math.wiki
Analysis › Derivatives › Differentiability and the derivative rules, proved
Sign in

Differentiability and the derivative rules, proved

Calculus proved the product, quotient and chain rules for functions defined around a point, with its limits taken on trust. What is a derivative at the end of an interval, why does a derivative force continuity, and which form of the definition makes every rule a two-line proof?

Where you are 3 in Derivatives

Why this exists

The derivative at a point defined f'(a) for f defined around a, and The product rule, The quotient rule and The chain rule proved the rules, the last repairing a division by zero. Analysis needs derivatives at the ends of intervals such as [0, \infty), rules resting on the proved limits of Limits of functions, epsilon and delta and Continuity, defined and proved, and a form of the definition, f(x) - f(c) = \varphi(x)(x - c) with \varphi continuous at c, from which continuity and every rule follow with no case split. It also shows what a derivative can be: \displaystyle x^2 \sin\frac{1}{x} is differentiable everywhere, yet its derivative is not continuous.

Definition

Throughout, I is an interval with more than one point. Every c \in I is a limit point of I: if d \in I and d \ne c, the points c + t(d - c) with 0 < t < 1 lie in I and come as close to c as we like.

Derivative. Let f : I \to \mathbb{R} and c \in I. The difference quotient of f at c is

Q(x) = \frac{f(x) - f(c)}{x - c} \quad \text{on } I \setminus \{c\}

If \lim_{x \to c} Q(x) exists (Limits of functions, epsilon and delta), f is differentiable at c and the limit is the derivative f'(c). If f is differentiable at every point of I, it is differentiable on I, with derivative f' : I \to \mathbb{R}; f'' is the derivative of f', and f^{(n)} the nth derivative. f is continuously differentiable on I if f' exists and is continuous on I.

Only points of I are tested, so at an endpoint of I the derivative is one-sided because of the domain. If I has points on both sides of c, the left derivative f'_-(c) and right derivative f'_+(c) are the left-hand and right-hand limits of Q at c, when they exist.

Intuition

Q(x) is the slope of the secant through (c, f(c)) and (x, f(x)). Differentiability says the secant-slope function, defined except at c, can be filled in at c so that it is continuous there; the filling value is the derivative (Figure 1). Calling the filled-in function \varphi and clearing the denominator gives f(x) - f(c) = \varphi(x)(x - c). Continuity at c asks only that f(x) - f(c) be small; differentiability asks that it be a continuous multiple of x - c.

x φ 1 2 2.5 5 10 15
\varphi(1)=5
f'(2)=\varphi(2)=10
\varphi(2.5)=13.25
Figure 1The slope function \varphi(x) = x^2 + 2x + 2 of f(x) = x^3 - 2x at c = 2 (Example 1). At x \ne 2 its height is the slope of the secant from (2, 4) to (x, f(x)); filled in at 2, it is continuous, and its value there is f'(2).

Results

Throughout, f, g : I \to \mathbb{R} and c \in I.

Theorem 1 (slope-function form). f is differentiable at c if and only if there is a function \varphi : I \to \mathbb{R}, continuous at c, with f(x) - f(c) = \varphi(x)(x - c) for every x \in I. Such a \varphi is unique, and \varphi(c) = f'(c). (This form is often named for Carathéodory.)

Theorem 2 (linear-approximation form). Let L \in \mathbb{R}. Then f'(c) = L if and only if

\lim_{x \to c} \frac{f(x) - f(c) - L(x - c)}{x - c} = 0

So the tangent line y = f(c) + f'(c)(x - c) is the only line through (c, f(c)) whose error, divided by x - c, tends to 0; for slope M the ratio tends to f'(c) - M.

Theorem 3 (differentiable implies continuous). If f is differentiable at c, then f is continuous at c. The converse is false.

