In a previous series of posts, we discussed the Elekes-Sharir(-Guth-Katz) framework. By using this framework, Guth and Katz derived the bound for the Erdős distinct distances problem (where denotes the number of distinct distances that are determined by pairs of points from , and ). The best known upper bound is , which leaves a gap of . The common belief is that , and thus, that there should exist a way to strengthen the analysis of the lower bound.
In this post we briefly discuss the existence of point sets that determine distinct distances, for which the Elekes-Sharir framework can only yield a bound of . Such configurations imply that to improve the Guth-Katz lower bound while by relying on the Elekes-Sharir framework, one either has to make a drastic change in the framework, or to separately handle the family of “problematic” point configurations.
This observation is one of the results in a recent paper by Javier Cilleruelo, Micha Sharir, and myself. Embarrassingly, I did not notice that a similar result was added to the third version of Guth and Katz’s paper. The aforementioned paper still contains other interesting results.
Let us first recall the lower bound construction of Erdős. This construction relies on the following theorem from number theory.
Theorem 1. (Landau-Ramanujan) The number of positive integers smaller than that are the sum of two squares is .
Consider a square of the integer lattice , as depicted in the above figure. Every distance between two points of is of the form , where . Thus, Theorem 1 immediately implies that .
Next, we recall that the Elekes-Sharir framework is centered around double-counting the cardinality of the set
Deriving the value of can be reduced to a problem about finding the number of pairs of intersecting lines in in a given set of lines (e.g., see this post). Guth and Katz relied on this reduction to obtain the bound (which holds for any set of points in , and not just in our case of a integer lattice).
We denote the set of (nonzero) distinct distances that are determined by as . Also, for , we set
Notice that since every ordered pair of is contained in a unique set , except for the pairs of the form . By applying the Cauchy-Schwarz inequality, we have
Combining this with Guth and Katz’s upper bound of implies the bound which is not tight, even though the reduction to lines in preserves the correct value of . When considering this, one thought that comes to mind is that perhaps, for the specific configuration , the upper bound of Guth and Katz is not asymptotically tight, and there is a smaller number of pairwise intersecting lines. This is not the case, and in fact the gap in the bound is a result of applying of the Cauchy-Schwarz inequality.
To see this, let us consider again. Since we know that , we have . However, before applying the Cauchy-Schwarz inequality, we have the expression , whose value is only . That is, by applying the Cauchy-Schwarz inequality, we lose a factor of . The actual value of matches Guth and Katz’s upper bound, implying that their bound is tight in this case. To derive this value of we rely on another theorem by Ramanujan.
Theorem 2. (Ramanujan) For a positive integer , we set . Then .
A simple use of Theorem 2 implies the bound (see the paper for more details). Therefore, to improve Guth and Katz’s upper bound by relying on the Elekes-Sharir framework, one either has to replace the use of the Cauchy-Schwarz inequality, or to characterize the set of problematic graphs and treat them separately.