Chen–Khare–Sahi's conjecture

What Lean checks

Lean proves the exact source Conjecture 5 counterexample, the whole rectangular-translate family, and the continuous small-positive-q family for every 3/4≤t<1. It verifies the full real semiring, mixed-degree reduction and saturation, and unique monic triangular eigenfunction characterizations including shifted supports. Strict compilation and isolated kernel replay passed; the full repository gate remains pending.

Statement and result

This note disproves Conjecture 5 in Hong Chen, Apoorva Khare and Siddhartha Sahi, Majorization via positivity of Jack and Macdonald polynomial differences, arXiv:2509.19649v2, §5. The implication from majorization to Muirhead-semiring membership fails. The weaker evaluation-positivity inequality is a different assertion and is not disproved here.

Work in three variables. For a partition ν\nu, put

Mν=mνmν(1,1,1),Dλ,μ=Mλ−Mμ(λ≽μ).M_\nu=\frac{m_\nu}{m_\nu(1,1,1)}, \qquad D_{\lambda,\mu}=M_\lambda-M_\mu \quad(\lambda\succeq\mu).

The Muirhead semiring ℳS\mathcal M_S consists of finite sums with nonnegative real coefficients of finite products of the MνM_\nu and Dλ,μD_{\lambda,\mu}. As in the source's §2.1, partitions have nonnegative integer parts. The partitions in a difference have equal size. Conjecture 5, assertion (2′)(2'), claims that

Pλ(x;q,t)Pλ(1,1,1;q,t)−Pμ(x;q,t)Pμ(1,1,1;q,t)∈ℳS\frac{P_\lambda(x;q,t)}{P_\lambda(1,1,1;q,t)}- \frac{P_\mu(x;q,t)}{P_\mu(1,1,1;q,t)}\in\mathcal M_S \tag{1}

for every 0<q,t<10\lt q,t<1 whenever λ≽μ\lambda\succeq\mu. Here PP denotes the monic Macdonald polynomial.

Theorem. At

q=110,t=910,λ=(4,2,0),μ=(4,1,1),q=\frac1{10},\qquad t=\frac9{10},\qquad \lambda=(4,2,0),\qquad\mu=(4,1,1), \tag{2}

the normalized difference in (1) does not belong to ℳS\mathcal M_S. The same conclusion holds, at the same parameters, for every pair

(r+4,r+2,r)≽(r+4,r+1,r+1),r∈ℤ≥0.(r+4,r+2,r)\succeq(r+4,r+1,r+1),\qquad r\in\mathbb Z_{\ge0}. \tag{3}

The proof gives a rational linear functional that is nonnegative on the entire relevant homogeneous part of the semiring and is strictly negative on (1). All computations below are rational identities; verify.py checks them using only Python's standard library.

1. A separating functional

On the degree-six symmetric polynomials define LL by

ν\nu 600 510 420 411 330 321 222
L(Mν)L(M_\nu) 62 28 9 9 2 2 0

Digits abbreviate partitions; for example, 420 means (4,2,0)(4,2,0). Since these seven MνM_\nu form a basis, this defines LL uniquely. Its values on the unnormalized monomial basis are

(L(m600),L(m510),L(m420),L(m411),L(m330),L(m321),L(m222))=(186,168,54,27,6,12,0).(L(m_{600}),L(m_{510}),L(m_{420}),L(m_{411}), L(m_{330}),L(m_{321}),L(m_{222})) =(186,168,54,27,6,12,0). \tag{4}

We will prove

f∈ℳS,f homogeneous of degree 6⇒L(f)≥0.f\in\mathcal M_S,\quad f\text{ homogeneous of degree }6 \quad\Longrightarrow\quad L(f)\ge0. \tag{5}

Exhaustive reduction of the semiring

Every generator of ℳS\mathcal M_S is homogeneous and nonnegative on the nonnegative orthant. For differences this is Muirhead's inequality. Consequently, if a homogeneous polynomial of degree six belongs to the semiring, it has a representation using only products of total degree six. Indeed, each homogeneous component of any other degree in a proposed representation is a nonnegative sum of nonnegative polynomials and is identically zero. Every nonzero summand in such a component would be positive somewhere on the positive orthant, which is impossible.

