4
$\begingroup$

Does the following integral have a closed form solution?

$$ \int_{0}^{y} \exp\left(\,-\sqrt{\,x(1-x)\,}\,\right)\,{\rm d}x $$

Or must I settle with an approximation?


Edit: Actual form of integral has an $\alpha$: $$ \int_{0}^{y} \exp\left(\, -\alpha \sqrt{\,x(1-x)\,}\,\right)\,{\rm d}x $$ I need the solution (or approximate solution) work with large $\alpha$ (and if just possible for small $\alpha$ too


2nd Edit: Is there analytic expression for the definite integral containing the $\alpha$? $$ \int_{0}^{1} \exp\left(\, -\alpha \sqrt{\,x(1-x)\,}\,\right)\,{\rm d}x $$


3rd Edit: Actually it is $-\alpha$ not $\alpha$

$\endgroup$
  • 2
    $\begingroup$ In terms of error and complementary error function yes. But in terms of a standard functions nope. $\endgroup$ – Chinny84 Jul 31 '14 at 0:12
  • 2
    $\begingroup$ One can argue the "error and complementary error function" aren't even "closed form" or "analytic", they are just symbols we assign something. See this $\endgroup$ – Squirtle Jul 31 '14 at 0:18
  • 2
    $\begingroup$ @Chinny84 You predict correctly. This can be shown using the substitution $u=2\sqrt{x(1-x)}$ and comparing the power series. $\endgroup$ – David H Aug 1 '14 at 19:59
  • 2
    $\begingroup$ @Hesam Ok, I think I see what your mistake was. It seems like you replaced the wrong "1/2" by $\alpha/2$. Here's a color coded version of the identity when $\alpha=1$: $\int_{0}^{1}\mathrm{e}^{\sqrt{x(1-x)}}= \color{blue}{\frac{1}{2}}\pi \left[L_{-1}\left(\color{red}{\frac{1}{2}}\right) + I_{1}\left(\color{red}{\frac{1}{2}}\right)\right]$. To generalize the formula to any $alpha$, replace only the $\color{red}{\text{red}}$ 1/2's by $\frac{\alpha}{2}$. $\endgroup$ – David H Aug 2 '14 at 7:19
  • 2
    $\begingroup$ @Hesam I.e., the general formula should read: $\int_{0}^{1}\mathrm{e}^{\color{red}{\alpha}\sqrt{x(1-x)}}= \color{blue}{\frac{1}{2}}\pi \left[L_{-1}\left(\color{red}{\frac{\alpha}{2}}\right) + I_{1}\left(\color{red}{\frac{\alpha}{2}}\right)\right]$. $\endgroup$ – David H Aug 2 '14 at 7:20
1
$\begingroup$

$\int_0^y e^{-\alpha\sqrt{x(1-x)}}~dx$

$=\int_0^y e^{-\alpha\sqrt{-(x^2-x)}}~dx$

$=\int_0^y e^{-\alpha\sqrt{-\left(x^2-x+\frac{1}{4}-\frac{1}{4}\right)}}~dx$

$=\int_0^y e^{-\alpha\sqrt{\frac{1}{4}-\left(x-\frac{1}{2}\right)^2}}~dx$

$=\int_{-\frac{1}{2}}^{y-\frac{1}{2}}e^{-\alpha\sqrt{\frac{1}{4}-x^2}}~dx$

$=\int_\pi^{\cos^{-1}(2y-1)}e^{-\alpha\sqrt{\frac{1}{4}-\left(\frac{\cos x}{2}\right)^2}}~d\left(\dfrac{\cos x}{2}\right)$

$=\dfrac{1}{2}\int_{\cos^{-1}(2y-1)}^\pi e^{-\frac{\alpha\sin x}{2}}\sin x~dx$

$=\int_{\cos^{-1}(2y-1)}^\pi\sum\limits_{n=0}^\infty\dfrac{\alpha^{2n}\sin^{2n+1}x}{2^{2n+1}(2n)!}dx-\int_{\cos^{-1}(2y-1)}^\pi\sum\limits_{n=0}^\infty\dfrac{\alpha^{2n+1}\sin^{2n+2}x}{4^{n+1}(2n+1)!}dx$

For $n$ is any non-negative integer,

$\int\sin^{2n+2}x~dx=\dfrac{(2n+2)!x}{4^{n+1}((n+1)!)^2}-\sum\limits_{k=0}^n\dfrac{(2n+2)!(k!)^2\sin^{2k+1}x\cos x}{4^{n-k+1}((n+1)!)^2(2k+1)!}+C$

This result can be done by successive integration by parts.

