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

I need to find the expectation of $\ln (x-\epsilon) $ with respect to a probability distribution $\mathbb{P}(x)$. A direct evaluation seems very difficult as the expression for $ \mathbb{P}(x)$ is very complicated. What I have however, is a closed form expression for the Stieltjes transform of $\mathbb{P}(x)$, $\mathbf{m}(\theta)$. I have derived an expression for this expectation. My questions are as follows

  1. Is the solution correct ?
  2. Any arguments that need further reasoning ?
  3. Is there a simpler way to do it ?
  4. Is there a way I can get rid of the $\epsilon$ ? using this technique I can't seem to get shake it off.

Here is the solution:

For any probability measure $ \\mathbb{P}(x) $ and a constant $ 0 < \epsilon < x $ the following identity is true \begin{equation} \int \ln (x- \epsilon)~ d\\mathbb{P}(x)= \ln \epsilon +\dfrac{1}{c} \ln (1+c\mathbf{m}(\epsilon))+\ln(-(1+\underline{\mathbf{m}}(\epsilon))) +\epsilon\mathbf{m}(\epsilon)\underline{\mathbf{m}}(\epsilon) \end{equation}

Define the Stieltjes transform of the probability measure $ \\mathbb{P} (x) $ as follows \begin{equation} \mathbf{m}(\theta)=\int \dfrac{1}{x-\theta} d\\mathbb{P}(x); ~ \theta \in \mathbb{C}^+ \end{equation} For $ 0 < c <1 $ define the function $ \underline{\mathbf{m}}(\theta)= c\mathbf{m}(\theta)-\dfrac{1-c}{\theta} $. Observe that \begin{equation} \mathbf{m}(\theta)=-\dfrac{1}{\theta(1+\mathbf{\underline{m}} (\theta)) } \end{equation} \begin{equation} \mathbf{\underline{m}} (\theta)=-\dfrac{1}{\theta(1+c\mathbf{m(\theta)})} \end{equation}

First observe that $ \ln(x-\epsilon) $ can be written as follows \begin{equation} \ln(x-\epsilon)= \ln(\epsilon) -i\pi + \int\limits_{\epsilon}^{\infty} \dfrac{1}{\theta} + \dfrac{1}{x-\theta} d\theta \end{equation} Integrating with respect to the probability measure $ \\mathbb{P}(x) $, we get \begin{equation} \int \ln(x-\epsilon)d\\mathbb{P}(x)= \ln(\epsilon) -i\pi + \int \int\limits_{\epsilon}^{\infty} \dfrac{1}{\theta} + \dfrac{1}{x-\theta} d\theta d\\mathbb{P}(x) \end{equation} On the R.H.S, applying Fubini's theorem and using the definition of $ \mathbf{m}(\theta)=\int \dfrac{1}{x-\theta}d\\mathbb{P}(x) $, we get \begin{equation} =\ln \epsilon -i\pi + \int\limits_{\epsilon}^{\infty} \dfrac{1}{\theta}+\mathbf{m}(\theta) d\theta \end{equation} Using the relation $ \underline{\mathbf{m}}(\theta)= c\mathbf{m}(\theta)-\dfrac{1-c}{\theta} $ we can write $ \dfrac{1+\theta\underline{\mathbf{m}}(\theta)}{\theta} $ as $ -\mathbf{m}(\theta)\underline{\mathbf{m}}(\theta) $. Hence we get \begin{equation} =\ln \epsilon -i\pi - \int\limits_{\epsilon}^{\infty} \mathbf{m}(\theta)\underline{\mathbf{m}}(\theta) d\theta \end{equation} Consider $ \Lambda(\theta,\mathbf{m}(\theta),\underline{\mathbf{m}}(\theta))=\dfrac{1}{c}\ln(1+c\mathbf{m}(\theta))+\ln(1+\underline{\mathbf{m}}(\theta))+\mathbf{m}(\theta)\underline{\mathbf{m}}(\theta)\theta $. It is not hard to see that the derivative of $ \Lambda $ w.r.t $ \theta $ is $ \mathbf{m}\underline{\mathbf{m}} $. Using this and plugging in the limits $ \epsilon, ~\infty $ and simplifying we get. \begin{equation} \int\ln(x-\epsilon)d\\mathbb{P}(x)= \ln(\epsilon) +\dfrac{1}{c} \ln(1+c\mathbf{m}(\epsilon))+\ln(-(1+\underline{\mathbf{m}}(\epsilon)))+\epsilon \mathbf{m}(\epsilon)\underline{\mathbf{m}}(\epsilon) \end{equation}

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.