Calculus3Rates and Approximation2 of 4
Rates and Approximation · 2

Linear Approximation and Differentials

1

Estimating with a tangent line

Finding the square root of 16.216.2 by hand sounds hard, but we know the square root of 1616 is 4\cv{4}. Since 16.216.2 is so close to 1616, 16.2\cv{\sqrt{16.2}} should be just a little more than 4\cv{4}. To find how much more, notice that close to (16,4)\cv{(16, 4)}, the curve y=x\cv{y = \sqrt{x}} is almost straight and stays very near its tangent line there. So the height of that tangent line at 16.216.2 makes a good estimate.

Estimate 16.2\cv{\sqrt{16.2}} using the tangent line to y=x\cv{y = \sqrt{x}} at x=16x = 16
4\htmlClass{lc-mute}{4}
9\htmlClass{lc-mute}{9}
16\htmlClass{lc-mute}{16}
25\htmlClass{lc-mute}{25}
1\htmlClass{lc-mute}{1}
2\htmlClass{lc-mute}{2}
3\htmlClass{lc-mute}{3}
4\htmlClass{lc-mute}{4}
5\htmlClass{lc-mute}{5}
x\htmlClass{lc-mute}{x}
y\htmlClass{lc-mute}{y}
y=x\htmlClass{lc-blue}{y = \sqrt{x}}
Text solution
The slope at 1616Write x\sqrt{x} as x12x^{\frac12} and use the power rule.
f′(x)\displaystyle f'(x) =12x\displaystyle = \frac{1}{2\sqrt{x}}
f′(16)\displaystyle \tn{f'(16)} =12⋅4\displaystyle = \frac{1}{2\cdot 4} =18\displaystyle = \tn{\frac18}
The tangent lineWrite the line through (16,4)\cv{(16, 4)} with slope 18\tn{\frac18} in point-slope form, then add 4\cv{4} to both sides.
L(x)\displaystyle \tn{L(x)} =4+18 (x−16)\displaystyle = \cv{4} + \tn{\frac18}\,\stp{(x - 16)}
The slope times the stepThe step from 1616 to 16.216.2 is 0.2\stp{0.2}, so the line rises 18(0.2)=0.025\tn{\frac18}\stp{(0.2)} = \tn{0.025}. Squaring 4.0254.025 gives 16.20062516.200625, very close to 16.216.2.
16.2≈L(16.2)\displaystyle \cv{\sqrt{16.2}} \approx \tn{L(16.2)} =4+0.025\displaystyle = \cv{4} + \tn{0.025} =4.025\displaystyle = 4.025

This tangent line only gives a good estimate when xx stays close to 1616. At x=25x = 25 it gives 5.125\tn{5.125}, but 25\cv{\sqrt{25}} is exactly 5\cv{5}. The line's slope stays 18\tn{\frac18}, while the curve's slope keeps decreasing, down to 110\cv{\frac{1}{10}} at 2525, so the line gets farther and farther above the curve.

4\htmlClass{lc-mute}{4}
9\htmlClass{lc-mute}{9}
16\htmlClass{lc-mute}{16}
25\htmlClass{lc-mute}{25}
1\htmlClass{lc-mute}{1}
2\htmlClass{lc-mute}{2}
3\htmlClass{lc-mute}{3}
4\htmlClass{lc-mute}{4}
5\htmlClass{lc-mute}{5}
x\htmlClass{lc-mute}{x}
y\htmlClass{lc-mute}{y}
slope 18\htmlClass{lc-pink}{\text{slope } \tfrac18}
y=x\htmlClass{lc-blue}{y = \sqrt{x}}
5.125\htmlClass{lc-pink}{5.125}
5\htmlClass{lc-blue}{5}
L(25)\displaystyle \tn{L(25)} =4+18 (9)\displaystyle = \cv{4} + \tn{\frac18}\,\stp{(9)} =5.125\displaystyle = \tn{5.125}
25=5\displaystyle \cv{\sqrt{25} = 5}
5.125−5\displaystyle \tn{5.125} - \cv{5} =0.125\displaystyle = \ovr{0.125}

The same idea works for any function that has a derivative at a point aa. We start at the known value f(a)\cv{f(a)} and add the slope f′(a)\tn{f'(a)} times the step x−a\stp{x - a}. The height on this tangent line, L(x)\tn{L(x)}, is called the linearization of ff at aa, and it needs only the value and the slope at aa.

16.2≈4+18 (0.2)\displaystyle \cv{\sqrt{16.2}} \approx \cv{4} + \tn{\frac18}\,\stp{(0.2)}
The linearization at a
L(x)\displaystyle \tn{L(x)} =f(a)+f′(a) (x−a)\displaystyle = \cv{f(a)} + \tn{f'(a)}\,\stp{(x - a)}
2

Choosing the function and the point

For 16.2\cv{\sqrt{16.2}}, we were given the function f(x)=x\cv{f(x) = \sqrt{x}} and the point a=16a = 16. But often we only have a number to estimate, like 263\cv{\sqrt[3]{26}}, and we choose both ourselves. Since we're taking a cube root, we let f(x)=x3\cv{f(x) = \sqrt[3]{x}}, and for aa we pick the nearest input where we know f(a)\cv{f(a)} and f′(a)\tn{f'(a)} exactly. After that, we'll estimate sin⁡31∘\cv{\sin 31^\circ}, and then a value of a function we have no formula for.