$\int\sin^{2n+1}x~dx$

$=-\int\sin^{2n}x~d(\cos x)$

$=-\int(1-\cos^2x)^n~d(\cos x)$

$=-\int\sum\limits_{k=0}^nC_k^n(-1)^k\cos^{2k}x~d(\cos x)$

$=\sum\limits_{k=0}^n\dfrac{(-1)^{k+1}n!\cos^{2k+1}x}{k!(n-k)!(2k+1)}+C$

$\therefore\int_{\cos^{-1}(2y-1)}^\pi\sum\limits_{n=0}^\infty\dfrac{\alpha^{2n}\sin^{2n+1}x}{2^{2n+1}(2n)!}dx-\int_{\cos^{-1}(2y-1)}^\pi\sum\limits_{n=0}^\infty\dfrac{\alpha^{2n+1}\sin^{2n+2}x}{4^{n+1}(2n+1)!}dx$

$=\left[\sum\limits_{n=0}^\infty\sum\limits_{k=0}^n\dfrac{(-1)^{k+1}n!\alpha^{2n}\cos^{2k+1}x}{2^{2n+1}(2n)!k!(n-k)!(2k+1)}\right]_{\cos^{-1}(2y-1)}^\pi-\left[\sum\limits_{n=0}^\infty\dfrac{\alpha^{2n+1}x}{2^{4n+3}n!(n+1)!}-\sum\limits_{n=0}^\infty\sum\limits_{k=0}^n\dfrac{(k!)^2\alpha^{2n+1}\sin^{2k+1}x\cos x}{2^{4n-2k+3}n!(n+1)!(2k+1)!}\right]_{\cos^{-1}(2y-1)}^\pi$

$=\sum\limits_{n=0}^\infty\sum\limits_{k=0}^n\dfrac{(-1)^kn!\alpha^{2n}((2y-1)^{2k+1}+1)}{2^{2n+1}(2n)!k!(n-k)!(2k+1)}+\sum\limits_{n=0}^\infty\dfrac{\alpha^{2n+1}(\cos^{-1}(2y-1)-\pi)}{2^{4n+3}n!(n+1)!}+\sum\limits_{n=0}^\infty\sum\limits_{k=0}^n\dfrac{(k!)^2\alpha^{2n+1}(2y-1)(1-(2y-1)^2)^{k+\frac{1}{2}}}{2^{4n-2k+3}n!(n+1)!(2k+1)!}$

$\endgroup$
7
+50
$\begingroup$

$\textbf{Possible direction}$

$$ \sqrt{x(1-x)} = \sqrt{\frac{1}{4}-\left(x-\frac{1}{2}\right)^2} = \frac{1}{2}\sqrt{1-4\left(x-\frac{1}{2}\right)^2}. $$ thus making a change of variable $$ \cos(t) = 2\left(x-\frac{1}{2}\right),\\ -\sin(t)dt = 2dx. $$ we can re-write the integral as $$ -\frac{1}{2}\int_{\pi}^{0} \mathrm{e}^{\frac{\alpha}{2}\sin(t)}\sin(t)dt = \frac{1}{2}\int_{0}^{\pi} \mathrm{e}^{\frac{\alpha}{2}\sin(t)}\sin(t)dt $$

so the original integral looks like $$ \frac{1}{2}\int_{0}^{\pi}\mathrm{e}^{\frac{\alpha}{2}\sin(s)}\sin(s)ds = \frac{1}{2}\pi \left[L_{-1}\left(\frac{\alpha}{2}\right) + I_{1}\left(\frac{\alpha}{2}\right)\right]\approx 1.48983 $$ $\textbf{update:}$ There seems to be a discrepancy between this "answer" and the numerical result obtained in the comments above. If you could please refrain from voting until the differences are accounted for that will be great. Cheers!

$\textbf{update 2}$ The last equation has been modified from my previous answer due to my silly mistake of converting the limits of the integration! Anyway, using mathematica (apologies) the final integral is of the form modified Struve function $L_{-1}(x)$. The answer is approximately 1.48983 for $\alpha =1$ and limits 0 to 1, which corresponds to previous comment above.

$\textbf{update 3}$ Modified answer as mentioned in the comments to the original question. Now try to find a way to represent the special functions with negative arguments.

