23

Calculate $$\displaystyle \int_{-\infty}^{\infty}\;\left( \frac{x^{2}}{1+4x+3x^{2}-4x^{3}-2x^{4}+2x^{5}+x^{6}}\right) \;dx$$

The answer given is $\pi$. How does one calculate this?

  • Where is this problem from? – bilaterus Dec 24 '15 at 12:53
  • I found it in another site, with no answer provided unfortunately. – Bmudtneduts Dec 24 '15 at 13:01
  • The integrand is a rational function, so there shouldn't be any difficulty in calculating the antiderivative. First you need to find a partial fraction expansion of the integrand. –  Dec 24 '15 at 13:22
  • It is not irreducible over the reals, every poly over the reals factors into linear and quadratic factor. I think you mean irreducible over the rationals, so that caculating the factors over the rationals is not practical. – Rene Schipperus Dec 24 '15 at 13:27
  • 3
    Another question is does it have a real root ? If not then you can use contour integration to say that the integral is the sum of the residues of the roots with positive imaginary part. But there is no clear way to find those roots... – Rene Schipperus Dec 24 '15 at 13:30
  • @ReneSchipperus: The same integral is in this post. – Tito Piezas III Dec 24 '15 at 18:05

3 Answers3

21

Let $F(x) = \frac{x^2}{P(x)}$ where $$P(x) = x^6+2x^5-2x^4-4x^3+3x^2+4x+1 = (x^3+x^2-2x-1)^2 + (x^2+x)^2$$

Change variable to $u = \frac{1}{x+1} \iff x = \frac{1}{u}-1$. The integral at hand becomes

$$\int_{-\infty}^\infty F(x) dx = \left(\int_{-\infty}^{-1^{-}} + \int_{-1^{+}}^\infty\right) F(x) dx = \left(\int_{0^{-}}^{-\infty} + \int_{+\infty}^{0^{+}}\right) F\left(\frac{1}{u} - 1\right)\left(-\frac{du}{u^2}\right)\\ = \int_{-\infty}^\infty \frac{1}{u^2} F\left(\frac{1}{u}-1\right) du $$ By direct substitution, we have $$\frac{1}{u^2}F\left(\frac{1}{u}-1\right) = \frac{(u^2-u)^2}{u^6-2u^5-2u^4+4u^3+3u^2-4u+1} = \frac{(u^2-u)^2}{(u^3-u^2-2u+1)^2+(u^2-u)^2}$$ Notice the function defined by $$g(u) \stackrel{def}{=} \frac{u^3-u^2-2u+1}{u^2-u} = u - \frac{1}{u}-\frac{1}{u-1}$$ has the form where Glasser's Master Theorem applies, we get

$$\int_{-\infty}^\infty F(x) dx = \int_{-\infty}^\infty \frac{du}{g(u)^2 + 1} = \int_{-\infty}^\infty \frac{dx}{x^2+1} = \pi $$

NOTE

Please note that the statement about Glasser's Master theorem in above link is slightly off. The coefficient $|\alpha|$ in front of $x$ there need to be $1$. Otherwise, there will be an extra scaling factor on RHS of the identity. When in doubt, please consult the original paper by Glasser,

Glasser, M. L. "A Remarkable Property of Definite Integrals." Math. Comput. 40, 561-563, 1983.

and an online copy of that paper can be found here.