Estimate 263\cv{\sqrt[3]{26}}, sin⁡31∘\cv{\sin 31^\circ}, and f(3.9)\cv{f(3.9)} using tangent lines
L(x)=f(a)+f′(a) (x−a)\htmlClass{lc-pink}{L(x)} = \htmlClass{lc-blue}{f(a)} + \htmlClass{lc-pink}{f'(a)}\,\htmlClass{lc-orange}{(x - a)}
1\htmlClass{lc-mute}{1}
2\htmlClass{lc-mute}{2}
3\htmlClass{lc-mute}{3}
4\htmlClass{lc-mute}{4}
x\htmlClass{lc-mute}{x}
y\htmlClass{lc-mute}{y}
2626
263\htmlClass{lc-blue}{\sqrt[3]{26}}
y=x3\htmlClass{lc-blue}{y = \sqrt[3]{x}}
Text solution
263\cv{\sqrt[3]{26}}: a step to the leftThe nearest perfect cube is 2727, so a=27a = 27, f(27)=3\cv{f(27) = 3} and f′(27)=13⋅19=127\tn{f'(27)} = \frac13\cdot\frac19 = \tn{\frac{1}{27}}. Since 2626 is to the left of 2727, the step is −1\stp{-1}.
263≈3+127 (−1)\displaystyle \cv{\sqrt[3]{26}} \approx \cv{3} + \tn{\frac{1}{27}}\,\stp{(-1)} =8027≈2.963\displaystyle = \frac{80}{27} \approx 2.963
sin⁡31∘\cv{\sin 31^\circ}: the step in radiansWith a=30∘a = 30^\circ, a step of 1\stp{1} gives about 1.371.37, more than any sine can be. The derivative of sin⁡x\sin x is cos⁡x\cos x only when xx is in radians, so the step is 1∘=π180\stp{1^\circ = \frac{\pi}{180}}.
sin⁡31∘≈12+32⋅π180≈0.5151\displaystyle \cv{\sin 31^\circ} \approx \cv{\frac12} + \tn{\frac{\sqrt3}{2}}\cdot\stp{\frac{\pi}{180}} \approx 0.5151
f(3.9)\cv{f(3.9)}: no formula neededThe tangent line needs only f(4)=2\cv{f(4) = 2} and f′(4)=−3\tn{f'(4) = -3}. The step 3.9−4=−0.1\stp{3.9 - 4 = -0.1} is negative, and the negative slope times the negative step gives a positive rise.
f(3.9)≈2+(−3) (−0.1)\displaystyle \cv{f(3.9)} \approx \cv{2} + \tn{(-3)}\,\stp{(-0.1)} =2.3\displaystyle = 2.3
3

Overestimates and underestimates

The tangent line at 1616 gave 16.2≈4.025\cv{\sqrt{16.2}} \approx \tn{4.025}. Is 4.025\tn{4.025} bigger than 16.2\cv{\sqrt{16.2}}, or smaller? We could square 4.0254.025 to find out, but we can also tell from the slope of the curve, without squaring anything.

Decide whether a linear approximation is an overestimate or an underestimate from the second derivative
9\htmlClass{lc-mute}{9}
16\htmlClass{lc-mute}{16}
25\htmlClass{lc-mute}{25}
3\htmlClass{lc-mute}{3}
4\htmlClass{lc-mute}{4}
5\htmlClass{lc-mute}{5}
x\htmlClass{lc-mute}{x}
y\htmlClass{lc-mute}{y}
4.025  ?  16.2\htmlClass{lc-pink}{4.025}\ \,?\ \,\htmlClass{lc-blue}{\sqrt{16.2}}
slope 18\htmlClass{lc-pink}{\text{slope } \tfrac18}
y=x\htmlClass{lc-blue}{y = \sqrt{x}}
Text solution
Compare the slopesThe line's slope is 18\tn{\frac18} everywhere, but the curve's slope decreases as xx increases. To the right of 1616, the curve climbs slower than the line. To the left, it climbs faster, so to meet the line at (16,4)\cv{(16, 4)} it had to start below it.
f′(9)=16\displaystyle \cv{f'(9) = \frac16}
f′(16)=18\displaystyle \tn{f'(16) = \frac18}
f′(25)=110\displaystyle \cv{f'(25) = \frac{1}{10}}
The sign of f′′f''A decreasing slope means f′′f'' is negative. The curve is below its tangent line on both sides of 1616, so 4.025\tn{4.025} is an overestimate.
f′′(x)\displaystyle f''(x) =−14x32<0\displaystyle = -\frac{1}{4x^{\frac32}} < 0
ex\cv{e^x} at 00Here f′′(x)=exf''(x) = e^x is positive, so the curve is above its tangent line L(x)=1+x\tn{L(x)} = \cv{1} + \stp{x} on both sides, and 1.1\tn{1.1} is an underestimate.
e0.1≈1.105>1.1\displaystyle \cv{e^{0.1} \approx 1.105} > \tn{1.1}
f(3.9)\cv{f(3.9)} when f′′<0f'' < 0The curve is still below its tangent line, even though ff is decreasing and we stepped to the left.
f(3.9)<L(3.9)\displaystyle \cv{f(3.9)} < \tn{L(3.9)} =2.3\displaystyle = 2.3

