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

I am looking for a version of the Bayesian Linear Model in Banach spaces.

Background: the finite dimensional case

The following is a well-known theorem about normal correlation for finite-dimensional random vectors.

Suppose that $X$ and $Y$ are random vectors with means $\mu_X$, $\mu_Y$, variance matrices $\Lambda_X$, $\Lambda_Y$ and cross-covariance $\Lambda_{XY}$. If $X$ and $Y$ are jointly Normally distributed, then the marginal distribution of $X$ is Normal and the conditional distribution of $Y$ given $X$ is Normal with mean $$\mu_Y + \Lambda_{YX}\Lambda_X^{-1}(X-\mu_X)$$ and variance $$\Lambda_Y - \Lambda_{YX}\Lambda_X^{-1}\Lambda_{XY}.$$ Conversely, if the distribution of $X$ is normal and the conditional distribution of $Y$ given $X$ is Normal with mean $AX$ (for some matrix $A$) and constant variance $\Lambda_{Y|X}$, then the joint distribution of $X$ and $Y$ is Normal.

From the previous result one can derive the following corollary.

Suppose that $$X\sim N(\mu_X, \Lambda_X)$$ and $$Y|X\sim N(AX, V).$$ Then $$X|Y\sim N(\mu_{X|Y}, \Lambda_{X|Y}),$$ where $$\mu_{X|Y} = \mu_X + K(Y - A\mu_X),$$ $$\Lambda_{X|Y} = \Lambda_X - K(A\Lambda_XA' + V)K',$$ with $K = \Lambda_XA'(A\Lambda_XA' + V)^{-1}$.

In all the above formulas, a generalized inverse can replace the inverse in case the inverse does not exist.

The infinite dimensional case

This is actually what I am looking for. More specifically,

Is anybody aware of a similar result for Banach space-valued random vectors? If not, what are the closest results that I can look at?

From a statistical point of view, the result above provides an updating formula for the Normal-Normal linear model.

Is there a Bayesian version of the linear model for Banach space-valued random vectors?

In all of my questions, I would be happy with results or pointers for separable Banach spaces and also for the simpler case of Hilbert space-valued random variables.

share|improve this question
add comment

Your Answer

 
discard

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

Browse other questions tagged or ask your own question.