Theorem 4 (algebra). If f and g are differentiable at c and k \in \mathbb{R}, then f + g, kf and fg are differentiable at c, with (f + g)'(c) = f'(c) + g'(c), (kf)'(c) = kf'(c) and

(fg)'(c) = f'(c)g(c) + f(c)g'(c)

If also g(x) \ne 0 for every x \in I, then \displaystyle \frac{f}{g} is differentiable at c and

\left(\frac{f}{g}\right)'(c) = \frac{f'(c)g(c) - f(c)g'(c)}{g(c)^2}

Theorem 5 (chain rule). Let J be an interval with more than one point, g : I \to \mathbb{R} with g(I) \subseteq J, and f : J \to \mathbb{R}. If g is differentiable at c and f at g(c), then f \circ g is differentiable at c and (f \circ g)'(c) = f'(g(c))\,g'(c).

Theorem 6 (changing the domain). Let J \subseteq I be an interval with more than one point, and c \in J. (a) If f is differentiable at c, so is f|_J, with the same derivative. (b) If J contains every point of I within some distance \alpha > 0 of c and f|_J is differentiable at c, so is f, with the same derivative. (c) If I has points on both sides of c, then f is differentiable at c if and only if f'_-(c) and f'_+(c) exist and are equal.

Why it is true

Theorem 1. Proof. Suppose f'(c) exists. Let \varphi = Q on I \setminus \{c\} and \varphi(c) = f'(c). For x \ne c the identity is Q(x) times x - c; at x = c both sides are 0. Since c is a limit point of I and \lim_{x \to c} \varphi(x) = f'(c) = \varphi(c), \varphi is continuous at c (Theorem 2 of Continuity, defined and proved).

Conversely, given \varphi, dividing by x - c gives Q(x) = \varphi(x) for x \ne c. By the same theorem \lim_{x \to c} \varphi(x) = \varphi(c), and a limit at c ignores the value at c, so Q(x) \to \varphi(c). Any two such functions equal Q off c and f'(c) at c. \square

Theorem 2. Proof. For x \ne c the quotient is Q(x) - L, which tends to 0 exactly when Q(x) \to L (Theorem 4 of Limits of functions, epsilon and delta). For slope M it is Q(x) - M \to f'(c) - M. \square

Theorem 3. Proof. By Theorem 1, f(x) = f(c) + \varphi(x)(x - c), and \varphi and x - c are continuous at c, so f is (Theorem 4 of Continuity, defined and proved). For the converse, \lvert x \rvert is continuous at 0, but its quotient there is 1 for x > 0 and -1 for x < 0, so it has no limit (Theorem 7 of Limits of functions, epsilon and delta). \square

Theorem 4. Proof. By Theorem 1 write f(x) - f(c) = \varphi(x)(x - c) and g(x) - g(c) = \psi(x)(x - c), with \varphi and \psi continuous at c. So f + g and kf have the slope functions \varphi + \psi and k\varphi. Adding and subtracting f(x)g(c),

\begin{aligned} &f(x)g(x) - f(c)g(c) \\ &= f(x)\big(g(x) - g(c)\big) + \big(f(x) - f(c)\big)g(c) \\ &= \big(f(x)\psi(x) + \varphi(x)g(c)\big)(x - c) \end{aligned}

Each factor in front of x - c is continuous at c (for the product, f is continuous at c by Theorem 3), so by Theorem 1 the derivatives are their values at c: \varphi(c) + \psi(c), k\varphi(c) and f(c)g'(c) + f'(c)g(c).

For the quotient, take f = 1 first. Since g never vanishes,

\frac{1}{g(x)} - \frac{1}{g(c)} = -\frac{\psi(x)}{g(x)g(c)}\,(x - c)

with a factor continuous at c, as g is continuous at c and g(c) \ne 0. So \displaystyle \left(\frac{1}{g}\right)'(c) = -\frac{g'(c)}{g(c)^2}, and the product rule applied to \displaystyle f \cdot \frac{1}{g} gives the stated formula. \square

Theorem 5. Proof. Let d = g(c). By Theorem 1, f(y) - f(d) = \varphi(y)(y - d) for y \in J, with \varphi continuous at d, and g(x) - g(c) = \psi(x)(x - c) for x \in I, with \psi continuous at c. Put y = g(x):

\begin{aligned} f(g(x)) - f(g(c)) &= \varphi(g(x))\big(g(x) - g(c)\big) \\ &= \varphi(g(x))\,\psi(x)\,(x - c) \end{aligned}

g is continuous at c (Theorem 3), so \varphi \circ g is continuous at c (Theorem 5 of Continuity, defined and proved), and so is (\varphi \circ g)\psi. Theorem 1 gives (f \circ g)'(c) = \varphi(d)\psi(c) = f'(g(c))\,g'(c). \square

Nothing divides by g(x) - g(c), so the case repaired in The chain rule needs no attention; its error function is \varphi - f'(d).