Here's y=x\cv{y = \sqrt{x}} with its tangent line at 1616. On both sides of 1616, the line is above the curve.

4\htmlClass{lc-mute}{4}
9\htmlClass{lc-mute}{9}
16\htmlClass{lc-mute}{16}
25\htmlClass{lc-mute}{25}
1\htmlClass{lc-mute}{1}
2\htmlClass{lc-mute}{2}
3\htmlClass{lc-mute}{3}
4\htmlClass{lc-mute}{4}
5\htmlClass{lc-mute}{5}
x\htmlClass{lc-mute}{x}
y\htmlClass{lc-mute}{y}
slope 18\htmlClass{lc-pink}{\text{slope } \tfrac18}
y=x\htmlClass{lc-blue}{y = \sqrt{x}}
3.125\htmlClass{lc-pink}{3.125}
3\htmlClass{lc-blue}{3}
5.125\htmlClass{lc-pink}{5.125}
5\htmlClass{lc-blue}{5}
L(9)\displaystyle \tn{L(9)} =4−78\displaystyle = \cv{4} - \tn{\frac78} =3.125\displaystyle = \tn{3.125}
3.125>3\displaystyle \tn{3.125} > \cv{3} =9\displaystyle = \cv{\sqrt{9}}
L(25)\displaystyle \tn{L(25)} =5.125\displaystyle = \tn{5.125}
5.125>5\displaystyle \tn{5.125} > \cv{5} =25\displaystyle = \cv{\sqrt{25}}

When f′′<0f'' < 0 on an interval, the slope is decreasing, and we say the curve is concave down. A tangent line to a concave-down curve gives an overestimate whether we step to the right or to the left. When f′′>0f'' > 0, the slope is increasing, and we say the curve is concave up. That's y=ex\cv{y = e^x}, which stays above its tangent line at 00.

−2\htmlClass{lc-mute}{-2}
−1\htmlClass{lc-mute}{-1}
1\htmlClass{lc-mute}{1}
2\htmlClass{lc-mute}{2}
3\htmlClass{lc-mute}{3}
4\htmlClass{lc-mute}{4}
x\htmlClass{lc-mute}{x}
y\htmlClass{lc-mute}{y}
1\htmlClass{lc-mute}{1}
e\htmlClass{lc-blue}{e}
2\htmlClass{lc-pink}{2}
e−1\htmlClass{lc-blue}{e^{-1}}
slope 1\htmlClass{lc-pink}{\text{slope } 1}
y=ex\htmlClass{lc-blue}{y = e^x}
f(x)=ex\displaystyle \cv{f(x) = e^x}a=0\displaystyle a = 0
L(x)\displaystyle \tn{L(x)} =1+x\displaystyle = \cv{1} + \stp{x}
e0.1≈L(0.1)\displaystyle \cv{e^{\stp{0.1}}} \approx \tn{L(\stp{0.1})} =1.1\displaystyle = \tn{1.1}
f′′(x)\displaystyle f''(x) =ex>0\displaystyle = e^x > 0
e0.1≈1.105>1.1\displaystyle \cv{e^{0.1} \approx 1.105} > \tn{1.1}

Does moving left of the tangent point change an overestimate into an underestimate?

4

Differentials and the change in yy

If xx changes by a small amount, how much does y=f(x)\cv{y = f(x)} change?

f(x+dx)−f(x)≈  ?\displaystyle \cv{f(x + \stp{dx}) - f(x)} \approx \;?

To answer that, let's give the changes names. We call the change in xx, dx\stp{dx}. The change in height along the curve y=f(x)\cv{y = f(x)} is Δy\cv{\Delta y}, and the rise along the tangent line is dy\tn{dy}. We call the line's rise dy\tn{dy} because dydx\frac{\tn{dy}}{\stp{dx}} is the slope of the tangent line, f′(x)\tn{f'(x)}. Slope is rise over run, so with run dx\stp{dx}, the rise is f′(x) dx\tn{f'(x)}\,\stp{dx}.

