Skip to content

Unit 3: Differentiation: Composite, Implicit, and Inverse Functions

Differentiating Inverse Functions

Find the derivative of an inverse function at a point from the original function alone, using the matching point and the reciprocal of its slope.

Topic 3.3 · 9 min read

The function f(x)=x3+xf(x) = x^3 + x has an inverse gg, but you cannot easily write a formula for it: finding g(2)g(2) means solving x3+x=2x^3 + x = 2. Yet you can find the slope of gg at x=2x = 2 in one line. Since f(1)=2f(1) = 2, the inverse sends 2 back to 1, and since f′(1)=3+1=4f'(1) = 3 + 1 = 4, the slope of gg at 2 is 14\frac{1}{4}. This works whenever the slope of ff at the matching input is not 0, and the reason is the chain rule from topic 3.1 applied to the equation that defines an inverse.

What an inverse undoes

If ff sends aa to bb, its inverse f−1f^{-1} sends bb back to aa:

f−1(b)=aexactly whenf(a)=b.f^{-1}(b) = a \quad\text{exactly when}\quad f(a) = b.

For that to make sense, ff must be one-to-one: no two inputs share an output, so each bb comes from a single aa. A function whose derivative is positive everywhere on an interval is increasing there, so it never repeats an output and has an inverse on that interval. The same is true when the derivative is negative everywhere. Topic 5.3 proves the link between the sign of f′f' and increasing or decreasing.

Two facts follow directly from the definition.

  • Undoing. f(f−1(x))=xf\left(f^{-1}(x)\right) = x for every xx in the domain of f−1f^{-1}, and f−1(f(x))=xf^{-1}\left(f(x)\right) = x for every xx in the domain of ff.
  • Swapped points. The point (a,b)(a, b) is on the graph of ff exactly when (b,a)(b, a) is on the graph of f−1f^{-1}. Swapping coordinates is reflection across the line y=xy = x, so the two graphs are mirror images in that line.

The notation f−1(x)f^{-1}(x) means the inverse function, not the reciprocal 1f(x)\frac{1}{f(x)}. The two are unrelated: for f(x)=x3f(x) = x^3, the inverse is x3\sqrt[3]{x}, while the reciprocal is 1x3\frac{1}{x^3}.

Why the slopes are reciprocals

Reflecting across y=xy = x swaps horizontal and vertical. A segment that rises 4 units over a run of 1 becomes, after reflection, a segment that rises 1 over a run of 4. So if the tangent to ff at (a,b)(a, b) has slope mm, the reflected tangent, which touches f−1f^{-1} at (b,a)(b, a), has slope 1m\frac{1}{m}.

The chain rule turns that picture into a proof. Write g=f−1g = f^{-1} and start from the undoing equation,

f(g(x))=x.f\left(g(x)\right) = x.

Differentiate both sides. The left side is a composition with outside ff and inside gg, so by the chain rule

f′(g(x))⋅g′(x)=1,sog′(x)=1f′(g(x)).f'\left(g(x)\right)\cdot g'(x) = 1, \qquad\text{so}\qquad g'(x) = \frac{1}{f'\left(g(x)\right)}.

Dividing is allowed only when f′(g(x))≠0f'(g(x)) \neq 0. The same result comes from implicit differentiation (topic 3.2): y=g(x)y = g(x) means f(y)=xf(y) = x, and differentiating with respect to xx gives f′(y)dydx=1f'(y)\frac{dy}{dx} = 1. In Leibniz notation, with x=f(y)x = f(y), this reads dydx=1dx/dy\frac{dy}{dx} = \frac{1}{dx/dy}: the rate of the inverse is the reciprocal of the original rate, measured at the same point of the curve.

The step that takes thought is finding aa. You are given bb, an output of ff, and you need the input that produced it, so you solve f(a)=bf(a) = b. With a formula, try small integers first, as with x3+x=2x^3 + x = 2 in the opening; on a calculator question, use the calculator's solver. With a table, look for bb in the f(x)f(x) column, not the xx column. With a graph, find the point of the graph whose height is bb. For instance, if the graph of ff passes through (2,5)(2, 5) with a tangent slope of 3 there, then the graph of f−1f^{-1} passes through (5,2)(5, 2) with slope 13\frac{1}{3}.

When the inverse has no derivative

If f′(a)=0f'(a) = 0, the formula would divide by zero, and the inverse really does fail to have a derivative at bb. The picture explains it: the tangent to ff at (a,b)(a, b) is horizontal, and reflecting a horizontal line across y=xy = x gives a vertical one. So the graph of f−1f^{-1} has a vertical tangent at (b,a)(b, a).

