The expected value of $1/X$ for $X\sim\mathcal{N}(\mu,\sigma^{2})$ is formally written as \begin{equation} \mathbb{E}[X^{-1}]=\frac{1}{\sqrt{2\pi}\sigma}\int_{-\infty}^{\infty}\frac{1}{x}\exp\left(-\frac{(x-\mu)^{2}}{2\sigma^{2}}\right)\,\mathrm{d}x, \end{equation} which obviously does not converge. I am interested in the Cauchy principal value of this integral, i.e. \begin{equation} \mathbb{E}[X^{-1}]=\frac{1}{\sqrt{2\pi}\sigma}\lim_{\epsilon\nearrow 0}\left(\int_{-\infty}^{-\epsilon}+\int_\epsilon^\infty\right)\frac{1}{x}\exp\left(-\frac{(x-\mu)^{2}}{2\sigma^{2}}\right)\,\mathrm{d}x. \end{equation}

I would have thought that this has been worked out a long time ago but could not find a solution anywhere. Does anyone have a source detailing a solution?


After much looking I found a solution and accompanying proof:

Quenouille, M.H., 1956. Notes on bias in estimation. Biometrika 43, 353–360

For $X\sim\mathcal{N}(\mu,\sigma^{2})$, the Cauchy principal value of the mean is \begin{equation} \mathrm{PV}(\mathbb{E}[X^{-1}])=\frac{\sqrt{2}}{\sigma}\,\mathcal{D}\left(\frac{\mu}{\sqrt{2}\sigma}\right), \end{equation}

where \begin{equation} \mathcal{D}(z)=e^{-z^{2}}\int_{0}^{z}e^{t^{2}}\,\mathrm{d}t, \end{equation} is the Dawson integral. For those interested, I also found a paper that derives $\mathrm{PV}(\mathbb{E}[X^{-1}])$ for the closely related skew-normal distribution:

The first negative moment in the sense of the Cauchy principal value (Chien-Yu Peng)

as well as the skew-t and generalized student-t:

The First Negative Moment of Skew-t and Generalized Student's t-Distributions in the Principal Value Sense (Chien-Yu Peng)


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.