There are two further reductions.

  1. A difference Mλ−MμM_\lambda-M_\mu is a sum of differences along a saturated chain from λ\lambda to μ\mu in dominance order. Thus it suffices to use cover differences.
  2. A product of ordinary MM-factors has nonnegative coefficients in the monomial basis and is symmetric. It is therefore a nonnegative linear combination of single MρM_\rho. Thus a product containing differences needs at most one ordinary MM-factor.

Every nonzero difference has degree at least two. A degree-six product therefore has at most three difference factors. The exhaustive remaining list consists of:

  • 27 products with one cover difference;
  • 10 products with two cover differences;
  • the cube of the unique degree-two difference;
  • the seven ordinary degree-six monomials MνM_\nu.

For completeness, the cover differences in degrees two through six are

Degree Covers λ⋗μ\lambda\gtrdot\mu
2 200–110
3 300–210, 210–111
4 400–310, 310–220, 220–211
5 500–410, 410–320, 320–311, 311–221
6 600–510, 510–420, 420–411, 420–330, 411–321, 330–321, 321–222

The following table checks all 38 products. Write Da,b=Ma−MbD_{a,b}=M_a-M_b and M000=1M_{000}=1.

Product LL value
D200,110M400D_{200,110}M_{400} 5
D200,110M310D_{200,110}M_{310} 4
D200,110M220D_{200,110}M_{220} 4
D200,110M211D_{200,110}M_{211} 3
D300,210M300D_{300,210}M_{300} 9
D300,210M210D_{300,210}M_{210} 9
D300,210M111D_{300,210}M_{111} 7
D210,111M300D_{210,111}M_{300} 4
D210,111M210D_{210,111}M_{210} 2
D210,111M111D_{210,111}M_{111} 2
D400,310M200D_{400,310}M_{200} 16
D400,310M110D_{400,310}M_{110} 15
D310,220M200D_{310,220}M_{200} 14/314/3
D310,220M110D_{310,220}M_{110} 14/314/3
D220,211M200D_{220,211}M_{200} 5/35/3
D220,211M110D_{220,211}M_{110} 2/32/3
D500,410M100D_{500,410}M_{100} 24
D410,320M100D_{410,320}M_{100} 11
D320,311M100D_{320,311}M_{100} 0
D311,221M100D_{311,221}M_{100} 3
D600,510D_{600,510} 34
D510,420D_{510,420} 19
D420,411D_{420,411} 0
D420,330D_{420,330} 7
D411,321D_{411,321} 7
D330,321D_{330,321} 0
D321,222D_{321,222} 2
D200,1102M200D_{200,110}^2M_{200} 2/32/3
D200,1102M110D_{200,110}^2M_{110} 1/31/3
D200,110D300,210M100D_{200,110}D_{300,210}M_{100} 2/32/3
D200,110D210,111M100D_{200,110}D_{210,111}M_{100} 2/32/3
D200,110D400,310D_{200,110}D_{400,310} 1
D200,110D310,220D_{200,110}D_{310,220} 0
D200,110D220,211D_{200,110}D_{220,211} 1
D300,2102D_{300,210}^2 0
D300,210D210,111D_{300,210}D_{210,111} 2
D210,1112D_{210,111}^2 0
D200,1103D_{200,110}^3 1/31/3

These entries can be obtained by expanding Mν=|Orb(ν)|−1∑e∈Orb(ν)xeM_\nu=|\mathop{\mathrm{Orb}}(\nu)|^{-1} \sum_{e\in\mathop{\mathrm{Orb}}(\nu)}x^e and applying the first table. The verifier generates the cover list independently from dominance inequalities, enumerates precisely the cases above, and recomputes every entry. All are nonnegative. The ordinary monomial generators also have nonnegative LL-values, proving (5).

