | \cdot ) : Y \to X $$ is a continuous function.

I'll give some simple examples to illustrate the problem:

Example 1: Assume $X = Y = \mathbb{R}$, let $x$ have a gaussian prior with mean 0 and unit variance and let $y$ be gaussian with mean $x$ and unit variance. In this case we have $$ \mathbb{E}(x | y) = \frac{y}{2} $$ which is continuous.

Example 2: Assume $X = Y = \ell_2$. Let $x$ have a gaussian prior with $x_i \sim N(0, 1/i)$ and let $y_i ~ \sim N(x_i / i, 1 / i^3)$. In this case we have $$ [\mathbb{E}(x | y)]_i = \left( \frac{1}{i} i^3 \frac{1}{i} + i \right)^{-1} \frac{1}{i} i^3 y_i = \frac{i}{2} y_i $$ This is a linear operator, but unbounded, hence $\mathbb{E}(x | \cdot)$ is not continuous.

  • $\begingroup$ Your question is not well posed. Apparently, $x$ is a fixed random variable with values in $X$. But what is then $y\mapsto E(x|y)$? $\endgroup$ – Jochen Wengenroth Feb 13 '18 at 8:39
  • $\begingroup$ I was sloppy on the notation to save some space. To be more strict, the idea is that $X/Y$ are $\mathcal{X}/\mathcal{Y}$ valued random variables and hence $E(X|Y)$ is an $\mathcal{X}$ valued random variable which is a function of a $\mathcal{Y}$ valued random variable. Thus the function I'm looking for is thus more strictly $y \to E(X|Y=y)$. $\endgroup$ – Jonas Adler Feb 13 '18 at 8:47

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.