4

I am attempting to evaluate

$$\int_{-\infty}^{\infty}\dfrac{b\tan^{-1}\Big(\dfrac{\sqrt{x^2+a^2}}{b}\Big)}{(x^2+b^2)(\sqrt{x^2+a^2})}\,dx. $$

I have tried using the residue formula to calculate the residues at $\pm ib,\pm ia,$ but it got messy very quickly. Then I tried to use a trigonometric substitution $x=a\tan(\theta)$; $dx=a\sec^{2}(\theta)\,d\theta$ which led me to the integral $$\int_{-\infty}^{\infty}\dfrac{b\tan^{-1}\Big(\dfrac{a\sec(\theta)}{b}\Big)\sec(\theta)}{(a^2\tan^{2}(\theta)+b^2)}\,d\theta.$$ The bounds for this integral seem incorrect, but I am more worried about the actual expression before I deal with the bounds, which may have to be changed into a double integral where $0\leq\theta\leq2\pi$ and the second bound would range from $-\infty$ to $\infty$. I am wondering if there is some kind of substitution I have missed, but I have hit the wall. The OP of this problem said there were cases that would come into play, but when I asked him whether or not those cases arose from $b<0$ and $b>0$ he told me they did not. The cases most likely arise from whether $a$ and $b$ are positive or negative, because the case where $b=0$ is trivial, and in the case where $a=0$ I used wolframalpha and the integral evaluates to $\dfrac{\pi\ln(2)\lvert b \rvert}{b^2}$ for $\Im(b)=0 \land \Re(b)\neq0.$ Contour integration may be necessary. I am stuck on this problem and I would greatly appreciate the help. Thank you for your time.

  • 1
    I have changed the formatting of the title so as to make it take up less vertical space -- this is a policy to ensure that the scarce space on the main page is distributed evenly over the questions. See here for more information. Please take this into consideration for future questions. Thanks in advance. – GNUSupporter 8964民主女神 地下教會 Apr 24 '18 at 15:19
  • 1
    Have you tried using Feynman's Trick? – Mark Viola Apr 24 '18 at 15:22
  • @MarkViola I have not attempted to use Feynman's Trick (manipulation under the integral sign) only because I have very little experience using that technique, but I will review notes and see if I can make it work. – JohnColtraneisJC Apr 24 '18 at 15:26
  • @MarkViola Thank you for pointing that out. – JohnColtraneisJC Apr 24 '18 at 15:27
  • 2
    You might try letting $$F(c)=b\int_{-\infty}^\infty \frac{\arctan\left(\frac{\sqrt{x^2+a^2}}{c}\right)}{(x^2+b^2)\sqrt{x^2+a^2}},dx$$ $$F'(c)=-b\int_{-\infty}^\infty \frac{1}{(x^2+b^2)(x^2+a^2+c^2)},dx$$The integral that represents $F'(c)$ can be obtained in closed form. And it's antiderivative might be found in closed form also. – Mark Viola Apr 24 '18 at 15:38
  • @MarkViola Taking another look at Feynman's Trick (differentiation under the integral sign), I believe the parameter will solely be $b$ since the expression will depend on it being positive or negative but we may allow $a$ to be positive or negative and in our expression, since $a$ is always squared, $a$ being positive or negative won't matter – JohnColtraneisJC Apr 24 '18 at 15:39

3 Answers3

6

In the following we shall assume that $a$, $b$, and $c$ are real valued and that $a>b>0$.

Let $F(c)$ be represented by the integral

$$F(c)=b\int_{-\infty}^\infty \frac{\arctan\left(\frac{\sqrt{x^2+a^2}}{c}\right)}{(x^2+b^2)\sqrt{x^2+a^2}}\,dx\tag 1$$

Differentiating $(1)$ reveals

$$\begin{align} F'(c)&=-b\int_{-\infty}^\infty \frac{1}{(x^2+b^2)(x^2+a^2+c^2)}\,dx\\\\ &=-\frac{\pi}{c^2+a^2+b\sqrt{c^2+a^2}}\tag2 \end{align}$$

Integrating $(2)$ and using $\lim_{c\to \infty}F(c)=0$, we find that

$$\begin{align} F(c)&=\pi\,\left(\frac{\arctan\left(\frac{bc}{\sqrt{a^2-b^2}\sqrt{a^2+c^2}}\right)-\arctan\left(\frac{b}{\sqrt{a^2-b^2}}\right)+\pi/2-\arctan\left(\frac{c}{\sqrt{a^2-b^2}}\right)}{\sqrt{a^2-b^2}}\right) \end{align}$$