2. Exact Macdonald polynomials

At the rational parameters (2), the two required monic polynomials are

P420=m420+1191(m411+m330)+11430989092629m321+2705349603390189787051m222,P411=m411+111991m321+122190181m222.\begin{aligned} P_{420}={}&m_{420} +\frac{11}{91}(m_{411}+m_{330}) +\frac{1143098}{9092629}m_{321} +\frac{27053496033}{90189787051}m_{222},\\ P_{411}={}&m_{411} +\frac{111}{991}m_{321} +\frac{1221}{90181}m_{222}. \end{aligned} \tag{6}

Here is an independent way to verify their identification that does not use the branching implementation that found the example. The first Macdonald difference operator is

𝒟f(x)=∑i=13∏j≠itxi−xjxi−xjf(x1,…,qxi,…,x3).\mathscr D f(x)=\sum_{i=1}^3 \prod_{j\ne i}\frac{tx_i-x_j}{x_i-x_j}\, f(x_1,\ldots,qx_i,\ldots,x_3). \tag{7}

It is triangular on the monomial basis, with diagonal eigenvalues

ην=qν1t2+qν2t+qν3.\eta_\nu=q^{\nu_1}t^2+q^{\nu_2}t+q^{\nu_3}.

The monic triangular eigenpolynomial with eigenvalue ην\eta_\nu is PνP_\nu. At (2), the seven degree-six eigenvalues are distinct. Direct substitution of (6) into (7) gives

𝒟P420=1009081106P420,𝒟P411=190081106P411.\mathscr D P_{420}=\frac{1009081}{10^6}P_{420}, \qquad \mathscr D P_{411}=\frac{190081}{10^6}P_{411}. \tag{8}

These checks are polynomial identities after multiplication by Δ=(x1−x2)(x1−x3)(x2−x3)\Delta=(x_1-x_2)(x_1-x_3)(x_2-x_3). For example, writing the variables as x,y,zx,y,z, the three coefficients after this multiplication are

W1=(y−z)(tx−y)(tx−z),W2=−(x−z)(ty−x)(ty−z),W3=(x−y)(tz−x)(tz−y).\begin{aligned} W_1&=(y-z)(tx-y)(tx-z),\\ W_2&=-(x-z)(ty-x)(ty-z),\\ W_3&=(x-y)(tz-x)(tz-y). \end{aligned}

The verifier expands ∑iWiP(x1,…,qxi,…,x3)\sum_i W_iP(x_1,\ldots,qx_i,\ldots,x_3) and compares every rational monomial coefficient with that of ηνΔP\eta_\nu\Delta P. It also checks distinctness of all seven eigenvalues, monicity, and triangular support. This proves (6) using the defining eigenfunction characterization.

The evaluations and functional values are

P420(1,1,1)=70163492573790189787051,L(P420P420(1,1,1))=19874360078325986478731,P411(1,1,1)=33237090181,L(P411P411(1,1,1))=28401136930.\begin{aligned} P_{420}(1,1,1)&=\frac{701634925737}{90189787051}, &L\!\left(\frac{P_{420}}{P_{420}(1,1,1)}\right) &=\frac{198743600783}{25986478731},\\ P_{411}(1,1,1)&=\frac{332370}{90181}, &L\!\left(\frac{P_{411}}{P_{411}(1,1,1)}\right) &=\frac{284011}{36930}. \end{aligned}

Subtracting gives

L(P420P420(1,1,1)−P411P411(1,1,1))=−4538292661539106631184392870<0.L\!\left(\frac{P_{420}}{P_{420}(1,1,1)} -\frac{P_{411}}{P_{411}(1,1,1)}\right) =-\frac{4538292661539}{106631184392870}<0. \tag{9}

Because 420≽411420\succeq411, equations (5) and (9) contradict (1). This proves the first assertion of the theorem.

3. Rectangular translates give an infinite counterfamily

Write e3=x1x2x3e_3=x_1x_2x_3. We first prove the following property of the whole Muirhead semiring:

e3rF∈ℳS⇒F∈ℳS(r∈ℤ≥0).e_3^rF\in\mathcal M_S\quad\Longrightarrow\quad F\in\mathcal M_S\qquad(r\in\mathbb Z_{\ge0}). \tag{10}

For a monomial generator MνM_\nu, its order of vanishing in x3x_3 is ν3\nu_3, and

Mν=e3ν3Mν−ν3(1,1,1).M_\nu=e_3^{\nu_3}M_{\nu-\nu_3(1,1,1)}.

