Uncertainty Propagation Calculation¶
Calculation method — instantiates Noise-Bounded Measurement Interpretation
Carries the uncertainty of raw inputs through the formula that combines them, so a derived quantity inherits an honest error bar instead of acquiring fake precision on the way out.
Measurements rarely get used raw; they are divided, multiplied, and fed into formulas to compute something else — and the derived result is only as certain as the inputs it was built from. Uncertainty Propagation Calculation is the method that carries input uncertainties through the transform to the output, so a computed quantity arrives with a defensible uncertainty rather than a spuriously exact digit string. Its defining move is to treat the formula itself as an uncertainty amplifier or attenuator: dividing two nearly-equal noisy numbers can blow a small input error into a huge output error, while averaging many can shrink it. The calculation makes that transformation explicit, mapping how each input's uncertainty — scaled by how sensitively the output depends on it — combines into the output's.
Example¶
A plant meters natural-gas flow through an orifice plate, where flow rate is computed from a formula involving the square root of the differential pressure across the plate and the gas density. Both inputs are measured with their own uncertainties — the pressure transmitter to ±0.2%, the density to ±0.5%. A naïve flow computer reports "12.473 kg/s" to four figures and moves on. The propagation calculation instead asks how the flow output responds to each input: because flow depends on the square root of pressure, a 0.2% pressure error contributes only about 0.1% to the flow, while the density term passes through more directly. Combining the scaled contributions gives a flow uncertainty of roughly ±0.5%, so the honest result is "12.47 ± 0.06 kg/s" — the fourth figure was never real.
The payoff shows up in what depends on that flow figure. Custody transfer, emissions accounting, and billing all rest on it, and knowing it is ±0.5% rather than exact changes what can be claimed and how tightly two months' totals can be compared. The propagation also reveals which input to improve: tightening density metrology buys more than chasing the pressure sensor, because the formula passes density through more heavily.
How it works¶
- Map each input's sensitivity. For each input, determine how strongly the output moves when that input moves — the local slope of the transform (analytically, or numerically by perturbing it).
- Scale each uncertainty by its sensitivity. An input's contribution to output uncertainty is its own uncertainty multiplied by that sensitivity, so heavily-weighted inputs matter more.
- Combine the contributions. Independent contributions are summed in quadrature; correlated inputs are handled with their covariance so shared errors aren't miscounted.
- Report the derived uncertainty and its drivers. The output carries its combined uncertainty and a ranking of which input dominates it.
What distinguishes it is that it operates across a transformation: it is not about characterizing one measurement but about how uncertainty flows from inputs to a computed output, catching both dangerous amplifications and reassuring cancellations.
Tuning parameters¶
- Method — first-order (linear/delta-method) propagation versus a Monte Carlo simulation of the transform. Linear is fast and transparent but wrong for strong nonlinearity or wide inputs; Monte Carlo handles those at a compute cost.
- Correlation handling — whether inputs are treated as independent or with a covariance model. Assuming independence is simpler but understates uncertainty when inputs share a common error.
- Sensitivity evaluation point — where the local slopes are taken; for a curved transform the answer depends on operating point, and a poorly chosen one misstates the amplification.
- Truncation order — whether second-order terms are kept for strongly curved transforms. Keeping them improves accuracy near nonlinearity but adds work.
- Contributor reporting depth — whether only the total is returned or the per-input breakdown too. The breakdown costs nothing extra to expose and shows where to improve.