Rise along the tangent line
dy\displaystyle \tn{dy} =f′(x) dx\displaystyle = \tn{f'(x)}\,\stp{dx}
Rise along the curve
Δy\displaystyle \cv{\Delta y} =f(x+dx)−f(x)\displaystyle = \cv{f(x + \stp{dx}) - f(x)}

When we estimated 16.2\cv{\sqrt{16.2}}, the step 0.2\stp{0.2} was dx\stp{dx}, and the rise along the tangent line, 0.025\tn{0.025}, was dy\tn{dy}. So the linear approximation is just f(a)+dy\cv{f(a)} + \tn{dy}.

x at 16 ⁣:\displaystyle \cv{\sqrt{x}} \text{ at } 16\colondx=0.2\displaystyle \stp{dx = 0.2}dy=18 (0.2)=0.025\displaystyle \tn{dy} = \tn{\frac18}\,\stp{(0.2)} = \tn{0.025}4.025=4+dy\displaystyle 4.025 = \cv{4} + \tn{dy}

But dy\tn{dy} is the rise along the line, not the curve. On y=x2\cv{y = x^2} at x=3x = 3, with dx=1\stp{dx = 1}, the line rises 6\tn{6} and the curve rises 7\cv{7}.

2\htmlClass{lc-mute}{2}
3\htmlClass{lc-mute}{3}
4\htmlClass{lc-mute}{4}
x\htmlClass{lc-mute}{x}
dx=1\htmlClass{lc-orange}{dx = 1}
dy=6\htmlClass{lc-pink}{dy = 6}
Δy=7\htmlClass{lc-blue}{\Delta y = 7}
y=x2\htmlClass{lc-blue}{y = x^2}
y=x2\displaystyle \cv{y = x^2}x=3\displaystyle x = 3dx=1\displaystyle \stp{dx = 1}
dy\displaystyle \tn{dy} =2(3) (1)\displaystyle = \tn{2(3)}\,\stp{(1)} =6\displaystyle = \tn{6}
Δy\displaystyle \cv{\Delta y} =42−32\displaystyle = 4^2 - 3^2 =7\displaystyle = \cv{7}
Δy−dy\displaystyle \cv{\Delta y} - \tn{dy} =1\displaystyle = \und{1}

Why did dy\tn{dy} miss that 1\und{1}, and what happens to the difference when dx\stp{dx} is small?

Compare dy\tn{dy} and Δy\cv{\Delta y} for y=x2\cv{y = x^2} at x=3x = 3
2\htmlClass{lc-mute}{2}
3\htmlClass{lc-mute}{3}
4\htmlClass{lc-mute}{4}
x\htmlClass{lc-mute}{x}
dx=1\htmlClass{lc-orange}{dx = 1}
dy\htmlClass{lc-pink}{dy}
Δy\htmlClass{lc-blue}{\Delta y}
y=x2\htmlClass{lc-blue}{y = x^2}
Text solution
Two strips and a cornerThink of x2x^2 as the area of a square with side xx. Growing the side from 33 to 3+dx3 + \stp{dx} adds two strips, which make dy\tn{dy}, and a corner, which dy\tn{dy} leaves out.
(3+dx)2−32\displaystyle \cv{(3 + \stp{dx})^2 - 3^2} =2⋅3 dx+(dx)2\displaystyle = \tn{2\cdot 3\,\stp{dx}} + \und{(dx)^2}
Make dx\stp{dx} smallWith dx=0.1\stp{dx = 0.1}, the strips shrink by a factor of 1010, but the corner shrinks by a factor of 100100. So when dx\stp{dx} is small, dy\tn{dy} is almost all of Δy\cv{\Delta y}.
0.6+0.01\displaystyle \tn{0.6} + \und{0.01} =0.61\displaystyle = \cv{0.61}

Here's the square for dx=1\stp{dx = 1}. The corner is the 1\und{1} that dy\tn{dy} missed.

33
dx=1\htmlClass{lc-orange}{dx = 1}
3 dx\htmlClass{lc-pink}{3\,}\htmlClass{lc-orange}{dx}
3 dx\htmlClass{lc-pink}{3\,}\htmlClass{lc-orange}{dx}
(dx)2\htmlClass{lc-green}{(dx)^2}
33
dx=0.1\htmlClass{lc-orange}{dx = 0.1}
3 dx\htmlClass{lc-pink}{3\,}\htmlClass{lc-orange}{dx}
3 dx\htmlClass{lc-pink}{3\,}\htmlClass{lc-orange}{dx}
(dx)2\htmlClass{lc-green}{(dx)^2}
dx=1 ⁣:6+1\displaystyle \stp{dx = 1}\colon \quad \tn{6} + \und{1} =7\displaystyle = \cv{7}

To find dy\tn{dy} for any function, take the derivative and multiply by dx\stp{dx}, evaluating the derivative at the starting value xx, not x+dxx + \stp{dx}. For y=x2sin⁡x\cv{y = x^2\sin x}, we find the derivative with the product rule.

y=x2sin⁡xdy=(2xsin⁡x+x2cos⁡x) dx\displaystyle \begin{aligned} \cv{y} &= \cv{x^2\sin x}\\ \tn{dy} &= \tn{(2x\sin x + x^2\cos x)}\,\stp{dx} \end{aligned}
5

Estimating error and relative error

Say we measure the radius of a sphere to be 10 cm10\text{ cm}, but our measurement could be off by up to 0.1 cm\stp{0.1}\text{ cm}. How much could that error change the computed volume?