Theorem 6. Proof. The quotient of f|_J at c is the restriction of Q to J \setminus \{c\}, and c is a limit point of J. So (a) and (b) are Theorem 6(a) and 6(b) of Limits of functions, epsilon and delta applied to Q. In (c), c is a limit point of both I \cap (-\infty, c) and I \cap (c, \infty), and Theorem 7 there applies. \square

Worked examples

Example 1

Slope functions of a polynomial and a rational function

For f(x) = x^3 - 2x at c = 2 and for g(x) = \dfrac{1}{x^2 + 1} at c = 1, find the slope function \varphi of Theorem 1 and read off the derivative.

Solution

f(2) = 4, and 2 is a root of x^3 - 2x - 4, so x - 2 is a factor:

f(x) - f(2) = (x^2 + 2x + 2)(x - 2)

So \varphi(x) = x^2 + 2x + 2, a polynomial, continuous at 2, and f'(2) = \varphi(2) = 10. In Figure 1, \varphi(1) = 5 and \varphi(2.5) = 13.25 are the slopes of two secants.

For g, \displaystyle g(1) = \frac{1}{2}, and over a common denominator

g(x) - \frac{1}{2} = \frac{1 - x^2}{2(x^2 + 1)} = -\frac{x + 1}{2(x^2 + 1)}\,(x - 1)

The factor is a rational function continuous at 1, so \displaystyle g'(1) = -\frac{2}{4} = -\frac{1}{2}.

Example 2

A derivative at an endpoint

(a) Let \displaystyle f(x) = x\sqrt{x} on [0, \infty). Find f'(0) and f'(c) for c > 0, and show that \displaystyle r(x) = \sqrt{x} is not differentiable at 0. (b) Compare \lvert x \rvert on [0, 1] with \lvert x \rvert on [-1, 1] at 0.

Solution

(a) At 0, \displaystyle f(x) - f(0) = \sqrt{x}\,(x - 0), and \displaystyle \sqrt{x} is continuous at 0 on [0, \infty) (Example 2 of Continuity, defined and proved). By Theorem 1, \displaystyle f'(0) = \sqrt{0} = 0, a derivative made one-sided by the domain.

For c > 0, rationalizing gives \displaystyle \sqrt{x} - \sqrt{c} = \frac{1}{\sqrt{x} + \sqrt{c}}\,(x - c), with a factor continuous at c, so \displaystyle r'(c) = \frac{1}{2\sqrt{c}}. The product rule gives

f'(c) = 1 \cdot \sqrt{c} + c \cdot \frac{1}{2\sqrt{c}} = \frac{3}{2}\sqrt{c}

At 0 the quotient of r is \displaystyle \frac{1}{\sqrt{x}}, equal to n at \displaystyle x_n = \frac{1}{n^2}; it diverges, so the limit fails (Theorem 3 of Limits of functions, epsilon and delta).

(b) On [0, 1] the quotient at 0 is \displaystyle \frac{x}{x} = 1, so the derivative is 1. On [-1, 1] the points x < 0 give -1, and there is none. Shrinking the domain can create a derivative, but by Theorem 6(a) never destroys or changes one.

Example 3

The tangent line as the best line

For f(x) = x^3 at c = 1, write f(x) = 1 + 3(x - 1) + E(x)(x - 1). Show that E(x) \to 0 by finding a \delta that makes \lvert E(x) \rvert < \varepsilon, and for \varepsilon = 0.01 compare it with the largest such \delta. What happens with slope 2.9?

Solution

x^3 - 1 - 3(x - 1) = x^3 - 3x + 2 = (x - 1)^2(x + 2), so E(x) = (x - 1)(x + 2), which is \varphi - 3 for the slope function \varphi(x) = x^2 + x + 1.

Given \varepsilon > 0, let \displaystyle \delta = \min\left\{1, \frac{\varepsilon}{4}\right\}. If \lvert x - 1 \rvert < \delta, then 0 < x < 2, so \lvert x + 2 \rvert < 4 and \lvert E(x) \rvert < 4\delta \le \varepsilon. So E(x) \to 0 and f'(1) = 3 by Theorem 2. For \varepsilon = 0.01, \delta = 0.0025.

Exactly, \lvert x^2 + x - 2 \rvert < 0.01 near 1 means

\frac{-1 + \sqrt{8.96}}{2} < x < \frac{-1 + \sqrt{9.04}}{2}

The right end is nearer, at \displaystyle \frac{\sqrt{9.04} - 3}{2} \approx 0.003330, against \displaystyle \frac{3 - \sqrt{8.96}}{2} \approx 0.003337. So the largest \delta is about 0.003330.

With slope 2.9 the ratio is x^2 + x + 1 - 2.9 \to 0.1, not 0.

Example 4

Aisha’s snowboard jump

Aisha shapes the jumps at a snowboard park, and Jin rides them. With x the horizontal distance in meters, her profile has height

h(x) = \begin{cases} 0, & 0 \le x \le 1 \\ a(x - 1)^2, & 1 < x \le 3 \end{cases}

meters, with the takeoff edge, the lip, at x = 3. (a) Find a for a lip 0.8 meters high, and show that h is differentiable on [0, 3]. (b) Find the slope at the lip and the takeoff angle. (c) Is h' differentiable at 1? (d) Jin suggests a straight ramp, g(x) = 0 on [0, 1] and g(x) = 0.4(x - 1) on (1, 3]. Is g differentiable on [0, 3]?

Solution

(a) h(3) = 4a = 0.8, so a = 0.2. On [1, 3], h agrees with p(x) = 0.2(x - 1)^2, and [1, 3] contains every point of [0, 3] within c - 1 of any c \in (1, 3]; so Theorem 6 ((a) for p, (b) for h) gives h'(c) = p'(c) = 0.4(c - 1). Likewise h = 0 on [0, 1] gives h'(c) = 0 for c \in [0, 1). At 1 the left quotient is 0 and the right one is \displaystyle \frac{0.2(x - 1)^2}{x - 1} = 0.2(x - 1) \to 0, so h'(1) = 0 by Theorem 6(c). Thus h' = 0 on [0, 1] and h'(x) = 0.4(x - 1) on (1, 3], continuous on [0, 3].

(b) h'(3) = 0.8, a derivative at an endpoint: the snow ends at the lip. The takeoff angle is \arctan 0.8 \approx 38.7^\circ.

(c) No. The quotient of h' at 1 is 0 on the left and 0.4 on the right, so h''(1) does not exist, though h' is continuous.

(d) No. g is continuous at 1, but g'_-(1) = 0 and g'_+(1) = 0.4: a corner, where the board’s angle changes at once (Figure 2). Its takeoff angle is \arctan 0.4, about 21.8^\circ.

x y 1 2 3 0.8 h g
Figure 2Example 4. Aisha’s profile h (solid) leaves the flat snow with slope 0; Jin’s straight ramp g (dashed) leaves it with a corner at x = 1. Both reach the lip at height 0.8.
Example 5

Differentiable at exactly one point

Let q(x) = x^2 for x \in \mathbb{Q} and q(x) = -x^2 otherwise. Prove that q is differentiable at 0 and at no other point.

Solution

Proof. Write q(x) - q(0) = \varphi(x)\,x with \displaystyle \varphi(x) = \frac{q(x)}{x} for x \ne 0 and \varphi(0) = 0. Since \lvert q(x) \rvert = x^2, \lvert \varphi(x) \rvert = \lvert x \rvert, so \varphi is continuous at 0 and q'(0) = 0 by Theorem 1.

