6

My attempt is as follows:-

  1. Derivative of $\sec^{-1}x$

    Let $\theta=\sec^{-1}x,$ where $\theta\in [0,\pi]-{\dfrac{\pi}{2}}.$

    $$\sec\theta=x.$$

    Differentiating both sides with respect to $x:$

    $$\sec\theta\cdot\tan\theta\cdot\dfrac{\mathrm d\theta}{\mathrm dx}=1\\ \dfrac{\mathrm d\theta}{\mathrm dx}=\dfrac{1}{x\sqrt{x^2-1}}.$$

    As $\sec^{-1}x$ is a strictly increasing function, its derivative should be positive, hence we write $x$ as $|x|$ to ensure that $\dfrac{\mathrm d\theta}{\mathrm dx}$ will not be negative if $x$ is negative. But I wonder why I didn't get $\dfrac{\mathrm d\theta}{\mathrm dx}=\dfrac{1}{|x|\sqrt{x^2-1}}$ in the above calculation?

  2. Integral of $\dfrac{1}{x\sqrt{x^2-1}}$

    Case $1:x>0$

    Then the integral is definitely $\sec^{-1}x.$

    Case $2: x<0$

    Then the integral is $-\sec^{-1}x.$

    But many textbooks write that $\displaystyle\int\frac{1}{x\sqrt{x^2-1}}\,\mathrm dx=\sec^{-1}x+C.$

    Shouldn't $\displaystyle\int\frac{1}{|x|\sqrt{x^2-1}}\,\mathrm dx=\sec^{-1}x+C\:?$ What am I missing here?

ryang
  • 38,879
  • 14
  • 81
  • 179
user3290550
  • 3,452
  • 1
    Note that, when $|x|<1$, $\sqrt{x^2-1}\not\in\mathbb R$ – J. W. Tanner Dec 23 '19 at 16:09
  • why so, there is $x^2$, which will make quantity positive inside, if $|x|>1$ – user3290550 Dec 23 '19 at 16:10
  • 2
    Many textbooks do not take enough care of taking only the principal values into account while differentiating inverse trigonometric functions. Hence the technically wrong answers. Indeed, $\frac{d}{dx}(\sec^{-1}x)=\frac1{|x|\sqrt{x^2-1}}$ for $|x|>1$ which is same as saying $\int\frac{dx}{x\sqrt{x^2-1}}=\sec^{-1}|x|+c$ for $|x|>1$. – StubbornAtom Dec 24 '19 at 18:01
  • There is another version of this problem: derivative of $\sec^{-1} z$ for complex $z$. For that version, $\frac{d}{dz}(\sec^{-1}z)=\frac1{|z|\sqrt{z^2-1}}$ is definitely wrong. – GEdgar Jan 08 '22 at 14:06

4 Answers4

3

The reason you didn't get the absolute value when you differentiated is that $$\newcommand{\sgn}{\text{sgn}} \tan(\theta)=\tan(\sec^{-1}x)=\sgn(x)\sqrt{x^2-1} $$ so the derivative is $$\frac{1}{\sgn(x)x\sqrt{x^2-1}}=\frac{1}{|x|\sqrt{x^2-1}} $$

Sometimes the principal range of $\sec^{-1}x$ is assumed to be $[0,\frac \pi2)\cup [\pi, \frac{3\pi}{2})$. This convention is popular when doing integration with $\sec^{-1}x$ substitution and avoids the issue with the absolute value. Under that convention, $$\int \frac{1}{x\sqrt{x^2-1}}=\sec^{-1}(x)+C $$

If you don't like redefining the range of $\sec^{-1}(x)$, then $$\int \frac{1}{x\sqrt{x^2-1}}=\sec^{-1}(|x|)+C $$ as @YvesDaoust wrote.

