5
$\begingroup$

I am trying to compute the following double integral:

Integrate[(r BesselJ[0, r])/(
2 (r^2 + z^2)^(1/2)), {z, -Infinity, Infinity}, {r, 0, Infinity}]

Which evaluates to 1. However, changing the order of integration yields an error (integral does not converge).

Can it be assured that the result is indeed 1 or that the integral diverges? How could I verify the result? I guess a numerical integration to Infinity is not very precise.

$\endgroup$
  • $\begingroup$ Both results are "correct" in that Fubini's Theorem does not apply to the double integral and you have to choose a definition for the improper integral (such as b.gatessucks's). $\endgroup$ – Michael E2 Apr 30 '15 at 13:25
6
$\begingroup$

The inverse square root term is responsible for the errors as it diverges logarithmically at infinity. You could do one step at a time and see if you agree with the procedure of introducing a regularizing term, doing the integral and then removing it.

int1 = Integrate[(r BesselJ[0, r])/(2 (r^2 + z^2)^(1/2)) Exp[-α z^2], 
                 {z, -Infinity, Infinity}, Assumptions -> {α > 0, r > 0}]

(* 1/2 E^((r^2 α)/2) r BesselJ[0, r] BesselK[0, (r^2 α)/2] *)

int2 = Integrate[int1, {r, 0, Infinity}]
(* (E^((1/4)/α) Sqrt[π] Erfc[1/(2 Sqrt[α])])/(2 Sqrt[α]) *)

Finally:

Limit[int2, α -> 0]
(* 1 *)
$\endgroup$
4
$\begingroup$

Here's another approach just to illustrate the difficulty with multiple integrals that are not absolutely (i.e. $L^1$) convergent.

The integral on any finite disk centered at the origin is zero.

So it would be natural to conclude that the integral over the whole {r,z} plane is zero.

In terms of polar coordinates:

Integrate[
 ((r BesselJ[0, r])/(2 (r^2 + z^2)^(1/2)) /. {z -> ρ Sin[θ], r -> ρ Cos[θ]}) ρ,
 {ρ, 0, a}, {θ, 0, 2 Pi}, Assumptions -> a > 0]
(*  0  *)

Integrate[
  ((r BesselJ[0, r])/(2 (r^2 + z^2)^(1/2)) /. {z -> ρ Sin[θ], r -> ρ Cos[θ]}) ρ,
  {θ, 0, 2 Pi}, {ρ, 0, a}, Assumptions -> a > 0]
(*  Integrate[1/2 a BesselJ[1, a Cos[θ]], {θ, 0, 2 π}, Assumptions -> a > 0]  *)

Quiet@Block[{a = #}, N@%] & /@ RandomReal[{0, 100}, 100] // Abs // Max
(*  3.9635*10^-14  *)

In cartesian {r, z} coordinates:

Integrate[
 (r BesselJ[0, r])/(2 (r^2 + z^2)^(1/2)),
 {r, -a, a}, {z, Sqrt[a^2 - r^2], Sqrt[a^2 - r^2]}, Assumptions -> a > 0]
(*  0  *)

Integrate[
 (r BesselJ[0, r])/(2 (r^2 + z^2)^(1/2)),
 {r, z} ∈ Disk[{0, 0}, a], Assumptions -> a > 0]
(*  0  *)

Etc.

$\endgroup$

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.