Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated

Semistable simplex-boundary lattices and uniform extraction


The finite family and theorem

For 2≤k<n2\leq k<n, let En,kE_{n,k} consist of the oriented boundary columns of all kk-subsets of [n][n], with coordinates on the (k−1)(k-1)-subsets. These are distinct nonzero integer vectors. Fix 0<C<∞0<C<\infty and a sequence with

nk∼Clog⁡n.(1)\frac nk\sim C\log n. \tag{1}

Theorem. These families have one positive hereditary full-dissociation extraction constant, depending on CC and at most a finite prefix, but not on the stage or subset.

This settles the extraction side of the earlier simplicial checks: their extraction gap in (1) is resolved. Its arbitrary-color gap is not. In particular, neither unbounded rational coloring nor avoidance of only elementary simplex boundaries is substituted for actual coloring.

1. The exact integer relation lattice

Give exterior powers of Rn\mathbb R^n the Euclidean metric with increasing standard wedges eIe_I orthonormal. Put

u=∑i=1nei,W=u⊥,An−1=W∩Zn,N=(nk).u=\sum_{i=1}^n e_i,\qquad W=u^\perp,\qquad A_{n-1}=W\cap\mathbb Z^n,\qquad N=\binom nk.

The boundary matrix is contraction ∂=ιu\partial=\iota_u from ⋀kZn\bigwedge^k\mathbb Z^n to ⋀k−1Zn\bigwedge^{k-1}\mathbb Z^n. The vectors fi=ei−enf_i=e_i-e_n, i<ni<n, together with ene_n, form a unimodular basis of Zn\mathbb Z^n. Thus every integral kk-vector has a unique expression x+en∧yx+e_n\wedge y, where x∈⋀kAn−1x\in\bigwedge^k A_{n-1} and y∈⋀k−1An−1y\in\bigwedge^{k-1}A_{n-1}. Since ⟨u,fi⟩=0\langle u,f_i\rangle=0 and ⟨u,en⟩=1\langle u,e_n\rangle=1,

∂(x+en∧y)=y.\partial(x+e_n\wedge y)=y.

Therefore the relation lattice is, as an equality of embedded lattices,

Kn,k=⋀kAn−1,m:=rank⁡Kn,k=(n−1k),r:=N−m=(n−1k−1).(2)K_{n,k}=\bigwedge^k A_{n-1},\qquad m:=\operatorname{rank}K_{n,k}=\binom{n-1}{k}, \quad r:=N-m=\binom{n-1}{k-1}. \tag{2}

The Gram matrix of the fif_i is In−1+Jn−1I_{n-1}+J_{n-1}, with eigenvalues n,1,…,1n,1,\ldots,1. The Gram matrix of their kk-fold wedges is its kk-th compound: diagonalize the original Gram matrix and take exterior powers to see its eigenvalues are all products of kk original eigenvalues. Consequently

det⁡(Kn,k)2=n(n−2k−1),σn,k:=det⁡(Kn,k)1/m=nk/[2(n−1)].(3)\det(K_{n,k})^2=n^{\binom{n-2}{k-1}},\qquad \sigma_{n,k}:=\det(K_{n,k})^{1/m} =n^{k/[2(n-1)]}. \tag{3}

These determinants are Euclidean covolumes, not indices in an ambient coordinate lattice.

2. Elementary irreducibility

The coordinate permutations of SnS_n act by signed orthogonal permutations on ⋀kRn\bigwedge^k\mathbb R^n. The representation on ker⁡∂=⋀kW\ker\partial=\bigwedge^k W is irreducible, as follows.

Let BB commute with every such permutation, with entries in the standard wedge basis. If ∣I∖J∣≥2|I\setminus J|\geq2, transpose two elements of I∖JI\setminus J. This fixes eJe_J and negates eIe_I, forcing BI,J=0B_{I,J}=0. There is one orbit of diagonal ordered pairs and one of adjacent ordered pairs ∣I∖J∣=1|I\setminus J|=1; their entries are determined by one scalar each, up to the fixed orientation signs. The commutant therefore has dimension at most two. It contains the independent operators II and ∂∗∂\partial^*\partial; the latter has nonzero adjacent entries, and ∂∗\partial^* is wedge multiplication by uu. Thus every commuting operator is

aI+b ∂∗∂.(4)aI+b\,\partial^*\partial. \tag{4}

If H⊆ker⁡∂H\subseteq\ker\partial is invariant, its orthogonal projector on the entire exterior space commutes with the group. By (4), its restriction to ker⁡∂\ker\partial is aIaI. A projection has a=0a=0 or a=1a=1, so HH is zero or the whole kernel.

