Propagate Measurement Uncertainty Across Formulas

Error propagation tells you how measurement uncertainties ripple through calculations—critical for lab reports, engineering tolerances, and any work where precision matters. A physics student measured length L = 10.0 ± 0.2 cm and width W = 5.0 ± 0.1 cm, then computed area A = L × W. He added the absolute uncertainties to get ±0.3 cm²—wrong. For products, you add relative uncertainties in quadrature: (0.02)² + (0.02)² = 0.0008, giving σ_A/A = 0.028, or about ±1.4 cm². The common mistake is applying the sum formula to products. When reading results, note which variable dominates the total uncertainty—improving that one has the biggest payoff.

The propagation formula depends on how variables combine. For sums and differences, add variances: σ_f = √(σ₁² + σ₂² + ...). For products and quotients, add squared relative uncertainties: (σ_f/f)² = (σ₁/x₁)² + (σ₂/x₂)² + .... For powers f = x^n, the exponent scales the relative uncertainty: σ_f/f = |n| × σ_x/x.

Perimeter (L₁ + L₂ + L₃) uses sum propagation. Area (L × W) uses product propagation. Volume of a sphere (4/3 π r³) uses power propagation for r³. Match the formula to the math.

For mixed expressions—say, (a + b) × c—break them into stages. First propagate a + b to get σ_(a+b), then treat that as a single term in the product propagation.

Absolute uncertainty (σ) carries units: 5.0 ± 0.1 cm has σ = 0.1 cm. Relative uncertainty (σ/x) is dimensionless: 0.1 / 5.0 = 0.02, or 2%. Percent uncertainty is just relative × 100.

Sum/difference propagation uses absolute uncertainties directly. Product/quotient and power propagation work in relative terms because scaling multiplies errors proportionally. After propagating, convert back to absolute if needed: σ_f = |f| × (relative uncertainty).

Comparing precision across different quantities requires relative uncertainty. A 0.1 g error matters more for a 1 g sample (10%) than for a 100 g sample (0.1%).

Variance contributions show each variable's share of the total. For products, each term contributes (aᵢ σᵢ/xᵢ)². The variable with the largest percentage is the dominant source. Reducing that uncertainty improves overall precision more than tweaking the others.

If time contributes 80% and distance 20% to velocity uncertainty, don't bother buying a better ruler. Get a better stopwatch.

Variance contributions also reveal hidden leverage. A variable with a large exponent (like r² in area calculations) may dominate even if its relative uncertainty looks small.

Uncertainty typically gets one or two significant figures. Round the uncertainty first, then round the result to match. If σ = 0.0234, round to 0.02. If the result is 12.3456, report 12.35 ± 0.02.

Don't report more precision than your uncertainty justifies. Writing 12.3456789 ± 0.02 is misleading—the last six digits are noise.

For relative uncertainty, two significant figures suffice: 2.3% is fine; 2.345678% is overkill. Scientific notation helps: (1.23 ± 0.02) × 10⁴ is cleaner than 12300 ± 200.

Standard formulas assume measurements are independent—knowing one tells you nothing about the others. If two quantities share a common error source (same instrument, same calibration), covariance terms appear. Ignoring them can over- or underestimate the true uncertainty.

The formulas are also first-order Taylor approximations. They work when σ/x is small (under 10%). For larger relative uncertainties or highly nonlinear functions, Monte Carlo simulation gives more accurate results.

Gaussian errors are assumed. If your errors follow a different distribution, the propagated distribution may not be symmetric, and credible intervals may differ from what these formulas suggest.