Wiki
Wiki

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

Updated

A counterexample to reflection augmentation


The following proposed lemma is false: if a dissociated set D⊂(0,M/2)D\subset(0,M/2) has size dd, then {M}∪D∪(M−D)\{M\}\cup D\cup(M-D) contains a dissociated (d+1)(d+1)-subset.

Take

D={2,4,10,11},M=23.D=\{2,4,10,11\},\qquad M=23.

The 16 subset sums of DD are

0,2,4,6,10,11,12,13,14,15,16,17,21,23,25,27,0,2,4,6,10,11,12,13,14,15,16,17,21,23,25,27,

so DD is dissociated, and D⊂(0,23/2)D\subset(0,23/2). Nevertheless,

P={2,4,10,11,12,13,19,21,23}P=\{2,4,10,11,12,13,19,21,23\}

has dissociation dimension four. A conceptual finite certificate that every five-subset is dependent follows.

The underlying configuration

Write the points symbolically as

labelABCDEFGHMvaluea2abca+ba+cb+c−ab+ca+b+c.\begin{array}{c|ccccccccc} \text{label}&A&B&C&D&E&F&G&H&M\\ \hline \text{value}&a&2a&b&c&a+b&a+c&b+c-a&b+c&a+b+c. \end{array}

The counterexample uses (a,b,c)=(2,10,11)(a,b,c)=(2,10,11). The argument below uses only these symbolic identities; it proves the same dependence conclusion whenever the nine displayed points are distinct.

There are four complementary pairs

{A,H},{B,G},{C,F},{D,E},(1)\{A,H\},\quad\{B,G\},\quad\{C,F\},\quad\{D,E\}, \tag{1}

each summing to MM. We shall also use

A+E=B+C,A+F=B+D,A+G=C+D,B+H=E+F,C+H=E+G,D+H=F+G.(2)\begin{array}{lll} A+E=B+C,& A+F=B+D,& A+G=C+D,\\ B+H=E+F,& C+H=E+G,& D+H=F+G. \end{array} \tag{2}

Reflection x↦M−xx\mapsto M-x exchanges the members of every pair in (1) and preserves equalities between two-element sums.

Five-subsets containing MM

A five-subset containing MM is dependent if it contains either complete pair in (1). Otherwise it contains exactly one element of each pair. Let rA,rB,rC,rD∈{0,1}r_A,r_B,r_C,r_D\in\{0,1\} indicate whether the chosen member is the reflected member, and let xA,xB,xC,xDx_A,x_B,x_C,x_D be the chosen values. Since M=A+C+DM=A+C+D, substitution gives

(1−rA−rC−rD)M=(−1)rAxA+(−1)rCxC+(−1)rDxD.(3)(1-r_A-r_C-r_D)M =(-1)^{r_A}x_A+(-1)^{r_C}x_C+(-1)^{r_D}x_D. \tag{3}

Unless rA=rC=rD=1r_A=r_C=r_D=1, every coefficient in (3) belongs to {−1,0,1}\{-1,0,1\}, providing a nontrivial ternary relation.

In the remaining case the chosen set contains H,F,EH,F,E. If its fourth non-MM member is BB, use B+H=E+FB+H=E+F. If that member is GG, use M+H=E+F+GM+H=E+F+G. Thus every five-subset containing MM is dependent.

Five-subsets not containing MM

Such a set contains a complete pair in (1). If it contains two complete pairs, their equal sums give a dependence. Otherwise it consists of one complete pair and one member of each of the other three pairs.

If its two-element sums are all distinct, (2) restricts the remaining three members as in the following table. The last column supplies a dependence in every surviving case.

Complete pairRemaining three membersRelation
A,HA,HB,C,DB,C,DC+D=HC+D=H
A,HA,HG,E,FG,E,FA+G=HA+G=H
B,GB,GA,E,FA,E,FA+B+G=E+FA+B+G=E+F
B,GB,GH,C,DH,C,DC+D=HC+D=H
C,FC,FA,G,EA,G,EA+C=EA+C=E
C,FC,FH,B,DH,B,DC+D=HC+D=H
D,ED,EA,G,FA,G,FA+D=FA+D=F
D,ED,EH,B,CH,B,CC+D=HC+D=H

For completeness, the restrictions in the table can be checked without enumerating all five-subsets. If A,HA,H are present and BB is chosen, the other two choices (C,E),(F,D),(F,E)(C,E),(F,D),(F,E) are excluded by the first, second, and fourth identities in (2), leaving C,DC,D. Reflection handles the choice GG. If B,GB,G are present and AA is chosen, the choices (C,D),(C,E),(F,D)(C,D),(C,E),(F,D) are excluded by the third, first, and second identities, leaving F,EF,E; reflection handles HH.

If C,FC,F are present and AA is chosen, choosing BB is impossible: either choice from {D,E}\{D,E\} triggers one of the first two identities. Hence GG is chosen; the third identity then excludes DD, leaving EE. Reflection handles HH. Interchanging b,cb,c gives the final two rows.

This covers every five-subset of PP. Hence d(P)=4d(P)=4, and the reflection-augmentation lemma fails. Since ∣P∣=9<24|P|=9<2^4, this example does not contradict the original extremal target.