-2
$\begingroup$

I am stuck with performing this integration. Can anyone help please?

$$ \mathbb{V}\mathrm{ar}[M_j] \stackrel{\text{def}}{=} \mathbb{E}[M_j^2] = e_j^T \left( \int_{\mathbb{R}^{n+1}} m^2 \pi^D_{\text{prior}}dm \right) e_j \stackrel{\text{def}}{=} \gamma^2 e_j^T \left(L_D^T L_D\right)^{-1} e_j $$

The $e_j$ is called: canonical basis $$ \pi^D_{\text{prior}}(m) \propto \exp \left( - \frac{1}{2 \gamma^2} || L_D m ||^2 \right) $$

$\endgroup$
  • 2
    $\begingroup$ Why do you need to integrate it? $\endgroup$ – Glen_b -Reinstate Monica Nov 5 '16 at 11:02
  • $\begingroup$ I want to estimate the variance as shown in Fig.1 $\endgroup$ – MBM Nov 5 '16 at 11:31
  • $\begingroup$ but then see that the term with the integral in it is followed by an equality (with $\text{def}$ above it, presumably indicating a definition). So it seems there's nothing to integrate, you just use the formula at the end. Someone integrated it for you. $\endgroup$ – Glen_b -Reinstate Monica Nov 5 '16 at 11:57
  • $\begingroup$ Thanks @Glen_b. I know that it is done. I am asking about the middle steps between?!! $\endgroup$ – MBM Nov 5 '16 at 12:23
  • $\begingroup$ So it's not that you want to estimate the variance (see your previous response to me) -- for which you could just use the formula -- but that you want to derive the result. Clarifying that issue (what you ultimately wanted to achieve) is why I asked. $\endgroup$ – Glen_b -Reinstate Monica Nov 6 '16 at 0:25
0
$\begingroup$

Presumably by $m^2$ you mean $mm^\top$ since $m$ is a vector. The integral computes the second moment matrix of a zero-mean multivariate normal distribution whose covariance is $\gamma^2 (L^\top_D L_D)^{-1}$. Since it is zero-mean, the second moment matrix is the covariance matrix. Write

$$\pi^D_{\mathrm{prior}}(m) \propto \exp\left( -\frac{1}{2} m^\top \left[ \left( \frac{\left[ L_d^\top L_D \right]}{\gamma^{2}} \right)^{-1} \right]^{-1} m \right).$$

Just compare this to the density given on the Wikipedia page of the multivariate normal.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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