I would like to now what is the relation between the Hessian Matrix calculated at the maximum point of the log(likelihood) function, $H_{f}$ and the Hessian Matrix calculated at the minimum point of the -log(likelihood) function, $H_{-f}$.

The first should be definite negative and the second should be definite positive. But, is it possible to calculate the first hessian matrix from the second one?

It should be $H_{f}=-H_{-f}$, but I am not sure. Is there any proof of it? Thank you


Let $\ell(\theta)$, for $\theta\in\Theta$, be the log-likelihood function. Then the maximum likelihood estimate is defined as

$$\hat\theta = \text{arg}\,\underset{\theta\in\Theta}{\min}-\ell(\theta)\\ \,\, = \text{arg}\, \underset{\theta\in\Theta}{\max}\ell(\theta)$$

Now, by the properties of the differentiation operator

$$\frac{d^2-\ell(\theta)}{d\theta d\theta^T} = -\frac{d^2\ell(\theta)}{d\theta d\theta^T}, $$

which holds for any $\theta$. In particular, for $\hat\theta$ we have that

$$ \frac{d^2-\ell(\theta)}{d\theta d\theta^T}|_{\theta=\hat\theta} = -\frac{d^2\ell(\theta)}{d\theta d\theta^T}\vert_{\theta=\hat\theta}. $$


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.