1
$\begingroup$

Consider the following integral in two complex variables $z_1$ and $z_2$:

$$\frac{1}{(2\pi i)^2}\oint_{{|z_1|=\epsilon}\atop{|z_2|=\epsilon}}dz_1 dz_2\frac{1}{z_1 z_2\left(1+a+\frac{z_1}{z_1+z_2}\right)\left(1+b+\frac{z_2}{z_1+z_2}\right)}$$

If I perform the contour integration in $z_1$ first, get the residue in $z_1$, and then perform the contour integration in $z_2$ to get the final residue, I get overall:

$$Res_{z_1=0,z_2=0}=\frac{1}{(1+a)(2+b)}$$

If I reverse the process and integrate the $z_2$ contour first, and then $z_1$, I end up with the final result:

$$Res_{z_1=0,z_2=0}=\frac{1}{(2+a)(1+b)}$$

These two overall residues are not the same, which implies that something must have gone wrong in the above calculations. How do I properly compute the actual residue of the integral above? Thanks for any suggestion.

EDIT

It was suggested by Robert Israel to rewrite

$$ \dfrac{1}{1+a+z_1/(z_1 + z_2)} = \dfrac{z_1 + z_2}{a z_1 + a z_2 + 2 z_1 + z_2}$$

and similar for the other term with $b$. This would give additional poles if these denominators vanish within $|z_1|=\epsilon$ and $|z_2|=\epsilon$. However, I meant these extra factors to be purely test functions that do not contribute any extra poles. Therefore, let us concentrate on the parameter region in $a$ and $b$ where the only poles in the above integrand are at $z_1=0$ and $z_2=0$. In this case the problem still stands unresolved.

$\endgroup$
3
$\begingroup$

(EDITED)

The poles at $z_1 = 0$ and $z_2=0$ are not the only ones inside the circles $|z|=\epsilon$. Note that $$ \dfrac{1}{z_1 z_2 (1+a+z_1/(z_1 + z_2)(1+b+z_2/(z_1+z_2))} = \dfrac{(z_1 + z_2)^2}{z_1 z_2 ((a+2) z_1 + (a+1) z_2)((b+1) z_1 + (b+2) z_2}$$ so (for fixed $z_2$ on the circle $|z|=\epsilon$) this has poles at $z_1 = 0$, $z_1 = -z_2 (a + 1)/(a + 2)$ and $z_1 = -z_2 (b+2)/(b+1)$. On the other hand, for fixed $z_1$ on the circle, it has poles at $z_2 = 0$, $z_2 = -z_1 (a+2)/(a+1)$ and $z_2 = -z_1 (b+1)/(b+2)$. Leaving out the cases $|a+2|=|a+1|$ and $|b+1|=|b+2|$ where there is a pole on the circle itself, one of $z_1 = -z_2 (a + 1)/(a + 2)$ and $z_2 = -z_1 (a+2)/(a+1)$ is inside the circle and the other is outside, and similarly for the poles involving $b$. The residues for poles inside the circle must be taken into account when doing the integrals.

For example, let's say $|a+1| > |a+2|$ and $|b+1| < |b+2|$. Then for the $z_1$ integral with fixed $z_2$ on the circle, we need only the pole at $z_1=0$:

$$ \eqalign{\frac{1}{2\pi i} \oint_{|z_1|=\epsilon} dz_1\; f(z_1, z_2) &= \text{res}(f(z_1,z_2); z_1 = 0) = \dfrac{1}{(b+2)(a+1) z_2}\cr \dfrac{1}{(2\pi i)^2} \oint_{|z_2|=\epsilon} dz_2 \oint_{|z_1|=\epsilon} dz_1\; f(z_1,z_2) &= \text{res}\left(\dfrac{1}{(b+2)(a+1)z_2}; z_2 = 0\right) = \dfrac{1}{(b+2)(a+1)}}$$ In the other order, for the $z_2$ integral with fixed $z_1$ on the circle we need the residues at $z_2 = 0$, $z_2 = -z_1 (a+2)/(a+1)$ and $z_2 = -z_1 (b+1)/(b+2)$.

$$ \eqalign{\frac{1}{2\pi i} \oint_{|z_2|=\epsilon} dz_2\; f(z_1, z_2) =& \text{res}(f(z_1,z_2); z_2 = 0) + \text{res}\left(f(z_1,z_2); z_2 = -z_1 \dfrac{a+2}{a+1}\right) \cr &+ \text{res}\left(f(z_1,z_2); z_2 = -z_1 \dfrac{b+1}{b+2}\right)\cr = {\frac {1}{z_{{1}} \left( b+1 \right) \left( a+2 \right) }}&+{\frac {1 }{z_{{1}} \left( a+2 \right) \left( a+1 \right) \left( a+b+3 \right) }}-{\frac {1}{z_{{1}} \left( b+2 \right) \left( b+1 \right) \left( a+b+3 \right) }}\cr &= \dfrac{1}{(b+2)(a+1) z_1}\cr \dfrac{1}{(2\pi i)^2} \oint_{|z_1|=\epsilon} dz_1 \oint_{|z_2|=\epsilon} dz_2\; f(z_1,z_2) &= \text{res}\left(\dfrac{1}{(b+2)(a+1) z_1}; z_1 = 0\right) = \dfrac{1}{(b+2)(a+1)} }$$ and there is no discrepancy.

$\endgroup$
  • $\begingroup$ So if we assume that $|a+1|/|a+2|>1$, then the problem would be that those are actually double poles and I should take a derivative to obtain the residue? Also, it seems to me that the above problem could be resolved in terms of a multivariate limit and a corresponding L'Hospital rule. Do you think this could work? $\endgroup$ – Kagaratsch Aug 13 '15 at 21:37
  • $\begingroup$ Oh, just found out, apparently there is no L'Hospital rule in multiple variables. :( $\endgroup$ – Kagaratsch Aug 13 '15 at 21:48
  • $\begingroup$ Actually, it turns out there is a typo in the above equality in your answer. The poles at $z_1=0$ and $z_2=0$ are the ones already present in the integral in my question. So no double poles appear, and if $|a+1|/|a+2|>1$ the problem still stands without having been resolved yet. $\endgroup$ – Kagaratsch Aug 13 '15 at 21:53
  • $\begingroup$ I'm editing to fix the typo and clarify. $\endgroup$ – Robert Israel Aug 13 '15 at 23:37

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.