achille hui
  • 122,701
  • Very nice. I observed some symmetry of the integrand which in principle allows residue calculation but your approach is much more efficient. – Start wearing purple Dec 24 '15 at 15:25
  • Thank you! Does the theorem you quote in the end also imply the step: $\displaystyle \int_{-\infty}^\infty F(x) dx = \int_{-\infty}^\infty \frac{1}{u^2} F\left(\frac{1}{u}-1\right) du$? – Bmudtneduts Dec 24 '15 at 15:41
  • @Bmudtneduts I don't think so (at least directly). In the quoted form of the theorem, the measure didn't change. In our case, the measure $du$ picks up an extra factor $-\frac{1}{u^2}$. – achille hui Dec 24 '15 at 15:55
  • @achillehui: The same integral is discussed in this post – Tito Piezas III Dec 24 '15 at 18:04
  • I think is not that easy at first sight to understand why the substitution $\frac{1}{1+x}=u$ was used. Perhaps somewhat more intuitive is to divide $P(x)$ by $(x^2+x)^2$, then to note that $\frac{x^3+x^2-2x-1}{x^2+x}=x-\frac{1}{x}-\frac{1}{1+x}$ also that $\frac{x^2}{(x^2+x)^2}=\frac{1}{(1+x)^2}$, and the last part is doing well with $\frac{d}{dx}\frac{1}{1+x}$. – Zacky Dec 14 '19 at 02:42
  • $$\int_{-\infty}^\infty \frac{x^2}{(x^2+x)^2+(x^3+x^2-2x-1)^2}dx=\int_{-\infty}^\infty \frac{\frac{x^2}{(x+x^2)^2}}{1+\frac{(x^3+x^2-2x-1)^2}{(x^2+x)^2}}dx$$ $$=\int_{-\infty}^\infty \frac{\frac{1}{(1+x)^2}}{1+\left(x-\frac{1}{x}-\frac{1}{1+x}\right)^2}dx \overset{1+x\to x}=\int_{-\infty}^\infty \frac{1}{1+\left(x-1-\frac{1}{x-1}-\frac{1}{x}\right)^2}\frac{dx}{x^2}$$ $$=\left(\int_{-\infty}^0+\int_0^\infty\right)\overset{\frac{1}{x}\to x}=\int_{-\infty}^\infty \frac{dx}{1+\left(x-\frac{1}{x}-\frac{1}{x-1}\right)^2} $$ – Zacky Dec 14 '19 at 02:42
11

There is an inner structure that enable this integral to be evaluated into such nice form.


Let $$f(x) = 1+4x+3x^2-4x^3-2x^4+2x^5+x^6$$ The first miracle is: $f(x)$ factorizes nicely in $\mathbb{Q}[i]$: $$f(x) = \underbrace{\left(x^3+(1-i) x^2-(2+i) x-1\right)}_{g(x)} \underbrace{\left(x^3+(1+i) x^2-(2-i) x-1\right)}_{h(x)}$$


The second miracle is: the root of $g(x)$ all lie in the same half plane. In this case, all roots of $g$ are in the upper plane. Denote them by $\alpha_1, \alpha_2, \alpha_3$, by contour integration $$I:=\int_{-\infty}^\infty \frac{x^2}{f(x)}dx = 2\pi i\left[ {\frac{{{\alpha _1}^2}}{{g'({\alpha _1})h({\alpha _1})}} + \frac{{{\alpha _2}^2}}{{g'({\alpha _2})h({\alpha _2})}} + \frac{{{\alpha _3}^2}}{{g'({\alpha _3})h({\alpha _3})}}} \right]$$ Now the right hand side is symmetric in $\alpha_i$, which are roots of $g$. Since $g,h\in \mathbb{Q}[i][x]$, we have $$\frac{I}{\pi} \in \mathbb{Q}$$ This explain the nice result of the integral. Note that the numerator $x^2$ can be replaced by any polynomial in $\mathbb{Q}[x]$, $I/\pi$ is still rational.


