Wiki
Wiki

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

Updated

Constant composition families


This page treats a class different from the uniform product layers in Arbitrary local layers: the entire distribution of local subsets is prescribed, rather than just their total cardinality. It gives an exact transversal formula and a proved three-quarter bound, with seed-dependent additive error, for every finite seed.

The conceptual proof in Arbitrary-seed coupling now establishes the polytope inclusion for every seed size. The earlier three-point exact rational computational certificate is retained below as an independent finite verification. No unrestricted reduction from arbitrary uniform families has been established.

1. The family and its exact transversal number

Let VV be a finite seed of size NN, and take qq disjoint, identified copies V1,…,VqV_1,\ldots,V_q. For each I⊆VI\subseteq V, prescribe a nonnegative integer mIm_I, with

∑I⊆VmI=q,k=∑I⊆V∣I∣mI>0.\sum_{I\subseteq V}m_I=q, \qquad k=\sum_{I\subseteq V}|I|m_I>0.

Let H(q,m)\mathcal H(q,m) consist of all subsets EE of the union of the copies for which exactly mIm_I blocks satisfy E∩Vj=IE\cap V_j=I. Thus the blocks may be permuted arbitrarily, but every local state has a fixed multiplicity. Every edge has size kk.

For D⊆VD\subseteq V put

hD=∑I:I∩D=∅mI.h_D=\sum_{I:I\cap D=\varnothing}m_I.

Then

τ(H(q,m))=min⁡∅≠D⊆VhD<q∣D∣(hD+1).(1)\boxed{\displaystyle \tau(\mathcal H(q,m)) =\min_{\substack{\varnothing\ne D\subseteq V\\h_D<q}} |D|(h_D+1).} \tag{1}

For the upper bound, delete a copy of DD in hD+1h_D+1 blocks. An edge avoiding these deletions would have to assign compatible states to all these blocks, although only hDh_D compatible state occurrences are available.

For the reverse bound, let TT be a transversal, and write Dj=T∩VjD_j=T\cap V_j. An edge avoiding TT is exactly a matching assigning the prescribed multiset of local states to the qq block positions, with a state II allowed at position jj when I∩Dj=∅I\cap D_j=\varnothing. Hall's theorem gives a set UU of positions whose union of compatible states has total multiplicity less than ∣U∣|U|. No DjD_j for j∈Uj\in U is empty, since an empty deletion admits every state. Choose j0∈Uj_0\in U minimizing ∣Dj∣|D_j|, and put D=Dj0D=D_{j_0}. Then

hD<∣U∣≤q,∣T∣≥∑j∈U∣Dj∣≥∣D∣∣U∣≥∣D∣(hD+1).h_D<|U|\le q, \qquad |T|\ge\sum_{j\in U}|D_j| \ge |D||U|\ge |D|(h_D+1).

This proves (1).

Delete unused seed points, meaning points belonging to no state II with mI>0m_I>0. This changes neither the family nor its transversal number. Thereafter hD<qh_D<q for every nonempty DD. With

αI=mI/q,a=∑I∣I∣αI=k/q,b=min⁡∅≠D⊆V∣D∣∑I:I∩D=∅αI,\alpha_I=m_I/q,\qquad a=\sum_I|I|\alpha_I=k/q,\qquad b=\min_{\varnothing\ne D\subseteq V} |D|\sum_{I:I\cap D=\varnothing}\alpha_I,

formula (1) gives the useful exact-error bounds

qb≤τ(H(q,m))≤qb+N.(2)qb\le\tau(\mathcal H(q,m))\le qb+N. \tag{2}

In particular, m∅=0m_\varnothing=0 implies τ≤N\tau\le N.

2. The finite Fano coupling problem

Index seven rows by the nonzero vectors of F23\mathbb F_2^3. Let R\mathcal R be the downward closure of the seven Fano caps, where a cap is the complement of a Fano line. For S,T∈RS,T\in\mathcal R,

S∪T≠[7].(3)S\cup T\ne[7]. \tag{3}

Indeed, their complements contain two Fano lines, which intersect. Consequently a seven-row array whose every point-spectrum lies in R\mathcal R has no two-point transversal.

