4

I want to show that the solution of a BVP: $\ u(x) = \ln{\frac{1+x}{1-x}}$ is in $L^2(0,1)$, so I need to show that the integral $$\int\limits_0^1\ln^2|\frac{x+1}{x-1}|dx < \infty$$

Just looking at the function, however, it's not even defined in the interval $[0,1]$, right? So can this not actually be a solution to the BVP? Even more generally about the integral itself, does that make it just $0$? And if not, can someone explain how an integral of a function can be defined in a region when the function itself is not?

Also, I know that there exist integral calculators online and have looked this up, specifically using: https://www.integral-calculator.com/

If you put in the given integral it says the integral could not be found and then gives a complex number approximation. What should be made of this? I have taken a complex variables course but don't really see how you could solve this using Residue Calculus since we don't have any symmetries.

EDIT: Note I had originally forgot the absolute value signs - my mistake.

Quanto
  • 97,352
Mjoseph
  • 989
  • 1
  • 9
  • 22
  • The bounds of the integral are $0$ and $1$, but you speak about the interval $[-1, 1]$ — is this just a typo or did you mean not defined in the interval $[0,1]$? What was the original BVP that you believe this function is a solution to? – Steven Stadnicki Apr 23 '21 at 16:28
  • I edited the typo although I think actually the function is still not defined in $[-1,1]$ at all – Mjoseph Apr 23 '21 at 16:30
  • Since $\ln^2\frac{z+1}{z-1}$ is non-analytic over $z\in[0,1]$, I suspect this integral be well-defined even in complex domain. – Mostafa Ayaz Apr 23 '21 at 16:32
  • Does $\ln^2(x)$ means $(\ln(x))^2$? $\ln(\ln(x))$? Presumably it's not supposed to mean $\ln(x)^2$. – user170231 Apr 23 '21 at 16:33
  • 1
    @user170231 Canonically $\ln^2(x)$ means $\left(\ln(x)\right)^2$. This is frequent notation with logarithms and with trig functions. – Steven Stadnicki Apr 23 '21 at 16:33
  • 1
    Oh I see, OP is trying to show $u(x)$ is square-integrable. – user170231 Apr 23 '21 at 16:34
  • It would really help to know the BVP that this is trying to solve. I suspect the solution you've given is incorrect, but for instance $u(x)=\ln\frac{1+x}{1-x}$ — which differs from your function by only an 'additive constant' — is in $L^2(0,1)$. – Steven Stadnicki Apr 23 '21 at 16:36
  • The BVP is the homogeneous Legendre Equation with $\mu=0$ over $[-1,1]$. My real objective is to show whether $x=1$ is a singular point of the limit point or limit circle type. I wanted to do so using a theorem form Weyl we were given in class but to do so I had to use an interval with one regular endpoint - hence why I'm talking about the interval $[0,1]$ – Mjoseph Apr 23 '21 at 16:38
  • To do so I need to find linearly independent solutions to $-[(1-x^2)u']'=0$ and see whether one or both are in $L^2_w$ but for this instance $w(x)=1$. From the rest of the work I've done on this problem, I suspect that both would be $L^2$ and that I'm overthinking the calculation of this integral; however, I am not quite sure. Thanks for any help! – Mjoseph Apr 23 '21 at 16:39
  • @joseph I would definitely ask your 'real question' rather than this one and show the work that you've done there. Given that $u_1=u+c$ will still satisfy $[(1-x^2)u']'=0$ I'm relatively sure that you've got the wrong 'branch' of the logarithm and that you want the $u$ I gave above rather than the form you have, but this would all be better as part of the question itself IMHO. – Steven Stadnicki Apr 23 '21 at 16:57
  • https://math.stackexchange.com/questions/4113811/determining-whether-a-singular-endpoint-is-of-limit-circle-or-limit-point-case – Mjoseph Apr 23 '21 at 17:33
  • I reposted the question (which you can find from the above link) including my entire thoughts on the subject @StevenStadnicki – Mjoseph Apr 23 '21 at 17:35