The simplest case is f(x)=x3f(x) = x^3, which is one-to-one with inverse g(x)=x3g(x) = \sqrt[3]{x}. Since f′(0)=0f'(0) = 0, the cube root has a vertical tangent at the origin. Directly, g′(x)=13x−2/3=13x23g'(x) = \frac{1}{3}x^{-2/3} = \frac{1}{3\sqrt[3]{x^2}}, which grows without bound as x→0x \to 0.

The chain-rule derivation assumed g′(b)g'(b) exists and then found its value. Slopes of secant lines show why it exists when f′(a)≠0f'(a) \neq 0. Take a point xx near bb and let y=g(x)y = g(x), so f(y)=xf(y) = x. The secant slope of gg between bb and xx is

g(x)−g(b)x−b=y−af(y)−f(a)=1f(y)−f(a)y−a,\frac{g(x) - g(b)}{x - b} = \frac{y - a}{f(y) - f(a)} = \frac{1}{\dfrac{f(y) - f(a)}{y - a}},

the reciprocal of a secant slope of ff between aa and yy. The graph of gg is the unbroken graph of ff reflected across y=xy = x, so as x→bx \to b, y→ay \to a, and the bottom fraction approaches f′(a)f'(a). When f′(a)≠0f'(a) \neq 0, the whole expression approaches 1f′(a)\frac{1}{f'(a)}, so g′(b)g'(b) exists and has that value. When f′(a)=0f'(a) = 0, the secant slopes of gg grow without bound in size, which is the vertical tangent. So whenever you use the formula, the nonzero denominator is the condition that makes the answer valid.

Derivatives you already use, explained

Two derivatives from Unit 2 come straight from this rule.

The natural log. ln⁡x\ln x is the inverse of f(x)=exf(x) = e^x, and f′(x)=exf'(x) = e^x. With g(x)=ln⁡xg(x) = \ln x,

g′(x)=1f′(ln⁡x)=1eln⁡x=1x.g'(x) = \frac{1}{f'(\ln x)} = \frac{1}{e^{\ln x}} = \frac{1}{x}.

The denominator eln⁡xe^{\ln x} is never 0, which is why ln⁡x\ln x is differentiable on its entire domain, x>0x > 0.

The square root. On x≥0x \geq 0, x\sqrt{x} is the inverse of f(x)=x2f(x) = x^2, and f′(x)=2xf'(x) = 2x. So

ddxx=12x,\frac{d}{dx}\sqrt{x} = \frac{1}{2\sqrt{x}},

which matches the power rule with exponent 12\frac{1}{2}. At x=0x = 0 the denominator is f′(0)=0f'(0) = 0, and the square root has a vertical tangent there.

The same method gives the derivatives of arcsin⁡x\arcsin x, arctan⁡x\arctan x and the other inverse trigonometric functions, which topic 3.4 covers.

Worked examples

Worked exampleExample 1: a formula you can't invert

Let f(x)=x5+3x−2f(x) = x^5 + 3x - 2, and let gg be the inverse of ff. Explain why gg exists, and write an equation for the line tangent to the graph of gg at x=2x = 2.

Show solution

Why the inverse exists. f′(x)=5x4+3≥3>0f'(x) = 5x^4 + 3 \geq 3 > 0 for every xx, so ff is increasing on all real numbers and never takes the same value twice.

The point. The tangent line to gg at x=2x = 2 needs g(2)g(2), the input of ff that gives 2. Solve x5+3x−2=2x^5 + 3x - 2 = 2 by trying small integers: f(1)=1+3−2=2f(1) = 1 + 3 - 2 = 2. So g(2)=1g(2) = 1, and the point is (2,1)(2, 1).

The slope. f′(1)=5+3=8f'(1) = 5 + 3 = 8, which is not 0, so

g′(2)=1f′(g(2))=1f′(1)=18.g'(2) = \frac{1}{f'(g(2))} = \frac{1}{f'(1)} = \frac{1}{8}.

The line.

y=1+18(x−2).y = 1 + \frac{1}{8}(x - 2).

As a check, reflect the tangent line to ff at (1,2)(1, 2), which is y=2+8(x−1)y = 2 + 8(x - 1). Swapping xx and yy gives x=2+8(y−1)x = 2 + 8(y - 1), and solving for yy gives the same line.