bjorn93
  • 6,787
  • Antiderivative ? –  Dec 23 '19 at 16:28
  • @YvesDaoust typo – bjorn93 Dec 23 '19 at 16:33
  • what? range of $\sec^{-1}x$ is always $[0,\pi]-\dfrac{\pi}{2}$, what are you saying bro? – user3290550 Dec 23 '19 at 16:39
  • 1
    @user3290550 Generally speaking, yes. But it's a matter of definition. When doing integrals, the alternative convention for the range of the arcsecant is popular. – bjorn93 Dec 23 '19 at 16:44
  • 3
    @user3290550 Note that $\sec^{-1}x$ is a multi-valued function. So we have to choose a principal range if we want to make it a single-valued function. Usually, that's $[0,\frac \pi2)\cup(\frac \pi2,\pi]$. But, it's still a matter of choice, and other choices are possible. – bjorn93 Dec 23 '19 at 19:06
  • no, $\sec^{-1}x$ is not a multi-valued function, see the graph. Its a invertible function which means it is one-one and onto. Its inverse is $\sec x$ – user3290550 Dec 23 '19 at 20:16
  • 2
    @user3290550 you can read here: https://en.wikipedia.org/wiki/Inverse_trigonometric_functions#Basic_properties – bjorn93 Dec 23 '19 at 21:44
  • yeah so over there see the graph of $\sec^{-1}x$ , notice in that graph , graph of $\mathrm{cosec}^{-1}x$ is also there, so don't get confused. Graph of $\sec^{-1}x$ is highlighted in red and it is varying from $0$ to $\pi$, discontinuous at $\dfrac{\pi}{2}$ and it is the one-one and onto function. – user3290550 Dec 24 '19 at 00:17
  • 2
    @user3290550 These functions are invertible when their range is restricted to principal values. – StubbornAtom Dec 24 '19 at 18:18
  • that's what I am talking about, I was always talking in respect of the principal range. – user3290550 Dec 24 '19 at 19:43
  • 2
    @user3290550 The principal range of an inverse function is a matter of convention. For example, Wikipedia and Wolfram Alpha define arccot's principal range differently. bjorn93's nswer has actually addressed all your queries, and then some. Also: How to Avoid the Inverse Secant (and Even the Secant Itself) – ryang Jan 08 '22 at 04:42
  • @bjorn93 "Note that $\sec^{-1}x$ is a multi-valued function." If it's multivalued, can we call it a function sir? Isn't that just a relation? – tryingtobeastoic Jan 24 '22 at 06:45
2

But many textbooks write that $\displaystyle\int\frac{1}{x\sqrt{x^2-1}}\,\mathrm dx=\sec^{-1}x+C.$

This is indeed wrong, since differentiating the equation at $x=-5$ gives $\displaystyle\frac{\sqrt6}{60}=-\frac{\sqrt6}{60}.$

Case $1:x>0$

the integral is definitely $\sec^{-1}x.$

Case $2: x<0$

the integral is $-\sec^{-1}x.$

Indeed, $\displaystyle\frac1{x\sqrt{x^2-1}}$ always has an antiderivative $$\frac x{|x|}\sec^{-1}x.\tag1$$ bjorn93 has also pointed out that it always has an antiderivative $$\sec^{-1}|x|.\tag{2}$$

Using the substitution $\displaystyle u=\frac1x$ and noting that $\sec^{-1}x+\operatorname{cosec}^{-1}x\equiv\frac{\pi}2,$ we can also obtain \begin{align}\int\frac{\mathrm dx}{x\sqrt{x^2-1}} &= \int\frac{-|u|}{u\sqrt{1-u^2}}\,\mathrm du\\ &= \begin{cases} \sin^{-1} u+C_1, &-1<u<0;\\ -\sin^{-1} u+C_2, &0<u<1\end{cases}\\ &= \begin{cases} \operatorname{cosec}^{-1}x+C_1, &x<-1;\\ -\operatorname{cosec}^{-1}x+C_2, &x>1\end{cases}\\ &= \begin{cases} \operatorname{cosec}^{-1}x+C_1, &x<-1;\\ \operatorname{sec}^{-1}x-\frac\pi2+C_2, &x>1\end{cases}\\ &= \begin{cases} \operatorname{cosec}^{-1}x+C, &x<-1;\\ \operatorname{sec}^{-1}x+D, &x>1.\tag3\end{cases}\end{align}

$(1),(2),(3)$ are consistent with one another because for $x<-1$ the expressions differ from one another by $\frac{\pi}2$ while for $x>1$ the expressions are identically equal. (Remember, each integration domain overlaps with just one of these intervals. Also, it is worth noting that when $C_1\ne C_2,$ differentiating the above result still returns the given integrand.)


