Likelihood function in multivariate analysis ( $H_0:\mu\mu^t=1$)

35 Views Asked by At

I am trying to compute the likelihood function for the test: $H_0:\mu\mu^t=1$ where my sample is of one element ($n=1$) of a $N_p(\mu,\sigma^2I)$ being $\sigma^2$ known.

So I should minimize $L(x,\mu,\Sigma)$ having in mind that $\mu\mu^t=1$.

Following the proof of the contrast of linear restrictions I have done as following:

Consider the function $a^+=a-n\lambda'(\mu\mu^t-1)$ where $\lambda$ is a Lagrange multiplicator, now differentiating with respect to $\mu$ I got:

$\dfrac{\partial{a^+}}{\partial{\mu}}=-2n\lambda'\mu+n(\bar{x}-\mu)\Sigma^{-1}=0$ this implies that: $\bar{x}-\mu=2n\lambda\mu$ $(1)$.

Using that $\mu\mu^t=1$ we get:

$(\bar{x}-\mu)\mu^t=\bar{x}\mu^t-1=2\lambda\mu\Sigma\mu^t \Longrightarrow \lambda=(\bar{x}\mu^t-1)(\mu\Sigma\mu^t)'$

Now, substituting $\lambda$ in $(1)$:

$\hat{\mu}=\bar{x}-2(\bar{x}\mu^t-1)(\mu^t\Sigma\mu)^{-1}\mu\Sigma$

Are these steps correct? I have had no lessons about Lagrange multiplicators so I am afraid I have done something wrong as the final result is not as compact as I thought it would be.

Any help would be appreciated