Given the integral $\int_{1}^{+\infty}\frac{1}{(1+a^2+x^2)^{\frac{\rho}{2}}(1+2\sqrt{E}x)}dx$, with $a,E>0$ and $\rho>2$, I want to know how to determine how it goes to 0 as $a\to +\infty$. Maybe the answer is trivial, but I prefer being cautious

  • $\begingroup$ Do you mean $\sqrt Ex$ or $\sqrt{Ex}$ ? $\endgroup$ – Yves Daoust Mar 26 '20 at 13:10
  • $\begingroup$ Root only on $E$ $\endgroup$ – SpuriousMatemagician Mar 27 '20 at 18:01
  • $\begingroup$ In such cases, don't clutter with a constant like $2\sqrt E$. $b$ is good enough. $\endgroup$ – Yves Daoust Mar 27 '20 at 19:37

I'll write $b=2\sqrt E$ to simplify things.

To get a systematic expansion, write

\begin{eqnarray} \int_1^a\frac1{(1+a^2+x^2)^\frac\rho2(1+bx)}\,\mathrm dx &=& \frac1{a^\rho b}\int_1^a\frac1{(1+\frac{x^2+1}{a^2})^\frac\rho2(1+\frac1{bx})x}\,\mathrm dx \\ &=& \frac1{a^\rho b}\int_1^a\sum_{j=0}^\infty\sum_{k=0}^\infty\binom{-\frac\rho 2}j\left(\frac{x^2+1}{a^2}\right)^j\left(-\frac1{bx}\right)^k\frac1x\,\mathrm dx \\ &=& \sum_{j=0}^\infty\sum_{k=0}^\infty\sum_{n=0}^j\frac{(-1)^k}{a^{\rho+2j}b^{k+1}}\binom{-\frac\rho 2}j\binom jn\int_1^ax^{2n-k-1}\,\mathrm dx\;. \\ &=& \sum_{j=0}^\infty\sum_{k=0}^\infty\sum_{n=0}^j\frac{(-1)^k}{a^{\rho+2j}b^{k+1}}\binom{-\frac\rho 2}j\binom jn \begin{cases} \ln a&2n-k=0\;,\\ \frac{a^{2n-k}-1}{2n-k}&\text{otherwise}\;. \end{cases} \end{eqnarray}

The leading term is the one for $j=k=0$, which yields

$$ \frac{\ln a}{a^\rho b}\;, $$

as in @stokes-line’s answer. The next term comes from contributions for $k=0$, $n=j\gt0$:

$$ \sum_{j=1}^\infty\frac1{a^{\rho+2j}b}\binom{-\frac\rho 2}j\frac{a^{2j}}{2j} = -\frac1{a^\rho b}\frac\rho4\,_3F_2\left(1,1,\frac\rho2+1;2,2;-1\right)\;, $$

where $_3F_2$ is a generalized hypergeometric series, and for $j=0$, $k\gt0$:

$$ \sum_{k=1}^\infty\frac{(-1)^k}{a^\rho b^{k+1}}\frac{-1}{-k} = -\frac{\ln\left(1+\frac1b\right)}{a^\rho b}\;. $$

The term after that comes from $k=1$, $n=j$:

$$ -\sum_{j=0}^\infty\frac1{a^{\rho+2j}b^2}\binom{-\frac\rho2} j\frac{a^{2j-1}}{2j-1} = \frac1{a^{\rho+1}b^2}\,_2F_1\left(-\frac12,\frac\rho2;\frac12;-1\right)\;. $$

For the part beyond $a$, we need to expand in $\frac1x$ instead, yielding

\begin{eqnarray} \int_a^\infty\frac1{(1+a^2+x^2)^\frac\rho2(1+bx)}\,\mathrm dx &=& \int_a^\infty\frac1{bx^{\rho+1}}\frac1{\left(1+\frac{a^2+1}{x^2}\right)^\frac\rho2\left(1+\frac1{bx}\right)}\,\mathrm dx \\ &=& \int_a^\infty\frac1{bx^{\rho+1}}\sum_{j=0}^\infty\sum_{k=0}^\infty\binom{-\frac\rho2}j\left(\frac{a^2+1}{x^2}\right)^j\left(-\frac1{bx}\right)^k\,\mathrm dx \\ &=& \sum_{j=0}^\infty\sum_{k=0}^\infty\binom{-\frac\rho2}j\frac{(-1)^k}{a^{\rho+2j+k}b^{k+1}}(a^2+1)^j\int_1^\infty t^{-(\rho+2j+k+1)}\,\mathrm dt \\ &=& \sum_{j=0}^\infty\sum_{k=0}^\infty\sum_{n=0}^j\binom{-\frac\rho2}j\binom jn\frac{(-1)^k}{a^{\rho+2j+k-2n}b^{k+1}}\frac1{\rho+2j+k}\;. \end{eqnarray}