P.S. It can be similarly derived that on $(-1,0)\cup(0,1)$ $$-\operatorname{sech}^{-1}|x| \text{ is an antiderivative of } \frac1{x\sqrt{1-x^2}},$$ and on $\mathbb R{\setminus}\{0\}$ $$-\operatorname{cosech}^{-1}|x| \text{ is an antiderivative of } \frac1{x\sqrt{1+x^2}}.$$

ryang
  • 38,879
  • 14
  • 81
  • 179
1

To avoid $\pm$ signs, proceed as follows:

  1. Derivative of $\sec^{-1}x$. Let $\theta=\sec^{-1}x,$ with $\theta\in [0,\pi]-{\dfrac{\pi}{2}}$. Then, $\sin \theta >0$

$$\cos\theta=\frac1{\sec\theta}=\frac1x\implies \sin\theta\cdot\dfrac{\mathrm d\theta}{\mathrm dx}=\frac1{x^2}\\ \dfrac{\mathrm d\theta}{\mathrm dx}=\frac{1}{x^2\sin\theta} =\frac{1}{x^2\sqrt{1-\frac1{x^2}}}= \frac{1}{\sqrt{x^2(x^2-1)}} $$

  1. Integral of $\dfrac{1}{x\sqrt{x^2-1}}$. For integration valid for the whole domain $|x|>1$

$$\int \frac{1}{x\sqrt{x^2-1}}dx = \int \frac{d(\sqrt{x^2-1})}{(x^2-1)+1}=\tan^{-1}\sqrt{x^2-1}+C $$

Quanto
  • 97,352
0

$$\left(\sec^{-1}(x)\right)'=\left(\cos^{-1}\left(\frac1x\right)\right)'=\frac1{x^2\sqrt{1-\dfrac1{x^2}}}.$$

This is obviously a positive function, and can indeed not be expressed as

$$\frac1{x\sqrt{x^2-1}}.$$

So the antiderivative of the latter is indeed $\text{sgn}(x)\sec^{-1}(x)$, or $\sec^{-1}(|x|)$. Note anyway that the integration interval may not overlap $(-1,1)$.

  • yeah, right but I am confused in the integration part? – user3290550 Dec 23 '19 at 16:23
  • @user3290550: wait, I am not done. –  Dec 23 '19 at 16:23
  • but if you most of the textbooks, they write integration of $\dfrac{1}{x\sqrt{1-x^2}}=sec^{-1}x$, but it should be $\text{sgn} (x)sec^{-1}x$ as you said – user3290550 Dec 23 '19 at 16:30
  • by the way you made a mistake, $\sec^{-1}(-x)=\pi-\sec^{-1}x$, so we cannot write integration of $\dfrac{1}{x\sqrt{1-x^2}}=\sec^{-1}|x|$, it will be $\text{sgn} (x)\sec^{-1}x$ – user3290550 Dec 23 '19 at 16:37
  • @user3290550 constant of integration takes care of the $\pi$. $\sec^{-1}(|x|)$ is correct. See my note about the range of arcsecant. – bjorn93 Dec 23 '19 at 16:40
  • ok what's your final answer for integration of $\dfrac{1}{x\sqrt{x^2-1}}$?, is it $sec^{-1}x$ or $\text{sgn} (x)sec^{-1}x$ – user3290550 Dec 23 '19 at 16:41
  • @bjorn93, how can you change the range of $sec^{-1}x$? – user3290550 Dec 23 '19 at 16:42
  • @user3290550: as said, both $\text{sgn}(x)\sec^{-1}(x)$ or $\sec^{-1}(|x|)$ are correct. –  Dec 23 '19 at 17:20