For a local array of seven subsets I1,…,I7⊆VI_1,\ldots,I_7\subseteq V whose columns belong to R\mathcal R, take its empirical distribution

17∑r=17δIr.\frac17\sum_{r=1}^7\delta_{I_r}.

Let PNP_N be the convex hull of these finitely many distributions. Any seven edges of a constant-composition family that have all their point-spectra in R\mathcal R give α∈PN\alpha\in P_N, by averaging over their blocks. Conversely, a rational α∈PN\alpha\in P_N has a rational convex decomposition; repeating its local arrays and symmetrizing the rows by an order-seven Fano automorphism realizes a bad seven-tuple at suitable multiples of the common denominator. The rounding argument below removes this divisibility issue under a strict transversal excess.

Here is a smaller sufficient polytope. For pairwise disjoint I,J,K⊆VI,J,K\subseteq V (empty states may repeat), put

Γ(I,J,K)=4δI+2δJ+δK7,KN=conv⁡{Γ(I,J,K):I,J,K pairwise disjoint}.(4)\Gamma(I,J,K)=\frac{4\delta_I+2\delta_J+\delta_K}{7}, \qquad K_N=\operatorname{conv}\{\Gamma(I,J,K):I,J,K\text{ pairwise disjoint}\}. \tag{4}

Then KN⊆PNK_N\subseteq P_N. To realize a generator, partition the seven rows into a Fano cap C1C_1 of size four, a two-element subset C2C_2 of its complementary line, and the remaining singleton C3C_3. Give the rows in these three classes the states I,J,KI,J,K, respectively. Every seed point has spectrum C1,C2,C3C_1,C_2,C_3, or the empty set. Each lies in R\mathcal R; for C2C_2, use the fact that a Fano line avoids any prescribed pair of points.

Define the probability polytope

QN={α: αI≥0, ∑IαI=1,∣D∣∑I:I∩D=∅αI≥34∑I∣I∣αI(∅≠D⊆V)}.(5)Q_N=\left\{\alpha:\ \alpha_I\ge0,\ \sum_I\alpha_I=1,\quad |D|\sum_{I:I\cap D=\varnothing}\alpha_I \ge\frac34\sum_I|I|\alpha_I \quad(\varnothing\ne D\subseteq V)\right\}. \tag{5}

The stronger inclusion QN⊆KNQ_N\subseteq K_N is proved for every NN in Arbitrary-seed coupling, by a conceptual separation argument. In particular, a distribution with b≥3a/4b\ge3a/4 belongs to PNP_N. The original verification for N≤3N\le3 follows.

3. The checked three-point polytope inclusion

Finite lemma. For 1≤N≤31\le N\le3,

QN⊆KN⊆PN.(6)\boxed{Q_N\subseteq K_N\subseteq P_N.} \tag{6}

Here are the full data for the nontrivial N=3N=3 inclusion. Identify subsets of the seed with bit masks 0,…,70,\ldots,7, in the order

∅, 1, 2, 12, 3, 13, 23, 123.\varnothing,\ 1,\ 2,\ 12,\ 3,\ 13,\ 23,\ 123.

Write Γijk=Γ(i,j,k)\Gamma_{ijk}=\Gamma(i,j,k). The polytope Q3Q_3 has 52 vertices, in the following 16 orbits under permutations of the three seed points. In the middle column, the displayed vector is divided by dd.

