普林斯顿微积分读本 II

Apr 12, 2021    链式法则   导数   简谐运动   隐函数   复利   双曲函数   反函数  

§微分

6.1 通过定义求导

定义: $$ f’(x)=\lim_{h\to 0}\frac{f(x+h)-f(x)}{h} $$

例如,$\frac{\mathrm d}{\mathrm d x}({1 \over x}) = -\frac{1}{x^2}$ 和 $\frac{\mathrm d}{\mathrm d x}(\sqrt x) = \frac{1}{2\sqrt x}$

幂函数 $f(x) = x^n$ 的导数的推导(略),当 $n > 0, n\in\Bbb{Z}$ 时

$$ \frac{\mathrm d}{\mathrm d x}(x^n) = nx^{n-1} $$

在上面的两个例子中,可见 $n = -1, n = 1/2$ 时,等式成立

事实上,对所有实数 $n\in\Bbb{R}$,等式成立(详见第9章

当 $n = 0$ 时,$x^n = 1$,$nx^{n-1} = 0$,$\frac{\mathrm d}{\mathrm d x}(1) = 0$

if $C$ is a constant, then $\frac{\mathrm d}{\mathrm d x}(C) = 0$

常数函数的导数为 $0$

当 $n = 1$ 时,$x^n = x$,$nx^{n-1} = x^0 = 1$,$\frac{\mathrm d}{\mathrm d x}(x) = 1$

6.2 用更好的方式求导

…we’ll see how to use simple operations|multiplication by a constant, adding and subtracting, multiplying, dividing, and composing functions|to build f from atoms of the form $x^a$, which we already know how to differentiate.

复合函数 —— 以 $x^a$ 为原子(简单函数)构造而成

6.2.1 常数倍的函数

It’s easy to deal with a constant multiple of a function: you just multiply by the constant after you differentiate.

常数倍的函数求导—— 常数乘以导数(先对幂函数求导,再乘以函数倍)

如,$\frac{\mathrm d}{\mathrm d x}-7x^2 = -7\times 2x = -14x$ (注意:适用于负的常数倍)

6.2.2 函数和与差

It’s even easier to differentiate sums and differences of functions: just differentiate each piece and then add or subtract.

对每一部分函数求导,然后进行加减

6.2.3 乘法法则

It turns out that you have to mix and match. That is, you take the derivative of $f$ and multiply it by $g$ (not the derivative of $g$). Then you also have to take the derivative of $g$ and multiply it by $f$.

函数的积 —— 乘法规则

Product rule (version 1):if $h(x) = f(x)g(x)$, then $$ h’(x) = f’(x)g(x) + f(x)g’(x) $$

Product rule (version 2):if $y = uv$, then $$ \frac{\mathrm dy}{\mathrm dx} = v\frac{\mathrm du}{\mathrm dx} + u\frac{\mathrm dv}{\mathrm dx} $$

You could multiply this out, but it would be even worse than multiplying out the original function h and then differentiating that.

注意:“有时候”,直接在原函数上完成乘法运算后再求导比运用乘法法则更加方便

Product rule (three variables):if $y = uvw$, then $$ \frac{\mathrm dy}{\mathrm dx} = vw\frac{\mathrm du}{\mathrm dx} + u\frac{\mathrm dv}{\mathrm dx}w + uv\frac{\mathrm dw}{\mathrm dx} $$

… here’s a tip for remembering the above formula: just add up $uvw$ three times, but put a $\mathrm d\over\mathrm dx$ in front of a different variable in each term. (The same trick works for four or more variables – every variable gets differentiated once!)

6.2.4 除法法则

函数的商 —— 除法法则

Quotient rule (version 1):if $h(x) = \frac{f(x)}{g(x)}$, then $$ h’(x) = \frac{f’(x)g(x) - f(x)g’(x)}{(g(x))^2} $$

Quotient rule (version 2):if $y = \frac{u}{v}$, then $$ \frac{\mathrm dy}{\mathrm dx} = \frac{\frac{\mathrm du}{\mathrm dx}v - u\frac{\mathrm dv}{\mathrm dx}}{v^2} $$

注意:the minus sign always be with the product which including the derivative of denominator.