We'll find that error in cubic centimetres, and also the relative error, which is the error divided by the whole volume.

Estimate the error and relative error in a sphere's volume from an error in its radius
Text solution
The error is a differentialFor a square of side xx, an error dx\stp{dx} gave dA=2x dx\tn{dA} = \tn{2x}\,\stp{dx}. Here the error dr=0.1\stp{dr = 0.1} gives dV=4πr2 dr\tn{dV} = \tn{4\pi r^2}\,\stp{dr}, evaluated at the measured radius r=10r = 10.
dV\displaystyle \tn{dV} =4π(10)2 (0.1)\displaystyle = \tn{4\pi(10)^2}\,\stp{(0.1)} =40π≈126 cm3\displaystyle = 40\pi \approx 126\text{ cm}^3
The relative errorDividing dV\tn{dV} by V=43πr3\cv{V = \frac43\pi r^3} leaves 3⋅drr3\cdot\frac{\stp{dr}}{r}, and 0.110=1%\frac{\stp{0.1}}{10} = 1\%. The 33 is the exponent the power rule brought down from r3r^3.
dVV\displaystyle \frac{\tn{dV}}{\cv{V}} =3⋅drr\displaystyle = 3\cdot\frac{\stp{dr}}{r} =3(1%)\displaystyle = 3(1\%) =3%\displaystyle = 3\%
If we measure the circumferenceWrite the area in terms of CC, then multiply its derivative by the error dC\stp{dC}.
r\displaystyle r =C2π  →  A\displaystyle = \frac{C}{2\pi} \;\to\; \cv{A} =πr2\displaystyle = \pi r^2 =C24π\displaystyle = \cv{\frac{C^2}{4\pi}}

The factor 4πr2\tn{4\pi r^2} is the surface area of the sphere. Just as growing the square added two thin strips, growing the radius from rr to r+drr + \stp{dr} adds a thin shell of thickness dr\stp{dr} over the whole sphere. The volume of that shell is about its surface area times its thickness, 4πr2⋅dr\tn{4\pi r^2}\cdot\stp{dr}.

Summary

The linearization at a point
The tangent line
L(x)\displaystyle \tn{L(x)} =f(a)+f′(a) (x−a)\displaystyle = \cv{f(a)} + \tn{f'(a)}\,\stp{(x - a)}

To estimate f(x)\cv{f(x)} near a known point aa, find the height at xx on the tangent line at aa. Start at the known height f(a)\cv{f(a)}, and add the slope f′(a)\tn{f'(a)} times the step x−a\stp{x - a}. The formula needs only the value and the derivative at aa, so two given numbers f(a)\cv{f(a)} and f′(a)\tn{f'(a)} are enough to estimate f(x)\cv{f(x)}. When ff is given only through an equation, use implicit differentiation or an equation for a rate like dWdt\frac{dW}{dt} to find the derivative, then evaluate at aa to get f′(a)\tn{f'(a)}.

Choosing the point and the step
A step to the left
263 ⁣:a=27x−a=−1\displaystyle \begin{gathered} \cv{\sqrt[3]{26}}\colon \quad a = 27\\ \stp{x - a = -1} \end{gathered}
An angle step in radians
sin⁡31∘ ⁣:a=30∘x−a=1∘=π180\displaystyle \begin{gathered} \cv{\sin 31^\circ}\colon \quad a = 30^\circ\\ \stp{x - a = 1^\circ = \frac{\pi}{180}} \end{gathered}

The number you want to estimate tells you the function ff. Choose aa as the nearest input where you know f(a)\cv{f(a)} and f′(a)\tn{f'(a)} exactly, like a=27a = 27 for 263\cv{\sqrt[3]{26}}. If x<ax < a, keep the step x−a\stp{x - a} negative. For trig functions, the derivative formulas only hold in radians, so always convert an angle step from degrees to radians using 1∘=π180\stp{1^\circ = \frac{\pi}{180}}.

Overestimates and underestimates
y=x\htmlClass{lc-blue}{y = \sqrt{x}}
y=ex\htmlClass{lc-blue}{y = e^x}

When f′′(x)<0f''(x) < 0 near aa, the curve is concave down and lies below its tangent line on both sides of aa, so L(x)\tn{L(x)} is an overestimate. When f′′(x)>0f''(x) > 0 near aa, the curve is concave up and lies above its tangent line on both sides of aa, so L(x)\tn{L(x)} is an underestimate. Whether the step goes to the left or to the right doesn't change the result.

