Zhu’s generalized Jacobi–Stirling conjecture

What Lean checks

Lean proves the original recurrence, the 3×3 determinant identity, the exact coefficient −4, the negation of coefficientwise positivity of all finite Hankel minors, and the stated counterexample region over every linearly ordered field. The standalone source passed strict compilation, an all-owned axiom audit, and stock-kernel replay. The combined repository gate has not yet passed for this registration.

Research date: 21 November 2026. Mathematical status: complete disproof. Prior-resolution review is recorded separately; no claim of exhaustive worldwide priority is made. Coordination: issue 163.

Source and exact statement

Bao-Xuan Zhu’s Coefficientwise Hankel-total positivity of the row-generating polynomials for the output matrices of certain production matrices, Advances in Mathematics 444 (2024), 109695, DOI 10.1016/j.aim.2024.109695, states the conjecture in §5.2, immediately after equation (5.10). In the author’s arXiv v2, this is printed page 24. The paper attributes the conjecture to the author in 2018.

Relabel the source’s parameters (x,y,q)(x,y,q) as (a,b,x)(a,b,x). The triangle is defined by

J0,0=1,Jn,k=0(k<0 or k>n),Jn,k=Jn−1,k−1+(ak2+bk)Jn−1,k.J_{0,0}=1,\qquad J_{n,k}=0\quad(k<0\text{ or }k>n), \qquad J_{n,k}=J_{n-1,k-1}+(ak^2+bk)J_{n-1,k}.

Its unweighted row polynomials are

Pn(x)=∑k=0nJn,kxk.P_n(x)=\sum_{k=0}^n J_{n,k}x^k.

The conjecture asserts that every finite minor of the Hankel matrix

H=(Pi+j(x))i,j≥0H=(P_{i+j}(x))_{i,j\ge0}

has nonnegative coefficients in the polynomial ring ℤ[a,b,x]\mathbb Z[a,b,x]. Deb’s current problem P1.9 states the same conjecture with precisely this recurrence and unweighted normalization.

Theorem

For the polynomials just defined, the following polynomial identity holds:

det(P0P1P2P1P2P3P2P3P4)=2x3[(a+b)(2a+b)(3a+b)−2a2x].\boxed{ \det\begin{pmatrix} P_0&P_1&P_2\\ P_1&P_2&P_3\\ P_2&P_3&P_4 \end{pmatrix} =2x^3\bigl[(a+b)(2a+b)(3a+b)-2a^2x\bigr]. }

Consequently, Zhu’s conjecture is false. Its first three rows and columns already have a minor with coefficient

[a2b0x4]detH[{0,1,2},{0,1,2}]=−4.[a^2b^0x^4]\det H[\{0,1,2\},\{0,1,2\}]=-4.

The same identity gives a full real counterexample region: for every a>0a>0 and b≥0b\ge0, the Hankel matrix is not totally nonnegative whenever

x>(a+b)(2a+b)(3a+b)2a2.x>\frac{(a+b)(2a+b)(3a+b)}{2a^2}.

This includes strictly positive parameters a,b,xa,b,x. It is an exact obstruction to the complete conjecture, without a limiting or numerical hypothesis.

Proof

Put

c=a+b,d=4a+2b,e=9a+3b.c=a+b,\qquad d=4a+2b,\qquad e=9a+3b.

These are the recurrence weights at k=1,2,3k=1,2,3. Repeated substitution into the defining recurrence gives

P0=1,P1=x,P2=cx+x2,P3=c2x+(c+d)x2+x3,P4=c3x+(c2+cd+d2)x2+(c+d+e)x3+x4.\begin{aligned} P_0&=1,\\ P_1&=x,\\ P_2&=cx+x^2,\\ P_3&=c^2x+(c+d)x^2+x^3,\\ P_4&=c^3x+(c^2+cd+d^2)x^2+(c+d+e)x^3+x^4. \end{aligned}

For completeness, the nonzero entries in rows two through four are respectively

(c,1),(c2,c+d,1),(c3,c2+cd+d2,c+d+e,1).(c,1),\qquad(c^2,c+d,1),\qquad (c^3,c^2+cd+d^2,c+d+e,1).

