Of Notes
Bayesian inference with a Cauchy likelihood via contour integration
Table of Contents
Problem
Setup
Suppose we observe measurements generated by a Cauchy distribution with an unknown location parameter
The likelihood of one observation is:
We place a Cauchy prior on
We observe a single measurement:
Our goal is to perform Bayesian inference and predict a future observation ^*$.
Part A: Construct the posterior
Using Bayes' theorem,
- Substitute the likelihood and prior.
- Express the posterior as a rational function of
. - Identify the normalisation constant (D)$.
Do not evaluate the integral yet.
Part B: Compute the evidence using complex analysis
The evidence requires evaluating:
- Extend the real variable
to ∈ \mathbb{C}$. - Identify the poles of the integrand.
- Choose a semicircular contour in the upper half-plane.
- Apply the residue theorem to evaluate the real integral.
- Use your result to write the fully normalised posterior.
Part C: Posterior predictive inference
Suppose we want to evaluate the predictive density at ^*=-1$.
We use:
- Substitute your posterior from Part B.
- Express the resulting integral as a rational function.
- Identify its poles in the complex plane.
- Use the residue theorem again to evaluate it.
Part D: Interpretation
Explain why integrating over