Setting $c=b$ yields the coveted result

$$\int_{-\infty}^\infty \frac{b\arctan\left(\frac{\sqrt{x^2+a^2}}{b}\right)}{(x^2+b^2)\sqrt{x^2+a^2}}\,dx=\pi\,\left(\frac{\arctan\left(\frac{b^2}{\sqrt{a^2-b^2}\sqrt{a^2+b^2}}\right)+\pi/2-2\arctan\left(\frac{b}{\sqrt{a^2-b^2}}\right)}{\sqrt{a^2-b^2}}\right)$$

Mark Viola
  • 179,405
  • Thank you for your answer. When I posted this thread to the comments section, OP informed me that the "answer is not correct to its truest form. The case still hasn’t been split." Do you know what case it is he is referring to? I asked him and he only replied "Look at the expression under the square root on the denominator." What could that mean? Please let me know when you have the chance and I look forward to your take on his comments. – JohnColtraneisJC Apr 25 '18 at 12:09
  • More information: "The result is different for |a|>|b| compared to |a|<|b| and this needs to be taken into consideration when evaluating the integral. It is not at all obvious from the problem alone that it will be the case but it is so." – JohnColtraneisJC Apr 25 '18 at 12:14
  • Also I would normally not withdraw the acceptance of an answer, but I figured that since you have a high reputation you are not as concerned with reputation points as you are the enjoyment of solving problems of this kind. – JohnColtraneisJC Apr 25 '18 at 12:50
  • Upon further investigation I believe integrating (2) with the assumption $\lvert,a,\rvert<\lvert,b,\rvert$ changes the integral's form completely, and using $a=2$ and $b=3$ as an example on Wolframalpha demonstrates this change http://www.wolframalpha.com/input/?i=integrate((-π%2F(x%5E2%2B2%5E2%2B3sqrt(x%5E2%2B2%5E2)) – JohnColtraneisJC Apr 25 '18 at 14:24
  • 1
    @JohnColtraneisJC If $a<b$, then we simply use $$\arctan(ix)=i\text{atanh}(x)=\frac{i}{2}\log\left(\frac{1+x}{1-x}\right)$$ – Mark Viola Apr 25 '18 at 16:25
  • Thank you for clarifying this for me, I really appreciate the time you spent to help me with this problem. – JohnColtraneisJC Apr 25 '18 at 17:08
  • You're welcome. My pleasure. – Mark Viola Apr 25 '18 at 17:26
1

$\newcommand{\bbx}[1]{\,\bbox[15px,border:1px groove navy]{\displaystyle{#1}}\,} \newcommand{\braces}[1]{\left\lbrace\,{#1}\,\right\rbrace} \newcommand{\bracks}[1]{\left\lbrack\,{#1}\,\right\rbrack} \newcommand{\dd}{\mathrm{d}} \newcommand{\ds}[1]{\displaystyle{#1}} \newcommand{\expo}[1]{\,\mathrm{e}^{#1}\,} \newcommand{\ic}{\mathrm{i}} \newcommand{\mc}[1]{\mathcal{#1}} \newcommand{\mrm}[1]{\mathrm{#1}} \newcommand{\pars}[1]{\left(\,{#1}\,\right)} \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\root}[2][]{\,\sqrt[#1]{\,{#2}\,}\,} \newcommand{\totald}[3][]{\frac{\mathrm{d}^{#1} #2}{\mathrm{d} #3^{#1}}} \newcommand{\verts}[1]{\left\vert\,{#1}\,\right\vert}$

Note that $\ds{{\arctan\pars{\xi} \over \xi} = \int_{1}^{\infty}{\dd t \over t^{2} + \xi^{2}}}$.

\begin{align} &\bbox[#ffd,10px]{\ds{\int_{-\infty}^{\infty}{b\arctan\pars{\root{x^{2} + a^{2}}/b} \over \pars{x^{2} + b^{2}}\root{x^{2} + a^{2}}}\,\dd x}} = 2\int_{0}^{\infty}{\arctan\pars{\root{x^{2} + a^{2}}/\verts{b}} \over \root{x^{2} + a^{2}}/\verts{b}}\,{\dd x \over x^{2} + b^{2}} \\[5mm] = &\ 2\int_{0}^{\infty}\int_{1}^{\infty}{\dd t \over t^{2} + \pars{x^{2} + a^{2}}/b^{2}}\,{\dd x \over x^{2} + b^{2}} = 2b^{2}\int_{1}^{\infty}\int_{0}^{\infty}{\dd x \over \pars{x^{2} + b^{2}t^{2} + a^{2}}\pars{x^{2} + b^{2}}}\,\dd t \\[5mm] = &\ {2 \over \verts{b}}\int_{1}^{\infty}\int_{0}^{\infty}{\dd x \over \pars{x^{2} + t^{2} + \mu^{2}}\pars{x^{2} + 1}}\,\dd t\,, \qquad\qquad\qquad\mu \equiv {a \over b} \end{align}


