HONG WANG · 2026 FIELDS MEDAL

How small can every direction hide?

A visual, step-by-step route from a rotating needle to the multiscale geometry that proved every Kakeya set in three-dimensional space has full dimension.

Begin with the map Open the proof engine

Audience: undergraduate → researcher
Reading path: 90–130 minutes
Interactive learning: 7 interactive labs · 1 visual bridge · 14 checkpoints · 29 questions

Projected tube field · conceptual, not a proof

THE 3D CONCLUSION

K Kakeya in ℝ³ ⟹ dim_HK = dim_MK = 3

Full dimension does not mean positive volume.

00 · ORIENTATION

One medal. Five connected frontiers.

The official citation recognizes a body of work in harmonic analysis and geometric measure theory—not one theorem alone.

It names planar-wave local smoothing, Fourier restriction, Falconer distance sets, planar Furstenberg sets, and the three-dimensional Kakeya problem. Thin objects, many scales, and the tension between concentration and spreading run through all five. [S1]

CORE STORY OF THIS GUIDE

3D Kakeya

Wang and Joshua Zahl proved that a set containing a unit segment in every direction cannot lose any fractal dimension.

Follow the theorem →

WAVES

Local smoothing

With Larry Guth and Ruixiang Zhang: the sharp cone square-function estimate and the conjectured smoothing range for waves in two space dimensions.

See the cone →

FOURIER

Restriction

From brooms and polynomial walls to a later decoupling–Furstenberg route approaching the conjectural exponent.

Trace the packets →

DISTANCES

Falconer

A good/bad wave-packet decomposition pushed the planar pinned-distance threshold to dimension 5/4.

Measure distances →

INCIDENCES

Furstenberg

With Kevin Ren: the sharp dimension of a planar set that is rich along a fractal family of lines.

Count line-rich dust →

CORE ROUTE · ABOUT 25 MINUTES

Read the question, the scaling ruler, the complete planar proof, and the 3D theorem decoder. The proof-engine chapter can then be read as a map.

RESEARCH-DEPTH ROUTE

Continue through every proof-engine node, the sticky and non-sticky branches, the Fourier bridge, and the primary-source ledger. Research bridges state dependencies without pretending to reproduce full papers.

JUST-IN-TIME PREREQUISITES

You need algebra, logarithms, and single-variable calculus. Indicator functions, Cauchy–Schwarz, asymptotic symbols, sphere nets, measure, and Fourier terms are defined when first used; the searchable glossary collects them.

Intuition Complete derivation Research bridge

01 · THE QUESTION

Needles started it. Sets transformed it.

The origin story asks where a needle can turn. The modern theorem asks how dimension behaves when a set contains a segment in every direction.

1917

Kakeya’s needle problem

Find the smallest-area planar region in which a unit needle can rotate through 180°. This is a motion problem: the needle’s positions must connect continuously.

MODERN

Besicovitch / Kakeya set

∀ω ∈ Sⁿ⁻¹, ∃a_ω : {a_ω + tω : 0≤t≤1} ⊂ K

A compact set contains one unit segment in every direction. No continuous motion between the segments is required. Besicovitch showed such a set can have zero area or volume.

LAB 1 · OVERLAP IS THE ONLY PLACE TO HIDE

Pack the same directional information into less visible area.

Every strip contributes the same “ink.” Moving their centers changes how much of that ink overlaps. The pixel calculation is a finite toy model of the union-volume problem.

Directions 36 Center-offset pattern 55%

summed strip ink: —
union footprint: —
average multiplicity: —

summed ink ≈ union footprint × average multiplicity

Grid estimate only. This one-parameter offset path is deliberately not a monotone optimizer: moving every center toward one point can create a large star-shaped footprint. The theorem must control every arrangement at arbitrarily small δ.

Static takeaway. With total strip ink fixed, a smaller occupied union requires larger average multiplicity. The controls only illustrate this union–overlap identity; they do not optimize a Kakeya arrangement.

02 · BUILD THE RULER

Dimension is a scaling exponent.

Blur a set at resolution δ. Count how many δ-boxes are needed. The rate at which that count grows is the dimension.

Cover at scale δ

Let N_δ(E) be the fewest cubes of side δ needed to cover a bounded set E⊂ℝⁿ.

Read a log–log slope

N_δ(E) ≈ δ^−d

This power law is a scaling model, not an exact identity required of every set. A line has slope 1, a sheet slope 2, and a solid cube slope 3 when log N_δ is plotted against log(1/δ).

Solve for the exponent

dim_ME = lim sup_δ→0 log N_δ(E) / log(1/δ)

This is upper Minkowski dimension. A lim inf gives lower Minkowski dimension.

Translate to blurred volume

|E_δ| ≈ N_δ(E)δⁿ ≈ δ^n−d

In ℝ³, lower Minkowski dimension 3 means this neighborhood volume decays more slowly than every fixed positive power of δ at all sufficiently small scales.

LAB 2 · READ A SLOPE

Model set Resolution δ = 1/16

N_δ(E): —
|E_δ| in ℝ³: —
dimension: —

Static takeaway. If the covering count grows like N_δ(E)≈δ^−d, then d is the log–log slope and |E_δ|≈δ^3−d in ℝ³.

A FINER RULER

Hausdorff dimension allows every covering set to choose its own size.

ℋ^s_δ(E) = inf { Σ_i(diam U_i)^s : E⊂⋃U_i, diam U_i≤δ }

For a unit interval, N pieces of length 1/N have total s-cost N(1/N)^s=N^1−s: it grows for s<1 and shrinks for s>1. The transition is dimension 1. The general definition takes an infimum over unequal covers. Always dim_HE≤lower dim_ME≤upper dim_ME, so a convenient sequence of equal scales alone cannot prove Hausdorff dimension.

THE DISCRETE DICTIONARY

Turn infinitely thin segments into countable tubes.

Directions

#(δ-net on S²) ≍ δ⁻²

The direction sphere is two-dimensional. Angular patches of diameter δ have area ≍δ².

One tube

|T| ≍ 1·δ·δ = δ²

A unit segment blurred by δ is a cylinder of length 1 and radius comparable to δ.

Summed mass

Σ_T|T| ≍ δ⁻²δ² ≍ 1

Before overlap, the total mass is already scale-invariant. The problem is entirely about how often tubes cover the same points.

Desired union

|⋃T| ≥ C_εδ^ε

For every ε>0. The δ^ε loss is weak enough to force dimension 3.

WHY δ^ε IS ENOUGH · COMPLETE

If K contains a unit segment in every direction, choose a δ-separated direction net Ω with |Ω|≍δ⁻².
The δ-neighborhood K_δ contains one δ-tube T_ω around each chosen segment.
The union theorem gives |K_δ|≥|⋃_ωT_ω|≥C_εδ^ε.
But |K_δ|≤C N_δ(K)δ³, because N_δ(K) enlarged cubes cover K_δ.
Therefore N_δ(K)≥c_εδ^−3+ε.
Because this holds at every sufficiently small δ, taking a liminf gives lower Minkowski dimension at least 3−ε. Since ε is arbitrary and K⊂ℝ³, both lower and upper Minkowski dimensions equal 3.