$\endgroup$
  • $\begingroup$ Thanks a lot, how I can change the integration (from x=0 to x=y) instead of (from x=0 to x=1)? $\endgroup$ – Hesam Jul 31 '14 at 1:12
  • 1
    $\begingroup$ Unfortunately I make use of the identity Eq(**) which (I have not tried to prove this one way or the other) works for 0 to $2\pi$. Maybe wait for a more complete answer or I can try to modify in the morning :). $\endgroup$ – Chinny84 Jul 31 '14 at 1:28
  • $\begingroup$ The value of integral from 0 to 1 seems to be 1.5 (numerical integration value). But I can not get that value from your solution. Is it right? $\endgroup$ – Hesam Jul 31 '14 at 1:28
  • 1
    $\begingroup$ How have you dealt with the negative argument in the modified Bessel function? I will have a further look at the numerical values compared with the "answer" above in the morning. $\endgroup$ – Chinny84 Jul 31 '14 at 1:31
  • 1
    $\begingroup$ But I am taking the derivative of $I_0$ not the Struve function. :/ $\endgroup$ – Chinny84 Aug 6 '14 at 7:43
5
$\begingroup$

For the range $0 \leq y \leq 1$, you may have a very accurate estimation of the integral expanding first the integrand as a Taylor series built at $x=0$. This gives $$e^{\sqrt{x(1-x)}}=1+\sqrt{x}+\frac{x}{2}-\frac{x^{3/2}}{3}-\frac{11 x^2}{24}-\frac{11 x^{5/2}}{30}-\frac{59 x^3}{720}-\frac{13 x^{7/2}}{630}+\frac{1513 x^4}{40320}-\frac{311 x^{9/2}}{22680}+\frac{14761 x^5}{3628800}-\frac{31417 x^{11/2}}{1247400}-\frac{594659 x^6}{479001600}-\frac{1877123 x^{13/2}}{97297200}-\frac{8409491 x^7}{87178291200}+O\left(x^{15/2}\right)$$ Then, for the integral, you have $$\int_{0}^{y} e^{\sqrt{x(1-x)}}\,{\rm d}x=y+\frac{2 y^{3/2}}{3}+\frac{y^2}{4}-\frac{2 y^{5/2}}{15}-\frac{11 y^3}{72}-\frac{11 y^{7/2}}{105}-\frac{59 y^4}{2880}-\frac{13 y^{9/2}}{2835}+\frac{1513 y^5}{201600}-\frac{311 y^{11/2}}{124740}+\frac{14761 y^6}{21772800}-\frac{31417 y^{13/2}}{8108100}-\frac{594659 y^7}{3353011200}-\frac{1877123 y^{15/2}}{729729000}-\frac{8409491 y^8}{697426329600}+O\left(y^{17/2}\right)$$ which is quite accurate.

Added later after OP's request

If we consider the case of $$\int_{0}^{y} e^{a\sqrt{x(1-x)}}\,{\rm d}x$$ assuming that $a$ is small, an identical procedure first leads to $$e^{a\sqrt{x(1-x)}}=1+a \sqrt{x}+\frac{a^2 x}{2}+\frac{1}{6} a \left(a^2-3\right) x^{3/2}+\frac{1}{24} a^2 \left(a^2-12\right) x^2+\frac{1}{120} a \left(a^4-30 a^2-15\right) x^{5/2}+\frac{1}{720} a^4 \left(a^2-60\right) x^3+\frac{a \left(a^6-105 a^4+315 a^2-315\right) x^{7/2}}{5040}+\frac{a^4 \left(a^4-168 a^2+1680\right) x^4}{40320}+\frac{a \left(a^8-252 a^6+5670 a^4+3780 a^2-14175\right) x^{9/2}}{362880}+\frac{a^6 \left(a^4-360 a^2+15120\right) x^5}{3628800}+\frac{a \left(a^{10}-495 a^8+34650 a^6-103950 a^4+155925 a^2-1091475\right) x^{11/2}}{39916800}+\frac{a^6 \left(a^6-660 a^4+71280 a^2-665280\right) x^6}{479001600}+\frac{a \left(a^{12}-858 a^{10}+135135 a^8-2702700 a^6-2027025 a^4+12162150 a^2-127702575\right) x^{13/2}}{6227020800}+\frac{a^8 \left(a^6-1092 a^4+240240 a^2-8648640\right) x^7}{87178291200}+O\left(x^{15/2}\right)$$ and, then, for the integral $$\int_{0}^{y} e^{a\sqrt{x(1-x)}}\,{\rm d}x=y+\frac{2}{3} a y^{3/2}+\frac{a^2 y^2}{4}+\frac{1}{15} a \left(a^2-3\right) y^{5/2}+\frac{1}{72} a^2 \left(a^2-12\right) y^3+\frac{1}{420} a \left(a^4-30 a^2-15\right) y^{7/2}+\frac{a^4 \left(a^2-60\right) y^4}{2880}+\frac{a \left(a^6-105 a^4+315 a^2-315\right) y^{9/2}}{22680}+\frac{a^4 \left(a^4-168 a^2+1680\right) y^5}{201600}+\frac{a \left(a^8-252 a^6+5670 a^4+3780 a^2-14175\right) y^{11/2}}{1995840}+\frac{a^6 \left(a^4-360 a^2+15120\right) y^6}{21772800}+\frac{a \left(a^{10}-495 a^8+34650 a^6-103950 a^4+155925 a^2-1091475\right) y^{13/2}}{259459200}+\frac{a^6 \left(a^6-660 a^4+71280 a^2-665280\right) y^7}{3353011200}+\frac{a \left(a^{12}-858 a^{10}+135135 a^8-2702700 a^6-2027025 a^4+12162150 a^2-127702575\right) y^{15/2}}{46702656000}+\frac{a^8 \left(a^6-1092 a^4+240240 a^2-8648640\right) y^8}{697426329600}+O\left(y^{17/2}\right)$$

