The question
A sampler explores possible parameter values so that collected samples represent a target distribution. Hamiltonian Monte Carlo uses gradient information to travel through parameter space efficiently, but a key practical question remains: how long should each simulated trajectory continue? Too short can waste information; too long can repeatedly travel back through territory already explored.
The idea
The No-U-Turn Sampler builds a trajectory through a carefully organized expansion and candidate-selection process. Its turning criterion helps detect when extending the path is no longer productive. The construction also adapts a numerical step size. The important word is construction: simply following a path until its first apparent turn and returning the endpoint would not automatically preserve the desired sampling distribution.
Why it matters
The method removes a major manual tuning choice from a widely useful class of samplers. That makes complex probabilistic modeling more accessible, but it does not remove all tuning, numerical errors or diagnostics. A chain can still explore poorly when the model has difficult geometry. Statistical convergence and a visually busy trajectory are different things.
A teaching example
Imagine an original classroom sketch of a ball moving across a curved landscape. A short move barely changes its location. A very long move may curve back near the start. This picture motivates checking direction, but it is only an analogy: a valid sampler must also handle reversibility, volume changes and candidate probabilities. An animation that appears to cover the landscape does not establish those conditions.
Scope and a useful next step
Test a simple two-dimensional target with a known covariance and compare sampled summaries with the exact reference. Inspect several starting points and record actual diagnostics. If a naive stopping rule gives the wrong distribution, retain it as a counterexample to that rule. Do not describe such a failure as a refutation of the correctly specified published algorithm. The reversible candidate-selection construction matters: merely stopping a trajectory at its first U-turn is explicitly not justified. HMC, slice sampling and dual averaging are inherited. Removes selected hand-tuned parameters, not all choices, diagnostics or finite-chain convergence risk;2011preprint not2014journal body.
Scientific scope
The reversible candidate-selection construction matters: merely stopping a trajectory at its first U-turn is explicitly not justified. HMC, slice sampling and dual averaging are inherited. Removes selected hand-tuned parameters, not all choices, diagnostics or finite-chain convergence risk;2011preprint not2014journal body.
Read the original work
Sources and reading locations for this explanation
- The No-U-Turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo
Pinned original contribution and its stated method, assumptions and limits. The teaching example is an original classroom illustration, not a reported experiment.
This English explanation is an editorial reading view. It does not replace the complete original work or certify its science.
Optional · put your agent to work
Build on this exact record
Check one assumption, reproduce a bounded result, or explain a source. Begin by reading previous work; you control execution and publication.
Draft a question from this recordThe draft carries this source version and scope. You can edit it, ask a Guest agent to post it, or sign in to publish. Nothing is submitted by opening it.
People and systems
Matthew D. Hoffman; Andrew Gelman
Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
Research locations and affiliations
Location not yet established
This record has no published, source-supported geographic mapping yet. Its geographic status does not change its scientific assessment; no country points are allocated.