Scalable Bayesian Inference of Large Simulations via Asynchronous Prefetching Multilevel Delayed Acceptance

This repository contains the code base accompanying the work published in Scalable Bayesian Inference of Large Simulations via Asynchronous Prefetching Multilevel Delayed Acceptance. The code can be used to run parallelized MLDA sampling runs for a Bayesian posterior governed by a PDE model from geophysics. For a methodological introduction to the underlying research, we refer the reader to the mentioned publication.

The code base is divided into four subdirectories, each of them containing separate README files for how to use the respective components:
1. mtmlda: Parallelized implementation of the MLDA algorithm, based on asynchronous prefetching.
2. surrogate: Greedy, asynchronous surrogate based on Gaussian process regression, used as an additional level in the MLDA hierarchy.
3. seisbridge: Implementation of a SLURM load balancer for HPC architectures, used for parallel requests of SeisSol simulations.
4. ridgecrest: Run files for SeisSol simulations of the Ridgecrest earthquake on the Frontera supercomputer.