- Practice
- About 20 min
- 5 practice questions
Linear approximation and absolute, relative and percentage errors
Replace a function near a point by its tangent line, use that linear approximation to estimate values without a calculator, and measure the quality of an estimate by absolute, relative and percentage error.
If something looks wrong, please tell us.
By the end of this lesson you can
- Write the linear approximation L(x) of a differentiable function at a point x₀.
- Choose a convenient function, point and Δx to estimate values such as √25.4 or ∛8.12 by hand.
- Estimate the change in a quantity from a small change in its input.
- Compute absolute, relative and percentage errors and say which estimate is better.
Chapter 8 of the Tamil Nadu textbook (Mathematics Volume II) asks a practical question: if an input changes a little, roughly how much does the output change? A heated metal plate grows slightly; how much does its area grow? Linear functions are easy to evaluate, so the idea is to replace a curved graph, near one point, by the straight line that hugs it best: the tangent line.
From the derivative to an approximation
If f is differentiable at x, then [f(x + Δx) − f(x)]/Δx → f′(x) as Δx → 0. So for small Δx the two are nearly equal, which gives two useful forms:
- Change in output: Δy = f(x + Δx) − f(x) ≈ f′(x) Δx.
- New value: f(x + Δx) ≈ f(x) + f′(x) Δx.
Linear approximation of f at x₀: L(x) = f(x₀) + f′(x₀)(x − x₀). Its graph is the tangent line at (x₀, f(x₀)).
The error f(x) − L(x) tends to 0 as x → x₀, and in fact faster than x − x₀ does, because [f(x₀ + h) − f(x₀) − f′(x₀)h]/h → f′(x₀) − f′(x₀) = 0. So L is an excellent approximation close to x₀ and gets worse further away. If f is itself linear, f(x) = mx + c, then L(x) = f(x) exactly.
Estimating a value by linear approximation
- Pick f The function whose value you need, e.g. f(x) = √x.
- Pick a nearby x₀ Where f(x₀) and f′(x₀) are easy, e.g. 25.
- Find Δx Δx = x − x₀, small.
- Compute f(x₀) + f′(x₀) Δx.
Worked example 1: a square root by hand
Estimate √25.4. Working: take f(x) = √x, x₀ = 25, Δx = 0.4. Then f(25) = 5 and f′(x) = 1/(2√x), so f′(25) = 1/10. The linear approximation at 25 is L(x) = 5 + (1/10)(x − 25), and √25.4 ≈ 5 + 0.1 × 0.4 = 5.04. A calculator gives 5.03984..., so the estimate is good to about 0.00016.
Worked example 2: a cube root
Estimate ∛8.12. Working: f(x) = x^(1/3), x₀ = 8, Δx = 0.12. f(8) = 2 and f′(x) = (1/3)x^(−2/3), so f′(8) = (1/3)(1/4) = 1/12. Hence ∛8.12 ≈ 2 + 0.12/12 = 2.01. Since 2.01³ = 8.120601, which is a little more than 8.12, the true root is slightly below 2.01 (it is 2.00995...). The choice of x₀ is not unique; any nearby point with an easy value works.
Absolute, relative and percentage error
The exact value of a quantity is its actual value; a value found by some approximation is the approximate value. Three measures of error are used:
| Measure | Formula | Units |
|---|---|---|
| Absolute error | |actual value − approximate value| | same units as the quantity |
| Relative error | |actual − approximate| / |actual| (actual ≠ 0) | no units |
| Percentage error | relative error × 100 | per cent |
The absolute error tells how much we are off, but not how good the estimate is. Being off by 2 when the actual value is 10 is a 20% error; being off by 2 when the actual value is 400 is only a 0.5% error. The relative or percentage error lets us compare the two. To compute either one, you need the actual value.
Worked example 3: errors in an estimated change
The edge of a metal cube expands from 10 cm to 10.1 cm. Estimate the increase in volume and find the errors of the estimate. Working: V(x) = x³, V′(x) = 3x², so ΔV ≈ V′(10) × 0.1 = 300 × 0.1 = 30 cm³. Exactly, ΔV = 10.1³ − 10³ = 1030.301 − 1000 = 30.301 cm³. Absolute error = 30.301 − 30 = 0.301 cm³. Relative error = 0.301/30.301 ≈ 0.00993. Percentage error ≈ 0.99%.
Worked example 4: how errors grow with powers
The radius of a circle is measured with a possible error of 1%. Roughly what percentage error does this cause in the area? Working: A = πr², so ΔA ≈ 2πr Δr and ΔA/A ≈ 2πr Δr/(πr²) = 2(Δr/r). The relative error doubles: about 2% in the area. In general, if y = xⁿ then Δy/y ≈ n(Δx/x), so the percentage error in xⁿ is about n times that in x; for a root, n = 1/k, it is about 1/k times as large.
Key takeaways: near x₀, f(x) ≈ f(x₀) + f′(x₀)(x − x₀), the tangent line; pick x₀ so that f(x₀) and f′(x₀) are easy; Δy ≈ f′(x) Δx estimates a change; absolute error has units, relative and percentage errors do not and are the right way to judge an approximation.
Check your understanding
-
1. The linear approximation of f(x) = x² at x₀ = 3 is
- L(x) = 6x − 9
- L(x) = 6x + 9
- L(x) = 3x
- L(x) = 9 + 2x
Show answer
Answer: A. f(3) = 9 and f′(3) = 6, so L(x) = 9 + 6(x − 3) = 6x − 9.
-
2. Using linear approximation at x₀ = 49, √50 is about
- 7.01
- 7.1
- 7.0714
- 7.05
Show answer
Answer: C. f′(x) = 1/(2√x) = 1/14 at 49, so √50 ≈ 7 + 1/14 = 7.0714. (The calculator value is 7.0711.)
-
3. The actual value is 80 and the approximate value is 76. The percentage error is
- 4%
- 5.26%
- 0.05%
- 5%
Show answer
Answer: D. Absolute error = 4; relative error = 4/80 = 0.05; percentage error = 0.05 × 100 = 5%. Divide by the actual value, not the approximate one.
-
4. A sphere's radius has a 1% measurement error. The approximate percentage error in its volume is
- 1%
- 3%
- 2%
- 4%
Show answer
Answer: B. V = (4/3)πr³, so ΔV/V ≈ 3(Δr/r) = 3 × 1% = 3%.
-
5. Which statement about errors is true?
- Relative error has the same units as the quantity
- Percentage error can be found without knowing the actual value
- Absolute error alone shows whether an approximation is good
- Relative error is undefined when the actual value is 0
Show answer
Answer: D. Relative error divides by the actual value, so it needs the actual value and is undefined if that value is 0. It has no units, and absolute error alone does not show how good the estimate is.
Sign in to keep track of the lessons you finish.