3. Integral irreducible lattices are semistable

Here is the needed general elementary observation. Suppose an integral Euclidean lattice LL has a finite group of orthogonal automorphisms acting irreducibly on its real span. Then

det⁡M1/rank⁡M≥det⁡L1/rank⁡L(0≠M⊆L).(5)\det M^{1/\operatorname{rank}M} \geq\det L^{1/\operatorname{rank}L} \quad(0\ne M\subseteq L). \tag{5}

To prove it, let α\alpha be the minimum normalized covolume of its nonzero sublattices. It is attained: squared covolumes are positive integers and there are only finitely many possible ranks. Every minimizer is saturated, since proper saturation lowers covolume. For saturated M1,M2M_1,M_2,

det⁡(M1+M2)det⁡(M1∩M2)≤det⁡M1det⁡M2.(6)\det(M_1+M_2)\det(M_1\cap M_2) \leq\det M_1\det M_2. \tag{6}

Extend a basis of the intersection in each lattice and project off its span. The volume of the combined projected vectors is at most the product of their separate volumes, proving (6). The intersection is primitive in both lattices, as required for the basis extensions.

If both lattices minimize normalized covolume, rank additivity and the defining lower bound det⁡M≥αrank⁡M\det M\geq\alpha^{\operatorname{rank}M} force their sum to be another minimizer. The maximal-rank minimizer is therefore unique and invariant under every lattice automorphism. Irreducibility makes its real span full; saturation then makes it LL. This proves (5).

Apply this argument to the integral lattice in (2), using Section 2. It follows that Kn,kK_{n,k} is semistable and that Kn,k/σn,kK_{n,k}/\sigma_{n,k} is stable, for every 2≤k<n2\leq k<n.

4. A fixed theta estimate in the critical range

Along (1), equations (2)--(3) give

σn,k⟶e1/(2C)>1,mN=n−kn⟶1.\sigma_{n,k}\longrightarrow e^{1/(2C)}>1,\qquad \frac mN=\frac{n-k}{n}\longrightarrow1.

Fix 1<σ0<e1/(2C)1<\sigma_0<e^{1/(2C)}. Choose TT in the weak stable-lattice theorem with ϵ(T)≤12log⁡σ0\epsilon(T)\leq\frac12\log\sigma_0, and put s=1/(σ0T)<1s=1/(\sigma_0T)<1. Eventually σn,k≥σ0\sigma_{n,k}\geq\sigma_0, so the theorem and monotonicity give

θKn,k(s)≤Tmeϵ(T)m.\theta_{K_{n,k}}(s)\leq T^m e^{\epsilon(T)m}.

Consequently

log⁡s+1Nlog⁡θKn,k(s)≤rNlog⁡s−mN(log⁡σ0−ϵ(T))≤−14log⁡σ0<0(7)\log s+\frac1N\log\theta_{K_{n,k}}(s) \leq\frac rN\log s-\frac mN(\log\sigma_0-\epsilon(T)) \leq-\tfrac14\log\sigma_0<0 \tag{7}

once m/N≥1/2m/N\geq1/2. The finite phase-entropy criterion supplies a fully dissociated subfamily of at least δN\delta N columns for one positive constant.

The SnS_n-action is transitive on column indices and carries columns to columns up to signs, through an ambient signed coordinate permutation. These operations preserve all ternary relation supports. Averaging intersections of a dissociated index set with its translates therefore gives the same δ\delta for every subset. This proves the theorem; a finite prefix is absorbed by its positive singleton fractions.

5. Scope and continuation

For one explicit sequence, take k=⌊n/(Clog⁡n)⌋k=\lfloor n/(C\log n)\rfloor for sufficiently large nn. The rational partition number is ⌈n/k⌉\lceil n/k\rceil, which grows. That fact alone says nothing sufficient about the actual partition number. The earlier entropy bound gives only lim sup⁡α(En,k)/N≤1/(2Clog⁡2)\limsup\alpha(E_{n,k})/N\leq1/(2C\log2), a fixed-density restriction, not divergence of χ1\chi_1.

Letting CC grow without bound to force a larger entropy lower bound on colors would make that extraction upper bound tend to zero. Such a change cannot supply the required uniform family.

The lattice identity, determinant, irreducibility, and semistability proofs are elementary and author-recorded; a separate reviewer derived them independently. No computation or external representation theorem was used. The still-open task is a color-adaptive flat cycle under every fixed finite palette at one fixed CC.

A second separate reviewer checked the exact embedded kernel, compound determinant, commutant, saturation argument, and signed transitivity, finding no gap. No verification record is filed, so the result remains author-recorded.