Neon Mie Inference

Source files:

• examples/neon-mie-lgpmd/
• Upstream LGPMD tutorial
• Associated article

Goal

This notebook-first example uses real liquid-neon radial distribution function (RDF) data from the LGPMD tutorial_v2.0 directory. It infers the epsilon, lambda, and sigma parameters of a lambda-6 Mie potential from the experimental RDF.

No molecular dynamics run is required. The committed upstream files include the simulation training set, held-out validation set, and experimental RDF. The fitting and learning cells use CUDA by default.

Run

Install the optional notebook tools once:

pip install "bfflearn[notebook]"

cd examples/neon-mie-lgpmd
jupyter lab

Open neon-mie-inference.ipynb and execute it from top to bottom. The notebook demonstrates the complete data-facing workflow:

1. load and interpolate upstream RDF data;
2. retain rows inside the declared physical inference domain;
3. construct and write a BFF QoIDataset;
4. build and validate a local Gaussian-process surrogate with a notebook-local Mie PMF mean;
5. estimate the effective number of resolved RDF features;
6. infer the Mie potential parameters and RDF discrepancy;
7. report the inferred Mie parameters in physical units;
8. plot the posterior, QoI-attributed marginals, and inferred RDF.

The RDF discrepancy is learned because the source data do not provide an experimental uncertainty. The callable PMF mean stays inside the notebook because it is specific to this Mie-potential example.

The copied-file inventory, source commit, article citation, and upstream license are recorded in SOURCE.md. All notebook-generated files are written under generated/. The notebook writes generated/qoi-marginals.pdf; with one RDF QoI its posterior filling has a single attribution color.