Wiki
Wiki

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

Updated


Source. Theorem 1, p. 248, of Paul Erdős and George Purdy, Some extremal problems in geometry, J. Combinatorial Theory 10 (1971), no. 3, 246--252, DOI 10.1016/0097-3165(71)90028-8, as identified on the source card.

Setting

The notation is that of Section 2 (p. 247). Points are distinct points of kk-dimensional Euclidean space EkE_k, and Δ>0\Delta>0. For n≥r+1n\ge r+1 and k≥rk\ge r, gk(r)(n;X1,…,Xn;Δ)g_k^{(r)}(n;X_1,\ldots,X_n;\Delta) is the number of rr-dimensional simplices Xi0⋯XirX_{i_0}\cdots X_{i_r} of volume Δ\Delta; its maximum over Δ\Delta is gk(r)(n;X1,…,Xn)g_k^{(r)}(n;X_1,\ldots,X_n), and its maximum over the points is gk(r)(n)g_k^{(r)}(n). For a fixed point X0X_0, n≥rn\ge r and k≥rk\ge r, Gk(r)(n;X0,…,Xn;Δ)G_k^{(r)}(n;X_0,\ldots,X_n;\Delta) counts only the simplices X0Xi1⋯XirX_0X_{i_1}\cdots X_{i_r} with vertex X0X_0, and Gk(r)(n)G_k^{(r)}(n) is defined from it by the same two maxima. The paper notes (p. 247) that gk(r)(n)≤nGk(r)(n−1)≤nGk(r)(n)g_k^{(r)}(n)\le nG_k^{(r)}(n-1)\le nG_k^{(r)}(n).

For r=k=2r=k=2: g2(2)(n)g_2^{(2)}(n) is the largest number of triangles of one common positive area spanned by nn points of the plane, and G2(2)(n)G_2^{(2)}(n) is the largest number of such triangles X0XiXjX_0X_iX_j through a fixed further point X0X_0.

Statement

Theorem 1 (p. 248). For every nn,

G2(2)(n)≤4n3/2,and thereforeg2(2)(n)≤4n5/2.G_2^{(2)}(n)\le 4n^{3/2}, \qquad\text{and therefore}\qquad g_2^{(2)}(n)\le 4n^{5/2}.

The second bound follows from the first by the inequality g2(2)(n)≤nG2(2)(n)g_2^{(2)}(n)\le nG_2^{(2)}(n) above. The constant 44 is explicit.

Proof pointer

Pages 248--249, a minimal-counterexample argument written here in outline. Take the least nn with G2(2)(n)>4n3/2G_2^{(2)}(n)>4n^{3/2} (then n≥4n\ge4), and the graph on X1,…,XnX_1,\ldots,X_n joining Xi,XjX_i,X_j when X0XiXjX_0X_iX_j has area Δ\Delta. Minimality forces every vertex to have degree at least ⌊4n⌋\lfloor4\sqrt n\rfloor, since deleting a vertex of smaller degree would leave more than 4(n−1)3/24(n-1)^{3/2} triangles. The neighbours of XiX_i lie on the two lines parallel to X0XiX_0X_i at the distance fixed by Δ\Delta, so one of them, SiS_i, holds at least half of them. Counting the points on ⌊n⌋\lfloor\sqrt n\rfloor such lines by inclusion and exclusion, two lines meeting in at most one point, gives more than nn points for n≥4n\ge4, a contradiction.

Dependencies

None outside the paper. Read depth: claims checked; the statement and the notation of Section 2 were read clause by clause on pp. 247--248, the proof on pp. 248--249 for its structure only.

Bears on

  • Problem 1086: the bound g2(2)(n)≤4n5/2g_2^{(2)}(n)\le4n^{5/2} is an upper bound for that problem's g(n)g(n), read as counting triangles of one common positive area, which is how the paper counts them. Later papers improved it (see the problem page's references); this page records only the 1971 bound.