MathLabs

Worked solution: The planar case: every Kakeya set in $\mathbb{R}^2$ has dimension 2 (Davies, 1971)

Step 3 of 4: Córdoba's L2L^2 trick: counting how tubes overlap
In plain words

Two δ×1\delta\times 1 tubes crossing at a shallow angle θ\theta overlap in a thin parallelogram of area about δ2/θ\delta^2/\theta — the more nearly parallel they are, the longer they run side by side. Adding up the characteristic functions of all NN tubes into a single 'multiplicity function' ff, the integral ∫f2\int f^2 exactly counts, with multiplicity, all pairwise overlaps ∣Ti∩Tj∣|T_i\cap T_j|. Because the NN directions are spread δ\delta-evenly across a range of angles of size ∼1\sim 1, summing δ2/∣θi−θj∣\delta^2/|\theta_i-\theta_j| over all pairs produces a harmonic-series divergence — a single factor of log⁡N\log N — rather than anything worse.

∣Ti∩Tj∣≲min⁡ ⁣(δ, δ2∣θi−θj∣),f=∑i=1NχTi|T_i \cap T_j| \lesssim \min\!\Big(\delta,\ \frac{\delta^2}{|\theta_i-\theta_j|}\Big),\qquad f = \sum_{i=1}^N \chi_{T_i}
Detailed analysis

Order the tubes so θi≈iδ\theta_i \approx i\delta for i=1,…,Ni=1,\dots,N. Fix TiT_i: for each k=1,…,Nk=1,\dots,N there are O(1)O(1) tubes TjT_j with ∣θi−θj∣≈kδ|\theta_i-\theta_j|\approx k\delta, each contributing overlap ≲δ2/(kδ)=δ/k\lesssim \delta^2/(k\delta)=\delta/k (capping at ∣Ti∩Ti∣=δ|T_i\cap T_i|=\delta for k=0k=0). So ∑j∣Ti∩Tj∣≲δ∑k=1N1k∼δlog⁡N\sum_j |T_i\cap T_j| \lesssim \delta\sum_{k=1}^{N}\frac{1}{k} \sim \delta\log N. Summing over all NN choices of ii, ∫f2=∑i,j∣Ti∩Tj∣≲Nδlog⁡N∼log⁡N\int f^2 = \sum_{i,j}|T_i\cap T_j| \lesssim N\delta\log N \sim \log N (using Nδ∼1N\delta\sim 1). Restoring the more standard length-NN, radius-11 tube normalization used in Guth's lecture notes (equivalent by rescaling) gives the frequently quoted form ∫f2≲N2log⁡N\int f^2 \lesssim N^2\log N.