---
title: 'Poly-Numbers: A Unified Mathematical Overview'
url: https://www.emergentmind.com/topics/poly-numbers
type: topic
---

# Poly-Numbers: A Unified Mathematical Overview

“Poly-Numbers” is used in contemporary mathematics for several distinct but partially overlapping classes of objects. In one major sense, it refers to poly-Bernoulli, multi-poly-Bernoulli, poly-Cauchy, and allied special-number sequences defined through polylogarithmic or multiple-polylogarithmic generating functions. In another, it refers to polytope numbers, which are integer sequences determined by the facial structure of polytopes. In a third, more computational sense, it appears in Wildberger’s polyseries framework, where “poly-numbers” are represented by truncatable coefficient sequences. The most developed strand is the poly-Bernoulli-centered one, whose recurring features are exponential generating functions, Stirling-transform formulas, duality for negative indices, zeta-type interpolation, and a large combinatorial ecology [1510.05765][1206.0511][2507.04231].

## 1. Core polylogarithmic families

The canonical starting point is Kaneko’s poly-Bernoulli numbers \(B_n^{(k)}\), defined for \(k\in\mathbb Z\) by the exponential generating function
\[
\frac{\operatorname{Li}_k(1-e^{-t})}{1-e^{-t}}
=\sum_{n=0}^{\infty} B_n^{(k)}\frac{t^n}{n!},
\]
where
\[
\operatorname{Li}_k(z)=\sum_{m=1}^{\infty}\frac{z^m}{m^k}.
\]
In the special case \(k=1\), one recovers the classical Bernoulli numbers, \(B_n^{(1)}=B_n\) [2603.15380]. This places poly-Bernoulli numbers among the standard polylogarithmic extensions of classical special-number sequences.

The associated polynomial families are obtained by adjoining an exponential factor. One standard form is
\[
\sum_{n\ge 0} B_n^{(k)}(z)\frac{t^n}{n!}
=
\frac{\operatorname{Li}_k(1-e^{-t})}{1-e^{-t}}e^{zt},
\]
while another equivalent convention appears as
\[
e^{-xt}\frac{\operatorname{Li}_\ell(1-e^{-t})}{1-e^{-t}}
=
\sum_{m=0}^\infty B_m^{(\ell)}(x)\frac{t^m}{m!}.
\]
These polynomial versions are important because they support translation formulas, Appell-type behavior, and parameter deformations [1903.05270][2003.12378].

A related bifurcation appears in the type \(B\) and type \(C\) poly-Bernoulli numbers. In the notation used for Schur-type generalizations, type \(B\) and type \(C\) are generated respectively by
\[
\frac{\operatorname{Li}_k(1-e^{-t})}{1-e^{-t}}
\quad\text{and}\quad
\frac{\operatorname{Li}_k(1-e^{-t})}{e^t-1}.
\]
This distinction is inherited by several later constructions, including Schur-type and zeta-interpolated variants [1812.10640].

The negative-index sector is singled out throughout the literature. For \(\ell\le 0\), the values \(B_m^{(\ell)}(k)\) are emphasized as integer-valued, and negative-index poly-Bernoulli numbers are repeatedly treated as intrinsically combinatorial rather than merely formal coefficients [2003.12378][1510.05765].

## 2. Explicit formulas, duality, and analytic realizations

A central structural fact is that poly-Bernoulli numbers admit explicit Stirling-number formulas. For nonnegative index \(k\),
\[
B_n^{(k)}
=
(-1)^n\sum_{m=1}^{n+1}
\frac{(-1)^{m-1}(m-1)!}{m^k}
\left\{\!\!\begin{matrix} n\\ m-1\end{matrix}\!\!\right\},
\]
while for negative index,
\[
B_n^{(-k)}
=
\sum_{j=0}^{\min(n,k)}
(j!)^2
\left\{\!\!\begin{matrix} n+1\\ j+1\end{matrix}\!\!\right\}
\left\{\!\!\begin{matrix} k+1\\ j+1\end{matrix}\!\!\right\}.
\]
These formulas place the theory inside the Stirling-transform apparatus and immediately explain integrality and symmetry phenomena in the negative-index regime [1212.3989].

The most classical symmetry is Kaneko’s duality
\[
\mathbb B_m^{(-k)}=\mathbb B_k^{(-m)},
\]
or equivalently \(B_n^{(-k)}=B_k^{(-n)}\) in the notation used by several later papers. This identity persists in many generalizations and is one of the defining structural traits of the subject [2603.15380][1503.04933].

Analytically, poly-Bernoulli numbers arise as special values of zeta-type objects. A symbolic treatment introduces the poly-Bernoulli umbra
\[
\mathcal{B}^{(k)}
:=
1+\sum_{l=1}^k \mathcal{B}_l \prod_{j=l}^{k-1}\mathcal{U}_j,
\]
with
\[
B_n^{(k)}(z)=\bigl(z+\mathcal{B}^{(k)}\bigr)^n.
\]
This representation yields new iterated integral formulas, Bernoulli-Barnes transforms, recurrences, and a symbolic realization of the Arakawa–Kaneko zeta values as negative moments of the companion umbra \(\mathcal C^{(k)}\) [1903.05270].

A newer analytic development identifies a shifted log-sine integral as a more intrinsic zeta-type source for poly-Bernoulli numbers. For \(0<\sigma\le \pi\) and \(\Re(s)>1\),
\[
\operatorname{SLs}(s;\sigma)
\coloneqq
\frac{1}{\Gamma(s-1)}
\int_0^\sigma
(\theta-\sigma)
\left(
-2\log \frac{\sin(\theta/2)}{\sin(\sigma/2)}
\right)^{s-2}
\,d\theta.
\]
After analytic continuation, its negative-integer values are governed by Lehmer-type polynomials, and at \(\sigma=\pi/3\) one obtains the exact identity
\[
\operatorname{SLs}(-n;\pi/3)
=
\frac13\sum_{k=0}^n B_{n-k}^{(-k)}.
\]
Thus anti-diagonal sums of negative-index poly-Bernoulli numbers emerge as special values of a log-sine integral rather than being built into the definition [2603.25392].

## 3. Multi-indexed, Schur-type, and function-field extensions

The one-index theory extends in several directions. A first step is the multi-poly-Bernoulli family
\[
\frac{\operatorname{Li}_{k_1,\ldots,k_r}(1-e^{-t})}{(1-e^{-t})^r}
=
\sum_{n=0}^{\infty} B_n^{(k_1,\ldots,k_r)}\frac{t^n}{n!},
\]
where
\[
\operatorname{Li}_{k_1,\ldots,k_r}(t)
=
\sum_{0<m_1<\cdots<m_r}
\frac{t^{m_r}}{m_1^{k_1}\cdots m_r^{k_r}}.
\]
Negative-index multi-poly-Bernoulli numbers admit power-sum expansions, special dualities, and explicit relations to ordinary negative-index poly-Bernoulli numbers [1503.04933].

A deeper generalization is the Kaneko–Tsumura multi-indexed theory, defined through the \(\sha\)-type multiple polylogarithm
\[
\mathrm{Li}_{s_1,\ldots,s_r}^{\sha}(z_1,\ldots,z_r)
=
\sum_{0<m_1<\cdots<m_r}
\frac{z_1^{m_1}z_2^{m_2-m_1}\cdots z_r^{m_r-m_{r-1}}}
{m_1^{s_1}\cdots m_r^{s_r}},
\]
and the generating function
\[
\frac{
\mathrm{Li}^\sha_{s_1,\ldots,s_r}
\bigl(
1-e^{-\sum_{v=1}^r t_v},\ldots,1-e^{-t_r}
\bigr)
}{
\prod_{j=1}^r \left(1-e^{-\sum_{v=j}^r t_v}\right)
}
=
\sum_{m_1,\ldots,m_r\ge 0}
\mathbb B_{m_1,\ldots,m_r}^{(s_1,\ldots,s_r)}
\frac{t_1^{m_1}\cdots t_r^{m_r}}{m_1!\cdots m_r!}.
\]
For arbitrary depth \(r\), these numbers now have an explicit formula in Stirling numbers of the second kind, and they satisfy the duality
\[
\mathbb B_{m_1,\ldots,m_r}^{(-k_1,\ldots,-k_r)}
=
\mathbb B_{k_1,\ldots,k_r}^{(-m_1,\ldots,-m_r)}.
\]
The same circle of ideas also supports restricted double-index formulas, \(p\)-adic periodicity for negative upper indices, and star-version analogues [2603.15380][2211.14549].

Another extension replaces integer index data by partition data. Schur-type poly-Bernoulli numbers are defined from a Schur-type polylogarithm \(Li_{\mathbf{k}^\lambda}\) built from semi-standard Young tableaux of shape \(\lambda\). For \(c=|C(\lambda)|\), the type \(B\) and type \(C\) Schur numbers are generated by
\[
\frac{Li_{\mathbf{k}^{\lambda}(1-e^{-z_1},\dots,1-e^{-z_c})}}
{(1-e^{-z_1})\cdots(1-e^{-z_c})}
=
\sum
\mathbb B^{\lambda,\mathbf{k}}_{m_1,\dots,m_c}
\frac{z_1^{m_1}\cdots z_c^{m_c}}{m_1!\cdots m_c!},
\]
and
\[
\frac{Li_{\mathbf{k}^{\lambda}(1-e^{-z_1},\dots,1-e^{-z_c})}}
{(e^{z_1}-1)\cdots(e^{z_c}-1)}
=
\sum
\mathbb C^{\lambda,\mathbf{k}}_{m_1,\dots,m_c}
\frac{z_1^{m_1}\cdots z_c^{m_c}}{m_1!\cdots m_c!}.
\]
These admit Arakawa–Kaneko-type and Kaneko–Tsumura-type interpolation theorems, hook-shape recurrences, and Stirling-number descriptions [1812.10640].

In positive characteristic, the function-field analogue is given by the multi-poly-Bernoulli-Carlitz numbers
\[
\sum_{n\ge 0} BC_n^{\mathfrak{s},\mathbf{j}} \frac{z^n}{\Pi(n)}
=
\frac{\operatorname{Li}_{\mathfrak{s}}\!\bigl(e_C(z)u_{1j_1},u_{2j_2},\dots,u_{rj_r}\bigr)}{e_C(z)},
\]
where \(e_C\) is the Carlitz exponential, \(\Pi(n)\) the Carlitz factorial, and the coefficients \(u_{ij}\) come from Anderson–Thakur polynomials. These numbers satisfy explicit formulas in Stirling-Carlitz numbers and recover the Bernoulli-Carlitz numbers in depth one [1801.06916].

## 4. Parameter deformations, \(q\)-analogues, and level-two theories

A broad deformation theory surrounds the core poly-Bernoulli family. One parameterized hierarchy passes from \(B_n^{(k)}\) to \(B_n^{(k)}(x)\), then to \(B_n^{(k)}(a,b)\), and finally to the fully generalized family \(B_n^{(k)}(x,a,b,c)\), defined by
\[
\frac{Li_k\!\left(1-(ab)^{-t}\right)}{b^t-a^{-t}c^{xt}}
=
\sum_{n=0}^{\infty} B_n^{(k)}(x,a,b,c)\frac{t^n}{n!}.
\]
Its key reduction formula is
\[
B_n^{(k)}(x,a,b,c)
=
(\ln a+\ln b)^n
B_n^{(k)}\!\left(\frac{x\ln c-\ln b}{\ln a+\ln b}\right),
\]
showing that the fully parameterized family is an affine-logarithmic rescaling of the classical one [1212.3989].

A \(q\)-deformed branch introduces \(q\)-poly-Bernoulli and \(q\)-poly-Cauchy polynomials with parameter \(\rho\). The \(q\)-poly-Bernoulli family is defined by
\[
\frac{\rho\, \operatorname{Li}_{k,q}\!\left(\frac{1-e^{-\rho t}}{\rho}\right)}{1-e^{-\rho t}}e^{-tz}
=
\sum_{n=0}^{\infty} B_{n,\rho,q}^{(k)}(z)\frac{t^n}{n!},
\]
and the two \(q\)-poly-Cauchy families are defined by Jackson \(q\)-integrals. These objects admit weighted-Stirling expansions, inversion formulas, and direct transform relations linking poly-Bernoulli and poly-Cauchy theories [1503.08394].