Differentials and the actual change
33
dx\htmlClass{lc-orange}{dx}
3 dx\htmlClass{lc-pink}{3\,}\htmlClass{lc-orange}{dx}
3 dx\htmlClass{lc-pink}{3\,}\htmlClass{lc-orange}{dx}
(dx)2\htmlClass{lc-green}{(dx)^2}
Along the tangent line and along the curve
dy=f′(x) dxΔy=f(x+dx)−f(x)\displaystyle \begin{aligned} \tn{dy} &= \tn{f'(x)}\,\stp{dx}\\ \cv{\Delta y} &= \cv{f(x + \stp{dx}) - f(x)} \end{aligned}

dx\stp{dx} is the change in xx. Over that run, dy=f′(x) dx\tn{dy} = \tn{f'(x)}\,\stp{dx} is the rise along the tangent line, while Δy=f(x+dx)−f(x)\cv{\Delta y} = \cv{f(x + \stp{dx}) - f(x)} is the actual rise along the curve. For a square of side 33 growing by dx\stp{dx}, the two side strips have total area 2⋅3 dx\tn{2 \cdot 3\,\stp{dx}}, which is dy\tn{dy}. The corner has area (dx)2\und{(dx)^2}, which is the difference Δy−dy\und{\Delta y - dy}. When dx\stp{dx} is small, the corner shrinks much faster than the strips, so dy≈Δy\tn{dy} \approx \cv{\Delta y}.

Estimating measurement error
Error in volume
dV=4πr2 dr\displaystyle \tn{dV} = \tn{4\pi r^2}\,\stp{dr}dVV=3⋅drr\displaystyle \frac{\tn{dV}}{\cv{V}} = 3\cdot\frac{\stp{dr}}{r}

Write the calculated quantity in terms of the measured value xx, and treat the measurement error as dx\stp{dx}. The estimated error in the quantity is the differential df=f′(x) dx\tn{df} = \tn{f'(x)}\,\stp{dx}, evaluated at the measured value. The relative error is the estimated error divided by the calculated value, dff(x)\frac{\tn{df}}{\cv{f(x)}}. For a sphere, the power rule brings down the exponent 33, so a 1%1\% error in the radius gives about a 3%3\% error in the volume.

6

Practice

Give decimals for estimates, and exact values where they exist.

1
Find the linearization L(x)L(x) of f(x)=x23f(x) = x^{\frac23} at a=8a = 8, and use it to estimate (8.06)23(8.06)^{\frac23}.

To estimate (8.06)23(8.06)^{\frac23}, what nearby number has an exact cube root?

The power rule gives f′(8)=23⋅8−13f'(8) = \frac23\cdot 8^{-\frac13}. What is 8−138^{-\frac13}?

The value at 8The cube root of 88 is 22, and squaring gives 4\cv{4}.
f(8)\displaystyle \cv{f(8)} =823\displaystyle = 8^{\frac23} =(83)2\displaystyle = \big(\sqrt[3]{8}\big)^2 =4\displaystyle = \cv{4}
The derivative at 8The power rule gives 23x−13\frac23 x^{-\frac13}, and 8−138^{-\frac13} is 183=12\frac{1}{\sqrt[3]{8}} = \frac12.
f′(x)\displaystyle f'(x) =23x−13,f′(8)\displaystyle = \frac23 x^{-\frac13}, \qquad \tn{f'(8)} =23⋅12\displaystyle = \frac23\cdot\frac12 =13\displaystyle = \tn{\frac13}
Put both into the linearization L(x)=f(a)+f′(a)(x−a)\tn{L(x)} = \cv{f(a)} + \tn{f'(a)}\stp{(x - a)}.
L(x)\displaystyle \tn{L(x)} =4+13(x−8)\displaystyle = \cv{4} + \tn{\frac13}\stp{(x - 8)}
Plug in x=8.06x = 8.06, so the step x−8\stp{x - 8} is 0.06\stp{0.06}.
(8.06)23≈L(8.06)\displaystyle (8.06)^{\frac23} \approx \tn{L(8.06)} =4+13(0.06)\displaystyle = \cv{4} + \tn{\frac13}\stp{(0.06)} =4.02\displaystyle = 4.02
2
(a) Use a linear approximation to estimate (1.02)10(1.02)^{10}. (b) Is your estimate an overestimate or an underestimate? Explain.

For f(x)=x10f(x) = x^{10}, what starting point aa is close to 1.021.02?

(b)\text{(b)}

Find f′′(x)f''(x). Is it positive or negative near 11?

