What is error propagation?
Every physical measurement carries an uncertainty. When you combine measured quantities through arithmetic, those uncertainties must be carried through to the final result. This error propagation calculator handles the four basic operations — addition, subtraction, multiplication and division — for two values A and B, each with its own uncertainty \(\delta A\) and \(\delta B\), assuming the errors are random and independent.
How to use it
Pick the operation, enter value A and its uncertainty \(\delta A\), then value B and its uncertainty \(\delta B\). The calculator returns the combined value Q, its absolute uncertainty \(\delta Q\), and the relative (percentage) uncertainty.
The formulas
For sums and differences, the absolute uncertainties combine in quadrature: $$\delta Q = \sqrt{\delta A^{2} + \delta B^{2}}$$ For products and quotients, the relative uncertainties combine in quadrature: $$\frac{\delta Q}{\lvert Q \rvert} = \sqrt{\left(\frac{\delta A}{A}\right)^{2} + \left(\frac{\delta B}{B}\right)^{2}}$$ and the absolute uncertainty is then \(\delta Q = \lvert Q \rvert \cdot\) that relative value.
Worked example
Multiply A = 10 ± 0.5 by B = 4 ± 0.3. The product is \(Q = 40\). The relative uncertainties are \(0.5/10 = 0.05\) and \(0.3/4 = 0.075\). Combining: $$\sqrt{0.05^{2} + 0.075^{2}} = \sqrt{0.0025 + 0.005625} = \sqrt{0.008125} \approx 0.090139$$ So \(\delta Q = 40 \times 0.090139 \approx 3.61\), giving Q = 40 ± 3.61 (about 9.0%).
More Worked Examples: Addition, Subtraction and Division
For addition and subtraction, absolute uncertainties combine in quadrature: \(\delta Q = \sqrt{\delta A^2 + \delta B^2}\). For multiplication and division, relative uncertainties combine in quadrature: \(\frac{\delta Q}{|Q|} = \sqrt{ \left(\frac{\delta A}{A}\right)^2 + \left(\frac{\delta B}{B}\right)^2 }\). All examples assume independent random errors.
Example 1: Addition
A = 12.0 ± 0.3 and B = 8.0 ± 0.4.
- Value: \(Q = 12.0 + 8.0 = 20.0\).
- Square each uncertainty: \(0.3^2 = 0.09\) and \(0.4^2 = 0.16\).
- Sum: \(0.09 + 0.16 = 0.25\).
- Square root: \(\delta Q = \sqrt{0.25} = \) 0.5.
- Relative uncertainty: \(0.5 / 20.0 = 2.5\%\).
Result: \(Q = 20.0 \pm 0.5\). Note that this is less than the simple sum of uncertainties (0.7), because independent errors partly cancel.
Example 2: Subtraction of nearly equal values
A = 50.0 ± 0.5 and B = 48.0 ± 0.4.
- Value: \(Q = 50.0 - 48.0 = 2.0\).
- Square each uncertainty: \(0.5^2 = 0.25\) and \(0.4^2 = 0.16\).
- Sum: \(0.25 + 0.16 = 0.41\).
- Square root: \(\delta Q = \sqrt{0.41} \approx \) 0.64.
- Relative uncertainty: \(0.64 / 2.0 \approx 32\%\).
Result: \(Q = 2.0 \pm 0.64\). Each input is known to about 1%, yet the difference is uncertain by about 32%, because the uncertainties add while the values cancel.
Example 3: Division
A = 100 ± 2 and B = 20 ± 1.
- Value: \(Q = 100 / 20 = 5.0\).
- Relative uncertainties: \(\delta A / A = 2/100 = 0.02\) and \(\delta B / B = 1/20 = 0.05\).
- Square each: \(0.02^2 = 0.0004\) and \(0.05^2 = 0.0025\).
- Sum: \(0.0004 + 0.0025 = 0.0029\).
- Square root: \(\sqrt{0.0029} \approx 0.0539\), so the relative uncertainty is about 5.4%.
- Absolute uncertainty: \(\delta Q = 5.0 \times 0.0539 \approx \) 0.27.
Result: \(Q = 5.0 \pm 0.27\) (about 5.4% relative uncertainty). The divisor B contributes most of it, since \(0.0025\) is about 86% of the total \(0.0029\).
Key Terms: Absolute, Relative and Percentage Uncertainty
- Absolute uncertainty (δA): the size of the uncertainty in a measured value, in the same units as the value (e.g. 0.3 cm in 12.0 ± 0.3 cm).
- Relative uncertainty: the absolute uncertainty divided by the magnitude of the value, \(\delta A / |A|\). It is dimensionless.
- Percentage uncertainty: relative uncertainty × 100%. For example, 0.3 / 12.0 = 0.025, or 2.5%.
- Propagated uncertainty: the uncertainty in a calculated result Q that follows from the uncertainties of the inputs used to compute it.
- Quadrature (root-sum-square): combining independent uncertainties by squaring each, adding, and taking the square root: \(\sqrt{x^2 + y^2}\).
- Independent (random) errors: errors that vary unpredictably from measurement to measurement and are unrelated between A and B. They can partly cancel, which is why quadrature is used.
- Systematic errors: consistent biases (e.g. a miscalibrated instrument) that shift results in the same direction. They do not average out and are not captured by quadrature formulas alone.
- Result Q with δQ: Q is the computed value (A + B, A − B, A × B or A ÷ B) and δQ is its propagated absolute uncertainty, written \(Q \pm \delta Q\).
Note: these formulas assume A and B are uncorrelated. If their errors are correlated (for example, both come from the same instrument or one depends on the other), the quadrature formulas do not apply and a covariance term must be included.
Interpreting the Propagated Uncertainty
What Q ± δQ means. If the input uncertainties are standard uncertainties (about 1 standard deviation) and the errors are roughly normally distributed, then \(Q \pm \delta Q\) is approximately a 68% interval for the true value. Doubling δQ gives roughly a 95% interval.
Rounding convention. Quote the uncertainty to 1 or 2 significant figures, then round Q to the same decimal place. For example, \(2.0 \pm 0.64\) is acceptable, and \(5.0 \pm 0.27\) can be written \(5.0 \pm 0.3\). Do not report more digits in Q than the uncertainty supports.
Which input dominates. Because terms are squared before being added, the largest term dominates. For addition and subtraction, compare the absolute uncertainties; for multiplication and division, compare the relative uncertainties. In 100 ± 2 ÷ 20 ± 1, the divisor's 5% outweighs the numerator's 2%, so improving the measurement of B reduces δQ most.
Subtracting nearly equal values. The absolute uncertainties add in quadrature while the values cancel, so the relative uncertainty of the difference can be large. In 50.0 ± 0.5 minus 48.0 ± 0.4, inputs of about 1% give a result with about 32% uncertainty. If possible, redesign the measurement to avoid differences of large, similar numbers.
Limitations.
- The errors in A and B must be independent; correlated errors need a covariance term.
- The multiplication and division formulas are first-order (linear) approximations that work best when relative uncertainties are small (roughly under 10%).
- Systematic errors are not included unless you add them into the input uncertainties.
- If a result is near zero or the uncertainties are very large, the 68% interpretation becomes unreliable.
FAQ
Why add in quadrature instead of just summing? Independent random errors partially cancel on average, so the statistically correct combination is the square root of the sum of squares rather than a simple sum.
Does subtraction reduce uncertainty? No — for A − B the absolute uncertainty is the same \(\sqrt{\delta A^{2} + \delta B^{2}}\) as for A + B, even though Q gets smaller. This is why subtracting nearly equal numbers is numerically dangerous.
What if a value is zero? Relative uncertainty is undefined when the value is zero, so the calculator treats that term as zero to avoid division by zero.