6.2.5 链式法则

复合函数求导 —— 链式法则

Chain rule (version 1):if $h(x) = f(g(x))$, then $$ h’(x) = f’(g(x))g’(x) $$

q1. $h(x) = \sqrt{g(x)}$ for some functions $g$ and $h$, and all you know about $g$ is that $g(5) = 4,g’(5) = 7$, how to find $h(5)$?

a1. set $f(x) = \sqrt x$, then $f’(x) = \frac{1}{2\sqrt x}$, use the chain rule (v1) and substitute $x = 5$ to get

$$h’(5) = f’(g(5))g’(5) = \frac{1}{2\sqrt 4}\times 7 = \frac{7}4$$

Chain rule (version 2):if $y$ is a function of $u$, and $u$ is a function of $x$, then $$ \frac{\mathrm dy}{\mathrm dx} = \frac{\mathrm dy}{\mathrm du}\frac{\mathrm du}{\mathrm dx} $$

…can’t you just cancel out the factor of $\mathrm du$? The answer is no – remember, expressions like $\frac{\mathrm dy}{\mathrm du}$ and $\frac{\mathrm du}{\mathrm dx}$ aren’t actually fractions, they are limits of fractions. The nice thing is that they often behave as if they were fractions|they certainly do in this case.

6.3 切线方程

… the tangent line through that point has slope $f’(x)$ and passes through the point $(x, f(x))$.

  1. find the slope, by finding the derivative and plugging in the given value of x;
  2. find a point on the line, by substituting the value of x into the function itself to get the y-coordinate. Put the coordinates together and call the resulting point $(x_0, y_0)$. Finally,
  3. use the point-slope form $y - y_0 = m(x - x_0)$ to find the equation.

过图像/曲线一点的切线斜率为该点上导函数的值

6.4 速度和加速度

… the term “velocity” will always refer to instantaneous velocity unless we actually say “average velocity.”

“速度”一般指瞬时速度

6.5 导数伪装的极限

如,

q1. $\lim\limits_{h\to 0}\frac{\sqrt[5]{32 + h} - 2}h$ ?

Note that $h$, not $x$, is the dummy variable here.

a1. set $f(x) = \sqrt[5]{x}$, then $f’(x) = \lim\limits_{h\to 0}\frac{\sqrt[5]{x + h} - \sqrt[5]{x}}h = \frac{1}5x^{-4/5}$,

put $x = 32$ to get $\lim\limits_{h\to 0}\frac{\sqrt[5]{32 + h} - 2}h = \frac{1}5\times 32^{-4/5} = 1/80$

If you get stuck on a limit, it might be a derivative in disguise. Telltale signs are that the dummy variable is by itself in the denominator, and the numerator is the difference of two quantities.

导数伪装的极限具有的特征:虚拟变量单独作为分母,分子为两个数值的差

注意:若虚拟变量单独作为分子,可以取其倒数进行计算(flip it over)

如,$\lim\limits_{h\to 0} = \frac{h}{(x+h)^6-x^6} = \frac{1}{6x^5}$

6.6 分段函数求导

Actually, this is only true if the left- and right-hand limits of the derivatives at the join points exist and are finite.

注意:断点(join points)处,导函数的左右极限相等时可微

6.7 导函数的图像

One important point: the height of the mountain itself isn’t relevant. It’s only the steepness that matters. In particular, you could shift the whole graph upward, and the climber would still be calling out the same degree of difficulty. A consequence of this is that if you are drawing the graph of a derivative from the graph of a function, the x-intercepts of the function are not important.

§三角函数的导数

7.1 三角函数的极限

7.1.1 The small case

There is one thing you have to be very careful of: when you say that $\sin(x)$ behaves like $x$ when $x$ is small, you should only use this fact in the context of products or quotients.

7.1.2 The large case

根据夹逼定理,得$\lim\limits_{x\to \infty}{\sin(x)\over x}=0$(见3.6

$$ \lim_{x\to \infty}{\sin(anything)\over x^\alpha}=0 $$

7.1.3 The other case

$x \to {\pi\over 2}$

A good general principle when dealing with a limit involving $x\to a$ for some $a \neq 0$ is to shift the problem to $t \to 0$ by substituting $t = x - a$. So in the above limit, set $t = x-\pi/2$. Then when $x \to \pi/2$, you can see that $t \to 0$.

证明

$\sin(x) < x < \tan(x)$ for $(0<x<{\pi\over 2})$

take reciprocals (switch the less-than signs to greater-than signs) and multiply by the positive quantity $\sin(x)$ (write it backward):

$\cos(x) < {\sin(x)\over x} < 1$

use the sandwich principle:

$\lim\limits_{x\to 0^+}\cos(x)=1, \lim\limits_{x\to 0^+}1=1$

$\implies \lim\limits_{x\to 0^+}{\sin(x)\over x}=1$,

To prove that the left-hand limit “$\lim\limits_{x\to 0^-}{\sin(x)\over x}=1$”

set $t = -x$, when $x \to 0^-$, then $t \to 0^+$

$\lim\limits_{t\to 0^+}{\sin(-t)\over -t}=\lim\limits_{t\to 0^+}{\sin(t)\over t}=1$

$$ \lim_{x\to 0}{\sin(x)\over x}=1 $$

Q.E.D.