Linearization at 1We choose a=1a = 1 because powers of 11 are easy, and 1.021.02 is only 0.02\stp{0.02} away.
f(x)=x10,a=1f(1)=1,f′(x)=10x9,f′(1)=10\displaystyle \begin{gathered} f(x) = x^{10}, \quad a = 1\\ \cv{f(1) = 1}, \qquad f'(x) = 10x^9, \quad \tn{f'(1) = 10} \end{gathered}
Multiply the step 0.02\stp{0.02} by the derivative 10\tn{10}, then add the starting value 1\cv{1}.
(1.02)10≈1+10(0.02)\displaystyle (1.02)^{10} \approx \cv{1} + \tn{10}\stp{(0.02)} =1.2\displaystyle = 1.2
Check the concavityThe second derivative 90x890x^8 is positive near 11, so the curve is concave up and the tangent line lies below it.
f′′(x)\displaystyle f''(x) =90x8>0  →  (1.02)10>1.2\displaystyle = 90x^8 > 0 \;\to\; (1.02)^{10} > 1.2
The true value is about 1.2191.219, so 1.21.2 is an underestimate.
(1.02)10≈1.219\displaystyle (1.02)^{10} \approx 1.219
3
The curve x2+xy+y2=7x^2 + xy + y^2 = 7 passes through the point (1,2)(1, 2). Use the tangent line at (1,2)(1, 2) to estimate yy when x=1.1x = 1.1.

Differentiate both sides with respect to xx to find dydx\frac{dy}{dx} at (1,2)(1, 2).

The starting yy is 22 and the step in xx is 0.10.1. How does yy change along the tangent line?

Differentiate implicitlyDifferentiate both sides with respect to xx, using the product rule on xyxy and the chain rule on y2y^2.
2x+y+xdydx+2ydydx\displaystyle 2x + y + x\frac{dy}{dx} + 2y\frac{dy}{dx} =0  →  dydx\displaystyle = 0 \;\to\; \frac{dy}{dx} =−2x+yx+2y\displaystyle = -\frac{2x + y}{x + 2y}
Plug in x=1x = 1 and y=2y = 2 to find the slope of the tangent line.
(1,2):dydx\displaystyle (1, 2): \qquad \tn{\frac{dy}{dx}} =−2+21+4\displaystyle = -\frac{2 + 2}{1 + 4} =−45\displaystyle = \tn{-\frac45}
The step in xx is 1.1−1=0.1\stp{1.1 - 1 = 0.1}, so multiply the slope by 0.1\stp{0.1} and add the starting yy-value.
y(1.1)≈2−45(0.1)\displaystyle y(1.1) \approx \cv{2} - \tn{\frac45}\stp{(0.1)} =2−0.08\displaystyle = \cv{2} - \tn{0.08} =1.92\displaystyle = 1.92
4
Let y=x3−2xy = x^3 - 2x. Find dydy. Then compute dydy and Δy\Delta y when x=2x = 2 and dx=0.1dx = 0.1.

The differential dydy is f′(x)f'(x) times what?

Δy\Delta y is the actual change f(x+dx)−f(x)f(x + dx) - f(x). What are x+dxx + dx and xx?

The differential dy\tn{dy} is the derivative times dx\stp{dx}.
dy\displaystyle \tn{dy} =(3x2−2) dx\displaystyle = \tn{(3x^2 - 2)}\,\stp{dx}
Plug in x=2x = 2 to get the slope 10\tn{10}, then multiply by dx=0.1\stp{dx = 0.1}.
x\displaystyle x =2, dx=0.1:dy\displaystyle = 2,\ \stp{dx = 0.1}: \qquad \tn{dy} =(3⋅4−2)(0.1)\displaystyle = \tn{(3\cdot 4 - 2)}\stp{(0.1)} =10(0.1)\displaystyle = \tn{10}\stp{(0.1)} =1\displaystyle = 1
The actual change Δy\cv{\Delta y} is the change on the curve from x=2x = 2 to x=2.1x = 2.1.
Δy\displaystyle \cv{\Delta y} =f(2.1)−f(2)\displaystyle = \cv{f(2.1) - f(2)} =(9.261−4.2)−(8−4)\displaystyle = (9.261 - 4.2) - (8 - 4) =5.061−4\displaystyle = 5.061 - 4 =1.061\displaystyle = 1.061
5
The circumference of a circle is measured as 8484 cm, with a possible error of 0.50.5 cm. (a) Use differentials to estimate the error in the computed area. (b) Estimate the relative error in the area.

The circumference CC was measured, so write the area AA in terms of CC.

The relative error in area is dAA\frac{dA}{A}. How does it simplify in terms of dCC\frac{dC}{C}?

