9
$\begingroup$

I'm trying to set straight various $q$-deformations of the standard geometric distribution.

The geometric distribution on $\left\{ 0,1,\ldots \right\}$ is well-known, it has $$ \mu_1(X=j)=(1-p)p^j,\qquad j=0,1,2,\ldots. $$ Here $0<p<1$ is the parameter.

This distribution can be $q$-deformed in at least two ways.


(1) One way uses the Heine's $q$-binomial theorem and has $$ \mu_2(X=j)=p^j\frac{(a;q)_{j}^{}}{(q;q)_{j}^{}}\frac{(p;q)_{\infty}^{}}{(ap;q)_{\infty}^{}},\qquad j=0,1,2,\ldots. $$ where $0<q<1$ is the main $q$-deformation parameter, and $0<a<1$ is an additional parameter. Here a$(z;q)_{k}^{}=(1-z)(1-zq)\ldots(1-zq_{}^{k-1})$ is the $q$-Pochhammer symbol, and it makes sense for $k=\infty$, too. The distribution $\mu_2$ should not be viewed as a $q$-deformation of $\mu_1$ because if one sets $q=0$ then $\mu_2$ does not become $\mu_1$.

EDIT: But when $a=q$ and for any $q$ $\mu_2$ does not depend on $q$ and becomes $\mu_1$.

$\mu_2$ is the orthogonality weight of the little $q$-Jacobi polynomials.


(1a) For $a=0$ the above distribution simplifies to $$ \mu_3(X=j)=\frac{p^j}{(q;q)_{j}^{}}(p;q)_{\infty}^{},\qquad j=0,1,2,\ldots. $$ And this for $q=0$ turns into $\mu_1$.

$\mu_1$ is the orthogonality weight of the little $q$-Laguerre / Wall polynomials or maybe $q$-Charlier polynomials.


(2) Now take another (probably more naive) way to $q$-deform $\mu_1$, and define $$ \mu_4(j)\propto\frac{p^j}{1-q^j},\qquad j=1,2,\ldots. $$ The normalization constant in $\mu_4$ is not that nice, it can be expressed through the $q$-digamma function. The distribution $\mu_4$ can in fact be obtained from $\mu_2$ by taking $a\to1$ and conditioning $\mu_2$ on $\left\{ 1,2,\ldots \right\}$. Clearly, for $q=0$ the distribution $\mu_4$ also becomes the standard geometric distribution (but on $\left\{ 1,2,\ldots \right\}$).

After a brief skimming through Koekoek et al ("Hypergeometric Orthogonal Polynomials and Their q-Analogues") I could not find orthogonal polynomials with the weight $\mu_4$.


My main questions are:

Are there any more or less standard names for $\mu_2,\mu_3$, and $\mu_4$? And which of $\mu_3$ or $\mu_4$ should be more naturally called the $q$-geometric distribution?

EDIT: I would like to have distributions which become the original ones as $q\to0$ and not $q\to1$, this is related to how the $q$-Whittaker polynomials (same as $t=0$ Macdonald polynomials) turn into the classical Schur polynomials when $q=0$.


PS. I do not even start with taking analogous distributions on a finite lattice, where there are more numerous interesting $q$-deformations of the binomial distribution. For a rather general $q$-deformation of the binomial distribution e.g., see formula (1.3.1) in https://arxiv.org/abs/1401.3321 - this is the orthogonality weight of the $q$-Hahn orthogonal polynomials.

$\endgroup$
0
3
$\begingroup$

Kupershmidt argues for $$\mu_{K}([j]_q)\propto(p;q)_{j-1}\,q^j,\;\;[j]_q=\frac{q^j-1}{q-1},$$ as the "natural" $q$-geometric distribution, because it produces the $q$-analog of the Pascal (negative binomial) distribution. It has a simple normalization, $$\mu_{K}([j]_q)=\frac{p}{1-(p;q)_\infty}\,(p;q)_{j-1}\,q^j.$$ Since $(p;1)_n=(1-p)^n$ and $[j]_1=j$, for $q\rightarrow 1$ the classical geometric distribution $\mu(j)=p(1-p)^{j-1}$ is recovered (the $\mu_1$ from the OP with $p\mapsto 1-p$).

$\endgroup$
1
  • $\begingroup$ Thank you, I didn't know about this distribution! But somehow I want things to become the original geometric distributions when $q=0$ and to $q\to1$ - this is related to Macdonald symmetric polynomials and how they degenerate to the Schur ones $\endgroup$ Dec 19 '16 at 13:52

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.