Let c \ne 0. If c is rational, the irrationals \displaystyle x_n = c + \frac{\sqrt{2}}{n} tend to c and q(x_n) = -x_n^2 \to -c^2 \ne q(c). If c is irrational, rationals \displaystyle r_n \in \left(c, c + \frac{1}{n}\right) (The Archimedean property and density of the rationals) tend to c and q(r_n) \to c^2 \ne q(c). Either way q is discontinuous at c (Theorem 3 of Continuity, defined and proved), so by Theorem 3 it is not differentiable there. \square

Example 6

A derivative that is not continuous

Let \displaystyle s(x) = x^2 \sin\frac{1}{x} for x \ne 0 and s(0) = 0. Prove that s is differentiable on \mathbb{R} and that s' is not continuous at 0.

Solution

Proof. At 0, s(x) - s(0) = \varphi(x)\,x with \displaystyle \varphi(x) = x\sin\frac{1}{x} and \varphi(0) = 0. As \lvert \varphi(x) \rvert \le \lvert x \rvert, \varphi is continuous at 0 and s'(0) = 0. The graph is trapped between y = x^2 and y = -x^2, both tangent to the axis at 0 (Figure 3).

x y
y=x^2
y=-x^2
Figure 3Example 6. \displaystyle s(x) = x^2 \sin\frac{1}{x} between y = x^2 and y = -x^2, drawn for -0.25 \le x \le 0.25 with the vertical scale doubled. Both parabolas are tangent to the axis at 0, so s'(0) = 0.

Near c \ne 0, s is built from x^2, \displaystyle \frac{1}{x} and \sin. With \sin' = \cos from Derivatives of trig, exponential and log functions (proved there from \displaystyle \frac{\sin h}{h} \to 1, The squeeze theorem and the limit of sine x over x), the product and chain rules give

s'(x) = 2x\sin\frac{1}{x} - \cos\frac{1}{x} \quad (x \ne 0)

Let \displaystyle x_k = \frac{1}{2k\pi} and \displaystyle y_k = \frac{1}{(2k + 1)\pi} for k \ge 1. Both tend to 0, and \sin vanishes at multiples of \pi, so

\begin{aligned} s'(x_k) &= -\cos(2k\pi) = -1 \\ s'(y_k) &= -\cos\big((2k + 1)\pi\big) = 1 \end{aligned}

Neither tends to s'(0) = 0, so s' is discontinuous at 0 (Theorem 3 of Continuity, defined and proved). \square

So s is differentiable but not continuously differentiable. Its derivative is bounded on [-1, 1], by 2\lvert x \rvert + 1 \le 3: the discontinuity is an oscillation, not a jump.

Boundaries

  • A point says nothing about its neighbors. Example 5 is differentiable at 0 and discontinuous everywhere else.
  • Continuous is not differentiable. Weierstrass gave a function continuous everywhere and differentiable nowhere, beyond this lesson.
  • A derivative can be discontinuous but cannot jump. Darboux’s theorem in The mean value theorem and its consequences gives every derivative the intermediate value property, so a step such as 0 on [0, 1] and 0.4 on (1, 3], the slope of Jin’s ramp away from its corner, is the derivative of no function on [0, 3].
  • The domain matters. \lvert x \rvert has a derivative at 0 on [0, 1] and none on [-1, 1] (Example 2); on \mathbb{Z} the quotient has nothing to approach.
  • The quotient rule asks g \ne 0 on I. If only g(c) \ne 0, continuity gives a subinterval around c where g \ne 0, and Theorem 6(b) applies there.

Where this goes next

The mean value theorem and its consequences starts from the definition: at an interior maximum the left quotients are \ge 0 and the right ones \le 0, forcing f'(c) = 0, and Rolle’s theorem, the mean value theorem and Darboux’s theorem follow. Taylor's theorem with remainder extends Theorem 2 to polynomials of higher degree. In several variables, where there is no quotient, Theorem 2 becomes the definition itself, as in Tangent planes and linear approximation.

