0
$\begingroup$

I am trying to make the following integration. But I am not sure about the result obtained as follows.

Is there some way to get more reliable and accurate results?

Thanks.

NIntegrate[
    (x1 x2 x3)/((1 + x1^2)^(3/2) (1 + x3^2)^(3/2) (1 + x2^2 - 7 x2 Cos[x4]) Sqrt[x1^2 + x2^2 - x1 x2 Cos[x5]] Sqrt[ 1 + 9 (x1^2 + x2^2 - x1 x2 Cos[x6])])
     , {x1, 0, ∞}
     , {x2, 0, ∞}
     , {x3, 0, ∞}
     , {x4, 0, 2 π}
     , {x5, 0, 2 π}
     , {x6, 0, 2 π}
]

Edit:

To avoid the singularity problem, I revise the above problem and make the integral as follows,

        fexp = (x1 x2 x3)/((1 + x1^2)^(3/2) (1 + x3^2)^(
        3/2) (1 + x2^2 - x2 Cos[x4]) Sqrt[
        3 + x1^2 + x2^2 - 2 x1 x2 Cos[x5]] Sqrt[
        1 + x1^2 + x2^2/4 - x1 x2 Cos[x6]]);

    NIntegrate[fexp
               , {x1, 0, ∞}
               , {x2, 0, ∞}
               , {x3, 0, ∞}
               , {x4, 0, 2 π}
               , {x5, 0, 2 π}
               , {x6, 0, 2 π}]

   Out=75.0157  (with warning messages, Numerical integration converging 
                too slowly; .... )

In addition, I want to get the results numerically by taking some proper algorithms. Of course, we can check the results analytically (partly).

$\endgroup$
4
  • 2
    $\begingroup$ The integrand has a singularity 1 + x2^2 - 7 x2 Cos[x4] == 0 in x2 and x4. I have strong doubts about the convergence of the improper integral under consideration. $\endgroup$
    – user64494
    Jul 7, 2022 at 13:33
  • $\begingroup$ I think this kind of singularity is not so serious in integrations. The difficult part is from the oscillation and multi-dimensional integral. $\endgroup$
    – Orders
    Jul 7, 2022 at 13:48
  • $\begingroup$ 1 + x2^2 - 7 x2 Cos[x4] ==(x2 - 1/2 (7 Cos[x4] - Sqrt[41 + 49 Cos[2 x4]]/Sqrt[2])*(x2 - 1/2 (7 Cos[x4] + Sqrt[41 + 49 Cos[2 x4]]/Sqrt[2]). Wee see non- integrable singularities at x2 ==-1/2 (7 Cos[x4] - Sqrt[41 + 49 Cos[2 x4]]/Sqrt[2] and x2 == 1/2 (7 Cos[x4] - Sqrt[41 + 49 Cos[2 x4]]/Sqrt[2] in x2. Don't hesitate to ask for further explanation in need. $\endgroup$
    – user64494
    Jul 7, 2022 at 13:59
  • $\begingroup$ @user64494 Thanks. Now, I think we cannot neglect singularity. But, that's not the intent of the question. I revise the questions. $\endgroup$
    – Orders
    Jul 12, 2022 at 2:24

1 Answer 1

1
$\begingroup$

You can do most of the inegrations analyticlly. As the other comments mention, you have to consider singularities.

f = (x1 x2 x3)/((1 + x1^2)^(3/2) (1 + x3^2)^(3/2) (1 + x2^2 - 
        7 x2 Cos[x4]) Sqrt[x1^2 + x2^2 - x1 x2 Cos[x5]] Sqrt[
       1 + 9 (x1^2 + x2^2 - x1 x2 Cos[x6])]) // Together;

int21 = Integrate[f, {x5, 0, 2 \[Pi]}, {x6, 0, 2 \[Pi]}, 
  Assumptions -> {x1 > 0, x2 > 0, x3 > 0, 0 < x4 < 2 Pi}]

(*   (16 x1 x2 x3 EllipticK[-((2 x1 x2)/(
     x1^2 - x1 x2 + x2^2))] EllipticK[-((18 x1 x2)/(
     1 + 9 (x1^2 - x1 x2 + x2^2)))])/(Sqrt[(x1^2 - x1 x2 + x2^2) (1 + 
      9 (x1^2 - x1 x2 + x2^2))] ((1 + x1^2) (1 + x3^2))^(
   3/2) (1 + x2^2 - 7 x2 Cos[x4]))   *)

int22 = Integrate[int21, {x3, 0, Infinity}, 
  Assumptions -> {x1 > 0, x2 > 0, x3 > 0, 0 < x4 < 2 Pi}]

(*   (16 x1 x2 EllipticK[-((2 x1 x2)/(x1^2 - x1 x2 + x2^2))] EllipticK[-((
   18 x1 x2)/(1 + 9 (x1^2 - x1 x2 + x2^2)))])/((1 + x1^2)^(
 3/2) Sqrt[(x1^2 - x1 x2 + x2^2) (1 + 9 (x1^2 - x1 x2 + x2^2))] (1 + 
   x2^2 - 7 x2 Cos[x4]))   *)

Singularities at

ContourPlot[(1 + x2^2 - 7 x2 Cos[x4]) == 0, {x2, 0, 10}, {x4, 0, 
  2 Pi}]

