Novelli–Thibon's conjecture

What Lean checks

Complete proposed proof of the intended monomial-product Conjecture 5.7 for all partitions and independent parameters, plus the adjacent all-size hook-coefficient formula. Two internal mathematical reviews and exact convention checks passed. The source index correction and priority limits are explicit; Schur positivity and general traces are not claimed.

Date: 21 November 2026. Research issue: #180.

Status: complete proposed proof of the intended monomial-product assertion in Conjecture 5.7, with two internal mathematical reviews and a documented public-source priority audit. Section 6 also proves the hook-coefficient formula stated immediately after that conjecture. These notes remain unrefereed. This note does not prove the separate Schur-positivity conjecture, Conjecture 5.4.

1. Source and indexing

The primary source is J.-C. Novelli and J.-Y. Thibon, Noncommutative chromatic quasi-symmetric functions, Macdonald polynomials, and the Yang–Baxter equation, arXiv:2502.09072v2, 4 November 2025; published in Journal of Combinatorial Theory, Series A 220 (2026), 106130, DOI 10.1016/j.jcta.2025.106130.

For partitions λ,μ⊢n\lambda,\mu\vdash n, write the Schur coefficient of the multi-tt modified Macdonald function as

H̃μ(X;q,𝐭)=∑λ⊢nK̃λμ(q,𝐭)sλ(X),K̃λμ=∑k,αck,αqk𝐭α.\widetilde H_\mu(X;q,\mathbf t) =\sum_{\lambda\vdash n}\widetilde K_{\lambda\mu}(q,\mathbf t)s_\lambda(X), \qquad \widetilde K_{\lambda\mu} =\sum_{k,\alpha}c_{k,\alpha}q^k\mathbf t^\alpha. \tag{1}

Let dλd_\lambda count standard tableaux of shape λ\lambda in French notation in which 2 is above 1. Put t0=1t_0=1. The intended conjecture is

∏k,α(qk𝐭α)ck,α=(qn(μ′)∏j≥1tjμj+1)dλ.\boxed{ \prod_{k,\alpha}(q^k\mathbf t^\alpha)^{c_{k,\alpha}} =\left(q^{n(\mu')}\prod_{j\ge1}t_j^{\mu_{j+1}}\right)^{d_\lambda}. } \tag{2}

Here n(μ′)=∑r(μr2)n(\mu')=\sum_r\binom{\mu_r}{2}. Equivalently the monomial inside parentheses is the product of qc−1tr−1q^{c-1}t_{r-1} over cells in column cc, row rr, counting rows from the bottom.

There is an indexing error in the source's equation (88): its exponent and the following tableau shape are printed as dμd_\mu and μ\mu. They must be dλd_\lambda and λ\lambda, where λ\lambda indexes the Schur coefficient. This is forced by the source's own example immediately afterward: the coefficient of s321s_{321} in H̃222\widetilde H_{222} has monomial product (q3t12t22)8(q^3t_1^2t_2^2)^8, with d321=8d_{321}=8, whereas d222=3d_{222}=3. It is also forced by the coefficient of sns_n, which is 1. We prove the full intended identity (2); we do not use the typographical error as a conjecture disproof.

The signed multiplicities ck,αc_{k,\alpha} can be treated as integers in a formal Laurent monomial product. No assertion that they are all nonnegative is used. Thus the proof is independent of Conjecture 5.4.

2. A single comparison identity

For σ∈Sn\sigma\in S_n, let

Fσ(X):=FDes(σ−1)(X)F_\sigma(X):=F_{\mathop{\mathrm{Des}}(\sigma^{-1})}(X)

be Gessel's fundamental quasisymmetric function. In particular,

Fσ(X)=∑std(w)=σxw1⋯xwn,F_\sigma(X) =\sum_{\mathop{\mathrm{std}}(w)=\sigma}x_{w_1}\cdots x_{w_n}, \tag{3}

where words have positive integer entries, and equal entries are standardized from left to right. Formula (3) follows directly by placing the entries in the order of increasing σ\sigma-label: they are weakly increasing, with a strict increase whenever the next label occurs earlier in the word.

For every pair of positions i<ji\lt j,

∑σ∈Sn[σi>σj]Fσ(X)=e2(X)p1(X)n−2.\sum_{\sigma\in S_n}[\sigma_i>\sigma_j]F_\sigma(X) =e_2(X)p_1(X)^{n-2}. \tag{4}

Indeed, left-to-right standardization gives [σi>σj]=[wi>wj][\sigma_i>\sigma_j]=[w_i>w_j] for i<ji\lt j. The other n−2n-2 letters are unrestricted. Therefore the left side equals

(∑a>bxaxb)(∑axa)n−2=e2p1n−2.\left(\sum_{a>b}x_ax_b\right) \left(\sum_a x_a\right)^{n-2} =e_2p_1^{n-2}.

Also ∑σFσ=p1n\sum_\sigma F_\sigma=p_1^n. These statements can be read in any finite alphabet; hence they hold as symmetric-function identities.

3. The exact filling formula for the source's modified function

We derive the needed formula from the source's Definition 5.3, equation (57), rather than assume that its untilded and modified conventions agree.

The graph conventions are those of Haglund and Wilson, Macdonald polynomials and chromatic quasisymmetric functions, EJC 27(3) (2020), P3.37, Definitions 1–3 and 14, pages 3–4 and 7, DOI 10.37236/9011. Draw μ\mu in French notation, with row lengths μ1≥μ2≥⋯\mu_1\ge\mu_2\ge\cdots. Number its cells by reading rows from top to bottom, and within each row from left to right. Let GG be the attacking graph. Its edges join:

  • two cells in the same row;
  • an upper cell to a cell in the immediately lower row strictly to its left.

All edges are written i<ji\lt j in reading order. Let UU be the cells outside the bottom row. For u∈Uu\in U, let v(u)v(u) be the cell immediately below it; uu precedes v(u)v(u) in reading order. Define

a(u)=arm(u),ℓ(u)=1+leg(u),δu(σ)=[σu>σv(u)],a(u)=\mathop{\mathrm{arm}}(u),\qquad \ell(u)=1+\mathop{\mathrm{leg}}(u),\qquad \delta_u(\sigma)=[\sigma_u>\sigma_{v(u)}],

and

IG(σ)=∑{i,j}∈E(G),i<j[σi>σj].I_G(\sigma)=\sum_{\{i,j\}\in E(G),\ i\lt j}[\sigma_i>\sigma_j].

The required expansion is

H̃μ(X;q,𝐭)=∑σ∈SnqIG(σ)−∑u∈Ua(u)δu(σ)∏u∈Utℓ(u)δu(σ)Fσ(X).\widetilde H_\mu(X;q,\mathbf t) =\sum_{\sigma\in S_n} q^{I_G(\sigma)-\sum_{u\in U}a(u)\delta_u(\sigma)} \prod_{u\in U}t_{\ell(u)}^{\delta_u(\sigma)}F_\sigma(X). \tag{5}

Here is an explicit derivation. In the source, G+G^+ adds the vertical edges {u,v(u)}\{u,v(u)\} to GG, and Lemma 4.2, equation (35), expands

ω𝐋𝐋𝐓̃H=∑σqdesH(σ)Φmin(σ).\omega\widetilde{\mathbf{LLT}}_H =\sum_\sigma q^{\mathop{\mathrm{des}}_H(\sigma)}\Phi_{\min(\sigma)}.

The commutative image of Φmin(σ)\Phi_{\min(\sigma)} is FσF_\sigma. For one optional vertical edge, the sum of the included and excluded contributions in Definition 5.3 is

(tℓ(u)−qa(u))qδu+(−tℓ(u)+qa(u)+1).(t_{\ell(u)}-q^{a(u)})q^{\delta_u} +(-t_{\ell(u)}+q^{a(u)+1}).

After division by q−1q-1, this is exactly

{qa(u),δu=0,tℓ(u),δu=1.\begin{cases} q^{a(u)},&\delta_u=0,\\ t_{\ell(u)},&\delta_u=1. \end{cases} \tag{6}

The global prefactor in (57) is q−A(q−1)−|U|q^{-A}(q-1)^{-|U|}, where

A=n(μ′)−(μ12)=∑u∈Ua(u).A=n(\mu')-\binom{\mu_1}{2} =\sum_{u\in U}a(u). \tag{7}

Thus multiplying (6) over uu and canceling qAq^A gives (5). This establishes the modified convention directly and removes the apparent ambiguity between equations (31), (32), and (50) in the source.

4. First derivatives at the unit parameter point

Write 𝟏\mathbf1 for q=t1=t2=⋯=1q=t_1=t_2=\cdots=1, and let

En(X)=e2(X)p1(X)n−2(n≥2),B=∑r(μr2)=n(μ′).E_n(X)=e_2(X)p_1(X)^{n-2}\quad(n\ge2), \qquad B=\sum_r\binom{\mu_r}{2}=n(\mu').

The attacking graph has BB same-row edges. For every row r≥2r\ge2, a cell in column cc attacks c−1c-1 cells in the next row below, so the cross-row count is

∑r≥2(μr2)=A.\sum_{r\ge2}\binom{\mu_r}{2}=A. \tag{8}

Hence |E(G)|=B+A|E(G)|=B+A. Differentiate (5), then apply (4) separately to every graph edge and every vertical edge. This gives

∂H̃μ∂q|𝟏=(|E(G)|−∑u∈Ua(u))En=BEn.\left.\frac{\partial\widetilde H_\mu}{\partial q}\right|_{\mathbf1} =\left(|E(G)|-\sum_{u\in U}a(u)\right)E_n =B E_n. \tag{9}

For the tjt_j-derivative, each column of height hh contributes exactly once when h≥j+1h\ge j+1, because the leg labels of its nonbottom cells are 1,2,…,h−11,2,\ldots,h-1. There are μj+1\mu_{j+1} columns of height at least j+1j+1. Consequently,

∂H̃μ∂tj|𝟏=μj+1En.\left.\frac{\partial\widetilde H_\mu}{\partial t_j}\right|_{\mathbf1} =\mu_{j+1}E_n. \tag{10}

Equations (9)–(10), together with H̃μ(X;𝟏)=p1n\widetilde H_\mu(X;\mathbf1)=p_1^n, give the stronger first-order identity

H̃μ(X;q,𝐭)≡p1n+(B(q−1)+∑j≥1μj+1(tj−1))e2p1n−2(mod𝔪2),\widetilde H_\mu(X;q,\mathbf t) \equiv p_1^n+ \left(B(q-1)+\sum_{j\ge1}\mu_{j+1}(t_j-1)\right)e_2p_1^{n-2} \pmod{\mathfrak m^2}, \tag{11}

where 𝔪=(q−1,t1−1,t2−1,…)\mathfrak m=(q-1,t_1-1,t_2-1,\ldots). Only finitely many parameters occur for fixed μ\mu.

5. Extracting Schur coefficients

Repeated multiplication by p1=s1p_1=s_1 adds one cell at each step by the one-cell Pieri rule. Since e2=s11e_2=s_{11},

[sλ]e2p1n−2=dλ.[s_\lambda]e_2p_1^{n-2}=d_\lambda. \tag{12}

Explicitly, a chain from the vertical two-cell shape to λ\lambda records the positions of entries 3,4,…,n3,4,\ldots,n of a standard tableau whose entries 1 and 2 initially occupy the bottom-left cell and the cell above it. This is a bijection with the tableaux defining dλd_\lambda.

Taking sλs_\lambda-coefficients in (9) and (10) gives

∑k,αkck,α=Bdλ,∑k,ααjck,α=μj+1dλ.\sum_{k,\alpha}k c_{k,\alpha}=B d_\lambda, \qquad \sum_{k,\alpha}\alpha_j c_{k,\alpha}=\mu_{j+1}d_\lambda. \tag{13}

The exponent of qq in the left side of (2) is the first sum in (13), and the exponent of tjt_j is the second. Thus every exponent agrees and (2) follows. For n=1n=1, both sides are 1, so the boundary case also holds.

This proves the intended Conjecture 5.7 for every pair of partitions and all independent parameters. No restriction on the number of rows, columns, or partition size is present.

6. The source's hook-coefficient formula

The final sentence following Conjecture 5.7 proposes a second identity. Let

Vμ={qc−1tr−1:(r,c)∈μ},t0=1.V_\mu=\{q^{c-1}t_{r-1}:(r,c)\in\mu\},\qquad t_0=1.

The symbol Vμ−1V_\mu-1 removes one copy of the weight 1 belonging to the bottom-left cell. For every 0≤k≤n−10\le k\le n-1, the proposed formula is

K̃(n−k,1k),μ(q,𝐭)=ek[Vμ−1].\widetilde K_{(n-k,1^k),\mu}(q,\mathbf t) =e_k[V_\mu-1]. \tag{14}

We prove the stronger plethystic evaluation

H̃μ[1−z;q,𝐭]=∏(r,c)∈μ(1−zqc−1tr−1).\widetilde H_\mu[1-z;q,\mathbf t] =\prod_{(r,c)\in\mu}(1-zq^{c-1}t_{r-1}). \tag{15}

Here 1−z1-z is a difference of alphabets: pj[1−z]=1−zjp_j[1-z]=1-z^j. Theorem 3.2, equation (20), of Novelli–Thibon gives the Carlsson–Mellit evaluation, with incoming degree bv(H)=|{i<v:{i,v}∈E(H)}|b_v(H)=|\{i\lt v:\{i,v\}\in E(H)\}|. After the inversion of qq in the source's modified LLT function, its commutative image satisfies

ω𝐋𝐋𝐓̃H[1−z;q]=(−1)nq|E(H)|𝐋𝐋𝐓H[z−1;q−1]=∏v(1−zqbv(H)).\omega\widetilde{\mathbf{LLT}}_H[1-z;q] =(-1)^nq^{|E(H)|}\mathbf{LLT}_H[z-1;q^{-1}] =\prod_v(1-zq^{b_v(H)}). \tag{16}

The first equality uses ωF[1−z]=(−1)nF[z−1]\omega F[1-z]=(-1)^nF[z-1] for a homogeneous symmetric function of degree nn. The second uses the cited evaluation 𝐋𝐋𝐓H[z−1;q−1]=∏v(z−q−bv(H))\mathbf{LLT}_H[z-1;q^{-1}]=\prod_v(z-q^{-b_v(H)}) and ∑vbv(H)=|E(H)|\sum_v b_v(H)=|E(H)|. Thus the incoming, rather than outgoing, degree convention is explicit.

Apply (16) to each optional-edge term of Definition 5.3. A vertical edge from uu to the cell vv immediately below it changes only the incoming degree of vv. Write a=a(u)a=a(u), b=bv(G)b=b_v(G), and t=tℓ(u)t=t_{\ell(u)}. Its local sum is

(t−qa)(1−zqb+1)+(−t+qa+1)(1−zqb)q−1=qa(1−zqb−at).\frac{(t-q^a)(1-zq^{b+1})+(-t+q^{a+1})(1-zq^b)}{q-1} =q^a(1-zq^{b-a}t). \tag{17}

Each cell is the lower endpoint of at most one optional edge, so these local sums multiply independently. The factors qaq^a cancel the global q−Aq^{-A}.

Suppose vv is in column cc and the row immediately above has length LL. The incoming attacking edges consist of the c−1c-1 earlier cells in its own row and the L−cL-c upper cells strictly to its right. Therefore b=L−1b=L-1, while the arm of uu is a=L−ca=L-c; hence b−a=c−1b-a=c-1. If vv has no cell above it, its incoming degree is simply c−1c-1.

In a column of height hh, the lower cell in row r<hr<h receives the label ℓ(u)=h−r\ell(u)=h-r; the top cell receives no optional factor, which is represented by t0=1t_0=1. The labels over the column are therefore h−1,h−2,…,0h-1,h-2,\ldots,0. Reversing them in the product gives exactly 0,1,…,h−10,1,\ldots,h-1, while qc−1q^{c-1} is constant in that column. This proves (15).

Finally, the Schur specialization sλ[1−z]s_\lambda[1-z] vanishes unless λ\lambda is a hook, and

s(n−k,1k)[1−z]=(1−z)(−z)k.s_{(n-k,1^k)}[1-z]=(1-z)(-z)^k.

The bottom-left cell supplies the factor 1−z1-z on the right side of (15). Dividing by it and comparing powers of −z-z proves (14) for every partition, including n=1n=1. This supplement is part of the same named package and is not counted separately.

7. Scope and audit record

Conjecture 5.7 is about first moments of exponent vectors. Its proof permits signed coefficients and therefore does not settle Conjecture 5.4. The supplement gives individual Schur coefficients only for hook shapes. No general claim is made about every Yang–Baxter trace mentioned before Conjecture 5.7; the displayed conjecture's defined family is H̃μ\widetilde H_\mu.

The argument uses the source's existing Definition 5.3, Lemma 4.2 and Theorem 3.2 with full attribution. The deductions are the universal first-order identity (11), its consequence (2), and the plethystic evaluation (15) giving (14). A separate source-audit document records the prior searches. Exact finite checks are included only to detect convention mistakes; the proof is the all-size argument (3)–(17).