For example, the second entry of row four is c2+d(c+d)c^2+d(c+d), and its third is (c+d)+e(c+d)+e. Thus the displayed expressions use the literal triangle recurrence, not an assumed generating-function formula.

Subtract P1P_1 times the first row from the second row of the displayed Hankel matrix, and P2P_2 times the first row from the third. Expanding down the first column yields

Δ3=(P2−P12)(P4−P22)−(P3−P1P2)2.\Delta_3=(P_2-P_1^2)(P_4-P_2^2)-(P_3-P_1P_2)^2.

The three differences are

P2−P12=cx,P3−P1P2=c2x+dx2,P4−P22=c3x+(cd+d2)x2+(−c+d+e)x3.\begin{aligned} P_2-P_1^2&=cx,\\ P_3-P_1P_2&=c^2x+dx^2,\\ P_4-P_2^2&=c^3x+(cd+d^2)x^2+(-c+d+e)x^3. \end{aligned}

Therefore

Δ3=cx[c3x+(cd+d2)x2+(−c+d+e)x3]−(c2x+dx2)2=cd(d−c)x3+(cd+ce−c2−d2)x4.\begin{aligned} \Delta_3 &=cx\bigl[c^3x+(cd+d^2)x^2+(-c+d+e)x^3\bigr] -(c^2x+dx^2)^2\\ &=cd(d-c)x^3+(cd+ce-c^2-d^2)x^4. \end{aligned}

Substitution of c,d,ec,d,e gives the two exact identities

cd(d−c)=2(a+b)(2a+b)(3a+b),cd+ce−c2−d2=−4a2.cd(d-c)=2(a+b)(2a+b)(3a+b),\qquad cd+ce-c^2-d^2=-4a^2.

This proves the determinant formula. Its negative coefficient disproves coefficientwise Hankel total positivity. For a>0,b≥0a>0,b\ge0 and xx above the stated threshold, 2x3>02x^3>0 and the bracket is negative, proving the real counterexample assertion. ∎

Normalization audit

The same source immediately introduces the factorial-weighted triangle

J̃n,k=k!Jn,k,P̃n(x)=∑k=0nk!Jn,kxk.\widetilde J_{n,k}=k!J_{n,k},\qquad \widetilde P_n(x)=\sum_{k=0}^n k!J_{n,k}x^k.

Its Proposition 5.7 proves coefficientwise Hankel total positivity for this weighted sequence. That sequence has a different recurrence: the coefficient of J̃n−1,k−1\widetilde J_{n-1,k-1} is kk. The present determinant uses coefficient one in that position throughout. This disproof concerns exactly the preceding unweighted conjecture.

The 2018 paper cited nearby, arXiv:1609.01544v2, proves strong log-convexity preservation for Jacobi–Stirling transformations in §4.2. Its full text contains no Hankel assertion. The date 2018 is therefore reported as Zhu’s retrospective attribution in the 2024 source, rather than as a claim that the conjecture appears verbatim in the 2018 article.

Exact illustrations and verification scope

At a=b=1a=b=1, the formula reduces to 4x3(12−x)4x^3(12-x), so every x>12x>12 gives a strictly negative determinant. For example, x=13x=13 gives −8788-8788. At a=1,b=0a=1,b=0, it reduces to 4x3(3−x)4x^3(3-x).

These evaluations illustrate the all-parameter identity; the proof does not depend on selecting a numerical witness. An independent sparse-integer-polynomial verifier reconstructs the triangle directly from its recurrence and expands the determinant by its six permutations. It checks the full polynomial identity and the negative monomial. That finite algebra check supplements the written proof.