What Does Umbrella Sampling Actually Sample?

Created on July 14, 2026 by Sungsoo Ahn , last updated on July 14, 2026

2026   ·   molecular-dynamics   umbrella-sampling   free-energy   wham   kups   ·   science

Note: This is an early draft page for the executable kUPS MD tutorial series. It is intentionally hidden from site navigation while the simulations, notebooks, figures, and review artifacts mature. This post follows the free-energy-estimator discussion by asking what biased umbrella windows actually sample, how adjacent overlap controls reconstruction, and how sparse windows can fail even with many samples. Corrections and replication issues should be tracked in sungsoo-ahn/kups-md-tutorials.

Introduction

Umbrella sampling does not sample the target PMF directly. Each window samples a biased ensemble. The unbiased PMF is reconstructed later, and that reconstruction only works when neighboring biased ensembles overlap enough to form a connected bridge across the collective variable.

For ML researchers working with MLIPs, this is the useful shift: window placement is a statistical design problem. A large trajectory in each window is not enough if the windows do not exchange probability mass through their tails. WHAM and MBAR formalize this reweighting problem across biased or intermediate states (Torrie & Valleau, 1977; Kumar et al., 1992; Shirts & Chodera, 2008).

This draft demonstrates the executable slice of the tenth tutorial with a known one-dimensional double-well PMF. Dense and sparse umbrella protocols are run against the same answer key, so the diagnostic can isolate overlap and window placement from physical-model error.

What Is Being Sampled?

The current diagnostic compares two harmonic umbrella protocols on the same double-well coordinate:

Protocol Windows Minimum adjacent overlap Barrier error
dense_windows 9 0.3552 0.0106
sparse_windows 4 0.0003 -0.2554

Both protocols draw many samples from every biased window. The difference is whether those biased samples connect neighboring regions of the collective variable.

What Should The Diagnostic Show?

The full run compares the known PMF to dense and sparse WHAM-style reconstructions. It also records adjacent histogram overlap and per-window sampling means. The sparse protocol intentionally skips the bridge through the barrier region, making the reconstruction less reliable even though every window has local support.

Umbrella-sampling diagnostics for the committed full profile. Dense windows maintain adjacent overlap and recover the known PMF, while sparse windows leave a near-zero-overlap bridge and bias the reconstructed barrier downward.

Reproduction

The current executable path is:

git clone https://github.com/sungsoo-ahn/kups-md-tutorials
cd kups-md-tutorials
uv sync
uv run kups-tutorial run 10 --profile smoke
uv run kups-tutorial verify 10 --profile smoke
uv run kups-tutorial run 10 --profile full
uv run kups-tutorial verify 10 --profile full
uv run jupyter execute notebooks/post-10-umbrella-sampling.ipynb --inplace

The notebook is deliberately not the implementation source. It imports the configuration loader, umbrella diagnostics, and figure generator from src/kups_md_tutorials/.

Current Status

This page is not the final article. The implemented pieces are:

  • smoke and full controlled umbrella-sampling workflows
  • committed compact summaries, PMF curves, and window-overlap outputs
  • executable notebook
  • generated SVG/PNG figure and snapshot review
  • self-review note covering code, science, notebook, and figure feedback

The missing pieces are:

  • final 3,500-10,000-word article prose
  • rendered desktop and mobile page snapshots
  • production MD context and final uncertainty diagnostics
  • final citation pass

The rule for this post is that umbrella sampling is only as trustworthy as the biased ensembles that connect the coordinate. Window placement, overlap, and replica consistency are part of the result, not optional diagnostics.

References

  • Torrie, G. M. & Valleau, J. P. (1977). Nonphysical sampling distributions in Monte Carlo free-energy estimation: umbrella sampling. Journal of Computational Physics, 23, 187-199. ↩
  • Kumar, S., Rosenberg, J. M., Bouzida, D., Swendsen, R. H. & Kollman, P. A. (1992). The weighted histogram analysis method for free-energy calculations on biomolecules. Journal of Computational Chemistry, 13, 1011-1021. ↩
  • Shirts, M. R. & Chodera, J. D. (2008). Statistically optimal analysis of samples from multiple equilibrium states. The Journal of Chemical Physics, 129, 124105. ↩