For a nonzero difference Dλ,μD_{\lambda,\mu}, dominance implies λ3≤μ3\lambda_3\le\mu_3, so

Dλ,μ=e3λ3Dλ−λ3(1,1,1),μ−λ3(1,1,1).D_{\lambda,\mu}=e_3^{\lambda_3} D_{\lambda-\lambda_3(1,1,1),\, \mu-\lambda_3(1,1,1)}. \tag{11}

The residual factor in (11) is again a Muirhead difference. Its restriction to x3=0x_3=0 is nonnegative and is not identically zero: the first partition has last part zero; the second either has positive last part, in which case its monomial restriction vanishes, or also has last part zero, in which case the two distinct monomial symmetric polynomials in the remaining variables cannot cancel. Thus the order in x3x_3 of this difference is exactly λ3\lambda_3.

It follows that every nonzero product of generators can be written e3aGe_3^aG, where G∈ℳSG\in\mathcal M_S and G(x1,x2,0)G(x_1,x_2,0) is a nonzero nonnegative polynomial. In a finite nonnegative sum of such products, terms of minimum order aa cannot cancel: their leading coefficients are nonnegative functions on the positive quadrant and are not all zero. Consequently, if the sum is divisible by x3rx_3^r, every product in the sum has a≥ra\ge r. Dividing each product by e3re_3^r leaves e3a−rG∈ℳSe_3^{a-r}G\in\mathcal M_S, proving (10).

The standard rectangular-translation identity is

Pν+r(1,1,1)(x;q,t)=e3rPν(x;q,t).P_{\nu+r(1,1,1)}(x;q,t)=e_3^rP_\nu(x;q,t).

Normalization at (1,1,1)(1,1,1) preserves this identity. Thus the difference for the pair (3) is e3re_3^r times the difference in (9). If any such translate belonged to the semiring, (10) would imply that the original difference did too. This contradiction proves the whole family (3).

4. Scope and validation

This is a counterexample to the complete equivalence asserted in Conjecture 5, through its majorization-to-semiring implication. It makes no claim that the ordinary evaluation difference is negative. In fact, Sra's October 2026 paper proves evaluation positivity for the very pair (4,2,0)≽(4,1,1)(4,2,0)\succeq(4,1,1), in Certificate 4.1.

The authors' Example 4.1 concerns a Schur difference outside the smaller Muirhead cone and explicitly represents it inside the semiring. Their Example 4.7 concerns a different Jack semiring and reverses the allowed parameter order. Neither is the counterexample proved here. The Jack-semiring Conjecture 4 remains untouched by this argument.

Validation performed:

  • exact discovery by Macdonald branching and a rational dual functional;
  • independent identification by the first Macdonald difference operator;
  • exhaustive reduction of the full degree-six semiring to 38 products;
  • an independently written standard-library check of the same 38 products;
  • exact rational evaluations in (6)–(9);
  • separate source and prior-status audits.

Run python verify.py. No numerical approximation or Lean verification is used in the certificate. The auxiliary discovery scripts use numerical linear programming only to propose functionals; that search is not part of the proof.

Sources

  1. Hong Chen, Apoorva Khare and Siddhartha Sahi, Majorization via positivity of Jack and Macdonald polynomial differences, arXiv:2509.19649v2, Conjecture 5, equation (5.7), and the Muirhead-semiring definition in §1. https://arxiv.org/html/2509.19649v2
  2. I. G. Macdonald, Symmetric Functions and Hall Polynomials, second edition, Oxford University Press, 1995, Chapter VI, §4, for the difference-operator characterization and rectangular translations. The operator and eigenfunction characterization also appear explicitly in A. M. Garsia and M. Haiman, A remarkable q,t-Catalan sequence and q-Lagrange inversion, §2, equations (2.1)–(2.5), printed page 14: https://math.berkeley.edu/~mhaiman/ftp/remarkable-qtcat/qtcat.pdf
  3. Suvrit Sra, On majorization inequalities for normalized Jack and Macdonald polynomials, 8 October 2026, Certificate 4.1 and §1.3. https://suvrit.github.io/papers/jackratio/jackratio-v1.pdf Also available at https://hexagonmath.org/pdf/2610.00725v1.

See SOURCE_AUDIT.md for the separate priority search and ANALYSIS_REVIEW.md for independent review.