6
$\begingroup$

$$ \int_{1}^{\infty} \frac{\sin^2 (\mu \sqrt{x^2 -1})}{(x+1)^{\frac{9}{2}} (x-1)^{\frac{3}{2}}} \,dx $$ Note: $\mu$ here is an extremely small constant.

I have tried:

  1. Estimating the integral by Taylor expansion of $\sin^2(\mu \sqrt(x^2 - 1)$ but the it diverges after few terms.
  2. I have also tried to split the integral into two regions, solving the two integrals differently and then adding them in the hope that the arbitrary parameter $\alpha$ will vanish (asymptotic matching/splitting). While I have been able to solve the two integrals but $\alpha$ does not vanish: $$\int_{1}^{\alpha} f(x)\,dx + \int_{\alpha}^{\infty} f(x)\,dx$$ Approximations in for the two regions:
    $1$ to $\alpha$ region: $x \tilde = 1$ and hence $(x+1) \tilde= 2$
    $\alpha$ to $\infty$ region: $(x^2 -1) \tilde= x^2$
$\endgroup$

3 Answers 3

6
$\begingroup$

Here is a log-log plot of $$\delta I=I_{\text{appr}}-\int_{1}^{\infty} \frac{\sin^2 (\mu \sqrt{x^2 -1})}{(x+1)^{\frac{9}{2}} (x-1)^{\frac{3}{2}}} \,dx,$$ as a function of $\mu$, with $$I_{\text{appr}}=\frac{2 \mu^2}{15}-\frac{\mu^4}{9}.$$ You write "$\mu$ is extremely small". Is this error acceptable?

This is a plot of the absolute error; the relative error is $\lesssim 10^{-6}$ for $\mu\lesssim 10^{-2}$.

$\endgroup$
2
  • 1
    $\begingroup$ May I ask you how you made it ? Thanks & cheers :-) $\endgroup$ Feb 4, 2023 at 9:32
  • 4
    $\begingroup$ Expand integrand in $\mu$ around 0, integrate term-wise. The first two terms are convergent. $\endgroup$
    – Fred Hucht
    Feb 4, 2023 at 9:57
5
$\begingroup$

It is more or less straightforward to write down an asymptotic expansion around $\mu\to0$, $$ I(\mu)\sim \frac{2 \mu ^2}{15}-\frac{\mu ^4}{9}+\frac{\pi \mu ^5}{15}+\frac{1}{450} \mu ^6 (60 \log (\mu )+60 \gamma -67)+\cdots $$ where $\gamma$ is Euler-Mascheroni. Given this result it should be rather clear why the naive Taylor expansion diverges after the first two terms: the next terms have odd powers of $\mu$ and even non-analytic factors of $\log(\mu)$, while your function $f$ only has even powers of $\mu$.

Here is a log-plot of the error:

enter image description here

(Blue means one-term approximation, Red is two-term, Green is three-term, and Orange is four-term.)

A simple Mathematica code to reproduce the formula above (and easily generalized to yield more terms):

Sin[µ Sqrt[x^2 - 1]]^2/((x + 1)^(9/2) (x - 1)^(3/2))
Series[%, {µ, 0, 10}]
small = Integrate[Normal[%], {x, 1, Λ}, Assumptions -> Λ > 1];
Sin[µ x]^2 Series[1/((x + 1)^(9/2) (x - 1)^(3/2)), {x, ∞, 8}] // Normal
large = Integrate[%, {x, Λ, ∞}, Assumptions -> Λ > 1 && 0 < µ < 1];
Assuming[Λ > 1 && 0 < µ < 1, Series[small + large, {µ, 0, 10}] // Normal] // Short
Assuming[Λ > 1 && 0 < µ < 1, Series[%, {Λ, ∞, 1}] // Normal]
$\endgroup$
3
$\begingroup$

I really like the double-series approach by @AccidentalFourierTransform, however I got remaining $\Lambda$s in the terms of order $O(\mu^8)$ onwards. Thinking about this approach, I located the problem in the neglection of higher order terms in $\sin^2(\mu x)$ in the series expansion in $x$ around infinity (line 4 in the MMA code, thanks for providing!). The problem can be eliminated by a change of variables in the original integral from $x \mapsto y \equiv \sqrt{x^2-1}$, such that $$ I = \int_0^\infty \mathrm dy \frac{y \sin^2(\mu y)} {x\,(x-1)^{3/2}\,(x+1)^{9/2}}, \quad \text{with} \quad x \equiv \sqrt{y^2+1}.\tag{1} $$ Now, the method can be applied: we split the integral at $\Lambda>0$, expand the first integrand in $\mu$ around 0 and the second integrand in $x$ around $\infty$. The sum of the integrated series expansions is again expanded around $\mu=0$ to get \begin{align} I&=\frac{2\mu^2}{15} - \frac{\mu^4}{9} + \frac{\pi\mu^5}{15} - \frac{[67 - 60(\gamma+\log\mu)]\,\mu^6}{450} - \frac{8\pi\mu^7}{315}\\ &\quad + \frac{[169 - 56 (\gamma+\log\mu)]\,\mu^8}{7056} + \frac{[913 - 360 (\gamma+\log\mu)]\,\mu^{10}}{2916000} \\ &\quad - \frac{[23797 - 9240 (\gamma+\log\mu)]\,\mu^{12}}{3841992000} + O\left(\mu ^{14}\right).\tag{2} \end{align} Note that now $\Lambda$ cancels in all terms (as suggested by the OP in 2.) and that the highest odd power seems to be $\mu^7$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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