How to run a propagation 01 Write the measured function. Any expression the calculator accepts, such as (1/2)*m*v^2; the variable names are discovered automatically. 02 Fill one row per variable. A value and a standard uncertainty σ in the same units; σ is one standard uncertainty, not a range or a tolerance. 03 Read the two headline numbers. f and its combined standard uncertainty σ_f, plus the relative uncertainty in percent. 04 Read the contribution cards. Each card shows the partial derivative ∂f/∂xᵢ and the percentage of the total variance it owns. 05 Fix the biggest term first. The percentages say where a better instrument or more repeats buy the most precision.
Worked example
Model
Inputs
Result
Shares
a - b
a = 100 ± 0.1, b = 99 ± 0.1
1 ± 0.14142136 — relative 14.142136%
a 50% , b 50%
(1/2)*m*v^2
m = 2 ± 0.002 kg, v = 30 ± 0.3 m/s
900 ± 18.02248596 J — relative 2.002498%
v 99.7506% (∂ = 60 , term 324 ), m 0.2494% (∂ = 450 , term 0.81 )
m/V
m = 250 ± 0.5 g, V = 100 ± 0.1 mL
2.5 ± 0.00559017 g/mL — relative 0.223607%
m 80% , V 20%
x + y
x = 10 ± 0.1, y = 20 ± 0.2
30 ± 0.2236068 — relative 0.745356%
y 80% , x 20%
The kinetic-energy row is the one to stare at. The mass partial derivative is 450 — much larger than the velocity's 60 — yet mass owns only 0.2494% of the variance, because its uncertainty is 0.1%, while velocity's 1% is doubled by the square. The pendulum default reads 9.81065219 ± 0.02188481 m/s² , 0.223072% relative, with the timing term at 79.904% and the length at 20.096% .
Derivatives are weights, not importance For independent inputs the first-order law is σ_f² = Σ (∂f/∂xᵢ)²·σᵢ². The derivative converts a small change in an input into a change in the output; multiplying by σᵢ turns geometry into uncertainty; squaring makes the sum a quadrature. Two shortcuts fall out of it. For a product or quotient, relative uncertainties add in quadrature: dividing by V contributes σ_V/V, and a power p contributes |p|·σ_x/x — which is why v² doubles the velocity's relative error. For a sum or difference, absolute uncertainties add in quadrature, and the relative error of the result is the absolute error divided by the result's own size.
Limits and what it does not do
First-order and independent. The formula is a linearisation at the measured values and assumes the inputs are uncorrelated; a shared calibration or a common drift needs a covariance term this panel does not model.
One standard uncertainty per input. Asymmetric errors, confidence intervals and distributions are rounded to a single σ before entry; a ±0.5% specification is not the same statement as a rectangular distribution.
The trap: subtraction is an uncertainty amplifier The same formula that makes quadrature gentle for sums makes differences brutal. Subtract two readings of about 100 with σ = 0.1 each and the absolute error is only 0.14142136 , but the result is 1, so the relative error is 14.142136% . Shrink the difference further — 1.02 minus 1.00, σ = 0.01 each — and the answer 0.02 carries 70.710678% . This is why a measurement should be designed so the quantity of interest is not a small difference of large numbers; no amount of arithmetic care can recover digits the experiment destroyed. A second, subtler trap: the partial derivatives are evaluated once, at the central values, so a strongly curved model can be wrong even when every input is independent.
Where it is used Instrument budgeting The share cards name the instrument to upgrade. On the pendulum, timing owns 79.904% of the variance: shortening σ_T from 2 ms to 1 ms does more than any improvement to the ruler.
Teaching the calculus of measurements Students compute one partial derivative by hand and check it against the card. The kinetic-energy example makes the chain rule concrete: differentiating ½mv² with respect to v gives mv = 60 , and the square's factor of two appears as the doubled relative error.
Privacy Parsing, differentiation and the quadrature sum run in your browser; the values and uncertainties you type are never uploaded.
References
BIPM JCGM, JCGM 100:2008 — the law of propagation of uncertainty. bipm.org (访问日期:2026-10-07)
NIST, Technical Note 1297 — evaluating and expressing uncertainty. nist.gov (访问日期:2026-10-07)
Wikipedia, Propagation of uncertainty — the first-order formula and product/quotient rules. en.wikipedia.org (访问日期:2026-10-07)
Wikipedia, Partial derivative — the local sensitivity weighting each input. en.wikipedia.org (访问日期:2026-10-07)
Wikipedia, Significant figures — digits lost in a difference. en.wikipedia.org (访问日期:2026-10-07)
Calculators in this hub
Hand-picked tools, one click away. The mini versions compute live and carry your values into the full calculator.
Sources & review
Reviewed by CalcX Editorial Team
Updated 2026-10-07