Here there is no logarithmic term; the first term comes from $k=0$, $n=j$:

$$ \sum_{j=0}^\infty\binom{-\frac\rho2}j\frac1{a^\rho b}\frac1{\rho+2j}=\frac1{a^\rho b}\frac1\rho\,_2F_1\left(\frac\rho2,\frac\rho2;\frac\rho2+1;-1\right)\;. $$

The next term comes from $k=1$, $n=j$:

$$ -\sum_{j=0}^\infty\binom{-\frac\rho2}j\frac1{a^{\rho+1}b^2}\frac1{\rho+2j+1}=-\frac1{a^{\rho+1}b^2}\frac1{\rho+1}\,_2F_1\left(\frac\rho2,\frac{\rho+1}2;\frac{\rho+3}2;-1\right)\;. $$

Thus, altogether we have the expansion

$$ \int_1^a\frac1{(1+a^2+x^2)^\frac\rho2(1+bx)}\,\mathrm dx \\ = \frac{\ln a-\ln\left(1+\frac1b\right)}{a^\rho b} + \left(\frac1\rho\,_2F_1\left(\frac\rho2,\frac\rho2;\frac\rho2+1;-1\right)-\frac\rho4\,_3F_2\left(1,1,\frac\rho2+1;2,2;-1\right)\right)\frac1{a^\rho b} \\ +\left(_2F_1\left(-\frac12,\frac\rho2;\frac12;-1\right)-\frac1{\rho+1}\,_2F_1\left(\frac\rho2,\frac{\rho+1}2;\frac{\rho+3}2;-1\right)\right)\frac1{a^{\rho+1}b^2} \\ +O\left(\frac1{a^{\rho+2}}\right)\;. $$

For integer values of $\rho$, the hypergeometric series yield simple expressions. For instance, for $\rho=2$ the expansion takes the form

$$ \int_1^a\frac1{(1+a^2+x^2)(1+bx)}\,\mathrm dx \\ = \frac{\ln a-\ln\left(1+\frac1b\right)}{a^2 b} + \left(\frac12\,_2F_1\left(1,1;2;-1\right)-\frac12\,_3F_2\left(1,1,2;2,2;-1\right)\right)\frac1{a^2 b} \\ +\left(_2F_1\left(-\frac12,1;\frac12;-1\right)-\frac13\,_2F_1\left(2,\frac32;\frac52;-1\right)\right)\frac1{a^3b^2} \\ +O\left(\frac1{a^4}\right)\;. \\ = \frac{\ln a-\ln\left(1+\frac1b\right)}{a^2 b} +\frac{\frac54+\frac\pi8}{a^3b^2} +O\left(\frac1{a^4}\right)\;. $$

The terms proportional to $\frac1{a^2b}$ cancel. For $\rho=3$, the result is

$$ \int_1^a\frac1{(1+a^2+x^2)^\frac32(1+bx)}\,\mathrm dx \\ = \frac{\ln a-\ln\left(1+\frac1b\right)}{a^3 b} + \left(\frac13\,_2F_1\left(\frac32,\frac32;\frac52;-1\right)-\frac34\,_3F_2\left(1,1,\frac52;2,2;-1\right)\right)\frac1{a^3 b} \\ +\left(_2F_1\left(-\frac12,\frac32;\frac12;-1\right)-\frac14\,_2F_1\left(\frac32,2;3;-1\right)\right)\frac1{a^4b^2} \\ +O\left(\frac1{a^5}\right)\;. \\ = \frac{\ln a-\ln\left(1+\frac1b\right)}{a^3 b} +\frac{\ln2-1}{a^3 b} +\frac2{a^4b^2} +O\left(\frac1{a^5}\right)\;. $$