Worked exampleExample 2: an inverse from a table (no calculator)

The function ff is differentiable and increasing for all real numbers, and gg is the inverse of ff. The table gives values of ff and f′f'.

x1234f(x)2367f′(x)122412\begin{array}{c|cccc} x & 1 & 2 & 3 & 4 \\ \hline f(x) & 2 & 3 & 6 & 7 \\ f'(x) & \tfrac{1}{2} & 2 & 4 & \tfrac{1}{2} \end{array}

(a) Find g′(3)g'(3).

(b) Find g′(6)g'(6).

(c) Let h(x)=x g(x)h(x) = x\,g(x). Find h′(2)h'(2).

(d) Write an equation for the line tangent to the graph of gg at x=7x = 7.

Show solution

Every part starts the same way: find the given number in the f(x)f(x) row, then read the xx above it. That xx is the value of gg.

(a) f(2)=3f(2) = 3, so g(3)=2g(3) = 2:

g′(3)=1f′(g(3))=1f′(2)=12.g'(3) = \frac{1}{f'(g(3))} = \frac{1}{f'(2)} = \frac{1}{2}.

(b) f(3)=6f(3) = 6, so g(6)=3g(6) = 3:

g′(6)=1f′(3)=14.g'(6) = \frac{1}{f'(3)} = \frac{1}{4}.

(c) hh is a product, so h′(x)=g(x)+x g′(x)h'(x) = g(x) + x\,g'(x). Since f(1)=2f(1) = 2, g(2)=1g(2) = 1 and g′(2)=1f′(1)=2g'(2) = \frac{1}{f'(1)} = 2. Then

h′(2)=g(2)+2 g′(2)=1+2(2)=5.h'(2) = g(2) + 2\,g'(2) = 1 + 2(2) = 5.

(d) f(4)=7f(4) = 7, so g(7)=4g(7) = 4 and g′(7)=1f′(4)=2g'(7) = \frac{1}{f'(4)} = 2. The tangent line is

y=4+2(x−7).y = 4 + 2(x - 7).

Notice that g′g' is large exactly where f′f' is small. Where ff is nearly flat, the reflected graph is nearly vertical.

Worked exampleExample 3: an exam-style inverse question (calculator for part c)

Let f(x)=2x+cos⁡xf(x) = 2x + \cos x.

(a) Explain why ff has an inverse function gg defined for all real numbers.

(b) Find g(1)g(1) and g′(1)g'(1).

(c) Find g′(3)g'(3). Give your answer to three decimal places.

(d) Show that 13≤g′(x)≤1\dfrac{1}{3} \leq g'(x) \leq 1 for every real number xx, and use this to check your answers to (b) and (c).

Show solution

(a) f′(x)=2−sin⁡xf'(x) = 2 - \sin x. Since sin⁡x≤1\sin x \leq 1, f′(x)≥1>0f'(x) \geq 1 > 0 for all xx, so ff is increasing and therefore one-to-one. Also 2x−1≤f(x)≤2x+12x - 1 \leq f(x) \leq 2x + 1, so f(x)f(x) takes arbitrarily large positive and negative values, and being continuous it takes every value in between. So gg is defined for all real numbers.

(b) Look for an input with f(a)=1f(a) = 1: f(0)=0+cos⁡0=1f(0) = 0 + \cos 0 = 1. So g(1)=0g(1) = 0, and since f′(0)=2−0=2f'(0) = 2 - 0 = 2,

g′(1)=1f′(g(1))=1f′(0)=12.g'(1) = \frac{1}{f'(g(1))} = \frac{1}{f'(0)} = \frac{1}{2}.

(c) The equation 2a+cos⁡a=32a + \cos a = 3 cannot be solved by algebra, so use the calculator's solver: a=g(3)≈1.42967a = g(3) \approx 1.42967. Store the full value rather than retyping the rounded one. Then

g′(3)=1f′(g(3))=12−sin⁡(1.42967…)≈0.990.g'(3) = \frac{1}{f'(g(3))} = \frac{1}{2 - \sin(1.42967\ldots)} \approx 0.990.

(d) For every xx, g′(x)=12−sin⁡(g(x))g'(x) = \frac{1}{2 - \sin(g(x))}. Because −1≤sin⁡(g(x))≤1-1 \leq \sin(g(x)) \leq 1, the denominator is between 1 and 3, so

13≤g′(x)≤1.\frac{1}{3} \leq g'(x) \leq 1.