When it helps, and when it misleads¶
Its strength is preventing derived quantities from acquiring precision they never earned — the polished dashboard number computed from noisy inputs that looks authoritative because it came out of a formula. It flags the dangerous cases, especially the subtraction of nearly-equal quantities, where a tiny input error becomes a catastrophic output error, and it tells you which input to tighten to shrink the result.[n1]
It misleads when its assumptions quietly fail. First-order propagation assumes the transform is roughly linear over the input's uncertainty range; across a sharp nonlinearity or with a wide input, that approximation can badly under- or over-state the true output uncertainty. Assuming independence when inputs share a common-mode error (the same calibration bias in two terms) miscounts the combination. And, like any propagation, it only carries forward the input uncertainties it was given — an unaccounted input source stays invisible in the output. The guarding discipline is to check the linearity assumption against the input's spread, model correlations where inputs share a source, and switch to simulation when the transform is sharply curved rather than trusting a slope taken at one point.
How it implements the components¶
uncertainty_propagation_map— it is the propagation: it maps how each input's uncertainty, scaled by the output's sensitivity to it, flows through the transform into the derived result.uncertainty_budget— it produces the derived quantity's budget, with the combined output uncertainty and a ranking of which input contributes most.
It builds the budget of a computed output but does not itemize the raw contributors of a single measurement into a static noise_source_inventory — that is Measurement Uncertainty Budget Table, its nearest twin, whose combined figure this calculation consumes as an input — nor does it render the result as a per-value claim_precision_limit on a display (that's Error Bar, Confidence Band, or Quality Flag).
Related¶
- Instantiates: Noise-Bounded Measurement Interpretation — it keeps uncertainty faithful as measurements are combined into derived outputs.
- Consumes: Measurement Uncertainty Budget Table supplies the input uncertainties this calculation propagates.
- Sibling mechanisms: Measurement Uncertainty Budget Table · Error Bar, Confidence Band, or Quality Flag · Signal-to-Noise Action Gate · Calibration-Curve Residual Report · Duplicate or Blind Remeasurement Check · Noise-Floor Estimation Protocol · Measurement Claim-Limitation Note
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Uncertainty Propagation Calculation operates as an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution because it carries the uncertainty of raw inputs through the formula that combines them, so a derived quantity inherits an honest error bar instead of acquiring fake precision on the way out.
Independent corroboration: The frozen evidence defines Uncertainty Propagation Calculation as 'Carries the uncertainty of raw inputs through the formula that combines them, so a derived quantity inherits an honest error bar instead of acquiring fake precision on the way out', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Universal
Rationale: Uncertainty propagation calculation is rooted in sampling, inference, measurement, and experimental design; historically, that field developed the core operation described here: carries the uncertainty of raw inputs through the formula that combines them, so a derived quantity inherits an honest error bar instead of acquiring fake precision on the way out.
Related originating lineages:
- Data Science & Analytics — Data modeling, telemetry, and analytic monitoring supplies a distinct formative lineage for the mechanism's uncertainty propagation calculation logic.
- Engineering & Design — Engineering design, reliability, and systems-safety practice supplies a parallel or contributing lineage for the mechanism's defining operation: carries the uncertainty of raw inputs through the formula that combines them, so a derived quantity inherits an honest error bar instead of acquiring fake precision on the way out.
- Mathematics — Mathematical modeling, proof, and abstract-structure practice supplies a parallel or contributing lineage for the mechanism's defining operation: carries the uncertainty of raw inputs through the formula that combines them, so a derived quantity inherits an honest error bar instead of acquiring fake precision on the way out.
Review resolution: The blind reviewers agree that statistics_experimental_design is the primary origin and differ only on alternate origin disagreement, origin mode disagreement, domain reach disagreement, encyclopedia synthesis disagreement. I preserve every independently explained alternate from both records rather than imposing a numeric cap. I retain cross_disciplinary_synthesis because the combined evidence shows material contributions from several lineages. The broader reach of universal records portability separately from historical provenance; encyclopedia_synthesis=true preserves the affirmative synthesis judgment where either reviewer identified one.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] The first-order law of propagation of uncertainty combines input contributions as the root-sum-of-squares of each input's uncertainty times its partial derivative of the output — which is why subtracting two nearly-equal large numbers, where the output is small but the sensitivities are large, produces a derived quantity whose relative uncertainty explodes. ↩