Find your next learning step
A polymer transition is often a probability distribution over plausible boundaries, not a single line that every replicate must obey.
Open the chapter contents · Polymer Science root · Gold Standard of Education · How X Works
Chapter contents
Choose the route closest to the question, or read from Chapter 1 for the complete argument. The chapters remain open on the page; these grouped routes are an index.
Route 1 · Define posterior transitions
Route 2 · Work the angular ledger
Route 3 · Align and transfer replicates
Route 4 · Check limits and evidence
Route 5 · Ask and continue
CHAPTER 1 OF 12 · DEFINE POSTERIOR TRANSITIONS
1. A change point should have a probability, not only a location
A polymer sweep often contains a transition zone rather than a perfectly sharp boundary. A single fitted change point compresses that uncertainty into one index. A Bayesian model instead assigns posterior probability to the number and location of changes, segment parameters and, when modelled, continuity across the boundary.
This matters when a physical interpretation depends on whether the transition occurs before or after yield, reversal or recovery. The honest result may be a range of plausible boundaries rather than one decisive cut.
CHAPTER 2 OF 12 · DEFINE POSTERIOR TRANSITIONS
2. Segments, priors and likelihoods
Divide an ordered experiment into segments within which a chosen parameter model is stable. A prior describes how many segments are plausible and how long they tend to persist. The likelihood describes measurements inside each segment, including noise and any covariance structure.
The posterior combines those pieces. It does not make prior assumptions disappear; it shows how strongly the data update them. Record the prior on segment length, the within-segment model and whether neighbouring segment parameters are independent or linked.
CHAPTER 3 OF 12 · WORK THE ANGULAR LEDGER
3. An original angular-path example
Suppose a dominant recovery subspace is represented by its angle to the loading axis. Six strain windows produce the following replicate summaries.
| Window | Replicate A angle | Replicate B angle | Replicate C angle | Median angle |
|---|---|---|---|---|
| 1 | 8° | 9° | 7° | 8° |
| 2 | 10° | 11° | 9° | 10° |
| 3 | 15° | 19° | 14° | 15° |
| 4 | 31° | 25° | 34° | 31° |
| 5 | 39° | 37° | 42° | 39° |
| 6 | 43° | 46° | 44° | 44° |
A step model may put the change between windows 3 and 4. A continuous-transition model may spread evidence across windows 2 to 5. Those answers describe different scientific hypotheses.
CHAPTER 4 OF 12 · WORK THE ANGULAR LEDGER
4. Posterior mass over competing boundaries
For a simple two-segment model, suppose the posterior probability of a boundary after windows 2, 3, 4 and 5 is 0.08, 0.47, 0.36 and 0.09. Reporting only the maximum at 3 hides almost half the mass elsewhere.
| Candidate statement | Posterior support | Defensible wording |
|---|---|---|
| Boundary after window 3 | 0.47 | Most supported single cut, but uncertain |
| Boundary after window 4 | 0.36 | Serious alternative |
| Boundary in the 3–4 region | 0.83 | Stronger interval statement |
| Exact boundary known | Not supported | Do not claim |
The numbers are illustrative, not an empirical polymer result. Their job is to show how uncertainty changes the sentence a researcher can defend.
CHAPTER 5 OF 12 · WORK THE ANGULAR LEDGER
5. Transitional regimes instead of forced steps
A transitional regime may have its own slope, variance or relaxation spectrum. Compare a two-segment step model with a three-regime model: stable, transition and stable. Posterior model probabilities, predictive checks and residual structure can show whether the middle regime earns its complexity.
Do not call every smooth curve a sequence of phases. A change-point model is a useful compression only when its segments correspond to reproducible behaviour and the residuals do not retain an obvious transition pattern.
CHAPTER 6 OF 12 · ALIGN AND TRANSFER REPLICATES
6. Aligning replicates without erasing disagreement
Replicates rarely change at the same sampled index. A hierarchical model can give each replicate its own boundary while sharing a population distribution. This separates true between-replicate timing variation from measurement noise.
Alignment should use the experimental coordinate—strain, waiting time, cycle count or temperature—not arbitrary row number. Keep unmatched windows visible. Forcing every replicate onto the same cut can create a false consensus similar to averaging incompatible subspaces.
CHAPTER 7 OF 12 · ALIGN AND TRANSFER REPLICATES
7. Changed-context transfer: up-ramp and down-ramp
Now fit the same material on an increasing ramp and a decreasing ramp. If the up-ramp transition is concentrated near 0.62 strain while the down-ramp transition lies near 0.48, the difference may indicate hysteresis. Compare posterior distributions of the two boundaries rather than subtracting two modal estimates.
The transfer check is to repeat the analysis under alternative priors, sampling grids and transition models. Connect the result to amplitude sweeps and yield onset and to the earlier change-point and hysteretic subspace path.
CHAPTER 8 OF 12 · ALIGN AND TRANSFER REPLICATES
8. Posterior predictive checks
Simulate angular or spectral paths from the fitted posterior and compare them with held-out replicates. Check transition width, within-regime variance, reversal location and the frequency of apparent extra changes. A model that locates a boundary cleanly but cannot reproduce the surrounding path is not adequate.
Report sensitivity to coordinate spacing. Dense sampling in one region gives that region more chances to host a cut unless the likelihood and prior account for the design.
CHAPTER 9 OF 12 · CHECK LIMITS AND EVIDENCE
9. Limits and common misconceptions
A posterior interval is conditional on the model family, prior and likelihood. It is not a model-free confidence guarantee. A narrow posterior can be misleading when correlated errors are treated as independent or when the transition model is misspecified.
Bayesian does not mean subjective guesswork, nor does it mean the data choose every assumption. Priors should be explicit, scientifically interpretable and stress-tested. More candidate change points also increase the opportunity to fit noise.
CHAPTER 10 OF 12 · CHECK LIMITS AND EVIDENCE
10. Practical evidence checklist
Preserve raw paths, replicate identifiers, coordinate values, preprocessing, subspace-estimation choices and uncertainty in each angle or projector. Save the prior on number and spacing of changes, segment likelihood, posterior sampler or recursion and convergence diagnostics.
Report posterior mass for competing boundaries, credible transition regions, posterior predictive checks and prior sensitivity. Refit after leaving out each replicate. If alignment changes the conclusion, show both aligned and unaligned results.
CHAPTER 11 OF 12 · ASK AND CONTINUE
11. Frequently asked questions
Must segment parameters be independent?
No. Independent segments simplify exact recursions, while dependent-segment models can represent continuity or gradual drift. Use the model that matches the scientific question.
Can a Bayesian model prove a new phase exists?
No. It compares specified models using specified data. Physical phase claims require independent evidence and reproducibility.
Why not average replicate change points?
An average hides multimodality and uncertainty. A hierarchical posterior can show the population centre, spread and replicate-specific evidence together.
CHAPTER 12 OF 12 · ASK AND CONTINUE
12. Sources and next reading
Fearnhead gives an exact recursive framework for Bayesian inference in multiple change-point models. Fearnhead and Liu extend the framework to dependence across segments, including continuous or discontinuous piecewise curves. Grassmannian geometry for subspace-valued measurements is developed by Edelman, Arias and Smith in the geometry of algorithms with orthogonality constraints. The angular and posterior ledgers above are original teaching examples.
Continue through the Polymer Science root, change-point paths, mixture subspaces and Grassmannian means. The Gold Standard of Education and How X Works provide the wider route.
Top · Contents · Previous · How X Works