Exercises

Slope functions and the best line

  1. Find the slope function \varphi of f(x) = x^2 - 5x at c = 3, and read off f'(3).
  2. Let g(x) = \dfrac{2x + 1}{x - 4} on (4, \infty). Find its slope function at c = 5 and g'(5).
  3. For f(x) = \dfrac{1}{x + 1} on (-1, \infty) at c = 1, find E(x) with \displaystyle f(x) = \frac{1}{2} - \frac{1}{4}(x - 1) + E(x)(x - 1). Show that \delta = \min\{1, 4\varepsilon\} makes \lvert E(x) \rvert < \varepsilon, and find the largest \delta for \varepsilon = 0.01.

Endpoints, domains and joins

  1. Let \displaystyle f(x) = x\sqrt{2 - x} on [0, 2]. Find f'(0) and f'(1), and show that f is not differentiable at 2.
  2. On [0, 9] let f(x) = ax^2 + b for 0 \le x \le 2 and \displaystyle f(x) = \sqrt{x + 7} for 2 < x \le 9. Find a and b making f differentiable on [0, 9], and find f'(0) and f'(9).
  3. For a smaller jump, Aisha uses h(x) = 0 on [0, 1], h(x) = a(x - 1)^2 on (1, 2] and h(x) = b + m(x - 2) on (2, 2.5], with the lip at x = 2.5 one meter high. Find a, b and m making h differentiable on [0, 2.5], find Jin’s takeoff angle, and decide whether h is twice differentiable on [0, 2.5].
  4. Find every point where f(x) = \lvert x^3 - x \rvert is not differentiable, and its one-sided derivatives there.

Oscillation

  1. Let \displaystyle t(x) = x^3 \sin\frac{1}{x} for x \ne 0 and t(0) = 0. Show that t is continuously differentiable on \mathbb{R} and that t''(0) does not exist.
  2. For p > 0 let \displaystyle w(x) = x^p \sin\frac{1}{x} for x > 0 and w(0) = 0, on [0, \infty). For which p is w continuous at 0? Differentiable at 0? Continuously differentiable on [0, \infty)?

Proofs

  1. Let g be differentiable at c with g(c) = 0, and let f : I \to \mathbb{R} be continuous at c. Prove that fg is differentiable at c with (fg)'(c) = f(c)g'(c), and show that continuity of f cannot be dropped.
  2. Suppose that for some M \ge 0, \lvert f(x) - f(c) \rvert \le M(x - c)^2 for every x \in I. Prove that f'(c) = 0.
  3. Prove that x^n has derivative nx^{n - 1} on \mathbb{R} for every integer n \ge 1, and on every interval not containing 0 for every integer n \le -1.
  4. Let f be differentiable at c, where I contains an open interval around c. Prove that \lim_{h \to 0} \dfrac{f(c + h) - f(c - h)}{2h} = f'(c), and show that the converse fails.
  5. Let J be an interval, f : I \to J a bijection differentiable at c with f'(c) \ne 0, and f^{-1} continuous at d = f(c). Prove that \displaystyle (f^{-1})'(d) = \frac{1}{f'(c)} (compare Derivatives of inverse functions).
  6. Let f : \mathbb{R} \to \mathbb{R} satisfy \lvert f(x) - f(y) \rvert \le (x - y)^2 for all x and y. Prove, without the mean value theorem, that f is constant.
  7. Refute: “If f is differentiable on \mathbb{R} and f'(0) > 0, then f is increasing on some interval (-\alpha, \alpha).” Use \displaystyle f(x) = x + 2x^2\sin\frac{1}{x}, f(0) = 0.
Answers
  1. \varphi(x) = x - 2 and f'(3) = 1. f(3) = -6, and x^2 - 5x + 6 = (x - 2)(x - 3).
  2. \varphi(x) = -\dfrac{9}{x - 4} and g'(5) = -9. g(5) = 11, and \displaystyle g(x) - 11 = \frac{-9x + 45}{x - 4}, with a factor continuous at 5.
  3. E(x) = \dfrac{x - 1}{4(x + 1)}; the largest \delta is \displaystyle \frac{1}{13} \approx 0.0769. Over 4(x + 1), \displaystyle f(x) - \frac{1}{2} + \frac{x - 1}{4} = \frac{(x - 1)^2}{4(x + 1)}. If \lvert x - 1 \rvert < \delta \le 1, then x + 1 > 1 and \displaystyle \lvert E(x) \rvert < \frac{\delta}{4} \le \varepsilon. For \varepsilon = 0.01, \lvert x - 1 \rvert < 0.04(x + 1) means \displaystyle \frac{12}{13} < x < \frac{13}{12}, at distances \displaystyle \frac{1}{13} and \displaystyle \frac{1}{12} from 1.
  4. \displaystyle f'(0) = \sqrt{2} and \displaystyle f'(1) = \frac{1}{2}. At 0 the quotient is \displaystyle \sqrt{2 - x} \to \sqrt{2}. For 0 < x < 2, \displaystyle f'(x) = \sqrt{2 - x} - \frac{x}{2\sqrt{2 - x}} = \frac{4 - 3x}{2\sqrt{2 - x}}. At 2 the quotient is \displaystyle -\frac{x}{\sqrt{2 - x}}; at \displaystyle x_n = 2 - \frac{1}{n^2} it is -nx_n, which diverges.
  5. \displaystyle a = \frac{1}{24}, \displaystyle b = \frac{17}{6}, f'(0) = 0, \displaystyle f'(9) = \frac{1}{8}. A derivative at 2 forces continuity: 4a + b = 3. The left quotient a(x + 2) tends to 4a, the right derivative is \displaystyle \frac{1}{2\sqrt{9}} = \frac{1}{6}, and Theorem 6(c) gives \displaystyle a = \frac{1}{24}, \displaystyle b = 3 - \frac{1}{6}. Then f'(0) = 2a \cdot 0 and \displaystyle f'(9) = \frac{1}{2\sqrt{16}}.
  6. \displaystyle a = b = \frac{1}{2}, m = 1, an angle of 45^\circ; not twice differentiable. At 1 both one-sided derivatives are 0 for every a. At 2, continuity gives b = a and equal one-sided derivatives give m = 2a; the lip height is b + 0.5m = 2a = 1. Then h'(2.5) = 1 and \arctan 1 = 45^\circ. h' is 0, x - 1, 1 on the three pieces, with one-sided derivatives 0 and 1 at 1, and 1 and 0 at 2, so h'' fails at both joins.
  7. At -1, 0 and 1: f'_-(-1) = -2, f'_+(-1) = 2, f'_-(0) = -1, f'_+(0) = 1, f'_-(1) = -2, f'_+(1) = 2. u(x) = x^3 - x is negative on (-\infty, -1) and (0, 1) and positive on (-1, 0) and (1, \infty), so near any other point f = \pm u, differentiable. At a root r, \displaystyle \frac{u(x)}{x - r} \to u'(r), which is 2, -1, 2 at -1, 0, 1, with the sign flipped on the side where u < 0.
  8. t'(0) = 0, since \displaystyle \left\lvert \frac{t(x)}{x} \right\rvert \le x^2. For x \ne 0, \displaystyle t'(x) = 3x^2 \sin\frac{1}{x} - x\cos\frac{1}{x}, and \lvert t'(x) \rvert \le 3x^2 + \lvert x \rvert \to 0, so t' is continuous at 0 too. The quotient \displaystyle \frac{t'(x)}{x} = 3x\sin\frac{1}{x} - \cos\frac{1}{x} is -1 at \displaystyle \frac{1}{2k\pi} and 1 at \displaystyle \frac{1}{(2k + 1)\pi}, so it has no limit at 0.
  9. Continuous for every p > 0, differentiable at 0 exactly when p > 1, continuously differentiable exactly when p > 2. \lvert w(x) \rvert \le x^p. The quotient at 0 is \displaystyle x^{p - 1}\sin\frac{1}{x}: it tends to 0 if p > 1; if p \le 1 it is 0 at \displaystyle \frac{1}{k\pi} but u_k^{p - 1}, which is 1 or tends to \infty, at \displaystyle u_k = \frac{1}{2k\pi + \pi/2}. For p > 1 and x > 0, \displaystyle w'(x) = px^{p - 1}\sin\frac{1}{x} - x^{p - 2}\cos\frac{1}{x}. If p > 2 both terms tend to 0 = w'(0); if 1 < p \le 2, at \displaystyle x_k = \frac{1}{2k\pi} it is -x_k^{p - 2}, which is -1 or tends to -\infty.
  10. Proof. By Theorem 1, g(x) = \psi(x)(x - c) with \psi continuous at c and \psi(c) = g'(c). Then f(x)g(x) - f(c)g(c) = f(x)\psi(x)(x - c), with f\psi continuous at c, so (fg)'(c) = f(c)g'(c) by Theorem 1. Without continuity: at c = 0 let g(x) = x, f(x) = 1 for x \ge 0 and f(x) = -1 for x < 0; then fg = \lvert x \rvert. \square
  11. Proof. Let \varphi = Q off c and \varphi(c) = 0. Then f(x) - f(c) = \varphi(x)(x - c) and \lvert \varphi(x) \rvert \le M\lvert x - c \rvert, so \lvert \varphi(x) \rvert < \varepsilon whenever \displaystyle \lvert x - c \rvert < \frac{\varepsilon}{M + 1}. Thus \varphi is continuous at c and f'(c) = 0 by Theorem 1. \square
  12. Proof. For n = 1 the quotient of x is 1. If (x^n)' = nx^{n - 1}, the product rule on x \cdot x^n gives x^n + nx^n = (n + 1)x^n; Proof by induction covers n \ge 1. For n = -m with m \ge 1, on an interval without 0, the quotient rule gives \displaystyle \left(\frac{1}{x^m}\right)' = \frac{-mx^{m - 1}}{x^{2m}} = nx^{n - 1}. \square
  13. Proof. With the slope function \varphi at c, f(c \pm h) - f(c) = \pm\varphi(c \pm h)h, so for small h \ne 0 the quotient is \displaystyle \frac{1}{2}\big(\varphi(c + h) + \varphi(c - h)\big). Both h \mapsto \varphi(c \pm h) are continuous at 0, so the quotient tends to \varphi(c) = f'(c). Converse: for \lvert x \rvert at 0 the quotient is \displaystyle \frac{\lvert h \rvert - \lvert -h \rvert}{2h} = 0, though there is no derivative. \square
  14. Proof. Let f(x) - f(c) = \varphi(x)(x - c) with \varphi continuous at c. For x \ne c, f(x) \ne f(c) since f is injective, so \varphi(x) \ne 0; and \varphi(c) = f'(c) \ne 0. Putting x = f^{-1}(y),
f^{-1}(y) - f^{-1}(d) = \frac{1}{\varphi(f^{-1}(y))}\,(y - d)