Orbit sizeddNumerator vector of α\alphaConvex decomposition
610(2,1,3,0,4,0,0,0)(2,1,3,0,4,0,0,0)(2Γ200+Γ402+Γ001+6Γ421)/10(2\Gamma_{200}+\Gamma_{402}+\Gamma_{001}+6\Gamma_{421})/10
319(4,2,6,0,6,0,1,0)(4,2,6,0,6,0,1,0)(7Γ240+7Γ420+3Γ106+2Γ061)/19(7\Gamma_{240}+7\Gamma_{420}+3\Gamma_{106}+2\Gamma_{061})/19
328(6,3,9,0,9,0,0,1)(6,3,9,0,9,0,0,1)(3Γ241+3Γ421+2Γ007)/8(3\Gamma_{241}+3\Gamma_{421}+2\Gamma_{007})/8
69(2,1,2,1,3,0,0,0)(2,1,2,1,3,0,0,0)(3Γ214+2Γ430+Γ421+3Γ043)/9(3\Gamma_{214}+2\Gamma_{430}+\Gamma_{421}+3\Gamma_{043})/9
313(3,2,2,2,4,0,0,0)(3,2,2,2,4,0,0,0)(2Γ340+2Γ102+3Γ210+6Γ403)/13(2\Gamma_{340}+2\Gamma_{102}+3\Gamma_{210}+6\Gamma_{403})/13
113(3,3,3,0,3,0,0,1)(3,3,3,0,3,0,0,1)(7Γ240+10Γ142+7Γ070+2Γ421)/26(7\Gamma_{240}+10\Gamma_{142}+7\Gamma_{070}+2\Gamma_{421})/26
611(3,0,3,0,5,0,0,0)(3,0,3,0,5,0,0,0)(3Γ204+Γ420+7Γ402)/11(3\Gamma_{204}+\Gamma_{420}+7\Gamma_{402})/11
310(3,0,3,0,3,0,1,0)(3,0,3,0,3,0,1,0)(2Γ600+9Γ240+3Γ060+6Γ402)/20(2\Gamma_{600}+9\Gamma_{240}+3\Gamma_{060}+6\Gamma_{402})/20
329(9,0,9,0,9,0,0,2)(9,0,9,0,9,0,0,2)(12Γ400+29Γ240+10Γ042+7Γ700)/58(12\Gamma_{400}+29\Gamma_{240}+10\Gamma_{042}+7\Gamma_{700})/58
69(3,0,1,2,3,0,0,0)(3,0,1,2,3,0,0,0)(3Γ420+8Γ042+7Γ340)/18(3\Gamma_{420}+8\Gamma_{042}+7\Gamma_{340})/18
313(5,0,0,4,4,0,0,0)(5,0,0,4,4,0,0,0)(4Γ304+3Γ430+6Γ043)/13(4\Gamma_{304}+3\Gamma_{430}+6\Gamma_{043})/13
37(3,0,0,0,4,0,0,0)(3,0,0,0,4,0,0,0)Γ400\Gamma_{400}
113(7,0,0,2,0,2,2,0)(7,0,0,2,0,2,2,0)(7Γ600+7Γ500+2Γ300+10Γ030)/26(7\Gamma_{600}+7\Gamma_{500}+2\Gamma_{300}+10\Gamma_{030})/26
35(3,0,0,0,0,0,2,0)(3,0,0,0,0,0,2,0)(2Γ600+3Γ060)/5(2\Gamma_{600}+3\Gamma_{060})/5
113(9,0,0,0,0,0,0,4)(9,0,0,0,0,0,0,4)(6Γ000+7Γ700)/13(6\Gamma_{000}+7\Gamma_{700})/13
11(1,0,0,0,0,0,0,0)(1,0,0,0,0,0,0,0)Γ000\Gamma_{000}

The displayed decompositions can be checked by direct rational arithmetic. The exhaustiveness of the vertex list is checked by evidence/main.py, using the machine-readable version of this table in evidence/assets/couplings.json. The verifier requires only the Python standard library and the shared failure harness.

For clarity, the enumeration is finite and exhaustive for the following reason. There are eight nonnegativity inequalities and seven inequalities in (5), in the affine space ∑IαI=1\sum_I\alpha_I=1 of dimension seven. The polytope has full dimension there: taking every nonempty coordinate sufficiently small and positive makes all inequalities strict. Every vertex therefore has seven linearly independent active inequalities. The verifier examines all (157)=6435\binom{15}{7}=6435 choices, solves each nonsingular system with exact rational arithmetic, retains precisely its feasible solutions, and checks that the result is exactly the 52 symmetry images in the table. It also checks every coefficient, every disjointness condition in (4), and every convex identity. This proves Q3⊆K3Q_3\subseteq K_3 by convexity.

For N<3N<3, add unused seed coordinates. Summing the singleton inequalities in (5) gives

a≤4N3N+4<43.a\le\frac{4N}{3N+4}<\frac43.