7.2 三角函数的导数

$$ \begin{aligned} {\mathrm{d}\over\mathrm{d}x}\sin(x)&=\cos(x)\ {\mathrm{d}\over\mathrm{d}x}\cos(x)&=-\sin(x)\ {\mathrm{d}\over\mathrm{d}x}\tan(x)&=\sec^2(x)\ {\mathrm{d}\over\mathrm{d}x}\sec(x)&=\sec(x)\tan(x)\ {\mathrm{d}\over\mathrm{d}x}\csc(x)&=-\csc(x)\cot(x)\ {\mathrm{d}\over\mathrm{d}x}\cot(x)&=-\csc^2(x) \end{aligned} $$

Notice that the three cofunctions ($\cos, \csc, \cot$) all have minus signs in front of them, and the derivatives are the co- versions of the regular ones.

如,$\csc(x)$(co- version)的导数是 $\sec(x)$ (regular function)的导数的 co- versions,并在前面加上负号(a minus sign)

函数 $\sin(x)$ 和 $\cos(x)$ 的二阶导是原函数的相反数(the negative of the original function)

$$ {\mathrm{d}\over\mathrm{d}x}\sin(ax)=a\cos(ax) $$

简谐运动(Simple harmonic motion)

可微但导数不连续的函数 $f(x)=x^2\sin({1\over x})$(见书7.2.3

§隐函数和相关变化率

8.1 隐函数

$${\mathrm{d} \over \mathrm{d}x}(y^2)={\mathrm{d}u \over \mathrm{d}x}={\mathrm{d}u \over \mathrm{d}y}{\mathrm{d}y \over \mathrm{d}x}=2y{\mathrm{d}y \over \mathrm{d}x}$$

If you want to know how the quantity $y^2$ changes when you change $x$ a little, then first you need to know something about how $y$ changes! (Again, if $y$ doesn’t depend on $x$, then $\mathrm{d}y\over \mathrm{d}x$ equals $0$ for all $x$, so $\mathrm{d}(y^2)/ \mathrm{d}x$ is also $0$ for all $x$. That is, $y^2$ doesn’t depend on $x$ either.)

8.1.1 方法与例子

In fact, all we have to do is whack a $\mathrm{d}\over \mathrm{d}x$ in front of both sides:

⒈直接解出切线方程

⒉通过隐函数解出 $\mathrm{d}y \over \mathrm{d}x$ 切点和切线斜率

here’s a brief summary of the above methods:

8.1.2 隐函数的二阶导

8.2 相关变化率

If you want to know how fast something is changing over time, you simply have to differentiate with respect to time. So, here’s the real definition: the rate of change of a quantity Q is the derivative of Q with respect to time.

变化率—— 对关于时间的函数求导

q1. 对球形的气球打气,每秒 $12\pi$ 立方英寸.当气球半径为 $2$ 英寸时,英寸的变化率为?当气球体积为 $36\pi$ 时,变化率为?

$${\mathrm{d}V \over \mathrm{d}t} = 4 \pi r{\mathrm{d}r \over \mathrm{d}t}$$

$\implies {\mathrm{d}r \over \mathrm{d}t}={3 \over r^3}$

q2. 高度为半径 $2$ 倍的圆锥形水罐(尖端朝下),以每秒 $8\pi$ 立方英尺的速度注水。当水的体积为 $18\pi$ 立方英尺时,水位变化率为?若水罐有一漏洞,罐中每立方英尺水则有每秒一立方英尺流出,此时水位变化率为?

$h = 2r$

$v = \frac1{3}\pi r^2h = \frac1{12}\pi h^3$

$${\mathrm{d}v \over \mathrm{d}t} = {\pi h^2 \over 4}{\mathrm{d}h \over \mathrm{d}t}$$

Part 1.

$\implies 8\pi = {\pi h^2 \over 4}{\mathrm{d}h \over \mathrm{d}t}$

when $18\pi = \frac1{12}\pi h^3$, so $h = 6$ …

Part 2.

