6
$\begingroup$

I am looking for orthogonal polynomials $P_{d,m,n}$ that have their values at integers $i$ specified by the following generating function

$(1-z)^i (1+z+z^2+ \ldots + z^d)^{n-i} = \sum_{m=0}^{i+d(n-i)} P_{d,m,n}(i) \cdot z^m$

That is,

$P_{d,m,n}(i) = \sum_{j=0}^{m} \big[ (-1)^j \binom{i}{j} \sum_{k_0+\cdots+k_d = n-i, \mbox{ } 1k_1+2k_2+\cdots+dk_d = m-j} \binom{n-i}{k_0,k_1\ldots,k_d} \big]$

The polynomials $P_{d,m,n}$ can be seen as a generalization of binary Krawtchouk Polynomials $K_{m,n,2}$ (see e.g. https://arxiv.org/abs/1101.1798) which satisfy

$(1-z)^i (1+z)^{n-i} = \sum_{m=0}^{n} K_{m,n,2}(i) \cdot z^m$

and

$K_{m,n,2}(i) = \sum_{j=0}^{m} (-1)^j \binom{i}{j}\binom{n-i}{m-j}$

I looked into multivariate Krawtchouk and Meixner polynomials but wasn't able to derive the above generating function from those. Can someone point me to existing literature or give me a hint? Even the family $P_{2,m,n}$, i.e. $d=2$, is of particular interest to me.

$\endgroup$
2
  • 1
    $\begingroup$ It's not clear what you want to derive from what. $\endgroup$ Commented Nov 22, 2017 at 13:03
  • $\begingroup$ The generating function in question connects to Hilbert series of certain ideals. In the case of Krawtchouk polynomials their orthogonality and hence distribution of zeros reflects on the Hilbert regularity. Therefore I am interested in the Polynomials $P_{d,m,n}$, in particular orthogonality, recurrence relations and distribution of zeros. $\endgroup$
    – user117316
    Commented Nov 22, 2017 at 19:44

1 Answer 1

1
$\begingroup$

From the formula for $P_{d,m,n}(i)$ it follows that $$P_{d,m,n}(i) = \sum_j (-1)^j \binom{i}{j} \sum_{k_0\geq 0} \binom{n-i}{k_0} \hat{B}_{m-j,n-i-k_0}(\underbrace{1,1,\ldots,1}_{d\ \text{ones}},0,0,\ldots),$$ where $\hat{B}_{m-j,n-i-k_0}()$ is the partial ordinary Bell polynomials. Recalling the generating function for ordinary Bell polynomials, one can further obtain that $P_{d,m,n}(i)$ equals the coefficient of $t^m$ in $$(1-t)^{2i-n}(1-t^{d+1})^{n-i}.$$ This further implies the following explicit formula: $$P_{d,m,n}(i) = \sum_{k=0}^{\lfloor m/(d+1)\rfloor} (-1)^{m-kd} \binom{n-i}{k} \binom{2i-n}{m-k(d+1)}.$$ It follows that $P_{d,m,n}(i)$ is a polynomial of degree $m$ in variable $i$.

$\endgroup$
3
  • $\begingroup$ This is certainly an interesting insight. The generating function $(1−t)^{2i−n}(1−t^{d+1})^{n−i}$ is exactly as the one above. Are the polynomials $P_{d,m,n}$ orthogonal, and if this is the case, what is there recurrence relation? $\endgroup$
    – user117316
    Commented Nov 22, 2017 at 19:52
  • $\begingroup$ Orthogonal : which weights do you use? $\endgroup$
    – esg
    Commented Nov 22, 2017 at 20:05
  • $\begingroup$ The binary Krawtchouk polynomials are orthogonal with respect to the inner product $⟨A,B⟩ = \sum_{i=0}^n \binom{n}{i} A(i)B(i)$. I did not try to figure out the inner product for the Polynomials $P_{d,m,n}$ so far since they might be known to the literature already. $\endgroup$
    – user117316
    Commented Nov 22, 2017 at 21:43

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .