Rank-two binomial positivity conjecture for Schur coefficients

From papers

For n=2n=2, write the Chern class in the Schur basis as

ck(2,d)=0jk/2Ak,j(d)s(kj,j)(x1,x2),c_k(2,d)=\sum_{0\le j\le\lfloor k/2\rfloor}A_{k,j}(d)s_{(k-j,j)}(x_1,x_2),

and expand each coefficient in the binomial basis:

Ak,j(d)=r0Bk,j,r(dr).A_{k,j}(d)=\sum_{r\ge0}B_{k,j,r}\binom dr.

A polynomial is binomially positive when all its binomial-basis coefficients are nonnegative. Rank-two binomial positivity conjecture. For all k,j,rk,j,r with 0jk/20\le j\le\lfloor k/2\rfloor,

Bk,j,r0.B_{k,j,r}\ge0.

Equivalently, for each fixed kk and jj, Ak,j(d)A_{k,j}(d) is binomially positive. The source gives no resolution status; it notes that this positivity is part of a broader rank-two phenomenon.

Progress summary

Partially solved

The conjecture is proved for the first three cases, but it remains open beyond them.

The conjecture asks whether all coefficients in the binomial expansions of the rank-two Schur coefficients are nonnegative. No proposer or original date is identified in the retrieved sources.

Known results

  • Positivity is proved for j=0j=0 and j=1j=1 (Theorem 3.6).
  • Positivity is proved for j=2j=2, namely Bk,2,r0B_{k,2,r}\ge0 for all k4k\ge4 and r0r\ge0 (Theorem 5.2).
  • The higher-jj cases remain open; a direct positive combinatorial interpretation for Bk,2,rB_{k,2,r} is also unknown.

May 2026 partial-progress paper

Gergely Bérczi and László M. Fehér report the results above and describe a staged AI-assisted project. The paper says ChatGPT 5.5 Pro produced the first accessible cases, but this attribution and the underlying proof process have not been independently verified.

Current status (as of August 2026): Positivity is settled for j=0,1,2j=0,1,2, while all cases with higher jj remain open.

Sources
Sources & referencesView supporting material

Primary source

Gergely Bérczi and László M. Fehér, “Positivity in classical enumerative geometry: a case study in synchronized AI-assisted mathematics”, arXiv:2605.25271 (2026).

Solutions 1

Proof

Proof for every Schur index

Define

Ck,d(u)=ek(i+(di)u:0id),Ak,j(d)=[uj](1u)Ck,d(u),C_{k,d}(u) =e_k\bigl(i+(d-i)u:0\le i\le d\bigr), \qquad A_{k,j}(d)=[u^j](1-u)C_{k,d}(u),

and write

Ak,j(d)=r0Bk,j,r(dr).A_{k,j}(d)=\sum_{r\ge0}B_{k,j,r}\binom dr.

We prove, simultaneously for every kk and 0jk/20\le j\le\lfloor k/2\rfloor,

Bk,j,r0,B_{k,j,r}\ge0,

together with the stronger support bounds

Bk,0,r=0(r<k),Bk,j,r=0(r<k1, j1).B_{k,0,r}=0\quad(r<k), \qquad B_{k,j,r}=0\quad(r<k-1,\ j\ge1).

First, if a polynomial F(x)F(x) has a nonnegative binomial expansion supported on indices at least II, then

(xc)(xr)=(rc)(xr)+(r+1)(xr+1)(1)(x-c)\binom xr =(r-c)\binom xr+(r+1)\binom x{r+1} \tag{1}

shows that multiplication by xcx-c preserves binomial nonnegativity whenever cIc\le I. If c=Ic=I, the minimum support increases by one.

For h0h\ge0 and t1t\ge1, set

Rh,t(x)=(xh+1t)=1t!v=0t1(x(h1+v)).R_{h,t}(x) =\binom{x-h+1}{t} =\frac1{t!}\prod_{v=0}^{t-1}(x-(h-1+v)).

Successive applications of (1) show that if FF has minimum support at least h1h-1, then Rh,tFR_{h,t}F is binomially nonnegative with minimum support at least h+t1h+t-1. The same conclusion holds when its initial support is at least hh.

For the zeroth index,

Ak,0(d)=ek(0,1,,d)=[d+1d+1k].A_{k,0}(d)=e_k(0,1,\ldots,d) =\genfrac{[}{]}{0pt}{}{d+1}{d+1-k}.

Grouping permutations by the size NN of their nonfixed support gives

Ak,0(d)=ND(N,k)((dN)+(dN1)),A_{k,0}(d) =\sum_N D(N,k) \left(\binom dN+\binom d{N-1}\right),

where D(N,k)0D(N,k)\ge0. A permutation of cycle defect k1k\ge1 has nonfixed support Nk+1N\ge k+1, proving the claimed zeroth-index positivity and support bound.

Now use the exact factorization

Ed(T,u):=i=0d(T+i+(di)u)=(T+d)Ed1(T+u,u).E_d(T,u) :=\prod_{i=0}^d(T+i+(d-i)u) =(T+d)E_{d-1}(T+u,u).

Comparing coefficients, multiplying by 1u1-u, extracting [uj][u^j], and writing x=d1x=d-1 gives the all-index recurrence

ΔAk,j(x)=t=1j(xk+t+1t)Akt,jt(x)+(x+1)t=0j(xk+t+2t)Ak1t,jt(x),(2)\begin{aligned} \Delta A_{k,j}(x) &= \sum_{t=1}^{j} \binom{x-k+t+1}{t}A_{k-t,j-t}(x)\\ &\quad+ (x+1)\sum_{t=0}^{j} \binom{x-k+t+2}{t}A_{k-1-t,j-t}(x), \end{aligned} \tag{2}

where ΔF(x)=F(x+1)F(x)\Delta F(x)=F(x+1)-F(x).

There is only one apparent out-of-range term: when k=2jk=2j, the t=0t=0 term in the second sum involves A2j1,jA_{2j-1,j}. Since Ch,d(u)C_{h,d}(u) is palindromic of degree hh,

Ah,t(d)=Ah,h+1t(d),A_{h,t}(d)=-A_{h,h+1-t}(d),

and therefore A2j1,j=0A_{2j-1,j}=0. All remaining terms in (2) involve legitimate lower-degree indices.

Induct on kk. In the first sum, put h=kth=k-t; the prefactor is Rh,t(x)R_{h,t}(x), so the support lemma shows that each summand is binomially nonnegative with minimum support at least

h+t1=k1.h+t-1=k-1.

In the second sum, for t1t\ge1, put h=k1th=k-1-t; each summand has minimum support at least

h+t1=k2.h+t-1=k-2.

Multiplication by x+1x+1 preserves positivity. The t=0t=0 term either vanishes by the palindromic boundary above or already has support at least k2k-2.

Hence ΔAk,j(x)\Delta A_{k,j}(x) has a nonnegative binomial expansion supported on indices at least k2k-2. But

ΔAk,j(x)=r0Bk,j,r+1(xr),\Delta A_{k,j}(x) =\sum_{r\ge0}B_{k,j,r+1}\binom xr,

and Bk,j,0=Ak,j(0)=0B_{k,j,0}=A_{k,j}(0)=0 for every positive kk. Therefore

Bk,j,r0for all k,j,r,Bk,j,r=0(r<k1, j1).B_{k,j,r}\ge0 \quad\text{for all }k,j,r, \qquad B_{k,j,r}=0\quad(r<k-1,\ j\ge1).

This proves Conjecture 3.2 for every Schur index, extending the previously established cases j=0,1,2j=0,1,2.

0 endorsements
Shivam Patel ·