$\implies 8\pi - v = -10\pi = {\pi h^2 \over 4}{\mathrm{d}h \over \mathrm{d}t}$

substitute $h = 6$ …

§指数函数和对数函数

9.1 基本知识

指数函数:$f(x)=b^x$

对数函数:$g(x)=\log_bx$($b > 0 \cap b \neq 1$)

$$ f(g(x))=x \iff b^{\log_b(x)}=x $$

对数 $\log_b(b^x)=x$(对所有实数 $x$ 和 $b>1$ 时成立,为什么?)

换底公式(Change of base rule):

$$ \log_b(x)={\log_c(x)\over\log_c(b)} \implies \log_b(x)=K\log_c(x) \implies K={1\over\log_c(b)} $$

证明

9.2 欧拉常数

关于复利的问题

Let’s suppose you have a bank account at a bank that pays interest at a generous rate of 12% annually, compounded once a year. You put in an initial deposit; every year, your fortune increases by 12%. This means that after n years, your fortune has increased by a factor of (1+0.12)^n^. In particular, after one year, your fortune is just (1 + 0.12) = 1.12 times the original amount. If you started with 100 dollars, you’d finish the year with 112 dollars. Now suppose you find another bank that also others an annual interest rate of 12%, but now it compounds twice a year. Of course you aren’t going to get 12% for half a year; you have to divide that by 2. Basically this means that you are getting 6% interest for every 6 months. So, if you put money into this bank account, then after one year it has compounded twice at 6%; the result is that your fortune has expanded by a factor of (1+0.06)^2^, which works out to be 1.1236. So if you started with 100 dollars, you’d finish with 112.36 dollars. …… the question is, where does it stop? If you compound more and more often at the same annual rate, do you get wads and wads of cash after a year, or is there some limitation on all this?

$$ L = \lim_{n \to \infty}(1+{r \over n})^n $$

let $h = r/n$, then

$$ \lim_{h \to 0^+}(1+h)^{r/h} = \lim_{h \to 0^+}(1+h^{1/h})^r = e^r $$

let $h = -t$, then

$$ \begin{aligned} \lim_{h\to0^-}(1+rh)^{1/h} &=\lim_{t\to0^+}(1-rt)^{-1/t}\ &=\lim_{t\to0^+}{1\over(1+(-r)t)^{1/t}}\ &={1\over e^{-r}}=e^r\ \implies \lim_{n \to \infty}(1+{x \over n})^n &= \lim_{h \to 0}(1+xh)^{1 \over h} = e^x \end{aligned} $$

当 $x =1$ 时,可得关于欧拉常数 $e$ 的方程

9.3 指数函数和对数函数的微分

9.3.1 对数函数的导数

根据导数的定义 $g’(x)=\lim\limits_{h\to0}\frac{g(x+h)-g(x)}{h}$ 推导

对数函数的导数为

$\lim\limits_{h\to0}\frac{\log_b(x+h)-\log_b(x)}{h}$

$$ \frac{\log_b(x+h)-\log_b(x)}{h}={1\over{h}}\cdot\log_b(\frac{x+h}{x})=\log_b(1+{h\over{x}})^{1\over{h}} $$

由欧拉常数的重要极限 $\lim\limits_{h\to0}(1+rh)^{1/h}=e^r$

令 $r=1/x$

可得 $\lim\limits_{h\to0}(1+{\frac{h}{x}})^{1/h}=e^{1/x}$

所以 $g’(x)=\log_b(e^{1/x})={1 \over x}\log_b(e)$

当底数 $b=e$

$$ {\mathrm{d}\over\mathrm{d}x}(\ln(x))={1 \over x} $$

根据换底公式 $\log_b(e)=\frac{\ln{e}}{\ln{b}}= {1\over\ln{b}}$ 可得

$$ g’(x)={1 \over x}\log_b(e)={1 \over x\ln(b)} $$

例如,${\mathrm{d}\over\mathrm{d}y}(\ln 8x)=1/x$

$\ln 8x = \ln 8+\ln x$,其中 $\ln 8$ 是一个常数

9.3.2 指数函数的导数

${\mathrm{d}\over\mathrm{d}y}(b^x)=b^x\ln(b)$

${\mathrm{d}\over\mathrm{d}y}(e^x)=e^x$

9.4 指数函数和对数函数的极限问题

9.4.1 与欧拉常数定义有关的极限问题

可以通过一个经典的极限来解决:

$$ \lim_{n \to \infty}(1+{1 \over n})^n = \lim_{h \to 0}(1+h)^{1 \over h} = e $$

Indeed, if you replace $h$ by any quantity that goes to $0$ as $h \to 0$

然而,如 $\lim\limits_{h\to0}(1+\cos(h))^{1\over\cos(h)}$,因为当 $h\to0$ 时,$\cos(h)\to1$,所以不适用于和欧拉常数有关的极限问题