For $\rho=4$, the result is

$$ \int_1^a\frac1{(1+a^2+x^2)^2(1+bx)}\,\mathrm dx \\ = \frac{\ln a-\ln\left(1+\frac1b\right)}{a^4 b} + \left(\frac14\,_2F_1\left(2,2;3;-1\right)-\,_3F_2\left(1,1,3;2,2;-1\right)\right)\frac1{a^4 b} \\ +\left(_2F_1\left(-\frac12,2;\frac12;-1\right)-\frac15\,_2F_1\left(2,\frac52;\frac72;-1\right)\right)\frac1{a^5b^2} \\ +O\left(\frac1{a^6}\right)\;. \\ = \frac{\ln a-\ln\left(1+\frac1b\right)}{a^4 b} -{\frac12}\frac1{a^4 b} +\frac{3\pi}4\frac1{a^5b^2} +O\left(\frac1{a^6}\right)\;. $$

For $\rho=5$, the result is

$$ \int_1^a\frac1{(1+a^2+x^2)^\frac52(1+bx)}\,\mathrm dx \\ = \frac{\ln a-\ln\left(1+\frac1b\right)}{a^5 b} + \left(\frac14\,_2F_1\left(\frac52,\frac52;\frac72;-1\right)-\,_3F_2\left(1,1,\frac72;2,2;-1\right)\right)\frac1{a^5 b} \\ +\left(_2F_1\left(-\frac12,\frac52;\frac12;-1\right)-\frac15\,_2F_1\left(\frac52,3;4;-1\right)\right)\frac1{a^6b^2} \\ +O\left(\frac1{a^7}\right)\;. \\ = \frac{\ln a-\ln\left(1+\frac1b\right)}{a^5 b} +\frac{\ln2-\frac43}{a^5 b} +\frac83\frac1{a^6b^2} +O\left(\frac1{a^7}\right)\;. $$

  • 1
    $\begingroup$ Thank you so much for this very detailed answer. I am interested only in the leading term, and this helped me understand the way it is computed. Instead, I didn't get why the next term comes from $k=0, n=j>0$ $\endgroup$ – SpuriousMatemagician Mar 27 '20 at 18:12
  • $\begingroup$ @SpuriousMatemagician: That's only the first part of the sentence; the next term comes from contributions from $k=0$, $n=j\gt0$ and from $j=0$, $k\gt0$. If you look at the powers of $a$ in the triple summation, there's $a^{\rho+2j}$ in the denominator, and the integration yields a term with $a^{2n-k}$ and a constant term. $\endgroup$ – joriki Mar 27 '20 at 18:21
  • $\begingroup$ Since $j$, $k$ and $n$ are all non-negative, for these exponents to add up to $a^\rho$ in the denominator, we must have $k=0$ and $n=j$ in the term $a^{2n-k}$ (and then $j\gt0$ because $k=j=0$ is the leading logarithmic term), or $j=0$ in the constant term (and then $k\gt0$ because $k=j=0$ is the leading logarithmic term). (And then of course there's a third contribution of order $a^{-\rho}$ that comes from the part of the integral beyond $a$.) $\endgroup$ – joriki Mar 27 '20 at 18:21
  • $\begingroup$ Ok. One more thing: let's say i am interested only in the leading term, When considering the part beyond $a$, after i have the series rapresentation, I can discard this part when related to the one below $a$, because here I have no logarithmic terms, right? $\endgroup$ – SpuriousMatemagician Mar 28 '20 at 3:03

The leading asymptotics can be found realizing that at large $a$ the integral comes from $1\lesssim x\lesssim a$ and is logarithmic in this range. As a result: $$ I\sim \frac{\ln a}{2 \sqrt{E} a^\rho}. $$

  • $\begingroup$ Can you go a little deeper in explaining this? If i gor it right, I have to split the integral in two parts: in $[1,a]$ and in $[a,+\infty]$. But then? Maybe in the first I can take out $a^2$ and write the Taylor expansion? Anyway i am not sure on how to proceed $\endgroup$ – SpuriousMatemagician Mar 25 '20 at 21:38
  • $\begingroup$ It depends on what accuracy you want. Thats the leading term, to get it you need to observe that the integrand quickly decays at $x>a$ and for $1\lesssim x \lesssim a$ you need to keep dependence on $x$ only in the second term in the denominator. The rest gives subleading corrections. $\endgroup$ – stokes-line Mar 25 '20 at 21:52

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.