STAGE CHECK 1 · DIMENSION RULER

Why would |K_δ|≳δ^ε for every ε>0 force full Minkowski dimension in ℝ³?

Hausdorff bridge. The robust shaded-tube theorem works for localized pieces and across arbitrary covers; a Frostman selection argument converts a hypothetical Hausdorff-dimension deficit into a shaded tube family that violates the theorem. This localization is the research-level step. [S4]

03 · A PROOF YOU CAN HOLD

The planar argument, line by line.

In two dimensions, Cauchy–Schwarz plus an angle-overlap calculation already gives the δ^ε bound. Seeing exactly why it works exposes what breaks in 3D. [S18]

LAB 3 · ANGLE CONTROLS OVERLAP

Crossing angle θ 30° Relative width δ 0.06

|T∩T′| ≲ min(δ, δ²/θ)

δ²/sin θ: —
capped by one strip: —

For infinite strips the overlap is a parallelogram: base δ/sinθ, height δ. Unit length caps the result by ≍δ.

Static takeaway. For transverse strips, overlap decays like δ²/sinθ until the unit-length cap δ takes over. This angular decay creates the planar harmonic sum.

Build the multiplicity function

F(x)=Σ_j=1^N1_Tⱼ(x), N≍δ⁻¹

The indicator 1_T(x) equals 1 inside T and 0 outside, so F(x) literally counts strips through x. Choose N unit δ-strips with δ-separated angles. Since each has area ≍δ,

∫F = Σ_j|T_j| ≍ Nδ ≍ 1.

Use Cauchy–Schwarz on the union

(∫_UF·1)² ≤ (∫_UF²)(∫_U1²)=|U|∫F²

Here U={F>0}=⋃T_j. Expanding the square turns analysis into pairwise geometry:

F²=Σ_i,j1_Tᵢ1_Tⱼ, ∫F²=Σ_i,j|T_i∩T_j|.

The diagonal i=j contributes Σ|T_i|≍1; the off-diagonal terms are controlled by crossing angles.

Sum overlaps by angular separation

For a fixed i, the k-th closest direction makes angle θ≈kδ. Hence

Σ_j|T_i∩T_j| ≲ δ + Σ_k=1^N δ²/(kδ) = δ(1+Σ_k=1^N1/k).

The harmonic sum is ≍log N. Summing over N≍δ⁻¹ choices of i gives ∫F²≲log(1/δ).

Convert the logarithm into dimension

|U| ≳ 1/log(1/δ).

For every ε>0, log(1/δ)≤C_εδ^−ε when δ is small. Thus |U|≥c_εδ^ε, which directly forces full lower and upper Minkowski dimension 2. Full Hausdorff dimension needs a cover-localized or maximal-function bridge. [S16]

WHY NOT REPEAT THIS IN 3D?

Pairwise angles forget collective geometry.

In ℝ³, directions form a two-dimensional sphere, so there are δ⁻² tubes. Around one fixed direction, the angular shell θ≈kδ contains at most O(k) separated directions, and this growth is sharp for a uniform net—not O(1) as in the planar circle. A pair of δ-tubes has overlap ≲δ³/θ, so one shell contributes O(k)·δ³/(kδ)=O(δ²). Summing ≍δ⁻¹ shells loses a power rather than only a logarithm.

2D shell: O(1)·1/k → Σ1/k | 3D shell: O(k)·1/k → Σ1

Wolff’s hairbrush method studies all tubes meeting one tube at once. Later work revealed planiness, graininess, stickiness, brooms, and planks—collective structures invisible to a bare pairwise sum.

LAB 4 · WHERE THE EXTRA POWER APPEARS

Advance one angular shell at a time.

Fix one tube direction and group all other directions by angular distance θ≈kδ. After factoring out the dimension-dependent tube-volume scale, the pairwise estimate contributes the same angular decay 1/k in both dimensions. What changes is how many directions fit in that shell.

Shell index k 6

2D directions in shell: O(1)
2D shell contribution: 1/6
3D directions in shell: O(k)
3D shell contribution: k · 1/k = 1
sum through shell K: H₆ ≈ 2.45 versus 6

Schematic diagnosis of pairwise failure—not a proof of a 3D theorem. Dots show only order of growth; constants are suppressed. Contributions are normalized by δ in 2D and δ² in 3D. On S¹, a shell has two short arcs and O(1) net directions; on S², its circumference grows like kδ, so it holds O(k).

Static takeaway. A 2D angular shell contains O(1) separated directions, so its normalized cost is O(1/k); a 3D shell contains O(k), canceling that decay and leaving constant cost per shell.

STAGE CHECK 2 · PLANAR PROTOTYPE

What creates the logarithm in the 2D proof?

04 · THE QUANTITATIVE THEOREM

One breakthrough, two precise formulations.

Wang–Zahl first proved the full three-dimensional Kakeya conjecture in 2025. Guth–Wang–Zahl then gave a shorter, reorganized proof in 2026. The two papers reach the same dimension theorem through related—but not literally identical—quantitative statements. [S4] [S5]

ORIGINAL FULL-PROOF FORMULATION · WANG–ZAHL 2025Theorem 1.2 of the 127-page paper

Read the original quantifiers left to right. For every ε>0, there is a number K=K(ε)>1 such that the following holds for every sufficiently small δ>0.

Let 𝕋 be a family of δ-tubes contained in the unit ball in ℝ³. Suppose every rectangular prism of dimensions a×b×2 contains at most

100abδ⁻² tubes from 𝕋.

For every T∈𝕋, let Y(T)⊂T be measurable and satisfy the per-tube fullness condition

|Y(T)| ≥ λ|T|.

Then

|⋃_T∈𝕋Y(T)| ≥ δ^ελ^K ∑_T∈𝕋|T|.

This is the quantitative theorem from which Wang–Zahl deduce that every Kakeya set in ℝ³ has Hausdorff and Minkowski dimension 3. The stronger three-dimensional Kakeya maximal-function conjecture would give this inequality with K=3; the paper explicitly does not prove that conjecture. Verify in the original paper [S4] ↓

Y(T)⊂Tthe active part of tube T

U(𝕋,Y)=⋃Y(T)the active union whose volume is protected

λ(𝕋,Y)=Σ|Y(T)|/Σ|T|average active fraction

Δ_max(𝕋)=sup_{W convex}Σ_T⊂W|T|/|W|worst convex concentration

LATER STREAMLINED FORMULATION · GUTH–WANG–ZAHL 2026Theorem 1.1 of the 47-page paper; output exponent renamed α here

Read the streamlined quantifiers left to right. You choose any tolerated output loss α>0. The theorem chooses a hypothesis allowance η>0. Then, at every sufficiently fine scale δ, two checkable hypotheses force the conclusion. Constants may depend on α, but not on δ or the tube arrangement.

Let 𝕋 be a family of unit δ-tubes in ℝ³. If

Δ_max(𝕋) ≤ δ^−η

and a shading Y fills an average fraction

λ(𝕋,Y) ≥ δ^η,

then its active union obeys

|U(𝕋,Y)| ≥ δ^α #𝕋 · |T|.

