Skip to content

Root-Sum-Square Calculation

An accumulation model — instantiates Tolerance Stack Management

Combines independent contributors by the square root of the sum of their squared tolerances, giving a realistic statistical stack that is far tighter than the worst case because deviations rarely all align.

The worst case assumes every contributor conspires to deviate in the same harmful direction at once; Root-Sum-Square Calculation assumes they don't. When contributors vary independently, their deviations partly cancel, and the statistically expected combined spread is not the arithmetic sum of the tolerances but the square root of the sum of their squares. That single formula is the whole mechanism, and its defining consequence is quantitative generosity: the RSS stack is markedly smaller than the worst-case stack, which lets a designer either accept looser (cheaper) part tolerances for the same fit, or promise a tighter fit from the same parts. It is the default statistical accumulation model — the cheapest way to stop paying the worst-case tax on variation that will almost never all line up.

Example

A shaft carries a stack of five spacers whose combined length sets an end clearance that must land near a target with about ±0.20 mm of allowable play. Each spacer holds ±0.05 mm. A worst-case sum would demand a budget of 5 × 0.05 = ±0.25 mm — over the limit, implying the design fails or the spacers must be re-toleranced tighter and dearer. The Root-Sum-Square Calculation instead combines them as √(5 × 0.05²) ≈ ±0.11 mm, comfortably inside the ±0.20 mm limit. The same physical parts now pass, with margin to spare, on the reasoning that all five spacers reaching their extreme in the same direction simultaneously is vanishingly rare. RSS is also what lets the team run the logic backwards to allocate: given the ±0.20 mm limit and five equal contributors, each spacer can be granted ±0.089 mm rather than a stingy ±0.04 mm.

How it works

The method squares each contributor's tolerance, sums the squares, and takes the square root — a formula that follows directly from how the variances of independent random quantities add. Because variances (not tolerances) are what sum, a contributor's influence grows with the square of its band, so the largest tolerance dominates the RSS total far more than it dominates the worst-case sum. Run forward, it predicts the combined spread; run backward from a fixed budget, it allocates each contributor a statistically-justified share. What distinguishes it from the worst-case sibling is the independence-and-cancellation premise — it prices the fact that deviations are random and rarely aligned.

Tuning parameters

  • Distribution/coverage assumption — what fraction of production the tolerance band is taken to represent (e.g. a ±3σ band). The stated confidence of the RSS result rides entirely on this premise.
  • Shift/inflation factor — an optional correction that widens the pure RSS to allow for real processes running slightly off-center rather than perfectly centered. It hedges the optimism of textbook RSS.
  • Independence scope — which contributors are treated as independent; grouping genuinely correlated contributors and summing those worst-case protects the result from the formula's fatal assumption.
  • Allocation direction — whether contributors share the budget equally or weighted by cost-of-precision when RSS is run backward to allocate.
  • Inclusion of systematics — whether a known one-directional bias (thermal growth, wear) is added linearly on top of the RSS random part rather than buried inside it.

When it helps, and when it misleads

Its strength is realism at almost no cost: for independent, roughly-centered contributors it predicts a combined spread much closer to what production actually delivers than the worst case, freeing real tolerance budget — often the difference between a manufacturable design and an over-constrained one.

Its failure mode is its headline assumption. If contributors are correlated — machined on the same fixture, sharing a thermal environment, cut by one worn tool — their deviations align, RSS badly under-predicts the real spread, and the formula becomes exactly the "false precision when variation is correlated" that the archetype warns against. It also silently assumes centered processes; a shifted process makes the true stack wider than RSS says, which is why the 1.5σ-style shift correction exists.[1] The discipline is to defend the independence assumption explicitly before trusting the number, add known systematic biases linearly rather than folding them into the RSS, and fall back to worst-case where the stakes forbid optimism.

How it implements the components

  • accumulation_model — it is the statistical combination rule: independent contributors combine as the root of the sum of squared tolerances.
  • variation_budget — run in reverse, the same formula converts a global fit limit into a statistically-justified allowance per contributor, sizing the budget the allocation artifacts then record.

It does not identify which contributors dominate or check the built result — sensitivity ranking is done by Statistical Tolerance Analysis and Monte Carlo Stack Simulation, and the conservative bound by Worst-Case Stack Calculation.

Notes

RSS and worst-case are the two poles of the accumulation choice, and picking between them is a risk decision, not a math one: RSS is right when contributors are independent and the cost of a rare stack failure is tolerable; worst-case is right when alignment is plausible or a single failure is unacceptable. When independence is doubtful but you still want statistical realism, that is the cue to escalate to a simulation rather than trust the closed form.

References

[1] Real processes rarely stay perfectly centered, so textbook RSS (which assumes centering) can under-predict the true spread. A dynamic-mean-shift correction — widely applied as a nominal 1.5σ allowance in Six Sigma practice — inflates the RSS result to account for expected drift off-center. ↩