Area in terms of circumferenceWrite the area directly in terms of CC because the measurement error 0.5 cm\stp{0.5}\text{ cm} is dC\stp{dC}.
r\displaystyle r =C2π  →  A\displaystyle = \frac{C}{2\pi} \;\to\; \cv{A} =πr2\displaystyle = \pi r^2 =C24π\displaystyle = \cv{\frac{C^2}{4\pi}}
Error in areaDifferentiate with respect to CC, then plug in C=84C = 84 and dC=0.5\stp{dC = 0.5}.
dA\displaystyle \tn{dA} =C2π dC\displaystyle = \tn{\frac{C}{2\pi}}\,\stp{dC} =84 (0.5)2π\displaystyle = \frac{84\,\stp{(0.5)}}{2\pi} =21π≈6.7 cm2\displaystyle = \frac{21}{\pi} \approx 6.7 \text{ cm}^2
Relative errorDivide dA\tn{dA} by A\cv{A}. Because A\cv{A} is proportional to C2C^2, the relative error in A\cv{A} is twice the relative error in CC.
dAA\displaystyle \frac{\tn{dA}}{\cv{A}} =C2π dCC24π\displaystyle = \frac{\tn{\frac{C}{2\pi}}\,\stp{dC}}{\cv{\frac{C^2}{4\pi}}} =2⋅dCC\displaystyle = 2\cdot\frac{\stp{dC}}{C} =2⋅0.584\displaystyle = 2\cdot\frac{\stp{0.5}}{84} =184≈1.2%\displaystyle = \frac{1}{84} \approx 1.2\%
6
A hemispherical dome 5050 m across gets a coat of paint 0.050.05 cm thick. Use differentials to estimate the amount of paint needed.

The paint is a thin shell over the dome. What is the radius of the dome if it is 50 m50\text{ m} across?

The thickness is dr=0.05 cmdr = 0.05\text{ cm}. How many meters is that?

The differential of volumeThe paint forms a thin shell over the hemisphere, so its volume is approximately the surface area 2πr2\tn{2\pi r^2} times the thickness dr\stp{dr}.
V=23πr3  →  dV\displaystyle \cv{V = \frac23\pi r^3} \;\to\; \tn{dV} =2πr2 dr\displaystyle = \tn{2\pi r^2}\,\stp{dr}
Radius and thicknessThe radius is half the diameter, and dividing 0.05 cm0.05\text{ cm} by 100100 converts the thickness to meters.
r\displaystyle r =502\displaystyle = \frac{50}{2} =25 m,dr=0.05 cm=0.0005 m\displaystyle = 25 \text{ m}, \qquad \stp{dr = 0.05 \text{ cm} = 0.0005 \text{ m}}
Plug r=25r = 25 and dr=0.0005\stp{dr = 0.0005} into dV\tn{dV}.
dV\displaystyle \tn{dV} =2π(25)2 (0.0005)\displaystyle = \tn{2\pi(25)^2}\,\stp{(0.0005)} =0.625π≈1.96 m3\displaystyle = 0.625\pi \approx 1.96 \text{ m}^3
7
(Challenge) At the start of 2010 a landfill holds 14001400 tons of waste. The amount WW, in tons, tt years after the start of 2010, is increasing and satisfies dWdt=W−30025\dfrac{dW}{dt} = \dfrac{W - 300}{25}. (a) Use the tangent line to the graph of WW at t=0t = 0 to estimate the amount of waste at the end of the first 3 months of 2010. (b) Find d2Wdt2\dfrac{d^2W}{dt^2} in terms of WW, and use it to decide whether your estimate in (a) is an underestimate or an overestimate.

Find W′(0)W'(0) from the differential equation. What fraction of a year is 33 months?

(b)\text{(b)}

Differentiate dWdt=W−30025\frac{dW}{dt} = \frac{W - 300}{25} with respect to tt using the chain rule.

Find the initial rate of changePlug W=1400W = 1400 into the differential equation to find W′(0)\tn{W'(0)}.
W′(0)\displaystyle \tn{W'(0)} =1400−30025\displaystyle = \frac{1400 - 300}{25} =44\displaystyle = \tn{44}
Estimate the waste at 3 monthsThree months is 14\stp{\frac14} of a year, so multiply the slope 44\tn{44} by 14\stp{\frac14} and add the starting amount 1400\cv{1400}.
3 months\displaystyle 3 \text{ months} =14 year:W(14)≈1400+44⋅14\displaystyle = \stp{\frac14} \text{ year}: \qquad W\Big(\frac14\Big) \approx \cv{1400} + \tn{44}\cdot\stp{\frac14} =1411 tons\displaystyle = 1411 \text{ tons}
Find the second derivativeDifferentiate W−30025\frac{W - 300}{25} with respect to tt using the chain rule, and substitute dWdt\frac{dW}{dt} back in.
d2Wdt2\displaystyle \frac{d^2W}{dt^2} =125dWdt\displaystyle = \frac{1}{25}\frac{dW}{dt} =W−300625\displaystyle = \frac{W - 300}{625}
Since WW starts at 14001400 and is increasing, W−300W - 300 is always positive, so the graph is concave up and the tangent line gives an underestimate.
W\displaystyle W ≥1400>300  →  d2Wdt2>0  →  W(14)>1411\displaystyle \ge 1400 > 300 \;\to\; \frac{d^2W}{dt^2} > 0 \;\to\; W\Big(\frac14\Big) > 1411