Conceptual relation, not literal identity. For equal-volume tubes, Δ_max is the streamlined language for the same kind of worst convex concentration measured by the original paper’s Katz–Tao convex Wolff constant. But the contracts differ: the 2025 theorem assumes a rectangular-prism count and per-tube λ-fullness, retaining λ^K; the 2026 theorem assumes convex-density control and average fullness, with a chosen δ^α loss. Compare the proof versions ↓

#𝕋number of tubes

|T|volume of one tube, ≍δ²

|W|, |U|three-dimensional volume

δ^−ηmild allowed clustering, not a constant bound

Streamlined-form toy scale check. If δ=10⁻³, α=0.1, and #𝕋·|T|≈1, the conclusion protects at least δ^α≈0.50 of the summed mass. This is only arithmetic—the theorem’s η and “sufficiently small” threshold are not numerically explicit.

OBJECT

Shading

Y(T)⊂T

Only the portion Y(T) where a tube is “active” counts. This is essential after localization and refinement: a coarse tube may be present but contribute only on part of itself.

ACTIVE UNION

What is occupied?

U(𝕋,Y)=⋃_T∈𝕋Y(T)

This is the geometric set whose volume the theorem protects.

FULLNESS · STREAMLINED FORM

How much survives?

λ = Σ|Y(T)| / Σ|T|

Average shaded fraction in the 2026 statement. The 2025 theorem above instead assumes |Y(T)|≥λ|T| for every tube.

CONVEX DENSITY

Where can tubes cluster?

Δ(𝕋,W)=Σ_T⊂W|T| / |W|

Test every convex container W, from thin tubes through planks and slabs to ball-like bodies.

WORST CONTAINER

The anti-clustering hypothesis

Δ_max(𝕋)=sup_{W convex}Δ(𝕋,W)

A large value exhibits a cheap explanation for overlap: many tubes were simply packed into one small convex region.

MULTIPLICITY

The dual quantity

μ(𝕋,Y)=Σ|Y(T)| / |U(𝕋,Y)|

Average number of active tubes covering an occupied point. A lower union-volume bound is exactly an upper multiplicity bound.

LAB 5 · FIND THE HIDING PLACE

Compare pointwise overlap with convex concentration.

Tube arrangement

tubes candidate convex hiding place

visual overlap: —
convex-density diagnosis: —
proof response: —

Schematic projection. The theorem optimizes over all 3D convex bodies, not only the drawn ellipse.

Static takeaway. High overlap may come from concentration inside a slab, plank, or hairbrush. The proof finds a convex hiding place and either normalizes it or spends its excess density.

WHY THE CLASSICAL CASE FITS

δ-separated directions satisfy the convex condition.

Replace a convex container W by a comparable John ellipsoid with semiaxes a≤b≤c.
Containing a unit δ-tube forces a,b≳δ and c≳1.
Admissible tube directions occupy an angular patch of area O(ab) on S².
A δ-separated direction set therefore contributes at most O(ab/δ²) tubes.
Multiplying by |T|≍δ² gives Σ_T⊂W|T|≲ab≲abc≍|W|, hence Δ_max≲1.

This standard geometric counting is why the robust convex theorem implies the ordinary Kakeya estimate. Boundary-scale constants are suppressed; they do not change the δ-exponent.

CLOSE THE LOOP · THEOREM TO DIMENSION

The robust tube bound returns to the set we started with.

1 · SAMPLE
At scale δ, choose one tube from the Kakeya set for each direction in a δ-separated net, so #𝕋≍δ⁻² and #𝕋·|T|≍1.
2 · VERIFY
Direction separation gives bounded convex density—and, in the original formulation, the required rectangular-prism tube count.
3 · APPLY
The streamlined theorem gives |U|≳δ^α; the original theorem gives |U|≳δ^ελ^K, with λ=1 for whole tubes.
4 · READ THE SLOPE
Since |K_δ|≲N_δ(K)δ³, these bounds force N_δ(K)≳δ^−3+γ for every γ>0, hence Minkowski dimension 3.

Hausdorff handoff. The equal-scale calculation proves the Minkowski conclusion. Localized shadings and a Frostman/covering selection keep the estimate alive inside variable-size covers, supplying the research-level bridge to Hausdorff dimension 3.

Optional source orientation · which paper is which?

WHICH PAPER IS WHICH?

Keep the award, original proof, and later streamlined proof separate.

AWARD SCOPE

No single “Fields paper”

A Fields Medal recognizes a body of work. The official citation highlights several contributions, including the proof of the Kakeya conjecture in dimension three; it does not designate one paper as an official “Fields paper.” Official citation [S1]

ORIGINAL · 2025

Wang–Zahl

Volume estimates for unions of convex sets, and the Kakeya set conjecture in three dimensions, 127 pages with 14 numbered figures, is the original full proof. Primary source [S4]

STREAMLINED · 2026

Guth–Wang–Zahl

A streamlined proof of the Kakeya set conjecture in ℝ³, 47 pages with no numbered figures, is a later reorganization of the same main result. Primary source [S5]

STAGE CHECK 3 · PAPER IDENTITY

Which attribution keeps the chronology, authors, and award scope straight?

THE THEOREM DOES / DOES NOT

DOES

settle Hausdorff and Minkowski dimension in ℝ³;
allow zero-volume examples;
control shaded and non-direction-separated families under convex nonconcentration;
credit Wang–Zahl for the full theorem.

DOES NOT

settle dimensions n≥4;
prove positive volume;
settle the 3D Kakeya maximal-function conjecture;
automatically settle Fourier restriction.

STAGE CHECK 4 · 3D THEOREM DECODER

What connects the robust convex theorem back to ordinary direction-separated Kakeya tubes?

05 · THE PROOF ENGINE · ROUTE MAP

One shared obstruction, then two complete logical routes.

Both proofs control active tube overlap by repeatedly localizing, measuring concentration, and testing whether a difficult family has become sticky. We first follow the dependency chain of Wang–Zahl’s original 2025 proof. An explicit divider then restarts the argument in the shorter Guth–Wang–Zahl 2026 organization.

SHARED LOGICAL PRESSURE

A near-extremal family must survive every attempted local gain. Repeated survival forces coherent multiscale structure; sticky Kakeya rules out a positive volume-loss exponent on that structure.

near extremizer ⟶ factor or gain ⟶ all-scale structure ⟶ sticky contradiction

Reference comparison · open the detailed 2025 / 2026 crosswalk

PROOF-ROUTE CROSSWALK

A shared trunk, two bookkeeping systems, one dimension theorem.

Read downward. Both proofs control active tube overlap and use sticky Kakeya as a rigid input. Their named assertions and improvement mechanisms are different.

SHARED GEOMETRIC TRUNK

1ShadingsKeep the active part Y(T) after localization.
2Convex anti-clusteringTest whether tube mass hides in anisotropic containers.
3Multiplicity + factoringSplit one hard overlap count into controlled inner and outer systems.
4Sticky inputMultiscale self-similarity cannot sustain a dimension deficit.

ORIGINAL FULL PROOF · 2025

Wang–Zahl: assertions D and E

Primary source [S4]

01
Start with the hairbrush.
A Wolff-style hairbrush argument proves D(1/2,0), recorded as Proposition 1.8.
02
Pass between two estimates.
Proposition 1.6 proves D(σ,ω) ⇔ E(σ,ω). D is the normalized estimate under mild Katz–Tao convex and Frostman slab errors; E retains the explicit errors m=C_KT-CW and ℓ=C_F-SW. D and E name assertions, not numeric values, and smaller σ or ω gives a stronger bound.
03
Build and repair a grains decomposition.
Polynomial partitioning enters Guth’s grains input. Proposition 7.5 repairs the two-scale geometry; Proposition 9.1 then inserts a third, internal scale and estimates fine, medium, and coarse multiplicities in that order.
04
Improve the scale loss.
Proposition 1.7 uses E(σ,ω) to obtain D(σ,ω−g(σ,ω)) for a definite g>0.
05
Trade the gain and close the parameter set.
A tube-count bound turns an ω-improvement into a left-neighbor σ-improvement. Upward monotonicity and ε-slack closedness then prove D(0,0) and E(0,0), as stated in Theorem 1.9.
06
Return to Kakeya sets.
Corollary 1.10 gives the convex-Wolff union estimate; Theorem 1.2 is its prism-controlled case, and the covering argument gives dimension 3.

LATER STREAMLINED PROOF · 2026

Guth–Wang–Zahl: K_KT and K_F

Primary source [S5]

01
Use the trivial starting exponent.
The pointwise bound μ≤#𝕋 gives K_KT(1).
02
Supply the relative-density input.
Main Lemma 1 turns K_KT(β) into K_F(β) at the same positive β.
03
Lower the exponent.
Main Lemma 2 combines K_KT(β) and K_F(β) to obtain K_KT(β−ν(β)).
04
Repeat with slack.
Iteration reaches K_KT(b) for an arbitrarily small target b>0; no β=0 endpoint is asserted.
05
Absorb the remaining tube count.
Convex-density cardinality control turns (#𝕋)^b into an arbitrarily small δ-power loss.
06
Return to Kakeya sets.
The streamlined Theorem 1.1 gives the robust union estimate, which feeds the Hausdorff and Minkowski dimension deductions.

ORIGINAL 2025Every Kakeya set in ℝ³ has dimension 3STREAMLINED 2026

ORIGINAL FULL PROOF · WANG–ZAHL 2025

Follow two loss parameters from the hairbrush seed to zero.

Now keep only the 2025 lane open. Theorem 1.2 is the destination; the original proof reaches it by constructing two induction-stable estimates and forcing their loss parameters to zero. Each module below states why it is needed, what enters, what the proof does, what exits, and the next unresolved question. [S4]

Reference ledger · track four invariants across the six original-proof stages

PROOF-STATE LEDGER

What changes—and what must survive—at every stage?

First pass: read the STAGE column and one row at a time. Symbols in later rows are previews; each is defined before it is used in the matching module below.

Six-stage ledger tracking active mass or shading, concentration controls, scale structure, and loss parameters through the original Wang–Zahl proof.
STAGE	ACTIVE MASS / SHADING	CONCENTRATION CONTROLS	SCALE STRUCTURE	LOSS PARAMETERS
1D / E contracts	A λ-dense shaded family (𝕋,Y)_δ; U=⋃Y(T).	Measure m=C_KT-CW and ℓ=C_F-SW; D assumes them mild, E prices them.	One fine scale δ; no parent scale has been chosen.	Seed D(1/2,0); track overlap loss σ and scale loss ω.
2Factor safely	Keep a substantial refinement and transport its surviving shading.	Construct outer Katz–Tao control and normalized inner Frostman control.	Convex containers outside; normalized fibers inside each container.	D⇔E at the same (σ,ω); E carries the explicit m,ℓ cost.
3Diagnose overlap	Regularize the active overlap so μ≈μ_fineμ_coarse.	Test the rescaled coarse Katz–Tao constant: low applies D; high must yield density or structure.	Fine δ-tubes inside ρ-parents; two anisotropic zooms reveal grains.	Seek a definite α-gain over the nearly sharp σ+ω benchmark.
4Repair grains	Pass to a δ^ζ-dense refined family (𝕋₁,Y₁).	Repair fine, parent, and local-grain controls; factor the parent cover.	Build ρ first; create θ only in the high-coarse-KT almost-tube branch.	Either gain a definite power now or keep ε₂, ε₃, and ζ small enough for the next induction.
5Force sticky	Iterate refinements; controlled rigid copies create enough ambient mass for NSP to select a Frostman all-scale subfamily.	Katz–Tao control persists on the copies; NSP separately selects a subfamily satisfying the Frostman hypothesis used by the sticky theorem.	A ladder δ=ρ_J<⋯<ρ₀=1 approximates every requested scale.	Compose rung losses, then obtain g(σ,ω)>0: E(σ,ω)⇒D(σ,ω−g).
6Close the endpoint	Return to arbitrary average λ-density; the theorem’s per-tube shading condition implies it.	E(0,0) retains m⁻¹; the prism hypothesis supplies m≤1000.	No new zoom: close σ for each fixed ω, then let ω↓0.	Trade g for a leftward σ-step and spend ε-slack to reach (0,0).

How to use this ledger. Rows 1–5 record the state handed to the next module; row 6 records the endpoint handed back to the theorem. A column lets you audit one invariant across the proof. It is a reading aid synthesized from the 2025 proof, not an additional lemma. [S4]

Reading arXiv:2502.17655 alongside the guide? Open the 12-part page map.

READ ALONGSIDE THE 127-PAGE PAPER

Where each paper block enters this guide.

The paper develops its tools in publication order. The website follows a pedagogical dependency order, not a page-by-page transcription: use these links to move from a paper block to the guide module that explains its role.

PP. 3–15
§1 · Statements, philosophy, vignetteGuide → Module 1 contracts
PP. 15–23
§2 · Proof sketchGuide → six-stage ledger
PP. 23–25
§3 · NotationGuide → Module 1 notation
PP. 25–36
§4 · Convex factoringGuide → Module 2 factoring
PP. 36–60
§§5–6 · Prisms + D/E equivalenceGuide → Module 2 equivalence
PP. 60–100
§§7–8 · Grains + three movesGuide → Module 4 repair
PP. 100–103
§9 · Refined inductionGuide → Module 4 induction
PP. 103–110
§10 · NSP + stickyGuide → Module 5 amplifier
PP. 110–114
§11 · Multiscale analysis and proof of Proposition 1.7Guide → Module 5 gain
PP. 114–115
§12 · Tube-doubling consequencesGuide → Module 6 coda
PP. 115–125
Appendix A · Grains decompositionGuide → Module 4 input
PP. 125–127
Appendix B · HairbrushGuide → independent seed

How to switch texts. Page ranges follow the original paper’s table of contents. Read a guide module for the causal role, then open its paper block for the proof. Appendices A and B deepen ingredients that the main chain already uses: grains in Module 4 and the hairbrush seed before Module 1. [S4, contents, pp. 1–3]

MODULE 1 · THE TWO CONTRACTS

D is normalized; E carries the price of clustering.

WHY

Zooming and localization can destroy a global nonclustering bound, so one estimate is not flexible enough for induction.

INPUT

A shaded family of essentially distinct δ-tubes and two measured concentration errors.

MOVE

Encode normalized families by D and arbitrary clustering penalties by E.

OUTPUT

Two precise volume contracts whose exponents can be improved.

N=#𝕋number of tubes

v=|T|≍δ²one tube’s volume

Q=Nv^1/2normalized tube-count scale

U=⋃Y(T)active union

Average density used by D and E. The pair (𝕋,Y)_δ is λ-dense when

Σ_T∈𝕋|Y(T)| ≥ λ Σ_T∈𝕋|T|.

This is weaker than Theorem 1.2’s per-tube requirement |Y(T)|≥λ|T|. The distinction matters because refinements may redistribute the surviving shading.

ABSOLUTE CONVEX CONCENTRATION

m=C_KT-CW(𝕋)

#𝕋[W] ≤ m |W| v⁻¹ for every convex W

A large m says many tubes can hide in a small convex body.

RELATIVE SLAB CONCENTRATION

ℓ=C_F-SW(𝕋)

#𝕋[W] ≤ ℓ |W| N for every slab W

A small ℓ says no slab captures much more than its volume fraction of the family.

NORMALIZED CONTRACT

Assertion D(σ,ω)

For every ε>0 there are κ,η>0 such that, for every δ>0, every δ^η-dense shading with m,ℓ≤δ^−η satisfies

|U| ≥ κ δ^ω+ε Nv · Q^−σ. (1.2)

TRANSPORT CONTRACT

Assertion E(σ,ω)

With the same quantifier order and only the δ^η-density hypothesis,

|U| ≥ κ δ^ω+ε m⁻¹Nv · [m^−3/2ℓQ]^−σ. (1.3)

Advanced · decode E’s three factors

FORMULA DECODER · THREE VISIBLE FACTORS

Why does E contain the factors m⁻¹, m^−3/2, and ℓ?

Recall Q=Nv^1/2. Two model configurations show how D’s normalized mass Nv and count scale Q transform when one changes scale.

E prices the transformed quantities as m⁻¹Nv · [m^−3/2ℓQ]^−σ.

A ρ-parent is a unit tube of thickness ρ containing fine δ-tubes. “Saturated” means that the relevant concentration upper bound is attained up to controlled constants.

MODEL A · SLAB PENALTY

A fine family hidden in one ρ-tube

GIVEN
Assume m≍1 and all N≍(ρ/δ)² fine tubes lie in one ρ-parent. A thickness-ρ slab contains that parent, so N≲ℓρN; saturation gives ℓ≍ρ⁻¹.
RESCALE
Dilate the parent by ρ⁻¹ in its two short directions. Fine tubes now have thickness δ/ρ and volume v_res≍ρ⁻²v; both concentration errors are ≍1.
UNDO THE ZOOM
The inverse map multiplies union volume by ρ², so ρ²Nv_res≍Nv, while Q_res=Nv_res^1/2≍ρ⁻¹Q≍ℓQ.
READ E
Because δ/ρ≥δ, D’s scale factor can be weakened to δ^ω+ε; its normalized bracket Q^−σ returns as [ℓQ]^−σ. This motivates the ℓ factor.

MODEL B · CONVEX PENALTIES

A saturated family of ρ-parents

GIVEN
Assume m≫1 and ℓ≍1, and that the ρ-parents form a balanced partitioning cover saturating the convex bound. The Katz–Tao inequality gives #𝕋[T_ρ]≲m|T_ρ|v⁻¹≍m(ρ/δ)² descendants per parent; saturation means this upper scale is attained.
CHOOSE SCALE
Set ρ=m^1/2δ, with ρ≤1. Saturation then gives m² fine tubes per parent, so N_ρ≍m⁻²N and v_ρ≍mv. Maximal essentially-distinct packing makes the fine union occupy a constant fraction of each saturated parent, which is why the parent union models its volume.
COMPUTE MASS + COUNT
N_ρv_ρ≍m⁻¹Nv, and Q_ρ=N_ρv_ρ^1/2≍m^−3/2Q.
READ E
Since ρ≥δ, D’s ρ^ω+ε factor can be weakened to δ^ω+ε. The parent mass is smaller by m⁻¹, while its count scale is m^−3/2Q. Thus the combined m-dependence is m^−1+3σ/2, nonincreasing for 0≤σ≤2/3.

MODEL, NOT PROOF. These are the motivating extremal configurations on pages 5–7; they explain the shape of E but do not prove D⇒E. Module 2 supplies the factoring and induction argument for that implication. [S4, pp. 5–7]

PARAMETER LANDSCAPE

Every successful move points left or down.

Left means smaller overlap loss σ; down means smaller scale loss ω. The arrows record implications between assertions, not a continuous numerical trajectory.

1 · SEED
D(1/2,0), hence D(1/2,ω) for fixed ω>0.
2 · TRANSPORT
Proposition 1.6 gives D(σ,ω) ⇔ E(σ,ω) at the same parameters.
3 · GAIN DOWN
Proposition 1.7 supplies some g(σ,ω)>0 and gives E(σ,ω) ⇒ D(σ,ω−g).
4 · TRADE LEFT SAFELY
For σ>0 set h=min{g,2σ}. The universal tube count converts the weaker h-gain into D(σ−h/4,ω), which stays inside σ≥0.
5 · CLOSE
Iterate in σ, use relative openness and closedness, then let ω↓0 to obtain D(0,0) and E(0,0).

Logic locator. Definitions 1.3 and 1.5, Propositions 1.6–1.7, and the closure argument on pages 7–8 and 110–114. This is an original implication map; spatial distances are not quantitative.

Read the parameters through the proved implications. The exponent σ is attached to the normalized overlap/count term; ω is the δ-scale loss. Formally, D(σ,ω) implies D(σ′,ω) whenever σ′∈[σ,2/3]. Here is the one-line reason even though Q need not be at least one: apply D(σ,ω) with ε/2 and shrink η if needed; ℓ≤δ^−η and Remark 1.4(C) give N≳δ^−1+η, hence Q=Nv^1/2≳δ^η, so Q^{−(σ′−σ)} costs at most δ^{−η(σ′−σ)} and fits in the remaining ε/2. The left/down map records this assertion-level monotonicity, Proposition 1.7, and the domain-safe tube-count trade. For the benchmark N≍δ⁻², D predicts |U|≳δ^σ+ω+ε.

Exact-source boundary. These are Definition 1.3 and Definition 1.5 on pages 4–5. D and E are assertions with quantifiers, not numerical functions. Keep C_F-SW distinct from the Frostman convex constant C_F-CW used later. [S4]

ORIGINAL ROUTE CHECK · MODULE 1 OUTPUT

Why can the proof not blindly reapply D after an arbitrary localization?

NEXT QUESTIONHow can the proof pass between D and E after rescaling?

MODULE 2 · FACTORING AND D ⇔ E

Turn uncontrolled clustering into controlled containers and fibers.

WHY

D cannot simply be reapplied when a zoom creates large m or ℓ.

INPUT

D(σ,ω), a weaker E(σ,ω′), and a family whose concentration may be large.

MOVE

Retain a substantial refinement, factor through convex containers, and use the condition that survives in each direction.

OUTPUT

The hard implication D⇒E, completing Proposition 1.6.

1 · FIX THE TARGET

Assume D(σ,ω)

D is available whenever a normalized piece has both mild concentration errors.

2 · CARRY A WEAKER SCAFFOLD

Know E(σ,ω′), with ω′≥ω+t

The descent begins at the trivial E(σ,2); this is already known, so using E at the larger exponent is not circular.

3 · IMPROVE ONE PASS

Lower ω′ by a uniform α

Factoring plus the trichotomy gives the auxiliary Ẽ(σ,ω′−α). Proposition 5.14 translates it back to E, ready for the next pass.

Two technical names. Ẽ is the auxiliary improvement form produced by the trichotomy; F is the companion prism form stable under anisotropic rescaling. Proposition 5.14 proves F⇔E⇔Ẽ. Write K_fac for Proposition 4.6’s subpolynomial factoring loss below; it is unrelated to Theorem 1.2’s exponent K(ε).

FACTORING SIEVE

Prune, package, then audit both directions.

The construction does more than inherit old bounds: it first keeps a substantial congruent subfamily, then chooses balanced convex containers whose outer and normalized inner systems satisfy the complementary estimates needed later.

Visible text equivalent. Read the sieve from left to right:

Raw 𝒰: start with congruent convex objects whose concentration may be uncontrolled.
Prune to 𝒰′: retain at least a K_fac⁻¹ fraction.
Package by 𝒲: group the retained objects into balanced convex containers W.
Audit separately: distinguish what follows automatically from what Proposition 4.6 constructs.

ReasonOuter containers 𝒲Normalized inner fibers 𝒰′[W]

Automatic inheritance

Frostman travels upward.

Katz–Tao travels downward.

Proposition 4.6 factoring guarantee

Katz–Tao factors from above.

Frostman factors from below.

Figure locator. Original code-native conceptual redraw after Wang–Zahl 2025, Proposition 4.6 and Figure 5, page 27. It reproduces no paper artwork or traced geometry; the audit labels are the logical content.

1 · RETAIN

Keep a subpolynomial fraction

#𝒰′ ≥ K_fac⁻¹#𝒰

The discarded mass is small enough to fit inside the allowed δ-loss.

2 · PACKAGE

Choose balanced convex containers

#𝒰′[W] ≥ K_fac⁻¹C_KT-CW(𝒰′)|W||A|⁻¹ (4.5)

Here A∈𝒰 and |A| is the common volume of one congruent input object. Large concentration becomes an explicit number of objects per container.

3 · AUDIT TWO REASONS

Do not swap inheritance and construction

Automatic for a balanced cover: C_F-CW(𝒲)≲K_fac C_F-CW(𝒰), C_F-SW(𝒲)≲K_fac C_F-SW(𝒰), and C_KT-CW(𝒰′[W])≤C_KT-CW(𝒰′).

Constructed by Proposition 4.6: C_KT-CW(𝒲)≤K_fac, while each normalized inner fiber satisfies C_F-CW(𝒰′[W])≤K_fac.

Frostman inheritance travels upward and Katz–Tao inheritance downward; the complementary pair is a new guarantee of the factoring sieve.

ALL-SCALE

Frostman convex control at every scale activates the imported sticky theorem.

SEPARATED SCALES

At τ≤δ^ζ₁/5ρ, apply E on the outside and inside and D at the normalized middle scale.

FLAT FACTOR

An a×b×1 prism with a≤δ^ζ₂/100b buys the favorable factor (b/a)^ω.

Formally, Proposition 6.3 supplies this trichotomy after a refinement. Fix 0<σ≤2/3 and ω,t>0. Lemma 6.4 chooses one α=α(σ,ω,t)>0, uniform for every ω′≥ω+t; weaken the improvement if necessary so that α≤t. Then

D(σ,ω) ∧ E(σ,ω′) ⟹ Ẽ(σ,ω′−α).

1 · IMPROVE THE AUXILIARY TARGET
Lemma 6.4 combines D(σ,ω) with a known weaker E(σ,ω′) to improve the auxiliary assertion Ẽ from ω′ to ω′−α.
2 · TRANSLATE THE CONTRACTS
Proposition 5.14 proves F(σ,ω) ⇔ E(σ,ω) ⇔ Ẽ(σ,ω). Here F is the companion prism contract that survives anisotropic rescaling.
3 · DESCEND WITH A BUFFER
Start from the trivial E(σ,2). Because α is uniform while ω′≥ω+t, finitely many passes reach some ω″≤ω+t; weakening gives E(σ,ω+t). Let t↓0 and use ε-slack closedness to reach E(σ,ω). The boundary cases follow from nearby positive parameters by the same slack.
4 · CONCLUDE
This proves D(σ,ω)⇒E(σ,ω). The reverse implication is immediate from the definitions when D’s two concentration errors are mild.

[S4, §§4–6]

ORIGINAL ROUTE CHECK · MODULE 2 OUTPUT

What exact statement can now enter the near-extremizer diagnostic?

NEXT QUESTIONOnce transport is legal, why must a nearly sharp family improve?

MODULE 3 · NEAR-EXTREMIZER DIAGNOSTIC

A hard family must defeat two different ways of winning.

WHY

A raw count of all coarse parents is too large when the family is nonsticky.

INPUT

The introductory model N≍δ⁻², σ,ω>0, and an almost-sharp D-family.

MOVE

Separate overlap within one parent from overlap among parents that are actually active.

OUTPUT

Either a multiplicity saving, a local-density gain, or a structured grains problem.

1 · NEARLY SHARP

Translate volume into overlap

Nv≍1, |U|≈δ^σ+ω ⟹ μ≈δ^−σ−ω

This is the proof-sketch benchmark after suppressing δ^±ε losses.

2 · STICKY BENCHMARK

Count descendants in one ρ-parent

#𝕋[T_ρ]≈(ρ/δ)²

Two angular parameters give the direction-separated, near-maximal branching count.

3 · NONSTICKY DEFICIT

Assume a fixed-power shortage

#𝕋[T_ρ]≈δ^ν(ρ/δ)²

Here ν>0 measures how far this parent falls below the sticky benchmark.

4 · APPLY D INSIDE

Convert the shortage into a saving

μ_fine≲δ^νσ(ρ/δ)^σ+ω

The factor δ^νσ is the gain. Only the active-parent multiplicity μ_coarse remains.

Temporary convention in the pages 9–14 proof sketch. The vignette sets Y(T)=T, regularizes the relevant multiplicities, and suppresses δ^±ε bookkeeping so the two possible gains stay visible. Module 4 restores general shadings, refinements, and every loss needed by the formal induction.

GLOBAL WIN

|⋃T| ≳ δ^σ+ω−α (1.5)

Equivalently, after regularizing multiplicity, prove μ≲δ^{−σ−ω+α}.

LOCAL WIN

|B_τ∩⋃T| ≳ δ^−α(δ/τ)^σ+ω|B_τ| (1.7)

Here B_τ is an intermediate ball of radius τ. Excess density inside a typical such ball also forces the desired global gain.

In the nonsticky model, a ρ-parent has only about δ^ν(ρ/δ)² descendants. Applying D after normalizing that parent gives

μ_fine ≲ δ^νσ(ρ/δ)^σ+ω, μ≈μ_fineμ_coarse, target: μ_coarse≲ρ^−σ−ω. (1.9)–(1.11)

The obstruction is now concentrated in μ_coarse. The introductory two-scale geometry below turns the active coarse overlap into a new tube problem at scale ρ. It is a proof vignette, not yet the formal Section 9 induction; the formal argument later inserts a medium θ-scale.

Dimension convention. Every grain dimension below is written shortest × middle × longest and means comparability up to controlled constants. Here c is the long in-parent grain scale produced by this decomposition—not a universal constant. In formal Proposition 7.5 the grains are a×b×c with δ≤a≤b≤c≤1 and parent scale ρ=b/c.

BEFORE THE TWO ZOOMS

Partition and refine until active overlap admits a grains cover.

This compresses a dependency, not one theorem call: broadness and polynomial partitioning feed refinements; the grains theorem then organizes the surviving active set into plate-like pieces.

1 · BROAD ACTIVE OVERLAP

Keep the broad, regularized portion where several direction classes contribute.

2 · PARTITION + REFINE

Cellular and wall cases force scale choices and refinements before the geometric cover is legal.

3 · GRAINS COVER

The surviving active union is covered by controlled plate-like grains inside a normalized parent.

Visible text equivalent. Broad regularized overlap enters polynomial partitioning; repeated case splits and refinements select an intermediate scale; the two-scale grains theorem outputs a mostly-disjoint cover by δ/ρ×c×c grains. The existing zoom panels begin from that output.

Dependency locator. Conceptual synthesis of Wang–Zahl 2025, Figure 2 on page 11, §2.2 on pages 19–21, and Appendix A on pages 115–124. The three cards are original code-native primitives and do not claim that a single partition directly produces the final cover.

ORIGINAL 2025 ROUTE · GEOMETRIC MICROSCOPE

Two zooms, one multiplicity product.

This original code-native diagram is redrawn conceptually after Wang–Zahl 2025, Figures 2–3, pages 11–12. It explains the proof vignette’s geometry without copying or tracing the paper artwork.

Read the original full proof [S4] ↓

1Normalize one parent

Inside a selected ρ-tube, anisotropic normalization makes each fine δ-tube a δ/ρ-tube. A grains theorem organizes the active union into plate-like pieces.

2Undo the parent zoom

Undoing that normalization changes each grain’s dimensions. The long and medium axes respond differently because the scaling was anisotropic.

3Normalize an intersection box

After intersecting grains whose tangent planes nearly agree are refined into a common box □—without requiring their long axes to be parallel—normalizing □ turns those grains into ρ×ρ×1 prisms, objects comparable to coarse ρ-tubes.

NORMALIZED PARENTgrain: δ/ρ × c × c

Fine tube width: δ/ρ.

UNDO T_ρ NORMALIZATIONgrain: δ × ρc × c

Return to the original coordinates.

NORMALIZE □box □: δ/ρ × c × c

Its grains become ρ × ρ × 1 prisms, comparable to ρ-tubes.

μ ≈ μ_fine · μ_coarse

μ: typical number of fine δ-tubes active at a point
μ_fine: typical active descendants from one ρ-parent through a point
μ_coarse: number of active coarse parents contributing there—not every parent that merely passes nearby

Visible text equivalent. First zoom into one coarse parent to expose fine-scale grains. Undo that zoom to place the grains back in the global arrangement. Then group suitably tangential grains and zoom into their common box, where they become coarse tube-like prisms. Bound overlap inside one parent and overlap among active parents separately, then multiply the two bounds.

The relation μ≈μ_fineμ_coarse is used after refinements make multiplicities nearly constant.
The proof vignette invokes broadness and tangency hypotheses; the drawing does not establish them.
Grains coming from different parents require further refinement before they can be grouped into common boxes.
All displayed dimensions are comparable up to controlled constants, not exact Euclidean equalities.

Figure provenance. A new code-native educational redraw conceptually based on Wang–Zahl 2025, Figures 2–3, pages 11–12. No paper image asset or traced geometry is reproduced; the source paper is available under CC BY 4.0.

CASE 1 · LOW LOCAL m

Apply D to the rescaled grains

If C_KT-CW(𝕋̃) is small and the vignette’s additional rescaled Frostman-slab bound holds, then the normalized grain family has both hypotheses needed to prove μ_coarse≲ρ^−σ−ω.

CASE 2 · HIGH m, THICK FACTOR

Turn concentration into density

|B_τ∩⋃T| ≳ C_KT-CW(𝕋̃)^σ/2(δ/τ)^σ+ω|B_τ| (1.12)

E converts the large concentration constant into the local win above.

CASE 3 · HIGH m, δ-THIN FACTOR

Enlarge and repeat

Replace the grains by wider factors. Repetition increases their dimensions; a saturated configuration yields a Córdoba-type L² gain.

Vignette boundary. Pages 9–14 deliberately impose simplifying broadness, uniformity, and tangency assumptions. If they fail, the full proof either finds another intermediate scale or obtains a direct Córdoba-type gain. The next module states the formal repair mechanism. [S4, pp. 9–14]

ORIGINAL ROUTE CHECK · MODULE 3 OUTPUT

What may the formal grains module legitimately take from this two-zoom diagnostic?

NEXT QUESTIONHow does the formal proof repair the vignette and make all three multiplicity scales legal?

MODULE 4 · GRAINS REPAIR + REFINED INDUCTION

Make every hypothesis needed by the next zoom hold at the same time.

WHY

The introductory grains picture does not by itself preserve density, scale range, tangency, and local Katz–Tao control simultaneously.

INPUT

The unresolved structured-grains branch from Module 3, restored to general shadings and refinements, together with E(σ,ω) and a δ^η-dense family satisfying C_KT-CW,C_F-SW≤δ^−η.

MOVE

Use Proposition 7.5 to produce or repair grains; only after factoring the parent cover can the proof introduce an internal θ-scale and multiply three E estimates.

OUTPUT

Either an immediate δ^ω−α volume gain or an induction-ready parent cover at an intermediate scale.

Advanced contract · Proposition 7.5 and the three terminating moves

Let ω>0, σ∈(0,2/3], and ζ∈(0,ω/1000), and suppose E(σ,ω) is true. Proposition 7.5 chooses α,η,κ>0 so that every δ>0 and every δ^η-dense family with C_KT-CW,C_F-SW≤δ^−η has one of two outputs.

OUTPUT A · WIN NOW

|U| ≥ κδ^ω−αNv·Q^−σ.

OUTPUT B · STRUCTURE

After a δ^ζ-refinement, write 𝕋_δ for the refined fine family. The refined shaded family is δ^ζ-dense and satisfies C_KT-CW(𝕋_δ),C_F-SW(𝕋_δ)≤δ^−ζ. There are δ≤a≤b≤c≤1 and a parent scale ρ=b/c satisfying

δ^1−ω/100 ≤ ρ ≤ δ^ω/100, c ≥ δ^ζ(ρ/δ)(#𝕋_ρ/#𝕋_δ).

