2
$\begingroup$

How can I find this integral $$I=\int_{0}^{1}\frac{2 \sin \pi x \cos \pi x}{1+x^2}dx$$

Any trick that could compute the definite integral is acceptable. However, it will be more challenging to find a primitive.

Any hint or help is appreciated.


My Work

I just wrote the integral as

$$I=\int_{0}^{1}\frac{\sin 2 \pi x}{1+x^2}dx$$

Then, I decided to introduce

$$J(\alpha)=\int_{0}^{1}\frac{\sin \alpha x}{1+x^2}dx$$

and solve a more general problem. Hence, we would have $I=J(2\pi)$ as a special result. But I don't know how to go further. I tried integration by parts and substitutions but to no avail!

$\endgroup$
  • $\begingroup$ Why do you think that this has a "simple" form solution? $\endgroup$ – Mark Viola Dec 19 '15 at 21:42
  • $\begingroup$ @Dr.MV: Maybe the primitive be complicated, But I feel that there should be a trick to just evaluate the definite integral without computing the primitive. Anyway, I have no satisfactory reasons for a simple solution to exist! At least, I want to know how it can be obtained! :) $\endgroup$ – H. R. Dec 19 '15 at 21:47
  • 2
    $\begingroup$ Have you tried differentiating under the integral sign? $\endgroup$ – Zain Patel Dec 19 '15 at 22:40
  • $\begingroup$ @ZainPatel: mmmm, Nope! :) $\endgroup$ – H. R. Dec 19 '15 at 22:41
2
$\begingroup$

We need

$$I = \int_0^1 \dfrac{\sin(2\pi x)}{1+x^2}dx = \sum_{k=0}^{\infty} (-1)^k \int_0^1 x^{2k}\sin(2\pi x)dx$$

We have

\begin{align} I_{k} & = \int_0^1 x^{2k}\sin(2\pi x) dx \\ &= -\dfrac1{2\pi} \int_0^1 x^{2k} d\left(\cos(2\pi x)\right) \\ &= -\dfrac1{2\pi}\left(1-2k\int_0^1 x^{2k-1}\cos(2\pi x)dx\right)\\ & = -\dfrac1{2\pi} \left(1-\dfrac{k}{\pi}\int_0^1x^{2k-1}d\left(\sin(2\pi x)\right)\right) \\ &= -\dfrac1{2\pi}\left(1+\dfrac{k(2k-1)}{\pi}I_{k-1}\right) \end{align} Hence, we obtain $$I = \sum_{k=0}^{\infty} (-1)^k I_k$$ where $I_k$ satisfies the recurrence $$I_k = -\dfrac1{2\pi}\left(1+\dfrac{k(2k-1)}{\pi}I_{k-1}\right)$$ with $I_0 = 0$.


Note for @ClaudeLeibovici: I just made sure that this is due to catastrophic cancellation in finite precision. You might already be familiar with this, but still. Below is the results I obtain using an adaptive Gaussian quadrature for $I_k$ accurate up to $14$ digits on a $64$ bit machine and using the recurrence done on a $64$ bit machine. One always needs to be careful evaluating even a simple recurrence as I have written here. The recurrence arising out of this question goes into my teaching list when I discuss evaluating recurrences in class the next time :-).

enter image description here

The C++ code evaluating the recurrence is shown below.

#include<iostream>
#include<iomanip>
#include<cstdlib>
#include<vector>

double PI   =   3.141592653589793238;

int main(int argc, char* argv[]) {
    int n   =   atoi(argv[1]);
    std::vector<double> I;
    I.push_back(0.0);
    for (int k=1; k<=n; ++k) {
        double temp =   -0.5/PI*(1.0+k*(2.0*k-1.0)/PI*I[k-1]);
        I.push_back(temp);
    }
    for (int j=0; j<=n; ++j) {
        std::cout << "Term " << j << " is: " << std::setprecision(15) << I[j] << "\n";
    }
    double S    =   0.0;
    for (int j=0; j<=n; j+=2) {
        S+=I[j];
    }
    for (int j=1; j<=n; j+=2) {
        S-=I[j];
    }
    std::cout << "\nThe integral is: " << S << "\n";
}
$\endgroup$
  • $\begingroup$ I was going to write this myself! :) Can we solve the recurrence relation? :) $\endgroup$ – H. R. Dec 20 '15 at 10:46
  • 1
    $\begingroup$ I dont know where I may have a mistake but, using the formula, the $I_k$ are just exploding : $I_{30}\approx -6.82906\times 10^{18}$, $I_{31}\approx 6.54218\times 10^{20}$, $I_{32}\approx -6.68165\times 10^{22}$, $I_{99}\approx 1.15444\times 10^{197}$, $I_{100}\approx -1.16384\times 10^{200}$. $\endgroup$ – Claude Leibovici Dec 20 '15 at 12:31
  • 1
    $\begingroup$ @ClaudeLeibovici This is due to round off. Try Wolfram Alpha instead. I ran into the same issue :-) $\endgroup$ – Leg Dec 20 '15 at 15:55
  • 1
    $\begingroup$ @H.R. There is no simple solution to the recurrence. Further, care needs to be taken while evaluating the recurrence. $\endgroup$ – Leg Dec 20 '15 at 15:56
  • 1
    $\begingroup$ @H.R. It is indeed due to round off. See my updated answer. $\endgroup$ – Leg Dec 20 '15 at 17:44
3
$\begingroup$

