0
$\begingroup$

I have a question about the derivative of a certain parameter integral. To fix notation, let $\phi_{\mu,\sigma}(x)=\frac{1}{\sqrt{2\pi}\sigma}\exp{\left(-\frac{(x-\mu)^2}{2 \sigma^2}\right)}$ denote the density function of a Gaussian distribution with mean $\mu\in\mathrm{R}$ and standard deviation $\sigma>0$. I am interested in the derivative \begin{equation} \frac{\partial}{\partial \sigma}\int_{0}^{\infty}\exp{(x)}\phi_{\mu,\sigma}(x)\,\mathrm{d}x. \end{equation} In particular, is this derivative always positive? If so, how can I show that (rigorously)? Any help, suggestions, or the like, are highly welcome. Many thanks in advance!

$\endgroup$
1
$\begingroup$

There is a general result: Let $(\Omega,\mathcal{A},\mathbb{P})$ a measure space. Let $u:(a,b) \times \Omega \to \mathbb{R}$ such that

  1. $x \mapsto u(\sigma,x) \in L^1(\mathbb{P})$ for all $\sigma \in (a,b)$
  2. $\sigma \mapsto u(\sigma,x)$ is differentiable for all $x \in \Omega$
  3. $\exists w \in L^1(\mathbb{P}) \, \forall (\sigma,x) \in (a,b) \times \Omega: |\partial_\sigma u(\sigma,x)| \leq w(x)$

Then $V(\sigma) := \int u(\sigma,x) \, d\mathbb{P}(x)$ is differentiable, $\partial_\sigma V(\sigma) = \int \partial_\sigma u(\sigma,x) \, d\mathbb{P}(x)$.

So in this case we choose $$u(\sigma,x) := e^x \cdot \varphi_{\mu,\sigma}(x).$$ We have to check the last property (the other ones are clearly fulfilled). By calculating the derivative one can show that

$$|\partial_\sigma u(\sigma,x)| \leq e^x \cdot \exp \left(- \frac{(x-\mu)^2}{a^2} \right) \cdot \left( \frac{1}{\sqrt{2\pi \cdot a^4}} + \frac{2}{\sqrt{2\pi a^2}} \cdot \frac{(x-\mu)^2}{a^3} \right)=:w(x) \in L^1$$

for all $\sigma \in (a,b)$ where $0<a<b$. Thus we can apply the theorem and obtain a formula for the derivative. By using the moments of the Gaussian distribution it is possible to calculate the integral.

$\endgroup$
  • $\begingroup$ thanks for the help! Following your suggestions, I managed to find the desired result. $\endgroup$ – s_2 Nov 17 '12 at 14:47

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.