Both earlier answers fit: 12\frac{1}{2} and 0.9900.990 are between 13\frac{1}{3} and 1. The value in (c) is close to the upper bound because g(3)≈1.430g(3) \approx 1.430 is close to π2≈1.571\frac{\pi}{2} \approx 1.571, where sin⁡\sin is nearly 1 and ff is at its flattest. A bound catches some slips but not all. Using f′(3)f'(3) instead of f′(g(3))f'(g(3)) would give 12−sin⁡3≈0.538\frac{1}{2 - \sin 3} \approx 0.538, which is also in range, so confirm the matching point too: f(1.42967)≈3.000f(1.42967) \approx 3.000.

Where students lose points

On the exam

Inverse-derivative questions give ff by a formula, a table, or a graph and ask for (f−1)′(b)(f^{-1})'(b) or a tangent line to f−1f^{-1}. They appear in multiple choice and as parts of free-response questions, with or without a calculator. The work a grader looks for is the matching point, written as f(a)=bf(a) = b or g(b)=ag(b) = a, then the reciprocal of f′(a)f'(a). If finding aa from f(a)=bf(a) = b needs a solver, the question will be calculator-active, and the answer is expected to three decimal places.

Check yourself

  1. Question 1

    Let f(x)=2x3+x−1f(x) = 2x^3 + x - 1, and let gg be the inverse of ff. What is g′(2)g'(2)?

    1. A

      17\dfrac{1}{7}

    2. B

      125\dfrac{1}{25}

    3. C

      77

    4. D

      −74-\dfrac{7}{4}

    Show answer

    A

    Solving f(a)=2f(a) = 2 gives a=1a = 1, since 2+1−1=22 + 1 - 1 = 2. Then g′(2)=1f′(1)=16+1=17g'(2) = \dfrac{1}{f'(1)} = \dfrac{1}{6 + 1} = \dfrac{1}{7}. Choice B is 1f′(2)\dfrac{1}{f'(2)}, the slope of ff at the wrong point; choice D is the derivative of 1f(x)\dfrac{1}{f(x)} at x=1x = 1, which confuses the inverse with the reciprocal.

  2. Question 2

    The function ff is differentiable and one-to-one, with f(4)=7f(4) = 7 and f′(4)=−23f'(4) = -\dfrac{2}{3}. Let gg be the inverse of ff. Find g′(7)g'(7), and say what its sign tells you about gg.

    Show answer

    g′(7)=−32g'(7) = -\dfrac{3}{2}; gg is decreasing at x=7x = 7.

    Since f(4)=7f(4) = 7, g(7)=4g(7) = 4, and g′(7)=1f′(4)=−32g'(7) = \dfrac{1}{f'(4)} = -\dfrac{3}{2}. Reflecting across y=xy = x never changes the sign of a slope, so the inverse of a decreasing function is decreasing.

  3. Question 3

    Let f(x)=(x−1)3+1f(x) = (x - 1)^3 + 1, which is one-to-one, and let gg be its inverse. Which statement about gg at x=1x = 1 is true?

    1. A

      g′(1)=0g'(1) = 0

    2. B

      g′(1)=13g'(1) = \dfrac{1}{3}

    3. C

      gg is not differentiable at x=1x = 1; its graph has a vertical tangent there.

    4. D

      gg is not defined at x=1x = 1.

    Show answer

    C

    f(1)=1f(1) = 1, so g(1)=1g(1) = 1, and f′(x)=3(x−1)2f'(x) = 3(x - 1)^2 gives f′(1)=0f'(1) = 0. The tangent to ff at (1,1)(1, 1) is horizontal, so its reflection, the tangent to gg at (1,1)(1, 1), is vertical. Choice A copies the slope of ff instead of taking its reciprocal; choice D fails because g(1)=1g(1) = 1 exists.

  4. Question 4

    The line tangent to the graph of a one-to-one differentiable function ff at x=3x = 3 is y=2x−1y = 2x - 1. Let gg be the inverse of ff. Write an equation for the line tangent to the graph of gg at x=5x = 5.

    Show answer

    y=3+12(x−5)y = 3 + \dfrac{1}{2}(x - 5)

    The tangent line touches ff at x=3x = 3, so f(3)=2(3)−1=5f(3) = 2(3) - 1 = 5 and f′(3)=2f'(3) = 2. Then g(5)=3g(5) = 3 and g′(5)=12g'(5) = \dfrac{1}{2}. Equivalently, swapping xx and yy in y=2x−1y = 2x - 1 reflects the whole tangent line across y=xy = x.

Go deeper