Let's try your last idea and start with : $$J(\alpha):=\int_{0}^{1}\frac{\sin \alpha x}{1+x^2}dx$$

Then $\;\displaystyle J''(\alpha)=-\int_{0}^{1}\frac{x^2\;\sin \alpha x}{1+x^2}dx\;$ and we obtain this ODE : $$J(\alpha)-J''(\alpha)=\frac {1-\cos(\alpha)}{\alpha}$$ The solution of the homogeneous ODE is simply $\;J(x)=ae^x+be^{-x}$.

Let's use variation of constant and start with $J(x)=a(x)e^x$ then : \begin{align} J(x)&=a(x)e^x\\ J'(x)&=(a'(x)+a(x))e^x\\ J''(x)&=(a''(x)+2a'(x)+a(x))e^x\\ J(x)-J''(x)&=-(a''(x)+2a'(x))e^x=\frac {1-\cos(x)}{x}\\ \end{align} For $b(x):=-a'(x)\,$ we have

\begin{align} b'(x)+2b(x)&=\frac {1-\cos(x)}{x}e^{-x}\\ \end{align}

For $b(x):=c(x)e^{-2x}\,$ this becomes for $\operatorname{Ei}$ the exponential integral :

\begin{align} c'(x)&=\frac {1-\cos(x)}{x}e^{x}\\ c(x)&=\int\frac {e^{x}}x\,dx-\int\frac {e^{x+ix}+e^{x-ix}}{2\,x}\,dx\\ c(x)&=C+\operatorname{Ei}(x)-\frac 12\operatorname{Ei}((1+i)x)-\frac 12\operatorname{Ei}((1-i)x)\\ \end{align}

Coming back to $a(x)$ :

\begin{align} a'(x)&=-\left(C+\operatorname{Ei}(x)-\frac 12\operatorname{Ei}((1+i)x)-\frac 12\operatorname{Ei}((1-i)x)\right)e^{-2\,x}\\ \end{align}

Now $\ \displaystyle \int \operatorname{Ei}(u\,x)e^{-vx}\,dx=\frac 1v\left(\operatorname{Ei}((u-v)x)-e^{-vx}\operatorname{Ei}(u x)\right)\;$ from Wolfram functions so that for $\,v=2$ and $u=1,1+i,1-i$ : $$a(x)=D-\frac 12\left(C_1\,e^{-2x}+\operatorname{Ei}(-x)-e^{-2x}\operatorname{Ei}(x)\\-\frac 12\left(\operatorname{Ei}((-1+i)x)-e^{-2x}\operatorname{Ei}((1+i)x)+\operatorname{Ei}((-1-i)x)-e^{-2x}\operatorname{Ei}((1-i)x)\right)\right)$$

Multiplying by $e^x$ we get the general formula for $J(x)$ : $$J(x)=De^x+D_1\,e^{-x}-\frac {e^x\operatorname{Ei}(-x)-e^{-x}\operatorname{Ei}(x)}2\\+\frac {e^x\operatorname{Ei}((-1+i)x)-e^{- x}\operatorname{Ei}((1+i)x)+e^x\operatorname{Ei}((-1-i)x)-e^{-x}\operatorname{Ei}((1-i)x)}4$$ (of course getting this with alpha is faster!)

Now we have to find $J(2\pi)$ knowing that $J(0)=0$ (I'll let you try too!).

$\endgroup$
  • $\begingroup$ (+1), Nice Job! And also hard! :) I think that we don't really need the homogeneous solution! Do we? :) $\endgroup$ – H. R. Dec 19 '15 at 23:10
  • $\begingroup$ Thanks @H.R.! Concerning the exponentials I am not sure (I don't exclude say a $\sinh(2\pi)$ at this point...) Btw if you have mathematica (I don't) could you try a FullSimplify for $x=2\pi$ on this. $\endgroup$ – Raymond Manzoni Dec 19 '15 at 23:14
  • 1
    $\begingroup$ They have to be fixed from your condition (here $J(0)=0$ and something else ($J'(0)$ or $J(\infty)$ or whatever). Well the $a,b$ were different constants then...! :-) $\endgroup$ – Raymond Manzoni Dec 19 '15 at 23:18
  • 1
    $\begingroup$ @H.R. Yes we could but I fear that it will be more complicated than doing the same thing on your initial integrand (I mean that integrating something like $\sin(2\pi x)x^{2n}$ may be simpler... I was unable to find something simple related to $\;\operatorname{Ei}(2\pi)$ or $\;\operatorname{Ei}(2\pi(1\pm i))$ ('symmetrized' combined with exp. like at the end or not) so no sure that much will get out of all this sorry... $\endgroup$ – Raymond Manzoni Dec 19 '15 at 23:54
  • 1
    $\begingroup$ @H.R. Good question! (the answer would require to read a whole book about symbolic integration again... :-) with Mma's special additions). Anyway the result is the same with nothing special at value $x=1$ (no closed form seem to be known for $\;\operatorname{Ei}(2\pi)$ or for other real values other than $0$). $\endgroup$ – Raymond Manzoni Dec 20 '15 at 0:25
2
$\begingroup$

