Statistical Methods

Preserve Sobol Sequence Balance in a Simulation Plan

Define the task

A Sobol sequence is a low-discrepancy design for numerical integration or simulation, not a sample of independent random observations. Define the integration dimension and transformation from the unit cube before generating points. The benefit depends on the integrand and construction; do not assume that replacing random numbers automatically improves every simulation. Preserve the engine configuration and software version so another analyst can reconstruct the design.

Carry out the check

Use power-of-two point counts through the documented balanced generation method. Avoid arbitrary thinning, dropping the first point or choosing an unrestricted count while claiming the same balance properties. If using randomized scrambling for uncertainty assessment, record independent replicate construction and summarize variation between replicate estimates rather than treating points within a sequence as independent. Check transformations near zero and one before applying an inverse distribution function, especially when an unscrambled sequence includes boundary points.

Worked hypothetical example

For a hypothetical integration plan with 2^10 points in each of eight independently scrambled replicates, there are 1,024 points per replicate and 8,192 evaluations overall. Compute eight replicate estimates before assessing between-replicate variation. The evaluation count alone does not supply an error bar or guarantee accuracy. Record convergence across planned sizes and a known-integral benchmark. Keep seed or generator details with the simulation record; a reproducible configuration is necessary but does not establish validity of the scientific model.

Method checked 19 November 2026 against Primary method documentation: Sobol.

Download the balanced simulation planning worksheet.

Related reading

Reading path: Analyze data. Guide 180 of 182. Choose any guide to suit your project.

Statistical Methods

Measure contiguous active runs in a binary sequence

A practical guide with worked examples: Label adjacent active observations and measure contiguous event runs.

Statistical Methods

Fit a nondecreasing sequence with isotonic regression

A worked guide to the assumptions and checks: Fit an ordered sequence under a monotonicity constraint.

Browse this reading pathTools for this stageGuides and downloads for this stage