\begin{align} &\bbox[#ffd,10px]{\ds{\int_{-\infty}^{\infty}{b\arctan\pars{\root{x^{2} + a^{2}}/b} \over \pars{x^{2} + b^{2}}\root{x^{2} + a^{2}}}\,\dd x}} \\[5mm] = &\ {2 \over \verts{b}}\int_{1}^{\infty}{1 \over t^{2} + \mu^{2} - 1} \int_{0}^{\infty}{\dd x \over x^{2} + 1}\,\dd t - {2 \over \verts{b}}\int_{1}^{\infty}{1 \over t^{2} + \mu^{2} - 1} \int_{0}^{\infty}{\dd x \over x^{2} + t^{2} + \mu^{2}}\,\dd t \\[5mm] = &\ {\pi \over \verts{b}}\int_{1}^{\infty}{\dd t \over t^{2} + \mu^{2} - 1} - {\pi \over \verts{b}}\int_{1}^{\infty}{\dd t \over \pars{t^{2} + \mu^{2} - 1}\root{t^{2} + \mu^{2}}} \\[5mm] = &\ \bbox[#ffe,10px,border:1px groove navy]{{\pi \over \verts{b}}\bracks{{\arctan\pars{\root{\mu^{2} - 1}} - \mrm{arccot}\pars{\root{\mu^{2} - 1}} + \mrm{arccot}\pars{\root{\mu^{4} - 1}} \over \root{\mu^{2} - 1}}}}\,, \quad\mu \equiv {a \over b} \end{align}

The second integral was evaluated as follows

\begin{align} &\bbox[10px,#ffd]{\ds{\int_{1}^{\infty}{\dd t \over \pars{t^{2} + \mu^{2} - 1}\root{t^{2} + \mu^{2}}}}} \,\,\,\stackrel{t\ \mapsto\ 1/t}{=}\,\,\, \int_{0}^{1}{t\,\dd t \over \bracks{\pars{\mu^{2} - 1}t^{2} + 1}\root{1 + \mu^{2}t^{2}}} \\[5mm] \stackrel{t^{2}\ \mapsto\ t}{=}\,\,\, &\ {1 \over 2}\int_{0}^{1}{\dd t \over \bracks{\pars{\mu^{2} - 1}t + 1}\root{1 + \mu^{2}t}} \,\,\,\stackrel{\root{1 + \mu^{2}t}\ \mapsto\ t}{=}\,\,\, \int_{1}^{\root{1 + \mu^{2}}}{\dd t \over \pars{\mu^{2} - 1}t^{2} + 1} \\[5mm] = &\ {\mrm{arccot}\pars{\root{\mu^{2} - 1}} - \mrm{arccot}\pars{\root{\mu^{4} - 1}} \over \root{\mu^{2} - 1}} \end{align}

Felix Marin
  • 89,464
0

We will use Mark Viola’s derivations to account for the case when $\lvert\,a\,\rvert< \lvert\,b\,\rvert$.

Integrating $(2)$ (See Mark Viola's answer above) and using $\lim_{c\to \infty}F(c)=0$, we find that

$$\begin{align} F(c)&=\pi\,\left(\frac{\arctan\left(\frac{bc}{\sqrt{a^2-b^2}\sqrt{a^2+c^2}}\right)-\arctan\left(\frac{b}{\sqrt{a^2-b^2}}\right)+\pi/2-\arctan\left(\frac{c}{\sqrt{a^2-b^2}}\right)}{\sqrt{a^2-b^2}}\right) \end{align}.$$

Now we will account for our case’s assumption and arrive at

$$\begin{align} F(c)&=\pi\,\left(\frac{-i\tanh^{-1}\left(\frac{bc}{\sqrt{b^2-a^2}\sqrt{a^2+c^2}}\right)+i\tanh^{-1}\left(\frac{b}{\sqrt{b^2-a^2}}\right)+\pi/2+i\tanh^{-1}\left(\frac{c}{\sqrt{b^2-a^2}}\right)}{i\sqrt{b^2-a^2}}\right) \\ &=\pi\,\left(\frac{-\tanh^{-1}\left(\frac{bc}{\sqrt{b^2-a^2}\sqrt{a^2+c^2}}\right)+\tanh^{-1}\left(\frac{b}{\sqrt{b^2-a^2}}\right)-i\pi/2+\tanh^{-1}\left(\frac{c}{\sqrt{b^2-a^2}}\right)}{\sqrt{b^2-a^2}}\right) \end{align}.$$