9.4.2 指数函数在 0 附近的极限

$$ \lim\limits_{x\to 0}e^x=1 $$

当 $\lim\limits_{h\to 0}{e^h-1 \over h}$ 时,虚拟变量 $h$ 独自作为分子,是一个极限伪装的导数问题,易得

$$ \lim\limits_{h\to 0}{e^h-1 \over h}=1 $$

9.4.3 对数函数在 1 附近的极限

$$ \lim\limits_{h\to 0}{\ln(1+h) \over h}=1 $$

例如,$\lim\limits_{h\to 0}{\ln(1-7h^2) \over 5h^2}=\lim\limits_{h\to 0}{\ln(1-7h^2) \over -7h^2} \times {-7h^2 \over 5h^2}=-{7 \over 5}$

9.4.4 指数函数在无穷处的极限

$$ \lim\limits_{x\to\infty}e^x=\infty,\lim\limits_{x\to-\infty}e^x=0 $$

注意:对于底不为 $e$ 的指数函数,根据底数和 $1$ 的大小关系,在正无穷处($x\to\infty$)有不同的极限

$$ \lim\limits_{x\to\infty}{x^n \over e^x}=0 $$

可得

$\lim\limits_{x\to\infty}(polynomial)\times e^{-x}=0$

This of course involves the behavior of $e^x$ near $-\infty$, but you can just throw the situation over to $+\infty$ by setting $t = -x$. We can see that as $x \to -\infty$ we have $t \to +\infty$.

9.4.5 对数函数在无穷处的极限

$$ \lim\limits_{x\to\infty}\ln x=\infty $$

与指数函数不同,对数函数增长趋势缓慢:

$\lim\limits_{x\to\infty}{\ln x \over x^a}=0$($a>0$)对所有底数大于 $1$ 的对数函数成立

令 $x=e^t$ 得,$\lim\limits_{x\to\infty}{\ln x \over x^a}=\lim\limits_{t\to\infty}{t \over e^at}$

9.4.6 对数函数在 0 处的极限

$$ \lim\limits_{x\to0^+}(x\ln x)=\lim\limits_{t\to\infty}-{\ln t \over t}=0 $$

对所有底数大于 $1$ 的对数函数,$\lim\limits_{x\to0^+}{x^a\ln x}=0$($a>0$)成立

9.5 取对数求导

Logarithmic differentiation is a useful technique for dealing with derivatives of things like $f(x)^{g(x)}$, where both the base and the exponent are functions of $x$.

对 $y$ 取对数,使 $u=\ln y$,

求导可得 ${\mathrm{d}u \over \mathrm{d}x}={\mathrm{d}u \over \mathrm{d}y}{\mathrm{d}y \over \mathrm{d}x}={1\over y}{\mathrm{d}y \over \mathrm{d}x}$

所以 ${\mathrm{d}y \over \mathrm{d}x} = {\mathrm{d}u \over \mathrm{d}x}\times y$

9.5.1 x^a 的导数

假设 $x>0$(使 $\ln x$ 有意义),对 $y=x^a$ 等式两边取对数求导,${\mathrm{d}\over\mathrm{d}y}(\ln y)=a\ln x$

you can only make sense of $x^a$ for $x<0$ when $a$ is a rational number with an odd denominator (after canceling out common factors). For example, $x^{5\over 3}$ makes sense for negative $x$ since you can always take a cube root–we’re OK because $3$ is odd.

注意:$x<0$ 的幂函数 $x^a$,对指数 $a$ 有如上要求

9.6 指数增长和衰减

if ${\mathrm{d}y \over \mathrm{d}x} = ky$, then $y = Ae^{kx}$ for some constant $A$ …… The number $k$, which is called the growth constant, controls how fast the rabbits are breeding in the first place.

指数函数的微分方程:${\mathrm{d}y \over \mathrm{d}x} = ky$

增长常数(growth constant):$k$

9.6.1 指数增长

人口问题:${\mathrm{d}P \over \mathrm{d}t}=kP \implies P(t)=P_0e^{kt}$

9.6.2 指数衰减

半衰期(half-life)问题:${\mathrm{d}P \over \mathrm{d}t}=-kP\implies P(t)=P_0e^{-kt}$

用 $t_{1/2}$ 表示半衰期 $\implies {1 \over 2}P_0=P_0e^{-kt_{1/2}}\implies k={\ln 2 \over t_{1/2}}$

9.7 双曲函数

双曲函数(hyperbolic functions)

微分:

*There is actually a branch of geometry called hyperbolic geometry, in which the triangles have wacky properties that lead to hyperbolic functions.