In this range, integral is zero.

int23 = Integrate[int22, {x4, 0, 2 Pi}, 
  Assumptions -> x1 > 0 && x2 > 0, PrincipalValue -> True]

(*   ConditionalExpression[0, 
 3 Sqrt[5] + 2 x2 >= 7 && 2 x2 <= 7 + 3 Sqrt[5]]   *)

{red = Reduce[int23[[2]], x2], red // N}

(*   {1/2 (7 - 3 Sqrt[5]) <= x2 <= 1/2 (7 + 3 Sqrt[5]), 
 0.145898 <= x2 <= 6.8541}   *)

int24 = Integrate[int22, {x4, 0, 2 Pi}, 
 Assumptions -> 
  x1 > 0 && (0 < x2 < 1/2 (7 - 3 Sqrt[5]) || 
     1/2 (7 + 3 Sqrt[5]) < x2)]

(*   (32 \[Pi] x1 x2 EllipticK[-((2 x1 x2)/(
   x1^2 - x1 x2 + x2^2))] EllipticK[-((18 x1 x2)/(
   1 + 9 (x1^2 - x1 x2 + x2^2)))])/((1 + x1^2)^(
 3/2) Sqrt[(x1^2 - x1 x2 + x2^2) (1 - 47 x2^2 + x2^4) (1 + 
    9 (x1^2 - x1 x2 + x2^2))])   *)

int25 = NIntegrate[
   int24, {x1, 0, Infinity}, {x2, 0, 1/2 (7 - 3 Sqrt[5])}, 
   WorkingPrecision -> 15] + 
  NIntegrate[
   int24, {x1, 0, Infinity}, {x2, 1/2 (7 + 3 Sqrt[5]), Infinity}, 
   WorkingPrecision -> 15]

(*   3.820016280311350   *)

Edit

What i did with int23, is regarding x1 and x2 as parameters and calculating the principal value for x4. This is affirmed by inserting parameters.

ii22[x1_, x2_, x4_] = int22; Manipulate[
 Plot[ii22[x1, x2, x4], {x4, 0, 2 Pi}, PlotPoints -> 50, 
  PlotRange -> 5], {{x1, 2}, 0, 10, 
  Appearance -> "Labeled"}, {{x2, 1/E}, 0, 10, 
  Appearance -> "Labeled"}]

enter image description here

int23b = Integrate[int22 /. {x2 -> 1/E}, {x4, 0, 2 Pi}, 
  PrincipalValue -> True, Assumptions -> x1 > 0]

(*   0   *)

Reduce[0 < x4 < 2 Pi && 0 == (1 + x2^2 - 7 x2 Cos[x4]) /. 
  x2 -> 1/E, x4, Reals]

(*   x4 == 2 \[Pi] - ArcCos[(1 + E^2)/(7 E)] || 
 x4 == ArcCos[(1 + E^2)/(7 E)]   *)

NIntegrate[
 int22 /. {x1 -> 2, x2 -> 1/E}, {x4, 0, ArcCos[(1 + E^2)/(7 E)], 
  2 Pi - ArcCos[(1 + E^2)/(7 E)], 2 Pi}, Method -> "PrincipalValue"]

(*   3.71023*10^-14   *)

I'm no pure mathematician to be able to proof this kind of proceeding. But otherwise you must state, integral is not integrable.

$\endgroup$
6
  • $\begingroup$ I have never seen the notion of principal value for multiple integrals. In fact, int23 is an iterated integral where the outer integral is treated as the principal value. I repeat, the double improper integral under consideration diverges and NIntegrate of the integrand under consideration bears witness to this, reporting of a possible singularity. $\endgroup$
    – user64494
    Jul 7, 2022 at 15:11
  • $\begingroup$ See here and here for info. $\endgroup$
    – user64494
    Jul 8, 2022 at 5:53
  • $\begingroup$ Can you kindly explain what you mean by "integral is not integrable"? TIA. $\endgroup$
    – user64494
    Jul 8, 2022 at 5:55
  • $\begingroup$ Example: Suppose $B \subset \mathbb{R}^2$ is the ball with radius 2 centered at the origin, set $f(x,y)=\frac{1}{x^2+y^2-1}$. Is $\int_Bf(x,y)dxdy$ defined? Clearly $\int_B |f(x,y)| dx dy$ is infinite. But using polar coordinates $x+iy=re^{i\varphi}$ (and using $dxdy = r dr d\varphi$ as a recipe, no theorem) and using PrincipalValue->True for the $r$-integral yields the result $\pi\log 3$. Is this a meaningful number? Can the notion of the integral be extended, with good properties, to such functions $f$? Cf. this 6-year-old post. $\endgroup$
    – user293787
    Jul 8, 2022 at 6:49
  • $\begingroup$ @user293787: Thank you for the link. As far as I understand it, that is an invention of Dr. Wolfgang Hintze only. The convergence of an improper integral should not depend on an exhaustion of the domain of the integration and coordinates (see G.H. Hardy, "A course of pure mathematics" , Cambridge Univ. Press (1975) for info.). Thank you anyway. $\endgroup$
    – user64494
    Jul 8, 2022 at 8:03

Your Answer

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

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