The uniform rectangle-creation rate conjecture for the finite k-chain

From papers

Fix kk and let Rk\mathcal{R}_k be the state space of the finite kk-chain. For a partition λRk\lambda\in\mathcal{R}_k, let λ+i\lambda^{+i} denote the state obtained by adding a box to a part of length i1i-1 as defined in the chain, and set

Rki={λRk:λ>λ+i}.\mathcal{R}_k^i=\{\lambda\in\mathcal{R}_k:|\lambda|>|\lambda^{+i}|\}.

If π\pi is the stationary distribution and PP is the transition matrix, define

ρi=λRkiπ(λ)P(λ,λ+i).\rho_i=\sum_{\lambda\in\mathcal{R}_k^i}\pi(\lambda)P(\lambda,\lambda^{+i}).

Here [k]={1,,k}[k]=\{1,\ldots,k\}. Uniform rectangle-creation rate conjecture.

ρi=1(k+23)\rho_i=\frac{1}{\binom{k+2}{3}}

for all i[k]i\in[k]. This would give equal asymptotic frequencies for the creation of rectangles of every relevant size and imply a particularly uniform limit-shape description; the statement is motivated by data for 1ik61\leq i\leq k\leq 6.

Progress summary

Nothing recorded yet. Refresh searches the literature and the public web for attempts on this problem, and writes the first summary here.

Sources & referencesView supporting material

Primary source

Svante Linusson and Alperen Özdemir, “The k-Plancherel measure and a Finite Markov Chain”, arXiv:2512.24346 (2025).

Solutions 1

Proof

The uniform rectangle-creation conjecture holds for every (k); the argument also proves the associated limit-shape conjecture.

Put K=k+1K=k+1 and Ri=(iKi)R_i=(i^{K-i}), 1ik1\le i\le k. We prove simultaneously that every stationary rectangle-creation rate is

ρi=(k+23)1\rho_i=\binom{k+2}{3}^{-1}

and that the limit shape is Dk+1D_{k+1}.

The rectangle property and unique rectangle decomposition give

sλRi(k)=sRisλ(k),Q[h1,,hk]=λRkQ[sR1,,sRk]sλ(k).s^{(k)}_{\lambda\cup R_i}=s_{R_i}s^{(k)}_\lambda,\qquad \mathbb Q[h_1,\ldots,h_k] =\bigoplus_{\lambda\in\mathcal R_k} \mathbb Q[s_{R_1},\ldots,s_{R_k}]s^{(k)}_\lambda.

Indeed, divide the multiplicity of each part ii by KiK-i; the remainder gives a unique λRk\lambda\in\mathcal R_k. Let A(r)A(r) denote multiplication by h1h_1 in this free basis, with ri=sRir_i=s_{R_i}. Each nonrectangle transition contributes 11, and each transition creating RiR_i contributes rir_i.

At exponential specialization p1=1, pj=0p_1=1,\ p_j=0 for j>1j>1, write

cλ=dλ(k)λ!,ri0=fRiRi!.c_\lambda=\frac{d_\lambda^{(k)}}{|\lambda|!},\qquad r_i^0=\frac{f^{R_i}}{|R_i|!}.

The source's transition matrix is

P=diag(c)1A(r0)diag(c).P=\operatorname{diag}(c)^{-1} A(r^0)\operatorname{diag}(c).

For each j=2,,kj=2,\ldots,k, vary pj=tp_j=t while keeping p1=1p_1=1. The specialized Pieri identity gives

A(r(t))c(t)=c(t).A(r(t))c(t)=c(t).

Positivity near t=0t=0 and irreducibility show that the Perron eigenvalue remains exactly 11. Differentiating against its stationary left eigenvector gives

i=1kρiDi,j=0,Di,j=pjlogsRiex=1jChj(Ri).(1)\sum_{i=1}^k\rho_iD_{i,j}=0,\qquad D_{i,j} =\left.\partial_{p_j}\log s_{R_i}\right|_{\operatorname{ex}} =\frac1j\operatorname{Ch}_j(R_i). \tag{1}

Here Chj\operatorname{Ch}_j is the normalized symmetric-group character of a jj-cycle.

Stanley's rectangular-character residue formula gives

Chj((Kq)×q)=1j[x1](x)j(xK)j(xq)j,\operatorname{Ch}_j((K-q)\times q) =-\frac1j[x^{-1}] \frac{(x)_j(x-K)_j}{(x-q)_j},

with falling factorials and expansion at infinity. For j2j\ge2,

q=1K11(xq)j=1j1(1(xK)j11(x1)j1).\sum_{q=1}^{K-1}\frac1{(x-q)_j} =\frac1{j-1} \left(\frac1{(x-K)_{j-1}} -\frac1{(x-1)_{j-1}}\right).

Multiplication by (x)j(xK)j(x)_j(x-K)_j makes both terms polynomials, so their x1x^{-1} coefficients vanish. Hence

i=1kDi,j=0(2jk).(2)\sum_{i=1}^kD_{i,j}=0 \qquad(2\le j\le k). \tag{2}

Moreover, Stanley's permutation-factorization formula and its Narayana top-degree term imply that

Qj(q)=Chj((Kq)×q)q(Kq)Q_j(q)= \frac{\operatorname{Ch}_j((K-q)\times q)} {q(K-q)}

is a polynomial of exact degree j1j-1, with nonzero leading coefficient Catj\operatorname{Cat}_j. Thus Q1,,QkQ_1,\ldots,Q_k form a polynomial basis; evaluating at q=1,,kq=1,\ldots,k gives the nonzero Vandermonde determinant

det[Di,j]1i,jk=1k!(i=1ki(Ki))(j=1kCatj)1i<rk(ri)>0.\det[D_{i,j}]_{1\le i,j\le k} = \frac1{k!} \left(\prod_{i=1}^ki(K-i)\right) \left(\prod_{j=1}^k\operatorname{Cat}_j\right) \prod_{1\le i<r\le k}(r-i)>0.

Therefore columns j=2,,kj=2,\ldots,k have rank k1k-1. By (2) their left kernel is exactly the span of (1,,1)(1,\ldots,1). Equation (1) forces

ρ1==ρk=ρ.\rho_1=\cdots=\rho_k=\rho.

Finally, each step adds one box, while a rectangle transition removes Ri=i(Ki)|R_i|=i(K-i) boxes from the residual state. Stationarity of residual size gives

0=1i=1ki(Ki)ρi=1(k+23)ρ.0=1-\sum_{i=1}^ki(K-i)\rho_i =1-\binom{k+2}{3}\rho.

Consequently

ρi=(k+23)1(1ik).\boxed{\rho_i=\binom{k+2}{3}^{-1}\quad(1\le i\le k).}

For k=1k=1, the same conclusion follows directly from this size-drift equation. The finite-state ergodic theorem gives equal asymptotic multiplicities of all rectangles; the geometric implication immediately following Conjecture 6 in the source then identifies the limiting boundary as

Ck=Dk+1.\boxed{C_k=D_{k+1}.}

Sources: S. Linusson and A. Özdemir, The kk-Plancherel Measure and a Finite Markov Chain, arXiv:2512.24346, Conjectures 5 and 6; R. Stanley, Irreducible Symmetric Group Characters of Rectangular Shape, arXiv:math/0109093, equations (8)–(9) and Theorem 1.

0 endorsements
Shivam Patel ·