The nonnegative-coefficient conjecture for the polynomial h(t,a;r)

From papers

Let rNr\in\mathbb N and let h(t,a;r)h(t,a;r) be the polynomial defined in the paper by

h(t,a;r):=[1+(3+a)t]r([1+(2+a)t]r+1[1+at]r+1)[1+(1+a)t]r([1+(4+a)t]r+1[1+(2+a)t]r+1).h(t,a;r):=[1+(3+a)t]^r\left([1+(2+a)t]^{r+1}-[1+at]^{r+1}\right)-[1+(1+a)t]^r\left([1+(4+a)t]^{r+1}-[1+(2+a)t]^{r+1}\right).

Nonnegative-coefficient conjecture. If rNr\in\mathbb N is arbitrary but fixed, then h(t,a;r)h(t,a;r), viewed as a polynomial in (t,a)(t,a), has nonnegative coefficients. This would considerably strengthen the stated inequality h(t,a;r)0h(t,a;r)\geq0 for real a,t[0,)a,t\in[0,\infty) and r[1,)r\in[1,\infty).

Progress summary

Open

No public discussion or published progress was found for this conjecture.

No public discussion or published progress was found.

Current status (as of August 2026): The conjecture appears open, with no recorded public activity or verified progress.

Sources & referencesView supporting material

Primary source

James Allen Fill and Svante Janson, “Positive autocorrelation at unit lag for stationary random walk Metropolis-Hastings in R^d”, arXiv:2601.19323 (2026).

Solutions 1

Proof

We prove that the polynomial has nonnegative joint coefficients in tt and aa for every integer r0r\ge0.

Introduce x=1+atx=1+at, u=t/xu=t/x, and

A=(1+3u)(1+2u),B=(1+u)(1+4u),A=(1+3u)(1+2u),\qquad B=(1+u)(1+4u), C=(1+u)(1+2u),D=1+3u.C=(1+u)(1+2u),\qquad D=1+3u.

Regrouping the four homogeneous products in the defining polynomial gives

h(t,a;r)=x2r+1gr(u),h(t,a;r)=x^{2r+1}g_r(u),

where

gr(u)=(1+2u)Ar(1+4u)Br+(1+2u)CrDr.g_r(u)=(1+2u)A^r-(1+4u)B^r+(1+2u)C^r-D^r.

The key identity is the following exact positive generating function:

r0gr(u)zr=1+2u1Az1+4u1Bz+1+2u1Cz11Dz=8u4z2(1+2u)(1Az)(1Bz)(1Cz)(1Dz).\begin{aligned} \sum_{r\ge0}g_r(u)z^r &=\frac{1+2u}{1-Az} -\frac{1+4u}{1-Bz} +\frac{1+2u}{1-Cz} -\frac1{1-Dz}\\ &=\frac{8u^4z^2(1+2u)} {(1-Az)(1-Bz)(1-Cz)(1-Dz)}. \end{aligned}

The identity follows simply by taking a common denominator.

Each of A,B,C,DA,B,C,D belongs to N[u]\mathbb N[u]. Therefore every factor

11Ez=j0E(u)jzj,E{A,B,C,D},\frac1{1-Ez}=\sum_{j\ge0}E(u)^jz^j, \qquad E\in\{A,B,C,D\},

has nonnegative coefficients in both uu and zz. The numerator 8u4z2(1+2u)8u^4z^2(1+2u) does also. Hence

gr(u)=j=02r+1cr,jujwith cr,jNg_r(u)=\sum_{j=0}^{2r+1}c_{r,j}u^j \qquad\text{with }c_{r,j}\in\mathbb N

for every r0r\ge0; the degree bound follows directly from the defining expression for grg_r.

Undoing the substitution yields the manifestly positive decomposition

h(t,a;r)=j=02r+1cr,jtj(1+at)2r+1j.h(t,a;r) =\sum_{j=0}^{2r+1}c_{r,j}\,t^j(1+at)^{2r+1-j}.

Thus

[tj+a]h(t,a;r)=cr,j(2r+1j)0[t^{j+\ell}a^\ell]h(t,a;r) =c_{r,j}\binom{2r+1-j}{\ell}\ge0

for all j,j,\ell. This proves the conjecture for every integer rr. In fact h(t,a;0)=h(t,a;1)=0h(t,a;0)=h(t,a;1)=0, and for every r2r\ge2 the polynomial lies in 8t4N[t,a]8t^4\mathbb N[t,a].

0 endorsements
Shivam Patel ·