1
$\begingroup$

Let us look at the abelian group $V$ of integer valued sequences modulo sequence which are zero almost everywhere.

Let me fix one constant $k\in \mathbb{N}$, which I will omit from the notation. One specific sequence is given by $a_n = \binom{n+k}{n}$. Let $\Sigma^m a$ be the shift of $a$ by $m$, i.e. $(\Sigma^ma)_n =a_{n-m}$. Now I am interested in zero combinations of those shifts.

Question: Given integers $\lambda_m$ (such that almost all,but not all integers are zero) and such that

\[\sum_m \lambda_m(\Sigma^ma)=0 \in V,\]

then $\sum_m |\lambda_m|\ge 2^{k+1}$.

A example where the (conjectured) minimum is obtained is the following. Let $T:V\rightarrow V$ be the difference operator, i.e. it is defined via $(Ta)_n=a_{n+1}-a_n$. Since $a$ is a polynomial of degree $k$, we have $T^{k+1}a =0$ and if we write the left hand side out, we get exactly $2^{k+1}$ as that sum.

$\endgroup$
3
  • $\begingroup$ The first sentence means "... of {\em integer} valued sequences ...", right? $\endgroup$
    – ThiKu
    May 30, 2017 at 14:21
  • 1
    $\begingroup$ Equivalent reformulation: if a polynomial $f(t)$ with integer coefficients is divisible by $(t-1)^{k}$, then the sum of absolute coefficients is not less than $2^{k+1}$. $\endgroup$ May 30, 2017 at 15:03
  • $\begingroup$ @ Fedor: I guess there is an index shift somewhere. In your reformulation it should be "less than $2^k$", since otherwise $p(t)=1$ would be an obvious counterexample. Furthermore it follows that $p = f/(t-1)^k$ must be divisible by $t+1$ (Otherwise plug in $t=-1$). $\endgroup$ May 30, 2017 at 16:36

1 Answer 1

1
$\begingroup$

Consider the Laurent polynomial $f(t)=\sum \lambda_m t^m$. It is easy to prove that your conditions are equivalent to a system of $k+1$ equations $f^{(i)}(1)=0$ for $i=0,\dots,k$. In other words, $f(t)=(t-1)^{k+1}h(t)$ for some Laurent polynomial $h$ with integer coefficients. If $h(-1)\ne 0$, we get $|f(-1)|\geqslant 2^{k+1}$, this implies your claim. But if $h(-1)\ne 0$, it may fail, for example, consider $k=2$ and $f(t)=(t-1)(t^2-1)(t^3-1)$.

$\endgroup$
5
  • $\begingroup$ Add the extra condition that the $\lambda$ are all $0,\pm 1.$ If I'm not mistaken it is an open problem if there is always a solution with the absolute values of these coefficients adding to $2k.$ $\endgroup$ May 30, 2017 at 15:56
  • $\begingroup$ Hm so what is the best lower bound possible? $\endgroup$ May 30, 2017 at 16:44
  • $\begingroup$ @Aaron, you're referring to the Tarry-Escott (or, multigrade) problem? $\endgroup$ May 31, 2017 at 0:24
  • $\begingroup$ @GerryMyerson Indeed, the Prohet-Tarry-Escott problem is what I had in mind. $\endgroup$ May 31, 2017 at 14:16
  • $\begingroup$ So on my comment I should have said $2k+2.$ Thst is a lower bound and is exact for small values. $\endgroup$ May 31, 2017 at 14:31

Your Answer

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

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