3
$\begingroup$

Set $X=\mathbb{R}^n$ and let $X^{I}$, the space of maps from the (bounded or unbounded) interval $I$ to $X$, be endowed with the locally convex topology of pointwise convergence.

Is it true that the subspace $C(I, X)$ of continuous maps is not in the Borel algebra of $X^I$?

It is clear from the proof of the above continuity statement that for each (dense) countable subset $D \subset I$, the space of maps that are continuous from $D$ to $X$ (or even Hölder-continuous for exponents $< 1/2$) is measurable and has measure $1$. However, the set of dense subsets is not countable, so we cannot take the intersection.

How to prove that the set of continuous functions are not measurable?

/Edit: Sorry about this. I meant to ask a somewhat different question but changed it while writing; then forgot to change the title.

Maybe some background: With Kolmogorovs Extension theorem, you get a unique measure on the above function space that is defined by its finite-dimensional distributions. This is the Wiener measure. Then there is the continuity theorem that says, you get a version of the associated stochastic process with continuous paths. From the measure perspective, this fits only together for me if the set of continuous paths is not measurable, as you could otherwise just restrict your measure to this space.

$\endgroup$
3
  • $\begingroup$ Sorry, I forgot to change the title... $\endgroup$ Commented Jun 3, 2013 at 20:49
  • $\begingroup$ Wouldn’t it fit together if the set has measure zero? $\endgroup$
    – The User
    Commented Jun 3, 2013 at 20:50
  • $\begingroup$ But you would still wonder what kind of canonical measure you constructed that does not have continuous paths and how you could transfer it to the set of continuous functions. But of course, this is only a heuristic argument. $\endgroup$ Commented Jun 3, 2013 at 21:24

1 Answer 1

2
$\begingroup$

For $(x_j)_{j\in\Bbb N}\subset I$ and $B\in\mathcal B((\Bbb R^n)^{\Bbb N})$, define $S((x_j)_{j\in\Bbb N},B):=\{f\in X^I,(f(x_j))_{j\in\Bbb N}\in B\}$. The collection of sets of this form contains the $\sigma$-algebra generated by the semi-norms $f\mapsto |f(x)|$, $x\in I$, which is the same as the $\sigma$-algebra generated by the evaluation maps $f\mapsto f(x)$.

We now show that the collection $\mathcal C=\{S(\mathbf x,B), \mathbf x\in\Bbb R^{\infty},B\in \Bbb R^{\infty}\}$ is a $\sigma$-algebra. The main difficulty is to show that this collection is stable for countable unions. Let $\tau\colon\Bbb N^2\to\Bbb N$ be a bijection, and $(S(\mathbf x_{k,\cdot},B_k))_{k\in\Bbb N}$ elements of $\mathcal C$. Then take $y_k:=x_{\tau^{-1}(k)}$ and $B\in\mathcal B((\Bbb R^n)^{\Bbb N}$ such that $B_k=\pi_{\{k\}\times\Bbb N}^{-1}(B)$, where $\pi_I((x_n)_{n\in\Bbb N})=(x_n,n\in I)$.

The subset of continuous functions cannot be expressed in this form, because we don't control the behavior of a function of $S((x_j)_{j\in\Bbb N},B)$ outside an uncountable set.

$\endgroup$
4
  • $\begingroup$ This seems false. You at least need to allow different sets $B$ for each $x_j$, otherwise the Sigma-Algebra is too small. $\endgroup$ Commented Jun 4, 2013 at 6:21
  • $\begingroup$ Also, the union of two such sets is generally not of this form anymore. $\endgroup$ Commented Jun 4, 2013 at 6:23
  • $\begingroup$ I probably missed something, but here $B$ is a Borel subset of the space of sequences of elements of $\BBb R^n$. And if $(x_j)_j$, $(y_j)_j$ are two sequences, and $B_1$, $B_2$ two such Borel sets, define $z=(x_1,y_1,x_2,\dots,)$, and we can write $S((x_j)_j,B_1)\cup S((y_j)_j,B_2)=\{f,(f(z_{2n+1})\in B_1\mbox{ or }f(z_{2n})\in B_2\}$. For countable unions, we can proceed similarly taking a bijection from $\Bbb N^2$ to $\Bbb N$. $\endgroup$ Commented Jun 4, 2013 at 8:21
  • 1
    $\begingroup$ Yes, sorry, I think I overread the $\mathbb{N}$ in the exponent. $\endgroup$ Commented Jun 4, 2013 at 12:34

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .