3D Kakeya Conjecture Solved | Generated by AI
Question: What problem did Hong Wang (Fields Medal 2026) solve?
Answer: She (with Joshua Zahl) proved the 3D Kakeya set conjecture in Euclidean space — a ~50-year-old problem in harmonic analysis / geometric measure theory. It resolves how small a set can be while still containing a unit length line segment pointing in every possible direction.
The actual math:
A Kakeya set in ℝⁿ is a compact set that contains a unit line segment in every direction. The classical surprise (Besicovitch, 1920s) is that such sets can have Lebesgue measure zero — you can rotate a needle 360° while sweeping through arbitrarily small area (the “Kakeya needle problem” in 2D, via the Perron tree construction).
But measure-zero ≠ small in every sense. The Kakeya conjecture asks about Minkowski/Hausdorff dimension:
Does every Kakeya set in ℝⁿ have full dimension n?
Status before this:
- n=2: proved by Davies (1971) — dimension 2, settled.
- n≥3: open. Best known lower bounds for n=3 stalled around dimension 2.5 for decades (Wolff’s bound via bush/hairbrush arguments, then incremental improvements by Katz-Tao, etc.)
Wang-Zahl’s 127-page proof (posted Feb 2025) shows the 3D case has dimension exactly 3 — i.e., full-dimensional, matching the conjecture.
Why it’s hard, in one line: you need to show that a union of tubes pointing in “spread out” directions can’t concentrate its mass too much — this requires controlling how tubes can overlap (multiplicity/incidence bounds) combined with polynomial method / algebraic geometry over ℝ techniques (following Guth-Katz’s approach for the related finite-field/Erdős distinct distances problems), plus induction on scales and careful decomposition of the tube configuration by “planiness” (how close a tube-bundle is to lying in a hyperplane).
Why it matters beyond pure math: Kakeya-type estimates feed directly into:
- Restriction/extension conjectures for the Fourier transform
- Bochner-Riesz conjecture
- Local smoothing for the wave equation
- Bounds on eigenfunction concentration (spectral theory)
So it’s not an isolated curiosity — it’s a load-bearing conjecture that other analysis results depend on.
Side note that’ll amuse you: Wang has a genuine co-first-authored NeurIPS 2019 paper — so she crossed into ML before going full pure-math.
References:
Generated by AI. Curating and sharing still takes effort.
If you find it useful, feel free to
donate.
WeChat: @lzwjava
·
X: @lzwjava ·
Say hi 👋