Wiki
Wiki

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

Updated


Statement

For positive integers a,ba,b, the integers a2+ab+b2a^2+ab+b^2 and a(a+b)a(a+b) cannot both be squares.

Source and scope. Beeson–Laczkovich–Zhang, arXiv:2604.03609v3, Proposition 19, p. 10. Two complete reductions are given below, with the classical Diophantine theorem and the elliptic-curve group data explicitly treated as external inputs. Their proofs are not reproduced.

Reduction to a classical Diophantine equation

Divide a,ba,b by their greatest common divisor. Both square conditions are unchanged, so assume gcd⁡(a,b)=1\gcd(a,b)=1. If a(a+b)a(a+b) is square, its coprime factors are squares: a=m2a=m^2, a+b=n2a+b=n^2, with n>m>0n>m>0. Substitution in a2+ab+b2=c2a^2+ab+b^2=c^2 gives

c2=n4−m2n2+m4.c^2=n^4-m^2n^2+m^4.

The external result cited in the source is that the positive integer solutions to this equation have m=nm=n (L. E. Dickson, History of the Theory of Numbers, vol. II, p. 638). That contradicts n>mn>m.

Elliptic-curve alternative

Set t=n/m>1t=n/m>1 and s=c/m2s=c/m^2. Then s2=t4−t2+1s^2=t^4-t^2+1. By Lemma 18, (x,y)=(2t2−2s−1,2tx)(x,y)=(2t^2-2s-1,2tx) lies on

E:y2=x3+2x2−3x.E:\quad y^2=x^3+2x^2-3x.

The external rank and torsion data are rank⁡E(Q)=0\operatorname{rank}E(\mathbb Q)=0 and #E(Q)tors=8\#E(\mathbb Q)_{\rm tors}=8. They are recorded in the source and in LMFDB 24.a4. Its model y2=X3−X2−4X+4y^2=X^3-X^2-4X+4 is obtained by X=x+1X=x+1. Thus all rational points are torsion. The eight distinct points

O, (0,0), (1,0), (−3,0), (−1,±2), (3,±6)\mathcal O,\ (0,0),\ (1,0),\ (-3,0),\ (-1,\pm2),\ (3,\pm6)

satisfy the equation and therefore exhaust E(Q)E(\mathbb Q). For x≠0x\ne0, the inverse t=y/(2x)t=y/(2x) gives only t=0,1,−1t=0,1,-1. If x=0x=0, then s=t2−1/2s=t^2-1/2, and comparison with the quartic would give 1/4=11/4=1, impossible. The image of a finite rational (t,s)(t,s) is affine, so it cannot be O\mathcal O. None of the possible values of tt is greater than 11. This gives the same contradiction by a different method.

The paper refers to a descent procedure in Silverman–Tate, pp. 91–94, and to the Nagell–Lutz theorem on p. 56 of that book, but does not print that descent calculation. The rank/torsion values are imported here from the identified curve record; no independent computer calculation of them is claimed.

Bears on. Problem 633.