注意:三角函数是非欧几何中的性质不同

双曲三角函数的微分:

水平渐近线:

§反函数和反三角函数

10.1 导数与反函数

参考1.2

10.1.1 通过微分证明反函数存在

In any case, if our function $f$ is always increasing, then it must satisfy the horizontal line test.
……
show that its derivative is always positive on its domain.

导函数 $f’(x)$ 在定义域内的一直为正数(或负数),原函数图像单调递增或递减

令 $f’(x)=0$,若方程在定义域内无实数解,导函数图像不经过 $x$ 轴,即 $f’(x)$ 的图像必全位于 $x$ 轴上方或下方(非负即正)

if $f$ is differentiable on its domain (a,b) and any of the following are true:

  1. $f’(x) > 0$ for all $x$ in $(a,b)$;
  2. $f’(x) < 0$ for all $x$ in $(a,b)$;
  3. $f’(x) \geq 0$ for all $x$ in $(a,b)$ and $f’(x) = 0$ for only a finite number of $x$ or
  4. $f’(x) \leq 0$ for all $x$ in $(a,b)$ and $f’(x) = 0$ for only a finite number of $x$, then $f$ has an inverse.

注意:导函数 $f’(x)=0$ 时,此时原函数仅有唯一的点 $(x,y)$ 处斜率为 $0$

10.1.2 导数和反函数可能出现的问题

注意:

⒈连续的分段函数中,存在常数函数时,斜率为 $0$ 的点不唯一,此时多个不同的 $x$ 对应相同的输出 $y$(常数函数的值),不符合水平线检验

⒉当函数不连续,或者垂直渐近线存在时,如 $f(x)=tan(x)$ 的导数 $f’(x)=sec^2(x)>0$,违背水平线检验

10.2 反三角函数

10.2.1 反正弦函数

if $f(x)=\sin(x)$ with domain $[-\pi/2,\pi/2]$, then it satisfies the horizontal line test, so it has an inverse $f^{-1}$. We’ll write $f^{-1}(x)$ as $\sin^{-1}(x)$ or $\arcsin(x)$.

$y=\sin^{-1}(x)=\arcsin(x)$ 求导

$x=\sin(y)$ 两边关于 $x$ 作隐函数可得 $1=\cos(y){\mathrm{d}y\over\mathrm{d}x}$

${\mathrm{d}y\over\mathrm{d}x}={1\over\cos(y)}={\mathrm{d}\over\mathrm{d}x}\sin^{-1}(x)$

由毕达哥拉斯定理得 $\cos(y)=\pm\sqrt{1-\sin^2(y)}$

所以 ${\mathrm{d}y\over\mathrm{d}x}={1\over\sqrt{1-x^2}}$

$y=\arcsin(x)$ 在区间 $(-1,1)$ 上单调递增,斜率为正,在端点处不可导

10.2.2 反余弦函数

限制定义域: $[0,\pi]$

${\mathrm{d}\over\mathrm{d}x}\cos^{-1}(x)=-{1\over\sqrt{1-x^2}}$

$y=\arccos(x)$ 在区间 $(-1,1)$ 上单调递减,既不是奇函数,也不是偶函数

综上可得, $$ {\mathrm{d} \over \mathrm{d}x}(\sin^{-1}(x)+\cos^{-1}(x))={\mathrm{d}\over\mathrm{d}x}\sin^{-1}(x)+{\mathrm{d} \over \mathrm{d}x}\cos^{-1}(x)=0 $$

在 $[-1,1]$ 上,$\sin^{-1}(x)+\cos^{-1}(x)={\pi\over2}$,导数为 $0$。

10.2.3 反正切函数

限制定义域: $(-\pi/2,\pi/2)$

${\mathrm{d}y\over\mathrm{d}x}={1\over\sec^2(y)}$

$\sec^2(y)=1+\tan^2(y)$

$\implies {\mathrm{d}\over\mathrm{d}x}\tan^{-1}(x)={1\over 1+x^2}$

$y=\arctan(x),x\in\Bbb{R}$ 单调递增,斜率为正,存在水平渐近线 $y=-\pi/2$ 和 $y=\pi/2$

注意:当 $x\to-\infty$ 时($x^2\to\infty$),$\tan^{-1}(x^2)$ 的极限为 $\pi/2$,而不是 $-\pi/2$

10.2.4 其他反三角函数

反正割函数:$y=\sec^{-1}(x)$

  1. 非奇非偶,分段,递增
  2. 定义域:$(-\infty,-1]\cup[1,\infty)$
  3. 值域:$[0,\pi]\setminus{\frac{\pi}{2}}$
  4. 水平渐近线:$y=\pi/2$
  5. 微分:$${\mathrm{d}\over\mathrm{d}x}\sec^{-1}(x)={1\over|x|\sqrt{x^2-1}},(-\infty,-1)\cup(1,\infty)$$

