Markov chain Monte Carlo on the active variables of a PDE inverse problem: chains, subspace and grid
Year: 2026
Type: dataset
Abstract: Markov chains for a Bayesian inverse problem with a 100-parameter Karhunen-Loeve log permeability field and an elliptic PDE, following P. G. Constantine, C. Kent and T. Bui-Thanh, "Accelerating Markov chain Monte Carlo with active subspaces", SIAM J. Sci. Comput. 38(5):A2779-A2805, 2016, doi:10.1137/15M1042127. pde_mcmc_chains.npz (load with numpy.load, no pickle) holds the synthetic data, the misfit gradients at 1000 prior samples and their active subspace, 20 random-walk Metropolis chains and 4 + 4 active-variable chains (reductions), and a grid of the misfit on the active variables. kl-short.npz is the Karhunen-Loeve basis the model uses. generate_data.py in the repository regenerates the chains from saved seeds. These files are deposited as the full-size archive of the data set uq_benchmarks/PDE_Inverse_Problem_MCMC in the as-data-sets repository (https://github.com/paulcon/as-data-sets/tree/main/uq_benchmarks/PDE_Inverse_Problem_MCMC), a collection of Jupyter notebooks that apply active subspaces (parameter-space dimension reduction) to science and engineering models. The notebook there analyzes these data; its DATA.md documents every file, column, unit and unknown. The repository keeps each file under 10 MB, so it ships the same arrays split into files under 10 MB. License: CC BY 4.0. Cite Constantine, Kent and Bui-Thanh (2016).
Author: Constantine, Paul G.
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225860584
