Wiki
Wiki

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

Updated

Proof

../


Throughout, qq is an odd prime power, μ∈Fq\mu\in\mathbb F_q is a non-square and, for t,u∈Fqt,u\in\mathbb F_q,

P(t)=(t,t2,0),C(u)=(0,μu2,u),A={P(t)}∪{C(u)}.P(t)=(t,t^2,0),\qquad C(u)=(0,\mu u^2,u),\qquad A=\{P(t)\}\cup\{C(u)\}.

Size. P(t)=C(u)P(t)=C(u) forces t=0t=0 and u=0u=0, so the two parabolas share only the origin P(0)=C(0)=0P(0)=C(0)=0 and ∣A∣=2q−1|A|=2q-1.

A pair lemma. For s,r∈Fqs,r\in\mathbb F_q the conditions t+t′=st+t'=s and t2+t′2=rt^2+t'^2=r determine the unordered pair {t,t′}\{t,t'\}: they give 2tt′=s2−r2tt'=s^2-r, so tt and t′t' are the two roots of X2−sX+(s2−r)/2X^2-sX+(s^2-r)/2, and 22 is invertible.

Strong Sidon. Every pair of elements of AA, doubles included, has a sum of one of the forms

P(t)+P(t′)=(t+t′, t2+t′2, 0),C(u)+C(u′)=(0, μ(u2+u′2), u+u′),P(t)+C(u)=(t, t2+μu2, u),P(t)+P(t')=(t+t',\,t^2+t'^2,\,0),\quad C(u)+C(u')=(0,\,\mu(u^2+u'^2),\,u+u'),\quad P(t)+C(u)=(t,\,t^2+\mu u^2,\,u),

where a pair containing the origin may be read in more than one form. Let σ=(σ1,σ2,σ3)\sigma=(\sigma_1,\sigma_2,\sigma_3) be such a sum. We show that σ\sigma determines the pair.

  1. If σ1≠0\sigma_1\ne0 and σ3≠0\sigma_3\ne0, the pair is not inside PP (third coordinate 00) and not inside CC (first coordinate 00). So it is {P(t),C(u)}\{P(t),C(u)\} with t=σ1t=\sigma_1 and u=σ3u=\sigma_3.
  2. If σ3=0\sigma_3=0, a pair meeting C∖{0}C\setminus\{0\} in some C(u)C(u) must be {C(u),C(−u)}\{C(u),C(-u)\}, since P(t)+C(u)P(t)+C(u) has third coordinate u≠0u\ne0. Otherwise the pair lies in PP and the pair lemma recovers it from (σ1,σ2)(\sigma_1,\sigma_2). A pair {C(u),C(−u)}\{C(u),C(-u)\} is recovered from σ2=2μu2\sigma_2=2\mu u^2. The two cases cannot share a sum: that would need t+t′=0t+t'=0 and 2t2=2μu22t^2=2\mu u^2 with u≠0u\ne0, so μ=(t/u)2\mu=(t/u)^2 would be a square.
  3. If σ1=0\sigma_1=0 and σ3≠0\sigma_3\ne0, a pair meeting P∖{0}P\setminus\{0\} in some P(t)P(t) would have first coordinate t≠0t\ne0 (with a point of CC) or third coordinate 00 (with a point of PP). So the pair lies in CC, and the pair lemma applied to u+u′=σ3u+u'=\sigma_3, u2+u′2=σ2/μu^2+u'^2=\sigma_2/\mu recovers it.

The cases are exhaustive and disjoint, so distinct pairs have distinct sums.

Maximality. Let x=(x1,x2,x3)∉Ax=(x_1,x_2,x_3)\notin A. We give a,b,c∈Aa,b,c\in A with c∉{a,b}c\notin\{a,b\} and x=a+b−cx=a+b-c.

  1. x1≠0x_1\ne0. Put σ=(x2−μx32)/x1\sigma=(x_2-\mu x_3^2)/x_1, t=(σ+x1)/2t=(\sigma+x_1)/2 and w=(σ−x1)/2w=(\sigma-x_1)/2, and take a=C(x3)a=C(x_3), b=P(t)b=P(t), c=P(w)c=P(w). Then a+b−c=(t−w, μx32+(t−w)(t+w), x3)=xa+b-c=(t-w,\ \mu x_3^2+(t-w)(t+w),\ x_3)=x. Here c≠bc\ne b because t−w=x1≠0t-w=x_1\ne0, and c=ac=a would force w=x3=0w=x_3=0 and x=P(t)∈Ax=P(t)\in A.
  2. x1=0x_1=0 and x3≠0x_3\ne0. Put σ=x2/(μx3)\sigma=x_2/(\mu x_3), u=(σ+x3)/2u=(\sigma+x_3)/2 and v=(σ−x3)/2v=(\sigma-x_3)/2, and take a=P(0)a=P(0), b=C(u)b=C(u), c=C(v)c=C(v). Then a+b−c=(0, μ(u−v)(u+v), u−v)=xa+b-c=(0,\ \mu(u-v)(u+v),\ u-v)=x. Here c≠bc\ne b because u≠vu\ne v, and c=a=0c=a=0 would force v=0v=0 and x=C(u)∈Ax=C(u)\in A.
  3. x1=x3=0x_1=x_3=0, so x2≠0x_2\ne0. Put t′=−x2/2t'=-x_2/2 and take a=P(1)a=P(1), b=P(t′)b=P(t'), c=P(1+t′)c=P(1+t'). Then a+b−c=(0, −2t′, 0)=xa+b-c=(0,\ -2t',\ 0)=x. Here c=ac=a would need t′=0t'=0, and c=bc=b would need 1=01=0.

In each case a+b=x+ca+b=x+c, and the pairs {a,b}\{a,b\} and {x,c}\{x,c\} differ because x∉Ax\notin A. Hence A∪{x}A\cup\{x\} is not strong Sidon, and AA is maximal. ■\blacksquare

Remarks. The midpoint condition 2x∈A+A2x\in A+A is never needed. The proof uses only that 22 is invertible and that μ\mu is not a square, so the same set works over every finite field of odd characteristic, and through Fq3≅(Z/p)3k\mathbb F_q^3\cong(\mathbb Z/p)^{3k} it gives maximal Sidon sets of size 2pk−12p^k-1 in elementary abelian pp-groups of rank 3k3k.