反余割函数:$y=\csc^{-1}(x)$

  1. 奇函数,分段,递减
  2. 定义域:$(-\infty,-1]\cup[1,\infty)$
  3. 值域:$[-\pi/2,\pi/2]\setminus {0}$
  4. 水平渐近线:$y=0$
  5. 微分:$${\mathrm{d}\over \mathrm{d}x}\csc^{-1}(x)=-{1\over|x|\sqrt{x^2-1}},(-\infty,-1)\cup(1,\infty)$$

反余切函数:$y=\cot^{-1}(x)$

  1. 非奇非偶,单调递减
  2. 定义域:$\Bbb{R}$
  3. 值域:$(0,\pi)$
  4. 水平渐近线:$y=0$
  5. 微分:$${\mathrm{d} \over \mathrm{d}x}\cot^{-1}(x)=-{1\over 1+x^2},x\in\Bbb{R}$$

10.3 反双曲函数

$y=\cosh^{-1}(x)=\ln{(x+\sqrt{x^2-1})}$

  1. 非奇非偶
  2. 定义域:$[1,\infty)$
  3. 值域:$[0,\infty)$
  4. 微分:${\mathrm{d} \over \mathrm{d}x}\cosh^{-1}(x)={1\over \sqrt{x^2-1}},x>1$

$y=\sinh^{-1}(x)=\ln{(x+\sqrt{x^2+1})}$

  1. 奇函数
  2. 定义域和值域:$\Bbb{R}$
  3. 微分:${\mathrm{d} \over \mathrm{d}x}\sinh^{-1}(x)={1\over \sqrt{x^2+1}},x\in\Bbb{R}$

10.3.1 其他反双曲函数

$y=\tanh^{-1}(x)$

  1. 奇函数
  2. 定义域:$(1,1)$
  3. 值域:$\Bbb{R}$
  4. 微分:${\mathrm{d} \over \mathrm{d}x}\tanh^{-1}(x)={1\over{1-x^2}}$

$y=\coth^{-1}(x)$

  1. 奇函数
  2. 定义域:$(-\infty,-1)\cup(1,\infty)$
  3. 值域:$\Bbb{R} \setminus {0}$
  4. 微分:${\mathrm{d} \over \mathrm{d}x}\coth^{-1}(x)={1\over{1-x^2}}$

$y=\mathrm{sech}^{-1}(x)$

  1. 非奇非偶
  2. 定义域:$(0,1]$
  3. 值域:$[0,\infty)$
  4. 微分:${\mathrm{d} \over \mathrm{d}x}\mathrm{sech}^{-1}(x)=-{1\over x\sqrt{1-x^2}},0<x<1$

$y=\mathrm{csch}^{-1}(x)$

  1. 奇函数
  2. 定义域和值域:$\Bbb{R} \setminus {0}$
  3. 微分:${\mathrm{d} \over \mathrm{d}x}\mathrm{csch}^{-1}(x)=-{1\over |x|\sqrt{1+x^2}}$

§导数与图像

11.1 极值

If we say that $x = a$ is an extremum of a function $f$, this means that $f$ has a maximum or minimum at $x = a$. (The plural of “extremum” is “extrema,” of course.)

11.1.1 全局和局部

Let’s say that a global maximum(or absolute maximum) occurs at $x = a$ if $f(a)$ is the highest value of $f$ on the entire domain of $f$.In symbols, we want $f(a) \geq f(x)$ for any value $x$ in the domain of $f$. …… Let’s say that a local maximum(or relative maximum) occurs at $x = a$ if $f(a)$ is the highest value of $f$ on some small interval containing $a$.

全局最大值:$f(a) \geq f(x)$

每一个全局最大值都是局部最大值

11.1.2 极值定理

最大值-最小值定理:连续函数在闭区间中必有全局最大值和最小值(见第五章)

极值定理(Extreme Value Theorem):函数 $f$ 的定义域 $(a,b)$ 上存在临界点 $c$ 使 $f’(c)=0$ 或者 $f’(c)$ 不存在(critical point),则 $f(c)$ 是函数的一个局部极值

local maxima and minima in an open interval occur only at critical points. But it’s not true that a critical point must be a local maximum or minimum!