We replace $c=b$ to obtain our result

$$\int_{-\infty}^\infty \frac{b\arctan\left(\frac{\sqrt{x^2+a^2}}{b}\right)}{(x^2+b^2)\sqrt{x^2+a^2}}\,dx=\\\pi\,\left(\frac{-\tanh^{-1}\left(\frac{b^2}{\sqrt{b^2-a^2}\sqrt{a^2+b^2}}\right)-i\pi/2+2\tanh^{-1}\left(\frac{b}{\sqrt{b^2-a^2}}\right)}{\sqrt{b^2-a^2}}\right).$$

  • After the first line, the term $\sqrt{a^2-b^2}$ should be replaced with $\sqrt{b^2-a^2}$. – Mark Viola Apr 25 '18 at 17:25
  • @Mark Viola You're correct – JohnColtraneisJC Apr 25 '18 at 17:27
  • This solution is incorrect. The integrand is a real to real function, there is no imaginary part in the final result. – Jack Tiger Lam May 02 '18 at 00:15
  • @Jack Lam I don’t see how the solution wouldn’t have an imaginary part when considering this specific case – JohnColtraneisJC May 02 '18 at 14:05
  • integrate differently. there is no imaginary part. choosing case values of a and b and putting them into Mathematica/Wolfram Alpha returns a real number. If you think it’s complex give me a specific example. – Jack Tiger Lam May 02 '18 at 14:34
  • @Jack Lam http://www.wolframalpha.com/input/?i=integrate&rawformassumption=%7B%22F%22,+%22Integral%22,+%22integrand%22%7D+-%3E%22-pi%2F(x%5E2%2B4%2B3sqrt(x%5E2%2B4))%22&rawformassumption=%7B%22F%22,+%22Integral%22,+%22variable%22%7D+-%3E%22x%22&rawformassumption=%7B%22F%22,+%22Integral%22,+%22rangestart%22%7D+-%3E%22-infinity%22&rawformassumption=%7B%22F%22,+%22Integral%22,+%22rangeend%22%7D+-%3E%22infinity%22&rawformassumption=%7B%22C%22,+%22integrate%22%7D+-%3E+%7B%22Calculator%22%7D You're right I am getting a real number-valued output, I am stuck on this problem and really hit the wall. – JohnColtraneisJC May 02 '18 at 15:14
  • @JackLam Would the $-\frac{\pi,i}{2}$ term stay as $\frac{\pi}{2}?$ – JohnColtraneisJC May 02 '18 at 15:31
  • Then the result from integrating couldn't possibly be wrong since Mark's answer was correct, it's just my manipulation of handling $\sqrt{a^2-b^2}$ terms that was done incorrectly. – JohnColtraneisJC May 02 '18 at 15:43
  • @MarkViola Would the step of $\lim_{c\to\infty}F(c)=0$ change the value of this integral for this case? – JohnColtraneisJC May 02 '18 at 16:50
  • @JohnColtraneisJC I don't understand the meaning of the question. – Mark Viola May 02 '18 at 16:57
  • @MarkViola OP is telling me I've arrived at an incorrect solution because an imaginary part is left in the answer, I was wondering whether you knew if the integral's value changes based on this different case or if I handled the imaginary number in the wrong way, specifically for the $\pi/2$ part because there are cancellations everywhere else. – JohnColtraneisJC May 02 '18 at 17:06
  • 1
    The OP is wrong. The inverse hyperbolic tangent with argument greater than $1$ will be a complex number with imaginary part equal too $\pi/2$. And both of the terms $b/\sqrt{b^2-a^2}$ and $b^2/(\sqrt{b^2+a^2}\sqrt{b^2-a^2})$ exceed $1$. – Mark Viola May 02 '18 at 17:16
  • http://www.wolframalpha.com/input/?i=simplify+(-i%2F2)%5Blog((1%2B((bx)%2F(sqrt(b%5E2-a%5E2)+sqrt(a%5E2%2Bx%5E2))))%2F(1-((bx)%2F(sqrt(b%5E2-a%5E2)+sqrt(a%5E2%2Bx%5E2)))))%5D%2B(i%2F2)log(b%2Fsqrt(b%5E2-a%5E2))%2B(i%2F2)log(x%2Fsqrt(b%5E2-a%5E2)) substituting this expression when $a,b,x\in\mathbb{R}$ for every arctan term we find cancellations in imaginary parts for all terms except $\pi/2$ – JohnColtraneisJC May 02 '18 at 17:39