Let $(n_1,n_2,n_3)$ a triple of non-negative integers summing up to $27$, i.e. $n_1+n_2+n_3=27$. I want to count how many triples there are satisfying this constraint, using the contour integral representation of the Kronecker delta. In principle, I should be able to count the number of triples in two different ways: $$ I_1=\sum_{n_1=0}^{27}\sum_{n_2=0}^{27}\sum_{n_3=0}^{27}\delta_{n_1+n_2+n_3,27} $$ $$ I_2=\sum_{n_1=0}^{\infty}\sum_{n_2=0}^{\infty}\sum_{n_3=0}^{\infty}\delta_{n_1+n_2+n_3,27} $$ (I believe I should be allowed to extend the range of the summations to $\infty$, landing on $I_2$, as clearly numbers $n_i>27$ cannot contribute to the counting). However, if I carry out the calculation in the two cases using the integral representation of the Kronecker delta function, the second 'version' ($I_2$) seems problematic (also for Mathematica). Using $$ \delta_{a,b}=\frac{1}{2\pi\mathrm{i}}\oint_{|z|=1}dz\ z^{a-b-1}\ , $$ I obtain $$ I_1 =\frac{1}{2\pi\mathrm{i}}\oint_{|z|=1}\frac{dz}{z^{28}}\left(\sum_{n=0}^{27} z^n\right)^3= \frac{1}{2\pi\mathrm{i}}\oint_{|z|=1}\frac{dz}{z^{28}}\left(\frac{-1+z^{28}}{-1+z}\right)^3 =\mathrm{Res}\left(\frac{1}{z^{28}}\left(\frac{-1+z^{28}}{-1+z}\right)^3;\{z,0\}\right) =406\ , $$ using Cauchy's residue theorem (there is a pole of order $28$ in zero and a removable singularity at $z=1$ on the contour). The answer is correct. However, if I attempt to compute $I_2$ $$ I_2 =\frac{1}{2\pi\mathrm{i}}\oint_{|z|=1}\frac{dz}{z^{28}}\left(\sum_{n=0}^{\infty} z^n\right)^3= \frac{1}{2\pi\mathrm{i}}\oint_{|z|=1}\frac{dz}{z^{28}}\left(\frac{1}{1-z}\right)^3 $$ I run into potential troubles, as the earlier removable singularity has now become a 3rd-order pole right on the contour. This presumably forces me to reinterpret it as a Cauchy principal value integral (which I am not super-familiar with). Mathematica has also some problems in evaluating the integral in $I_2$ (setting $z=\exp(\mathrm{i\theta})$), while the integral in $I_1$ is perfectly fine. Now, assuming that the $n_i$s can be safely extended to $\infty$ (I do not see why they shouldn't), I would be grateful if you could help me with the evaluation of the PV integral $I_2$ and confirm that it also yields 406. And if it doesn't, where do things go wrong? Many thanks in advance.


Imagine permitting $M$ terms in each sum in $I_1$. Then the integrand is

$$\frac1{z^{28}}\left (\frac{1-z^M}{1-z} \right )^3 = \frac1{z^{28}}\left (1-3 z^M + 3 z^{2 M}-z^{3 M} \right ) \sum_{n=0}^{\infty} \binom{n+2}{n} z^n$$

For example, if $M=7$, then we have contributions at $n=27$, $n=20$, $n=13$, and $n=6$, so that the total number of triples becomes

$$\binom{29}{27} - 3 \binom{22}{20} + 3 \binom{15}{13} - \binom{8}{6} = \frac12 (29 \cdot 28 - 3 \cdot 22 \cdot 21 + 3 \cdot 15 \cdot 14-8 \cdot 7) = 33$$

When $M \gt 27$, it no longer matters what $M$ is; the answer will remain $406$ as $M$ increases. Thus, this is how we view the Cauchy principal value: we take the limit as $M \to \infty$.

It should be noted that the actual integral represented by $I_2$ doesn't even exist as a Cauchy PV, as we have a triple pole on the contour. The only way I see around this problem is to consider $M$ very large but finite.

  • $\begingroup$ many thanks for your answer. However, I am still a bit 'puzzled' by the fact that I started from a seemingly legitimate expression $I_2$ (in terms of sums extending to infinity), and end up with an integral expression for $I_2$ that apparently does not make sense (even as a PV). Is this then a case where the limit ($M\to\infty$) and the contour integral cannot be exchanged? I would then love to learn which hypotheses of what theorem are violated in this case... $\endgroup$ – Pierpaolo Vivo Jan 4 '16 at 8:55
  • 1
    $\begingroup$ @PierpaoloVivo: To be honest, I am a bit puzzled too. But keeping $M$ finite and large is the only way I can reconcile how this all works. I guess the problem is that, near $z=1$, $z^M$ may not be as small as we would like. I'll bet that's the origin of the misunderstanding. $\endgroup$ – Ron Gordon Jan 4 '16 at 14:12

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.