如 $x=0$ 是 $f(x)=x^3$ 的临界点,但该处的函数值不是局部极值

注意:闭区间的端点处可能存在局部极值,此时上述定理不成立

11.1.3 求全局极值

根据极值定理(闭区间),通过比较各临界点和端点处的函数值大小求出全局极值

注意:临界点 $x=0$ 处非极值的情况

If the domain isn’t bounded, then the situation is a little more complicated.

定义域不是有限的情况下,有时候需要考虑(infimum):$$\lim\limits_{x\to\infty}f(x)$$

11.2 罗尔定理

罗尔定理(Rolle’s Theorem):函数 $f$ 在闭区间 $[a,b]$ 连续,在开区间 $(a,b)$ 可导,如果 $f(a)=f(b)$,那么开区间内至少存在一点 $c$ 使 $f’(c)=0$

注意:绝对值函数 $y=|x|$ 在临界点 $x=0$ 处不可导(spike at $x=0$)

$f"(x>0$ 时,$f(x)$ 的图像至多只有 $2$ 个 $x$ 轴上的截距

11.3 中值定理

中值定理(The Mean Value Theorem):函数 $f$ 在闭区间 $[a,b]$ 连续,在开区间 $(a,b)$ 可导,那么 $(a,b)$ 内至少存在一点 $c$ 使得 $f’(c)=\frac{f(b)-f(a)}{b-a}$

$f’(C)$ 的物理意义:从 $a \to b$ 的平均速度,$c$ 点的瞬时速度

11.3.1 中值定理的推论

  1. 证明:常数函数的导数为 $0$
  2. 如果 $f’(x)=g’(x)$,那么 $f(x)=g(x)+C$($C$ 为常数)
  3. 区间内的导数为正,单调递增

11.4 二阶导数和图像

假设 $(f’)’(x)>0$,那么 $f’(x)$ 单调递增,函数在区间内凹向上(concave up);反之如果 $(f’)’(x)<0$,则函数在区间内凹向下(concave down)

To the left of $x = c$, the curve is concave down, while to the right of $x = c$, the curve is concave up. We’ll say that the point $x = c$ is a point of inflection for $f$ because the concavity changes as you go from left to right through $c$.

拐点(point of inflection):$f’’(c) = 0$,且在 $x=c$ 左右,二阶导数 $f’’(x)$ 异号

注意:$f’’(c) = 0$,$c$ 不一定为函数的拐点。如 $f(x)=x^4$

拐点的一边,函数曲线必定在拐点切线之上,而另一边的曲线在拐点切线之下

11.5 导数为零的点

11.5.1 一阶导数

$f’(c)=0$ 的三种情况:

11.5.2 二阶导数

假设 $f’(c)=0$,

  1. 如果 $f’’(c)>0$,$x=c$ 为局部最小值
  2. 如果 $f’’(c)<0$,$x=c$ 为局部最大值
  3. 如果 $f’’(c)=0$,无法判断。如 $f(x)=x^3$ 和 $g(x)=x^4$

§绘制函数图像

12.1 符号表格

  1. 以递增的顺序列出零点(0)和不连续点(*)
  2. 每两个数之间及其前后任取一个数(间隔),记录函数的符号(+/-)
  3. 一阶导数:列出零点,以零点为间隔任取数字记录符号,在下一行中用斜线表示单调性
  4. 二阶导数:列出零点(估算数值),以零点为间隔任取数字记录符号,在下一行中将凹性表示出来

12.2 绘制图像

  1. 对称性(Symmetry):奇偶性,非奇非偶
  2. 截距(Intercepts):横轴和纵轴
  3. 定义域(Domain):反三角函数等
  4. 垂直渐近线(Vertical asymptotes):常出现在分母为零的位置(注意:可能为removable discontinuity)
  5. 符号(Sign of the function)
  6. 水平渐近线(Horizontal asymptotes):计算 $\lim\limits_{x\to\infty}f(x)$ 和$\lim\limits_{x\to-\infty}f(x)$,制成符号表格判断函数图像位于渐近线的哪一侧
  7. 导数的正负(Sign of the derivative):单调性,增函数或减函数
  8. 最大值和最小值(Maxima and minima)
  9. 二阶导数的正负(Sign of the second derivative):凹性
  10. 拐点(Points of inflection)

12.3 函数绘图实例