5 Answers5

7

$$ \begin{align} \int_0^1\log\left(\frac{1+x}{1-x}\right)^2\mathrm{d}x &=\int_1^\infty\log(u)^2\frac{2\,\mathrm{d}u}{(u+1)^2}\tag1\\ &=4\int_1^\infty\frac1{u+1}\frac{\log(u)}u\,\mathrm{d}u\tag2\\ &=-4\int_0^1\frac{\log(u)}{u+1}\,\mathrm{d}u\tag3\\ &=-4\sum_{n=0}^\infty\int_0^1(-1)^nu^n\log(u)\,\mathrm{d}u\tag4\\ &=4\sum_{n=0}^\infty\frac{(-1)^n}{(n+1)^2}\tag5\\ &=4\cdot\frac{\pi^2}{12}\tag6\\[3pt] &=\frac{\pi^2}3\tag7 \end{align} $$ Explanation:
$(1)$: $u=\frac{1+x}{1-x}\implies x=\frac{u-1}{u+1}$ and $\mathrm{d}x=\frac{2\,\mathrm{d}u}{(u+1)^2}$
$(2)$: integrate by parts
$(3)$: substitute $u\mapsto1/u$
$(4)$: apply the Taylor series for $\frac1{1+u}$
$(5)$: $\int_0^1u^n\log(u)\,\mathrm{d}u=-\frac1{(n+1)^2}$
$(6)$: $\sum\limits_{n=1}^\infty\frac{(-1)^{n-1}}{n^2}=\sum\limits_{n=1}^\infty\frac1{n^2}-2\sum\limits_{n=1}^\infty\frac1{(2n)^2}=\frac{\pi^2}6-\frac{\pi^2}{12}$
$(7)$: simplify

robjohn
  • 345,667
  • 1
    This has been a greatly helpful answer! Not only am I quite able to follow this but it's also been easy to verify each step - thank you! – Mjoseph Apr 25 '21 at 14:35
5

Substitute $\frac{1-x}{1+x}\to x $

\begin{align} \int_0^1\ln^2{\frac{1+x}{1-x}} dx = &\int_0^1\frac{2\ln^2{x}}{(1+x)^2}dx= \int_0^1{\ln^2{x}}\>d\left(\frac {2x}{1+x}\right)\\ =& -4\int_0^1\frac{\ln x}{1+x}dx=-4\cdot (-\frac{\pi^2}{12})=\frac{\pi^2}3 \end{align}

$ \int_0^1\frac{\ln x}{1+x}dx =-\int_0^1\frac{\ln (1+x)}xdx =-\frac{\pi^2}{12}$

Quanto
  • 97,352
5

Actually, this integral is very straightforward to evaluate using Cauchy's Theorem in the complex plane.

Consider the complex integral

$$\oint_C dz \, \log^3{\left (\frac{z+1}{z-1} \right )} $$

where $C$ is the following contour in the complex plane:

enter image description here

The outer arc of $C$ has a radius $R$ and the small circular pieces around the branch points at $z=\pm 1$ have radius $\epsilon$. The contour integral is then equal to

$$e^{i \pi} \int_R^{1+\epsilon} dx \, \log^3{\left (\frac{x-1}{x+1} \right )} + i \epsilon \int_{\pi}^0 dx \, e^{i \phi} \, \log^3{\left (\frac{-1+\epsilon e^{i \phi}+1}{-1+\epsilon e^{i \phi}-1} \right )} \\ + \int_{-1+\epsilon}^{1-\epsilon} dx \, \left [\log{\left (\frac{1+x}{1-x} \right )} -i \pi\right ]^3 + i \epsilon \int_{\pi}^{-\pi} d\phi \, e^{i \phi} \, \log^3{\left (\frac{1+\epsilon e^{i \phi}+1}{1+\epsilon e^{i \phi}-1} \right )} \\ - \int_{-1+\epsilon}^{1-\epsilon} dx \, \left [\log{\left (\frac{1+x}{1-x} \right )} +i \pi \right ]^3 + i \epsilon \int_{2 \pi}^{\pi} dx \, e^{i \phi} \, \log^3{\left (\frac{-1+\epsilon e^{i \phi}+1}{-1+\epsilon e^{i \phi}-1} \right )} \\ - e^{-i \pi} \int_R^{1+\epsilon} dx \, \log^3{\left (\frac{x-1}{x+1} \right )} + i R \int_{-\pi}^{\pi} d\theta \, e^{i \theta} \, \log^3{\left (\frac{R e^{i \theta}+1}{R e^{i \theta}-1} \right )} $$