An inequality involving at least one old coordinate is preserved when unused coordinates are added to its deletion set; an inequality involving only unused coordinates has left side at least one and right side 3a/4<13a/4<1. Thus the embedded distribution belongs to Q3Q_3. In a convex decomposition furnished by the table, no positive term can use an added coordinate, since its state probability is zero. This proves QN⊆KNQ_N\subseteq K_N for N<3N<3 as well.

4. Integral realization with a reserve of empty blocks

The following argument also explains why a strict asymptotic excess cannot hide in divisibility or in local states of bounded multiplicity.

Assume unused seed points have been removed, and let N≥1N\ge1. Suppose

η=b−34a>0,δ=η2(N+1)a.\eta=b-\frac34a>0, \qquad \delta=\frac{\eta}{2(N+1)a}.

Inflate every nonempty state probability by 1+δ1+\delta:

βI=(1+δ)αI(I≠∅),β∅=1−∑I≠∅βI.(7)\beta_I=(1+\delta)\alpha_I\quad(I\ne\varnothing), \qquad \beta_\varnothing=1-\sum_{I\ne\varnothing}\beta_I. \tag{7}

This is a probability distribution. Indeed, writing p=1−α∅≤ap=1-\alpha_\varnothing\le a, one has

δp≤δa=η2(N+1)≤Nα∅2(N+1)<α∅,\delta p\le\delta a=\frac{\eta}{2(N+1)} \le\frac{N\alpha_\varnothing}{2(N+1)}<\alpha_\varnothing,

where η≤b≤Nα∅\eta\le b\le N\alpha_\varnothing follows by taking D=VD=V.

For a nonempty DD, its avoidance probability changes by minus δPr⁡α(I∩D≠∅)\delta\Pr_\alpha(I\cap D\ne\varnothing). Therefore

∣D∣Pr⁡β(I∩D=∅)−34Eβ∣I∣≥η−δ(N+3/4)a≥η/2>0.(8)\begin{aligned} |D|\Pr_\beta(I\cap D=\varnothing)-\frac34\mathbb E_\beta|I| &\ge \eta-\delta(N+3/4)a\\ &\ge\eta/2>0. \end{aligned} \tag{8}

Here Pr⁡α(I∩D≠∅)≤a\Pr_\alpha(I\cap D\ne\varnothing)\le a was used. Hence β∈QN\beta\in Q_N.

By the arbitrary-seed inclusion and Carathéodory's theorem, write

β=∑j=1LλjΓj,L≤2N,λj≥0,∑jλj=1.\beta=\sum_{j=1}^{L}\lambda_j\Gamma_j, \qquad L\le2^N,\quad\lambda_j\ge0,\quad\sum_j\lambda_j=1.

Consequently

α=11+δ∑jλjΓj+δ1+δδ∅.(9)\alpha=\frac1{1+\delta}\sum_j\lambda_j\Gamma_j +\frac{\delta}{1+\delta}\delta_\varnothing. \tag{9}

Each Γj\Gamma_j has a seven-block realization in which each row has local state counts 7Γj7\Gamma_j. Take the local construction following (4) and apply all seven powers of an order-seven Fano automorphism to its row labels. For the binary labels 1,…,71,\ldots,7 used above, one such automorphism is the Singer cycle

(1 2 4 3 6 7 5).(1\ 2\ 4\ 3\ 6\ 7\ 5).

It is multiplication by xx in F2[x]/(x3+x+1)\mathbb F_2[x]/(x^3+x+1), so it preserves the Fano lines and caps.

Use

nj=⌊qλj7(1+δ)⌋n_j=\left\lfloor\frac{q\lambda_j}{7(1+\delta)}\right\rfloor

copies of this seven-block realization, filling all other positions initially with empty blocks. For each nonempty state II, every row has the same integer deficit

dI=mI−∑j7nj(Γj)I≥0.d_I=m_I-\sum_j7n_j(\Gamma_j)_I\ge0.

