Examples¶
This page summarizes the built-in gallery and highlights the principal numerical patterns across the current experiment set.
Gallery Overview¶
The built-in gallery currently contains twelve examples. These include directional payoffs, symmetric payoffs, widened marginals, a quadratic case whose interval nearly collapses in the present discretization, and causal multi-period studies.

Related files:
Summary Table¶
| Example | Lower | Upper | Width | Smallest eps value |
|---|---|---|---|---|
| Uniform absolute spread | 0.6087 | 0.9975 | 0.3888 | 0.9538 |
| Call on spread | 0.1894 | 0.3841 | 0.1947 | 0.3469 |
| Put on spread | 0.4394 | 0.6341 | 0.1947 | 0.5969 |
| Quadratic spread | 0.9969 | 0.9969 | 0.0000 | 0.9969 |
| Centered spread straddle | 0.6739 | 1.0894 | 0.4155 | 1.0289 |
| Centered call on spread | 0.0869 | 0.2947 | 0.2078 | 0.2462 |
| Wide absolute spread | 0.8754 | 1.3101 | 0.4347 | 1.2487 |
| Wide put on spread | 0.7133 | 0.9382 | 0.2249 | 0.8868 |
| Broad spread straddle | 0.6217 | 1.0194 | 0.3977 | 0.9587 |
| Causal absolute spread T3 | see generated gallery | see generated gallery | see generated gallery | see generated gallery |
| Causal call T4 | see generated gallery | see generated gallery | see generated gallery | see generated gallery |
| Causal convergence study | see generated gallery | see generated gallery | see generated gallery | n/a |
Reading Conventions¶
The examples are interpreted according to the following conventions:
- the exact LP bounds are the benchmark
- interval width is used as a measure of residual model uncertainty under the martingale restriction
- the smallest shipped
epsvalue is treated as an approximation diagnostic rather than as a substitute for the LP benchmark - structural diagnostics and convergence diagnostics are used together
Selected Comparisons¶
Reference Absolute Spread¶
S1 ~ Uniform[1, 3]S2 ~ Uniform[0, 4]- payoff
|S2 - S1| - exact interval
[0.6087, 0.9975]
This remains the baseline experiment for the repository. The interval is substantial, and the unrestricted countermonotone benchmark is materially larger than the exact martingale upper value.


Directional Spread Comparison¶
The call-on-spread and put-on-spread examples provide a useful directional comparison on the same support pair.
- call on spread: interval
[0.1894, 0.3841] - put on spread: interval
[0.4394, 0.6341]
The widths coincide in the current discretization, but the interval levels differ. This separates directional asymmetry from overall interval scale.


Centered Symmetric System¶
The centered straddle and centered call examples move to S1 ~ Uniform[-1, 1] and S2 ~ Uniform[-2, 2].
- centered straddle: interval
[0.6739, 1.0894] - centered call: interval
[0.0869, 0.2947]
These examples are useful when geometric symmetry is desirable without eliminating nontrivial pricing intervals.


Wider-Marginal Regime¶
The wide absolute-spread and wide put-on-spread examples increase the variance of the second marginal.
- wide absolute spread: interval
[0.8754, 1.3101] - wide put on spread: interval
[0.7133, 0.9382]
These examples show how increased dispersion in the second marginal broadens or shifts the robust pricing range.


Broad Straddle¶
The broad spread straddle retains the original support pair but changes the payoff to |(S2 - S1) - 0.25|.
- lower bound
0.6217 - upper bound
1.0194 - interval width
0.3977
This example complements the original absolute-spread benchmark by introducing a symmetric payoff around a nonzero strike.


Causal Multi-Period Examples¶
The causal gallery entries use chains of marginals rather than a single two-period pair.
- causal absolute spread T3: three uniform marginals widening from
[1, 3]to[0, 4] - causal call T4: four marginals with an adjacent call-on-spread payoff
- causal convergence study: selected
Tvalues for an additive absolute-spread time-step study
These entries write causal artifact names such as:
causal_transport_chain.pngcausal_summary.jsoncausal_experiment_report.mdcontinuous_limit.pngfor convergence studies
Diagnostic Figures¶
Useful diagnostics include:
uniform_abs_spread/regularization_path.pngwide_abs/regularization_path.pngbroad_straddle/regularization_path.pnguniform_abs_spread/structural_diagnostics.pngcentered_straddle/structural_diagnostics.pngwide_put/structural_diagnostics.png
The structural diagnostics combine marginal profiles, conditional dispersion, and convex-order call gaps in a single figure.
Per-Example Reports¶
Each example directory includes:
experiment_report.mdandsummary.jsonfor two-period entriescausal_experiment_report.mdandcausal_summary.jsonfor causal entriescontinuous_summary.jsonfor convergence entries- matching exact, regularization, causal, or continuous-limit plots
These files are linked from the generated Gallery Casebook.