4
$\begingroup$

I used the following result in another post without providing proof (because I couldn't prove it):

$$k^a=\sum_{m=1}^b\left ( c_m^a\prod_{n\neq m} \frac{k-c_n}{c_m-c_n} \right ),$$

where $a$ and $b$ are non-negative integers, $b>a$, each $c_j\in\mathbb{C}$, each $|c_j|>0$, and each $c_j$ is unique. The product runs from $n=1$ to $b$ but skips $m$.

How can we prove this?

$\endgroup$
  • 1
    $\begingroup$ The "each $|c_j|>0$" condition is clearly spurious. $\endgroup$ – darij grinberg Oct 8 '14 at 0:33
2
$\begingroup$

Given $\{c_n\}_{n=1}^b$ (where $c_n \not= c_m$ for $n\not = m$) we define

$$h(z) = \prod_{n=1}^b (z-c_n)$$ Then

$$h'(c_m) = \prod_{n\not = m}^b (c_m-c_n)$$

Using this the right hand side minus the left hand side of your equality can be written

$$g(z)\equiv \sum_{m=1}^bc_m^a\frac{h(z)}{h'(c_m)(z-c_m)} - z^a$$

which is a polynomial of degree $b-1$ satisfying (since $\lim_{z\to c_m} \frac{h(z)}{z-c_m} = h'(c_m)$)

$$g(c_n) = 0,~~~~n=1,2,\ldots,b$$

thus $g(z) \equiv 0$ as no non-zero polynomial of degree $b-1$ can have more than $b-1$ roots. Note that we do not need the restriction $|c_i|>0$.

For the case $b=a+1$ there is an even simpler proof. See this related question.

$\endgroup$
  • 1
    $\begingroup$ An elegant answer. $\endgroup$ – SDiv Oct 8 '14 at 8:34
2
$\begingroup$

Start by encoding the sum call it $S_b$ using residues. We have by inspection that $$S_b = \sum_{m=1}^b \mathrm{Res} \left(f(z); z=c_m\right)$$ where $$f(z) = \frac{z^a}{k-z} \prod_{n=1}^b \frac{k-c_n}{z-c_n}$$ and $c_m\ne k$ and $b>a.$

We can therefore collect $S_b$ by integrating f(z) around a contour that encloses the $b+1$ poles. We will then use the residue at infinity to evaluate the sum of the residues inside the contour.

Now the residue at infinity of $f(z)$ is given by $$\mathrm{Res} \left(-\frac{1}{z^2} f\left(\frac{1}{z}\right); z=0\right).$$ The functional term becomes $$-\frac{1}{z^2} \frac{1}{z^a}\frac{1}{k-1/z} \prod_{n=1}^b \frac{k-c_n}{1/z-c_n} = -\frac{1}{z^2} \frac{1}{z^a}\frac{z}{zk-1} \prod_{n=1}^b \frac{z(k-c_n)}{1-zc_n} \\ = -\frac{1}{z^2} \frac{z^{b+1}}{z^a}\frac{1}{zk-1} \prod_{n=1}^b \frac{k-c_n}{1-zc_n} = \frac{1}{z^{a-b+1}}\frac{1}{1-zk} \prod_{n=1}^b \frac{k-c_n}{1-zc_n}.$$

But we have $b>a$ and hence $b-a-1 > -1$ or $b-a-1\ge 0$ so the term is in fact $$ z^{b-a-1} \frac{1}{1-zk} \prod_{n=1}^b \frac{k-c_n}{1-zc_n}.$$ and the residue at zero of the substituted function is zero.

This means that $$S_b = -\mathrm{Res} \left(f(z); z=k\right)$$ which gives $$k^a \prod_{n=1}^b \frac{k-c_n}{k-c_n} = k^a,$$ done.

Addendum I. As an alternative to using the residue at infinity we could have used a circular contour enclosing all $b+1$ poles and observed that the integral along this contour goes to zero since on this circle we have

$$f(z) \in \Theta\left(R^{a-1} \times R^{-b}\right) = \Theta\left(R^{a-b-1}\right)$$ and $$\lim_{R\to\infty} 2\pi R \times R^{a-b-1} = 2\pi \lim_{R\to\infty} R^{a-b} = 0$$ because $b>a.$ This has the advantage of working for $a$ a positive real parameter as opposed to a positive integer only.

Addendum II. The case $k=c_m$ for some $m$ gives $$\sum_{m=1}^b\left(c_m^a\prod_{n\neq m} \frac{k-c_n}{c_m-c_n} \right) = k^a$$ because all the products vanish except the one for $c_m = k.$

A similar calculation can be found at this MSE link.

$\endgroup$
  • $\begingroup$ Thanks for the solution to a more general problem. It seems from your Addendum I that the expression in the question should hold for non-integer $a$. I played with some numbers and it only seems to hold for integers. Perhaps I have interpreted your answer incorrectly? $\endgroup$ – SDiv Oct 8 '14 at 8:38
  • $\begingroup$ Thanks for this observation. I also checked non-integer values but I get a match of the original formula and my residue formula for rational values, which seems intuitive by continuity. But the matching values are not equal to $k^a$ and this is because my proof does not go through in this case since $z^a = \exp(a\log(z))$ for non-integer $a$ introduces the branch cut of the logarithm, which crosses the contour and invalidates the residue calculation. $\endgroup$ – Marko Riedel Oct 8 '14 at 11:26

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.