I don't think you will get a simple answer judging by Mathematica's output:

Integrate[2 Sin[Pi x] Cos[Pi x]/(1 + x^2), {x, 0, 1}]

produces

Cosh[\[Pi]] (CosIntegral[(-2 + 2 I) \[Pi]] - 2 CosIntegral[2 I \[Pi]] + CosIntegral[(2 + 2 I) \[Pi]]) Sinh[\[Pi]] + 1/2 Cosh[2 \[Pi]] (2 SinhIntegral[2 \[Pi]] + I (SinIntegral[(-2 + 2 I) \[Pi]] + SinIntegral[(2 + 2 I) \[Pi]]))

Numerically,

NIntegrate[2 Sin[Pi x] Cos[Pi x]/(1 + x^2), {x, 0, 1}]

produces

0.0926961.

$\endgroup$
  • 2
    $\begingroup$ Thanks. :) I know it is hard. That's why I am putting it here. Humans do better than machines! :) Please use MathJax for formatting equations! :) $\endgroup$ – H. R. Dec 19 '15 at 21:13
2
$\begingroup$

Let us consider $$I=\int \frac{\sin(2 \pi x)}{1+x^2} \,dx$$ Now, since $$\frac{1}{1+x^2}=\frac i 2\left( \frac{1}{x+i}-\frac{1}{x-i} \right)$$ hence $$I=\frac i 2\left(\int \frac{\sin(2 \pi x)}{x+i} \,dx-\int \frac{\sin(2 \pi x)}{x-i} \,dx\right)$$ Now, considering the first integral, writing $2\pi x=2\pi(x+i)-2i\pi$ and expanding $\sin(a-b)$ ,$$\sin(2 \pi x)=\cosh (2\pi ) \sin (2 \pi (x+i))- i \sinh (2\pi) \cos (2 \pi (x+i))$$ Now, change variable $2\pi(x+i)=t$ to make $$\int \frac{\sin(2 \pi x)}{x+i} \,dx=\cosh(2\pi)\int \frac{\sin(t)}{t} \,dt-i \sinh(2\pi)\int \frac{\cos(t)}{t} \,dt$$ and using $$\int \frac{\sin(t)}{t} \,dt=\text{Si}(t)\quad , \quad \int \frac{\cos(t)}{t} \,dt=\text{Ci}(t)$$ Doing almost the same for the second integral, recombining and back to $x$ $$2I=\sinh (2 \pi ) (\text{Ci}(2 \pi (i-x))+\text{Ci}(2 \pi (i+x)))+i \cosh (2 \pi ) (\text{Si}(2 \pi (i+x))+\text{Si}(2 \pi (i- x))$$ from which you get E.tsukerman's result.

Edit

Using the same approach for $$J=\int\frac{\sin (\alpha x)}{1+x^2}\,dx$$ $$2J=\sinh (\alpha ) (\text{Ci}(\alpha(i-x) )+\text{Ci}(\alpha(i+x) ))+i \cosh (\alpha ) (\text{Si}(\alpha(i+x) )+\text{Si}( \alpha(i-x ))$$

Edit

Concerning the numerical value, it is awful (just as already said by other participants).

For $x=1$, the value is $\approx -0.077380457987+420.572696077718 i$

For $x=0$, the value is $\approx -0.170076586683+420.572696077718 i$

Edit

For the antiderivative, doing the same, it would have been faster to compute $$K=\int \frac{e^{i ax}}{1+x^2} \,dx=\frac i 2 \left( e^a \,\text{Ei}(a(ix-1))-e^{-a}\, \text{Ei}(a(i x+1))\right)$$

$\endgroup$
  • $\begingroup$ +1: I don't have much hope for a simple solution here. Cheers anyway! $\endgroup$ – Raymond Manzoni Dec 20 '15 at 9:00
  • $\begingroup$ @RaymondManzoni. This is simple, no ? Cheers. $\endgroup$ – Claude Leibovici Dec 20 '15 at 9:01
  • $\begingroup$ well for $x=1$ you still have $\operatorname{Ci}(2\pi(i+1))$ to evaluate : not much simpler than the initial integral I fear... $\endgroup$ – Raymond Manzoni Dec 20 '15 at 9:04
  • 1
    $\begingroup$ @RaymondManzoni. I totally agree with you concerning the problem of the evaluation. I just tried to show that the antiderivative problem was workable. $\endgroup$ – Claude Leibovici Dec 20 '15 at 9:06
  • 1
    $\begingroup$ @H.R. This was done using pen and paper !! Cheers :-) $\endgroup$ – Claude Leibovici Dec 20 '15 at 12:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.