Moreover

D∗:=∑I≠∅dI<7L.(10)D_*:=\sum_{I\ne\varnothing}d_I<7L. \tag{10}

At least qδ/(1+δ)q\delta/(1+\delta) blank blocks are available. A further seven-block construction corrects one unit of any specified dId_I: in block rr put II in row rr and the empty state in the other six rows. Every point-spectrum is a singleton, so it remains in R\mathcal R. Each row gains exactly one occurrence of II. Thus all deficits can be corrected using 7D∗<49L7D_*<49L blank blocks. All remaining blocks stay empty. Each row now has precisely the prescribed composition mm, while (3) excludes a two-point transversal.

In particular the construction works whenever

qδ1+δ≥49⋅2N.(11)\frac{q\delta}{1+\delta}\ge49\cdot2^N. \tag{11}

5. Consequence for the local seven-edge property

Theorem. Every constant-composition family with an NN-point seed and property P7P_7 satisfies

τ(H(q,m))≤34k+CN,CN=N+49⋅2N(2N(N+1)+1).(12)\boxed{\tau(\mathcal H(q,m))\le\frac34k+C_N,\qquad C_N=N+49\cdot2^N\bigl(2N(N+1)+1\bigr).} \tag{12}

In particular, C3=9803C_3=9803. For every sequence of these families with k→∞k\to\infty and fixed NN, or more generally N22N=o(k)N^2 2^N=o(k),

lim sup⁡τk≤34.\limsup\frac{\tau}{k}\le\frac34.

The composition may vary arbitrarily along the sequence.

If unused seed points were removed, apply the bound with their remaining number N′≤NN'\le N and use CN′≤CNC_{N'}\le C_N.

To verify the stated, nonoptimized constant, suppose τ>3k/4+CN\tau>3k/4+C_N. Formula (2) implies

qη=qb−3k/4>CN−N.q\eta=qb-3k/4>C_N-N.

Also a≤Na\le N and η≤b≤1\eta\le b\le1, the latter by taking any singleton deletion. Thus

qδ1+δ=qη2(N+1)a+η≥qη2N(N+1)+1>49⋅2N.\frac{q\delta}{1+\delta} =\frac{q\eta}{2(N+1)a+\eta} \ge\frac{q\eta}{2N(N+1)+1} >49\cdot2^N.

The integral construction gives seven edges without a two-cover, contradicting P7P_7. Repeated rows cause no problem: their underlying subfamily still has at most seven edges and no two-cover. This proves (12).

The large additive constant is only a convenient rounding allowance. The substantive finite statement is the arbitrary-seed polytope inclusion; no claim of optimal additive error is made. This remains a restricted-class theorem, since no local-property-preserving constant-composition completion of arbitrary uniform families is known here.

6. An arbitrary-seed constraint and its limitation

If b>3a/4b>3a/4, every pairwise-intersecting family F\mathcal F of nonempty states has probability mass x=α(F)<4/7x=\alpha(\mathcal F)<4/7. For x>0x>0, choose a minimum-size state I∈FI\in\mathcal F. Then ∣I∣≤a/x|I|\le a/x, and a state disjoint from II cannot lie in F\mathcal F. Consequently

b≤∣I∣Pr⁡(J∩I=∅)≤a(1−x)x,b\le |I|\Pr(J\cap I=\varnothing)\le\frac{a(1-x)}x,

which implies the claim. These are necessary inequalities for KNK_N: a disjoint triple contains at most one nonempty member of F\mathcal F, with coefficient at most 4/74/7. They do not establish membership in KNK_N.

The singleton inequalities alone must not replace the full hypothesis. They give pmax⁡+3a/4≤1p_{\max}+3a/4\le1, where pmax⁡=max⁡vPr⁡(v∈I)p_{\max}=\max_v\Pr(v\in I), but this does not imply membership in KNK_N. On five points, put mass 1/1001/100 on the whole seed, mass 99/50099/500 on each singleton, and zero on the empty state. Then a=26/25a=26/25, pmax⁡=26/125p_{\max}=26/125, and

pmax⁡+3a/4=247/250<1.p_{\max}+3a/4=247/250<1.

Any disjoint-triple template containing the whole seed has empty states in its other two slots, so a convex combination with positive whole-seed mass must also have positive empty mass. The displayed law is not in K5K_5. It has b=0b=0 and is not a counterexample to QN⊆KNQ_N\subseteq K_N, or to either assertion of Problem 644.