As $\epsilon \to 0$ and $R \to \infty$, all integrals vanish or cancel except the third and fifth; the contour integral is then equal to

$$\int_{-1}^1 dx \, \left \{\left [\log{\left (\frac{1+x}{1-x} \right )} -i \pi\right ]^3 -\left [\log{\left (\frac{1+x}{1-x} \right )} +i \pi\right ]^3 \right \} $$

By Cauchy's Theorem, the contour integral is equal to zero. Expanding the integrand, we immediately determine the value of the integral without further computation, because the cube and linear powers of the log cancel. That is, we are left with the log squared term sought, and a constant term that is trivially dealt with. The result is, using symmetry of an even function:

$$\int_0^1 dx \, \log^2{\left (\frac{1+x}{1-x} \right )} = \frac{\pi^2}{3}$$

Ron Gordon
  • 138,521
5

A handy substitution when our limits are from $x=0$ and $x=1$ is $$t=\frac {1-x}{1+x}\qquad\implies\qquad\mathrm dx=-\frac {2\,\mathrm dt}{(1+t)^2}$$ The integral becomes $$\begin{align*}\mathfrak{I} & =\int\limits_0^1\mathrm dx\,\log^2\left(\frac {1-x}{1+x}\right)\\ & =2\int\limits_0^1\mathrm dt\,\frac {\log^2t}{(1+t)^2}\end{align*}$$ Now recall the geometric series formula $$\sum\limits_{n\geq1}x^{n-1}=\frac 1{1-x}$$ Differentiating with respect to $x$ once gives $$\sum\limits_{n\geq1}nx^{n-1}=\frac 1{(1-x)^2}$$ Replacing the integrand with our modified geometric sequence, then $$\begin{align*}\mathfrak{I} & =2\sum\limits_{n\geq1}n(-1)^{n-1}\int\limits_0^1\mathrm dt\, t^{n-1}\log^2t\\ & \stackrel{\text{IBP}}{=}4\sum\limits_{n\geq1}\frac {(-1)^{n-1}}{n^2}\end{align*}$$ If you're familiar with Basel's problem, then it's easy to see through some infinite sum manipulation that $$\begin{align*}\sum\limits_{n\geq1}\frac {(-1)^{n-1}}{n^2} & =\sum\limits_{n\geq1}\frac 1{n^2}-2\sum\limits_{n\geq1}\frac 1{(2n)^2}\\ & =\frac {\pi^2}{12}\end{align*}$$ Hence $$\int\limits_0^1\mathrm dx\,\log^2\left(\frac {1-x}{1+x}\right)=4\left(\frac {\pi^2}{12}\right)=\frac {\pi^2}{3}$$

Frank W
  • 5,897
3

Do you know that :

$$f(x):=\operatorname{ln}\frac{1+x}{1-x}=2\operatorname{arctanh}(x) \ ? \ \tag{1}$$

Remark: $\operatorname{arctanh}$ is the same as $\operatorname{tanh}^{-1}$.

Besides, Wolfram Alpha gives

$$\int_0^1 \operatorname{arctanh}(x)^2 dx= \pi^2/12$$

Therefore your result is $\pi^2/48 < \infty$

Wolfram Alpha also gives an (awful) expression for a primitive function of $f$.

Jean Marie
  • 81,803
  • It sounds like you are aware of this, but on $(0,1)$, $\frac{x+1}{x-1}$ is negative. So it's not clear what $\ln\left(\frac{x+1}{x-1}\right)$ means for such $x$-values. – 2'5 9'2 Apr 23 '21 at 18:31
  • @JeanMarie is right I made a careless mistake in dropping the absolute value signs – Mjoseph Apr 23 '21 at 18:34
  • Hi, isn't it more standard to use the notation $\operatorname{artanh}$ instead of $\operatorname{arctanh}$? – A-Level Student Apr 25 '21 at 21:12
  • @A-level Student Thanks for having drawn my attention to the notation with "ar" (https://en.wikipedia.org/wiki/Inverse_hyperbolic_functions) that honestly I didn't know. In my country (France) we still use peculiar old notations (ch,sh,th instead of cosh, sinh, tanh, etc., argth instead of arctanh or arctanh or atanh) – Jean Marie Apr 25 '21 at 21:37
  • @JeanMarie Interesting! My pleasure :) – A-Level Student Apr 25 '21 at 21:40