Reversible Jump Markov Chain Monte Carlo

Reversible-Jump Markov-Chain Monte Carlo

At each iteration, one type of proposal is selected in the list below.

• Brith proposal: add a layer interface
• Death proposal: delete a layer interface
• Move proposal: Move a layer interface
• Perturbation proposal (Vs): Perturb Vs perturbation
• Perturbation proposal (Vp): Perturb Vp perturbation (optional)
• Perturbation proposal (sigma): Perturb data noise standard deviation

Each of these proposal types is selected with an equal probability.

To enhance the performance of MCMC sampling, one can tweak some tuning parameters through the input parameter file. These tuning parameters include iteration number, a standard deviation for proposed perturbation amount, and so on.

Note

When adding a layer interface via a birth proposal, the layer velocities and depths are randomly determined with their prior probability. This treatment makes computing acceptance criteria for the proposed model much easier. See Dosso et al. 2014 for further information.