Using similar construction, we obtain the analogous integrals:

Let $$f(x) = 4 + 8x - 11{x^2} - 18{x^3} + 13{x^4} + 8{x^5} + {x^6}$$ then $f$ satisfies the above two "mircales" so we have $$\int_{ - \infty }^\infty {\frac{1}{{f(x)}}dx} = \frac{{5\pi }}{6} \qquad \int_{ - \infty }^\infty {\frac{x}{{f(x)}}dx} = - \frac{\pi }{3} \qquad \int_{ - \infty }^\infty {\frac{{{x^2}}}{{f(x)}}dx} = \frac{\pi }{3}$$

Another example with

$$f(x) = 4 + 12x - 6{x^2} - 26{x^3} + 11{x^4} + 8{x^5} + {x^6}$$ $$\int_{ - \infty }^\infty {\frac{1}{{f(x)}}dx} = \frac{{3\pi }}{4} \qquad \int_{ - \infty }^\infty {\frac{x}{{f(x)}}dx} = - \frac{\pi }{4} \qquad \int_{ - \infty }^\infty {\frac{{{x^2}}}{{f(x)}}dx} = \frac{\pi }{4}$$

An octic example:

$$f(x) = 13 + 12 x + 7 x^4 + 2 x^5 - 3 x^6 + x^8$$ $$\int_{ - \infty }^\infty {\frac{1}{{f(x)}}dx} = \frac{{487\pi }}{4148} \qquad \int_{ - \infty }^\infty {\frac{x}{{f(x)}}dx} = - \frac{325\pi }{4148} \qquad \int_{ - \infty }^\infty {\frac{{{x^2}}}{{f(x)}}dx} = \frac{515\pi }{4148}$$

pisco
  • 18,983
5

More of a hint, but it might work:

Use the formula

$$\int_{-\infty}^{\infty}\frac{l (x +a)+ c}{(x+a)^2 + b^2}dx= \frac{c \pi}{b}$$

if $b>0$. Write $$\frac{x^{2}}{1+4x+3x^{2}-4x^{3}-2x^{4}+2x^{5}+x^{6}} = \sum_{k=1}^3 \frac{l_k (x +a_k)+ c_k}{(x+a_k)^2 + b_k^2}$$

where the $l_k$, $a_k$, $b_k$, $c_k$ satisfy some (symmetric) equalities.

Show that these equalities imply $\sum_{k=1}^3 \frac{c_k}{b_k} = 1$

This might not be hopeless with some software. One could show that at least one of the expressions $\sum_{k=1}^3 \pm\frac{c_k}{b_k} - 1$ is zero, or equivalently, their product. Now this is an algebraic thing that could be show to follow from the equations for the coefficients. As for showing that the one with all $+$ equals $1$, some approximations would be useful, say like the ones from the answer of @Dr. Sonnhard Graubner:

${\bf Added:}$ Just to see what happens with other cases, the slightly modified integral $$\int_{-\infty}^{\infty}\frac{x^{2}}{1+4x+4x^{2}-4x^{3}-2x^{4}+2x^{5}+x^{6}}dx=2 \pi \sqrt{ t}=1.442791771994468\ldots$$ where $t$ is the root of the equation

$$(2^{26}\cdot53^6\cdot419^6)t^{10}-714086275692025123245183700303872 t^9+17223872258514797331184452894720 t^8-95944433146175550843118419968 t^7+1052704800953003893513568256 t^6-112701726213711713166176256 t^5+2130836339803327583245568 t^4-5118952508328476790656 t^3-21564414502323395600 t^2-1183162373726451992 t+45434497^2 =0 $$ and $t \approx 0.0527288$

All these integrals are $\pi \times $ some algebraic number that can be in principle determined.

Zacky
  • 27,674
orangeskid
  • 53,909
  • 1
    It seems that the top should be something like $cx+d$ instead of just $c_k$. –  Dec 24 '15 at 14:08
  • @JohnMa: Yes, you are right, I have to change to that, thank you! – orangeskid Dec 24 '15 at 14:09
  • @Tito Piezas III: Thanks, that contracted the expression a bit, I tried factoring some more coefficients and factors $419$ did appear up to a point. I lifted the equality straight from wolframalpha, i suppose there is some rule for the degree of the equation in the general case. – orangeskid Dec 25 '15 at 03:21
  • @orangeskid: I have asked a related question here. – Tito Piezas III Dec 25 '15 at 04:32