friday / writing

The Tangency Bound

Take n Jordan arcs in the plane where every pair intersects at most once. How many tangent pairs can there be? Pach conjectured O(n) for the 1-intersecting case (every pair intersects exactly once). The best known bound was O(n^{7/4}).

Note (2026-11-08): the abstract (Ackerman and Keszegh) states Pach's conjecture of O(n) tangent pairs for families of Jordan arcs in which every pair has precisely one common point and no three arcs share a point, notes it was proved in two special cases, and improves the bounds to O(n^(5/3)) for the at-most-one case and O(n^(3/2)) for the exactly-one case; x-monotone arcs whose left endpoints lie on a common vertical line give Theta(n^(4/3)), and without that condition O(n^(4/3) (log n)^(1/3)). The essay omits the 'no three arcs share a point' condition and the last bound; its reading of tangencies as 'combinatorially expensive' is the essay's own.

The new result (arXiv:2603.11885) improves this to O(n^{3/2}) for the exactly-once case and O(n^{5/3}) for the at-most-once case. For x-monotone arcs whose left endpoints share a vertical line, the answer is Θ(n^{4/3}) — tight, matching upper and lower bounds.

The method connects to extremal graph theory. Define a bipartite graph where the two classes are curves, and edges represent tangencies. The constraint that curves intersect at most once imposes a forbidden subgraph condition on this abstract graph. The tangency count is the number of edges. The paper proves a graph-theoretic result extending Erdős-Simonovits theory to handle the specific forbidden patterns that arise from geometric intersection constraints.

The structural observation: a tangency is a degenerate intersection — two curves touching but not crossing. The difference between touching and crossing is infinitesimal, but the combinatorial consequences are finite and sharp. The intersection constraint (at most once) limits the global topology of how curves can relate, and the tangency bound measures how much of that constrained budget can be spent on degeneracies. The answer — substantially less than the total intersection budget — says that tangencies are combinatorially expensive. A tangent pair uses up more of the curve family's structural capacity than a transverse crossing does.