Errors and approximations
Errors and approximations: types and sources of error, significant figures and rounding, absolute/relative/percentage error, limits of accuracy, error propagation (sum, difference, product, quotient, powers; multivariable via differentials), truncation error in series, and practical estimation of er
Drafted with Aria, reviewed by the AiCanCode.org team · Updated . How we write and check our notes. Spotted an error? Use Give Feedback at the bottom of the page.
Why it matters
Engineers rarely know a “true value.” Every measurement and every computed result has some error. Understanding errors and approximations helps you decide how many significant figures to keep, how much trust to place in a result, and how measurement errors affect a final design calculation (like stress, flow rate, or power).
Key ideas
- Error: The difference between the true value T and approximate value A. Two common measures:
- Absolute error: |T − A|. Non-negative.
- Relative error: |T − A| / |T|, a ratio showing error size compared to the true value.
- Percentage error: Relative error × 100%.
- Accuracy vs precision:
- Accuracy: How close a value is to the true value.
- Precision: How fine the resolution or repeatability is (e.g., measured to nearest 0.01 mm). High precision does not guarantee accuracy.
- Significant figures (sig. figs.): Digits that carry meaningful information about precision. Rules used in exams:
- All non-zero digits are significant.
- Zeros between non-zero digits are significant (e.g., 1003 has 4 sig. figs.).
- Leading zeros are not significant (e.g., 0.00450 has 3 sig. figs.).
- Trailing zeros are significant only if there is a decimal point (e.g., 1500 has 2, but 1500. has 4 sig. figs.).
- Rounding rule: To round at a digit, look at the next digit. If it is 5 or more, increase the last kept digit by 1; otherwise leave it. (If your university specifies round-half-even, follow that. Here we assume “round-half-up.”)
- Types of errors:
- Gross errors: Mistakes by human/reading/recording.
- Systematic errors: Consistent bias due to instrument calibration, observation method, or environment.
- Random errors: Unpredictable variations with zero mean, analyzed statistically.
- Limits of accuracy (instrument resolution): If a quantity x is measured to the nearest unit u (e.g., 0.01 m), then the measurement x̄ is within ±(u/2) of the true x, i.e., |x − x̄| ≤ u/2. This bound is also called the maximum possible error.
- Propagation of errors (single-variable functions): If y = f(x) and x has a small absolute error Δx, then the approximate absolute error in y is
- Δy ≈ |f′(x)| · Δx. The approximate relative error is |Δy|/|y| ≈ |f′(x)|·|Δx|/|f(x)|.
- This comes from the differential dy = f′(x) dx (first-order Taylor approximation).
- Propagation of errors (several variables): If y = f(x₁, x₂, …, xₙ) and each xᵢ has small error Δxᵢ, then the first-order absolute error bound is
- |Δy| ≤ |∂f/∂x₁|Δx₁ + |∂f/∂x₂|Δx₂ + … + |∂f/∂xₙ|Δxₙ.
- For relative error, divide the above by |y| (when y ≠ 0).
- Common propagation shortcuts (first-order, worst-case):
- Sum/difference: z = x ± y → Δz ≤ Δx + Δy.
- Product/quotient: z = x·y or x/y → Δz/|z| ≤ Δx/|x| + Δy/|y|.
- Power: z = xⁿ → Δz/|z| ≤ |n|·(Δx/|x|).
- Truncation error: Error due to cutting off an infinite process (e.g., Taylor series, numerical method). If f(x) has a Taylor series, truncating after k terms leaves a remainder Rₖ₊₁(x). For small x or sufficiently many terms, Rₖ₊₁ can be bounded by a known formula.
- Bias vs uncertainty:
- Bias: Systematic shift from the true value (e.g., uncalibrated instrument). Reduces accuracy even if repeated many times.
- Uncertainty: Range of probable values due to unknown errors; reported as ± bound.
- Diagram to draw for differentials (if asked): Draw the curve y = f(x). At a point x, draw the tangent line. Show a small horizontal change Δx and the corresponding vertical change along the tangent f′(x)Δx as the linear approximation to the actual change in y. Label dy ≈ Δy.
- Diagram to draw for random errors (if asked): Draw a bell-shaped curve (normal distribution) centered at the true value T. Shade the central band to show most repeated measurements cluster near T; label mean, standard deviation σ, and indicate that random errors spread measurements around T.
Exact statements commonly asked:
- Definition: “Absolute error is the magnitude of the difference between the true value T and approximate value A, i.e., |T − A|.”
- Definition: “Relative error is the ratio of absolute error to the magnitude of the true value, i.e., |T − A|/|T|.”
- Statement (limits of accuracy): “If a measurement is recorded to the nearest u, the maximum possible error is u/2.”
- Proposition (first-order error propagation): “If y = f(x₁, …, xₙ) and the errors are small, the maximum absolute error in y is bounded by the sum of the absolute values of the first-order partial derivatives times the respective absolute errors.” Derivation sketch:
- Taylor-expand: f(x + Δx) ≈ f(x) + ∑(∂f/∂xᵢ)Δxᵢ.
- Take absolute values and apply triangle inequality: |Δy| = |∑(∂f/∂xᵢ)Δxᵢ| ≤ ∑|∂f/∂xᵢ|·|Δxᵢ|.
- Divide by |f(x)| (if nonzero) for relative error.
- Rule (products/quotients/powers) follow by applying the above to f(x) = ln z = ln x + ln y, etc., or by direct differentials: d(ln z) = dz/z = dx/x + dy/y; for z = xⁿ, dz/z = n·dx/x.
Formulas
Percentage error = (|T − A| / |T|) × 100%— Unitless percent. Applies when T ≠ 0.Maximum possible error = u / 2— u: measurement resolution (SI unit of quantity). Applies for “measured to the nearest u.”dy ≈ f′(x)·dx— y = f(x), dx: absolute error in x (SI unit of x), dy: absolute error in y (SI unit of y). First-order approximation for small dx.|Δy| ≤ |∂f/∂x₁|Δx₁ + |∂f/∂x₂|Δx₂ + … + |∂f/∂xₙ|Δxₙ— y = f(x₁,…,xₙ). Small errors; first-order worst-case bound.Sum/difference: Δz ≤ Δx + Δy— z = x ± y. Δx, Δy absolute error bounds.Product/quotient: Δz/|z| ≤ Δx/|x| + Δy/|y|— z = x·y or z = x/y, x ≠ 0, y ≠ 0.Power: Δz/|z| ≤ |n| · (Δx/|x|)— z = xⁿ, x ≠ 0.d(ln z) = dz/z— Useful for deriving relative error rules.Sphere volume: V = (4/3)·π·r³— r in m; V in m³. For error:dV = 4·π·r²·drandΔV/V ≈ 3·(Δr/r).Taylor remainder bound (Maclaurin): |Rₙ₊₁(x)| ≤ (M·|x|ⁿ⁺¹)/(n+1)!— M is a bound on |f⁽ⁿ⁺¹⁾(ξ)| for ξ between 0 and x. Applies to smooth functions near 0.
Worked examples
- Limits of accuracy and propagation (square of a length) Given: A square plate has side measured as L = 2.368 m to the nearest millimetre (0.001 m). Find the maximum possible error in area A = L² and the percentage error.
Steps:
- Resolution u = 0.001 m → Maximum possible error in length ΔL = u/2 = 0.0005 m.
- Area formula: A = L². Differential: dA = 2L·dL.
- Maximum absolute error (first-order): ΔA ≈ 2·L·ΔL = 2 × 2.368 m × 0.0005 m = 0.002368 m².
- Approximate area using measured L: A ≈ (2.368 m)² = 5.607424 m².
- Percentage error ≈ (ΔA/A) × 100% = (0.002368 / 5.607424) × 100% ≈ 0.04226%.
Answer: Maximum absolute error in area ≈ 0.002368 m²; percentage error ≈ 0.0423% (first-order).
- Error in computed quantity from product and quotient Given: Length L = 3.26 m measured to nearest 0.01 m; Width W = 4.1 m to nearest 0.1 m; Time t = 2.03 s to nearest 0.01 s. Compute Q = (L·W)/t and its maximum percentage error.
Steps:
- Maximum possible errors: ΔL = 0.005 m; ΔW = 0.05 m; Δt = 0.005 s.
- Compute Q using measured values:
- Numerator: L·W = 3.26 m × 4.1 m = 13.366 m².
- Q = 13.366 m² / 2.03 s ≈ 6.5842 m²·s⁻¹.
- Relative error bound (first-order): ΔQ/|Q| ≤ ΔL/|L| + ΔW/|W| + Δt/|t| = (0.005/3.26) + (0.05/4.1) + (0.005/2.03) ≈ 0.001533 + 0.012195 + 0.002463 = 0.016191.
- Percentage error bound ≈ 0.016191 × 100% ≈ 1.619%.
- Absolute error bound in Q: ΔQ ≤ 0.016191 × 6.5842 m²·s⁻¹ ≈ 0.1066 m²·s⁻¹.
Answer: Q ≈ 6.584 m²·s⁻¹ with maximum percentage error ≈ 1.62% (absolute error about ±0.107 m²·s⁻¹).
- Error in volume of a sphere from radius measurement Given: Radius r = 6.00 cm measured to nearest 0.01 cm. Find the maximum absolute and percentage errors in volume V.
Convert to SI: r = 0.0600 m; resolution u = 0.0001 m → Δr = u/2 = 0.00005 m.
Steps:
- V = (4/3)πr³; differential: dV = 4πr²dr.
- Absolute error (first-order): ΔV ≈ 4πr²Δr = 4π × (0.0600 m)² × 0.00005 m = 4π × 0.0036 m² × 0.00005 m = 4π × 1.8×10⁻⁷ m³ ≈ 2.26195×10⁻⁶ m³.
- Compute V using measured r: V ≈ (4/3)π × (0.0600 m)³ = 4.18879 × 2.16×10⁻⁴ m³ ≈ 9.04778×10⁻⁴ m³.
- Percentage error ≈ (ΔV/V) × 100% ≈ (2.26195×10⁻⁶ / 9.04778×10⁻⁴) × 100% ≈ 0.25%.
Answer: Maximum absolute error ≈ 2.262×10⁻⁶ m³; percentage error ≈ 0.25% (first-order).
- Truncation error using Taylor (Maclaurin) series Approximate sin x at x = 0.3 rad using 3-term Maclaurin polynomial and bound the error.
Steps:
- Maclaurin series: sin x = x − x³/3! + x⁵/5! − … . Truncate after x − x³/3!.
- Approximation: T₃(x) = x − x³/6. For x = 0.3, T₃(0.3) = 0.3 − 0.027/6 = 0.3 − 0.0045 = 0.2955.
- Remainder bound: Next term magnitude ≤ |x|⁵/5! = 0.3⁵/120 = 0.00243/120 ≈ 2.025×10⁻⁵.
- True sin(0.3) ≈ 0.2955202067 (for checking). Actual error ≈ |0.2955202067 − 0.2955| ≈ 2.0207×10⁻⁵, within the bound.
- Percentage error ≈ (2.0207×10⁻⁵ / 0.2955202067) × 100% ≈ 0.00684%.
Answer: sin(0.3) ≈ 0.2955 with truncation error bounded by about 2.03×10⁻⁵ (≈0.0068%).
Common mistakes
- Dropping units mid-calculation; always carry SI units through every step.
- Over-rounding intermediate results; round only at the end to the correct significant figures.
- Using sum/difference error rules for products/quotients or vice versa. Remember: for products/quotients, relative errors add; for sums/differences, absolute errors add.
- Ignoring sign when combining errors; bounds use magnitudes (worst case), so use absolute values.
- Applying first-order approximations for large errors; these rules assume small errors (linearization valid). For large uncertainties, higher-order terms may matter.
- Confusing precision with accuracy; many correct decimal places do not guarantee correctness if there is bias.
For your exam
- Derivations: First-order propagation via differentials (single and multivariable); relative error for product/quotient/power using logarithmic differentiation; Taylor remainder bound statement.
- Numericals: Compute maximum possible error from instrument resolution; propagate errors in formulas (area, volume, density, flow, power); percentage error of functions like xⁿ, sin x, etc., using differentials; truncation error in series approximations with remainder bounds.
- Diagrams: Tangent-line approximation for dy ≈ f′(x)dx; bell-curve sketch for random errors; symmetric interval around a measured value showing ±u/2.
- Full-marks answers include: clear formula before substitution, correct units, correctly halved resolution for maximum error, correct use of absolute vs relative error, clean rounding at the end, and a brief justification (“using first-order approximation” or “using product relative error rule”).
Quick check
- A length is recorded as 12.34 cm to the nearest 0.01 cm. What is the maximum possible error and the relative error bound?
- If z = x³ and x has a relative error of 0.4%, what is the approximate relative error in z?
- Using T₃(x) = x − x³/6 to approximate sin x at x = 0.2 rad, give a bound on the truncation error using the next term.
Answers: (1) ΔL = 0.005 cm; relative error bound ≈ 0.005/12.34 ≈ 0.000405 (0.0405%). (2) ≈ 3 × 0.4% = 1.2%. (3) Next term magnitude ≤ x⁵/5! = 0.2⁵/120 = 0.00032/120 ≈ 2.667×10⁻⁶.
Finished this topic? Mark it so your progress, study plan and readiness keep up.
Stuck on something here?