\varphi \circ f^{-1} is continuous at d (Theorem 5 of Continuity, defined and proved) and never 0, so its reciprocal is continuous at d, and Theorem 1 gives \displaystyle (f^{-1})'(d) = \frac{1}{f'(c)}. \square 15. Proof. Let x < y, n \ge 1 and \displaystyle x_k = x + \frac{k(y - x)}{n} for 0 \le k \le n. By the triangle inequality,

\begin{aligned} \lvert f(y) - f(x) \rvert &\le \sum_{k=1}^{n} \lvert f(x_k) - f(x_{k-1}) \rvert \\ &\le n\left(\frac{y - x}{n}\right)^2 = \frac{(y - x)^2}{n} \end{aligned}

A number \ge 0 below \displaystyle \frac{(y - x)^2}{n} for every n \ge 1 is 0 (The Archimedean property and density of the rationals), so f(y) = f(x). \square 16. Proof. \displaystyle f(x) = \big(1 + 2x\sin\frac{1}{x}\big)x with a factor continuous at 0 of value 1, so f'(0) = 1. For x \ne 0, \displaystyle f'(x) = 1 + 4x\sin\frac{1}{x} - 2\cos\frac{1}{x}, so f'(x_k) = -1 at \displaystyle x_k = \frac{1}{2k\pi}. As the quotient at x_k tends to -1, some z > x_k, as close to x_k as we like, has f(z) < f(x_k). Every (-\alpha, \alpha) contains such x_k < z, so f is increasing on none of them. \square

math.wiki · CC-BY-SA-4.0