A distinct “level two” theory replaces the ordinary polylogarithm by its odd-part analogue
\[
\operatorname{Li}_{2,k}(z)=\sum_{n=0}^\infty \frac{z^{2n+1}}{(2n+1)^k}.
\]
The level-two poly-Bernoulli numbers are defined by
\[
\frac{\operatorname{Li}_{2,k}(2\sin(x/2))}{2\sin(x/2)}
=
\sum_{n=0}^\infty B_{2n}^{(k)}\frac{x^{2n}}{(2n)!},
\]
and satisfy the explicit level-two Stirling formula
\[
B_{2n}^{(k)}
=
\sum_{m=0}^n
\left\{\!\!\begin{matrix} n\\ m\end{matrix}\!\!\right\}_2
\frac{(-1)^{n-m}(2m)!}{(2m+1)^k}.
\]
Here \(\left\{\begin{smallmatrix}n\\m\end{smallmatrix}\right\}_2\) are the Stirling numbers of the second kind with level \(2\), identified in the paper with the central factorial numbers \(T(2n,2k)\) [2104.09726].

The same level-two setting also produces polycosecant and polycotangent numbers. They are defined by
\[
\frac{A_k(\tanh (t/2))}{\sinh t}
=
\sum_{n=0}^{\infty} D_n^{(k)}\frac{t^n}{n!},
\qquad
\frac{A_k(\tanh (t/2))}{\tanh t}
=
\sum_{n=0}^{\infty} B_n^{(k)}\frac{t^n}{n!},
\]
with \(A_k(z)=\operatorname{Li}_k(z)-\operatorname{Li}_k(-z)\). They satisfy dualities analogous to poly-Bernoulli duality,
\[
D_{2m}^{(-2l-1)}=D_{2l}^{(-2m-1)},
\qquad
B_{2m}^{(-2l)}=B_{2l}^{(-2m)},
\]
but the interpolation theory is asymmetrical: the cited paper explicitly notes that a zeta function interpolating polycotangent numbers at non-positive integers has not yet been constructed [2205.05247].

## 5. Combinatorial models and structural behavior

The negative-index poly-Bernoulli numbers have an unusually rich combinatorial life. They count lonesum \(0\)-\(1\) matrices of size \(n\times k\), Callan permutations, max-ascending permutations, Vesztergombi permutations, and acyclic orientations of the complete bipartite graph \(K_{n,k}\). The same paper adds a new interpretation in terms of \(T\)-free \(0\)-\(1\) matrices, where the forbidden \(T\)-configuration is a \(2\times2\) submatrix of the form
\[
\begin{pmatrix}1&1\\1&*\end{pmatrix}.
\]
This model yields a direct combinatorial proof of Kaneko’s negative-index recurrence
\[
\mathbb{B}_n^{(-k)}
=
\mathbb{B}_n^{(-(k-1))}
+
\sum_{j=1}^{n}\binom{n}{j}\,
\mathbb{B}_{n-(j-1)}^{(-(k-1))}.
\]
The same source also stresses the symmetry
\[
\mathbb B_n^{(-k)}=\mathbb B_k^{(-n)}
\]
as visible in all major combinatorial realizations [1510.05765].

A more refined construction is the symmetrized poly-Bernoulli number. In that setting, a Stirling-weighted symmetrization restores the \(m\leftrightarrow \ell\) symmetry uniformly in an auxiliary parameter \(k\), and the alternating diagonal sum
\[
\sum_{\ell=0}^n (-1)^\ell B_{n-\ell}^{(-\ell)}(k)
\]
is identified with
\[
k!\cdot (-1)^{n/2} G_n(1,1,k),
\]
where \(G_n(x,y,z)\) is the Dumont–Foata polynomial. This places symmetrized poly-Bernoulli numbers inside the combinatorics of pistols, Genocchi numbers, and Gandhi polynomials [2003.12378].

The broader “poly-number” ecology also includes structural positivity phenomena. For poly-Cauchy numbers, the sign-normalized sequences
\[
\sigma_n^{(k)}=(-1)^{n-1}c_n^{(k)},
\qquad
\omega_n^{(k)}=(-1)^n\widehat c_n^{(k)}
\]
are log-convex, and the same remains true for multiparameter-poly-Cauchy numbers under the hypotheses
\[
\alpha_j\ge 1,\qquad \alpha_j-\alpha_{j-1}\ge 1.
\]
The paper additionally proves unimodality for certain eventually constant parameter patterns [1603.06725].

Negative-index multi-poly-Bernoulli numbers also acquire direct combinatorial interpretations. In particular,
\[
B_n^{(-2,0,\dots,0)}=2(3+b)^n-(2+b)^n
\]
counts restricted barred preferential arrangements in which one of two fixed restricted sections is empty. The same paper proves that the last digits of restricted barred preferential-arrangement counts, negative-index multi-poly-Bernoulli numbers, and related numbers \(U_n^{(-j_1,\dots,-j_b)}\) have a four-cycle modulo \(10\) [1509.07352].

## 6. Geometric and computational senses of “poly-number”

Outside the polylogarithmic line, “poly-number” language also appears in geometric combinatorics. Polytope numbers are sequences \(P(n)\) and interior sequences \(P(n)^\sharp\) attached to a polytope \(P\), defined recursively from facial data and a distinguished vertex. For simplices,
\[
\alpha^d(n)=\binom{n+d-1}{d},
\qquad
\alpha^d(n)^\sharp=\alpha^d(n-d-1)=\binom{n-2}{d}.
\]
A central theorem states that every polytope admits a pointed triangulation, and that its polytope numbers can be decomposed in several ways into sums of simplex numbers, including
\[
P(n)=\alpha^d(n)+\sum_{i\in[d-1]} a_i\,\alpha^d(n-i)
\]
and
\[
P(n)=\sum_{k\in[d]_0} d_k\,\alpha^k\bigl(n-(k+1)\bigr).
\]
This makes polytope numbers a higher-dimensional figurate-number theory controlled by shellings, face lattices, and simplex decompositions [1206.0511].

A particularly explicit subfamily is given by rectified simplex polytope numbers. If \(\lambda_r^d(n)\) denotes the number sequence of the \(r\)-rectified \(d\)-simplex, then
\[
\lambda_r^d(n)
=
\sum_{i=0}^{r} (-1)^{r-i} \binom{d+1}{r-i}\alpha^d\bigl((i+1)n-r\bigr),
\]
with a parallel formula for the interior sequence. The alternating signs are not formal artifacts: they directly encode the inclusion–exclusion geometry of rectification [1507.01654].

A third usage appears in Wildberger’s polyseries framework. There a polyseries is written
\[
\mathbb{A}=\left[a_1,a_2,\ldots\right\rangle,
\]
with truncation
\[
\mathbb A_k=\mathbf T_k(\mathbb A).
\]
The paper emphasizes data-structure operations such as length, index lookup, value extraction, truncation, and sorting, rather than a full general algebra of series operations. Its principal application is a Catalan-number expansion for the quadratic congruence
\[
t+ax+x^2\equiv 0\pmod{t^n},
\]
leading to the truncated polyseries solution
\[
x \equiv \sum_{k=1}^{n-1} C_{k-1}\frac{t^k}{a^{2k-1}} \pmod{t^n}.
\]
In this computational usage, a “poly-number” is effectively a coefficient-stream object manipulated through finite truncations [2507.04231].

Taken together, these strands show that “Poly-Numbers” is best understood as a family resemblance concept rather than a single universally fixed definition. In the strongest and most developed sense, it designates polylogarithm-generated special sequences—especially poly-Bernoulli and poly-Cauchy families—together with their multi-indexed, \(q\)-deformed, level-two, Schur-type, and function-field analogues. In parallel, the same term extends to geometric figurate theories and to coefficient-sequence formalisms, indicating that the prefix “poly-” marks not one object but a recurrent pattern of enrichment by higher structure.

Source: https://www.emergentmind.com/topics/poly-numbers