Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Is it true that if $Y_1,Y_2,\cdots,Y_n$ are jointly Gaussian with $E[Y_i]=0~\forall i$ then

$$\mathrm{span}(Y_1,Y_2,\cdots,Y_n) = \mathcal{L}^2(\Omega,\sigma(Y_1,Y_2,\cdots,Y_n),P) ?$$

I wanted to know if this is true based on the following evidence:

If $Y_1,Y_2,\cdots,Y_n$ are jointly Gaussian with $E[Y_i]=0~\forall i$, then the Conditional Expectation defined by

$$E[Y_1 | Y_2, Y_3, \cdots, Y_n] := E[Y_1 | \sigma(Y_2, Y_3, \cdots, Y_n)]$$

is just the projection of $Y_1$ on $\mathcal{L}^2(\Omega,\sigma(Y_2,\cdots,Y_n),P)$ as $Y_1 \in \mathcal{L}^2(\Omega,\sigma(Y_1,Y_2,\cdots,Y_n),P)$.

On the other hand it can also be shown that the conditional expectation is just the projection of $Y_1$ on $\mathrm{span}(Y_2,Y_3,\cdots,Y_n)$ (for example, this is used in the derivation of the Kalman FIlter) so that the conditional expectation is just a linear combination of $Y_i$, i.e.

$$E[Y_1 | \sigma(Y_2, Y_3, \cdots, Y_n)] = \sum_{i=2}^n \alpha_i Y_i ~a.s.$$

for some $\alpha_i \in \mathbb{R}$.

I have asked a similar question at stackexchange but did not get much help.

Thanks, Phanindra

share|improve this question
    
$\mathcal{L}^2(\Omega, \sigma(Y_1, \dots, Y_n), P)$ is infinite dimensional. –  Nate Eldredge Mar 29 '12 at 17:00

1 Answer 1

up vote 8 down vote accepted

It is certainly not true, there are $L^2$ variables measurable in the $(Y_i)$ which are not Gaussian (e.g. the sign of $Y_1$) whereas the span of the $(Y_i)$ is composed of Gaussians, which is the definition of being jointly Gaussian.

That some vector has the same projection on two sub-spaces is rather weak evidence that those two sub-spaces are equal ...

OTOH, you can look into Skorokhod embedding, which allows you to realize any centered, square-integrable random variable as the value of Brownian motion at a well-chosen stopping time. This is not the same as what you were asking, but it does tell you that $L^2$ is somehow not that far from being spanned by Gaussian variables.

share|improve this answer
    
@Vincent: Thanks for your answer. I have understood the difference between the two. –  jpv Mar 29 '12 at 12:03

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.