$\endgroup$
  • $\begingroup$ Thanks, Could you mention what if we have a constant coefficient alpha in the exponent: exp(alpha * (x*(1-x))^.5) $\endgroup$ – Hesam Jul 31 '14 at 3:29
  • 1
    $\begingroup$ I shall update my answer right now. $\endgroup$ – Claude Leibovici Jul 31 '14 at 3:36
  • $\begingroup$ Sure, I am waiting to see if someone can provide analytic short form solution. $\endgroup$ – Hesam Jul 31 '14 at 5:01
  • 1
    $\begingroup$ It is normal that large $a$'s will cause problems. Let me think about it. $\endgroup$ – Claude Leibovici Jul 31 '14 at 8:27
  • 1
    $\begingroup$ @Chinny84. Thank you ! I really appreciate this comment from you. Cheers :) $\endgroup$ – Claude Leibovici Jul 31 '14 at 11:21
1
$\begingroup$

I'm sorry I'm not able to write good mathematical content on this site yet, I've just signed up.

I think it is possible to find an analytical solution to this. First the function : $x --> x*(1-x)$ is bijective when x is between 0 and $\frac{1}{2}$. So we can assume that x verifies that, or we seperate the integral in two, one part from 0 to $\frac{1}{2}$ and the other from $\frac{1}{2}$ to y knowing that the method to calculate it is the same.

First we change variables with $u^2 = x*(1-x)$ that gives $x= (1-\frac{\sqrt{1-4*u^2}}{2})$ (two solutions but only one of them is smaller than $\frac{1}{2}$))

replace it in the integral and use an integration by parts on $4*\frac{u}{\sqrt{(1-4*u^2)}}$ that should show up after the correct re-writing of the integral following the change of variable.

Then to simplify notation, use another one : $v=2*u$ ( we could do it before the integration by part)

If I'm right you should obtain a certain function of y (the main variable) which is made of roots and exponentials so is known) and $\int \sqrt{(1-v^2)}*e^{\frac{v}{2}}$ delimited by the proper values ( here 0 and $\frac{1}{2}*\sqrt{y(1-y)}$ ). Then you use a sin change of variable : v=sin(t) and it gives you :

the same term function of y, and $\int cos(t)^2*e^{\frac{sin(t)}{2}}$ From here, you use two integration by parts to sort the thing out and it should give you the analytical solution, ugly as it is :) . I may have been mistaken at some point so please don't hesitate to tell me if it turns out to be so. Thank you to those who read this to the end :)

$\endgroup$
  • 3
    $\begingroup$ Welcome to MSE! You'll want to read this tutorial on Mathjax so that you can make your mathematics as accessible and clear as possible. In the same vein, I'd suggest you convert as much of your words into specific mathematical content (integrals, changes of variables, etc.) That'll make it much easier for everyone to assess your argument. $\endgroup$ – Semiclassical Jul 31 '14 at 1:55

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.