The ρ-tubes form a balanced partitioning cover that factors from above and below for the Frostman slab axioms; the a×b×c grains form a robust two-scale decomposition and obey C^loc_KT-CW(𝒢)≤δ^−ζ.

MOVE 1

Restore grain length

If the lower bound for c has fallen behind the current parent count, reapply the two-scale grains theorem. The scale ρ stays fixed and c increases.

MOVE 2

Lengthen, then retest

If ρ is too large for the induction window, find substantially longer grains and retest the resulting scale. This move may repeat before ρ returns to the allowed range.

MOVE 3

Repair local KT control

If C^loc_KT-CW(𝒢) is large, widen the grains. Their length does not decrease, while ρ substantially increases.

Output A stops with the gain. A nonterminal update either increases c by a fixed power, or increases ρ by a fixed power while c does not decrease; Output B advances to the next required check. Since c,ρ≤1, only boundedly many updates can occur.

OUTPUT A · STOPOUTPUT B · ADVANCEc ↑ by a fixed powerρ ↑; c does not decrease

THREE DISTINCT CHECKS

One admissibility condition, one gain exit, one grouping condition.

LONG-END EXIT · INSIDE ONE GRAIN

Required internal geometry: an active tube crosses a grain through its long ends.

TRANSVERSE · DIRECT GAIN EXIT

If tangent planes are transverse, the Córdoba-type L² argument supplies the gain branch.

TANGENTIAL · GROUP AND CONTINUE

If tangent planes nearly agree, refine into a common box and continue the multiscale argument.

Geometry locator. Original code-native schematic for the proof vignette on pages 12–14 and the formal repair in Proposition 7.5 and §§7–9. Long-end exit is not a synonym for transversality; the three cards intentionally lead to different proof states.

Printed text equivalent · Proposition 7.5 repair dashboard

Two outputs. Output A is |U|≥κδ^ω−αNv·Q^−σ. Output B is a δ^ζ-dense refinement satisfying C_KT-CW,C_F-SW≤δ^−ζ, with δ≤a≤b≤c≤1, ρ=b/c, δ^1−ω/100≤ρ≤δ^ω/100, c≥δ^ζ(ρ/δ)(#𝕋_ρ/#𝕋_δ), a balanced partitioning parent cover that factors above and below for the Frostman slab axioms, robust a×b×c grains, and C^loc_KT-CW(𝒢)≤δ^−ζ.

Move 1: restore the grain-length bound, keeping ρ fixed while c increases.
Move 2: find substantially longer grains and retest the resulting scale; the move may repeat before ρ returns to the induction window.
Move 3: repair large local Katz–Tao concentration by widening grains; c does not decrease and ρ substantially increases.
Termination: Output A stops; Output B advances to the next check; every nonterminal update either increases c by a fixed power or increases ρ by a fixed power while c does not decrease. Since c,ρ≤1, only boundedly many updates occur.

Keep the geometry distinct. Long-end exit is an admissibility condition inside one grain. A transverse interaction gives a Córdoba-type L² gain. Nearly tangential grains may instead be grouped in a common box for the next zoom.

FORMAL HANDOFF · PROPOSITION 9.1

The proof vignette has two zooms; the induction has three multiplicity levels.

Proposition 9.1 does not begin with a θ-scale. It reaches that scale only after Proposition 7.5 has produced the two-scale structure and Proposition 4.6 has tested the resulting parent cover.

1 · BUILD
Proposition 7.5 first supplies refined fine tubes (𝕋₁,Y₁), a parent scale ρ, a ρ-cover 𝕋_ρ, and grains.
2 · FACTOR
Proposition 4.6 factors 𝕋_ρ through congruent convex containers Z.
3 · TEST COARSE KT
If C_KT-CW(𝕋_ρ)≤δ^−ζ, stop with Proposition 9.1’s structured output: factoring above and below for both Katz–Tao convex and Frostman slab axioms.
4 · CREATE θ ONLY IF NEEDED
Otherwise coarse Katz–Tao concentration is large. Proposition 5.2 gives the direct gain or makes Z almost tube-like. Only in the high-coarse-KT, almost-tube branch are the Z replaced by coaxial θ-tubes, producing 𝕋_θ.
5 · MULTIPLY
Apply E at δ/ρ to the fine system, at ρ/θ to medium grains, and at θ to the coarse system; then multiply the three bounds.

Notation before calculation. 𝕋₁ is the refined δ-tube family from Proposition 7.5; ρ is its parent scale; θ is the width of the almost-tube factors and is introduced only in Step 4; 𝕋_θ is their balanced θ-tube cover. The nested losses satisfy ε₂≪ε₃ and ultimately ε₃<σζ/10.

QUESTION 1 · INSIDE ONE ρ-PARENT

How many fine δ-tubes are active? Answer with μ_fine.

QUESTION 2 · INSIDE ONE θ-TUBE

How many grains are active? Answer with μ_medium.

QUESTION 3 · AT THE POINT

How many θ-tubes are active? Answer with μ_coarse.

Multiplying those three answers bounds the total fine-tube multiplicity.

FINE · δ TO ρ

μ_fine ≲_δ δ^{−2ε₃+σζ}(δ/ρ)^−ω[(#𝕋₁/#𝕋_θ)(δρ/θ²)]^σ (9.7)

MEDIUM · ρ TO θ

μ_medium ≲_δ δ^−2ε₃(ρ/θ)^−ω−σ (9.8)

COARSE · θ TO 1

μ_coarse ≲ θ^{−ω−ε₂−ε₃}[(#𝕋_θ)θ]^σ (9.9)

1 · SCALE LOSS TELESCOPES

(δ/ρ)^−ω(ρ/θ)^−ωθ^−ω=δ^−ω

2 · INTERMEDIATE COUNTS CANCEL

The factor #𝕋_θ^−σ from the fine estimate cancels #𝕋_θ^σ from the coarse estimate.

3 · THE σ-FACTORS COLLAPSE

[(N/#𝕋_θ)(δρ/θ²)]^σ(ρ/θ)^−σ[(#𝕋_θ)θ]^σ=(Nδ)^σ

4 · THE REFINEMENT SAVES A POWER

δ^{σζ−4ε₃}, ε₃<σζ/10 ⟹ σζ−4ε₃>0

The nested ε₂≪ε₃ choice absorbs the remaining auxiliary losses into the displayed ε₃ budget.

μ_fineμ_mediumμ_coarse ≲ δ^{−4ε₃+σζ}δ^−ω(Nδ)^σ. (9.10)

The ledger shows the main cancellations explicitly. With ε₂ chosen sufficiently small relative to ε₃ and then ε₃<σζ/10, the residual exponent σζ−4ε₃ is positive, so this high-Katz–Tao branch gives the direct gain. This is what converts the grains geometry into the direct-gain-or-factorization dichotomy. [S4, §§7–9]

ORIGINAL ROUTE CHECK · MODULE 4 OUTPUT

Which branch is allowed to become the next rung of the scale ladder?

NEXT QUESTIONWhat does repeated failure of the gain alternative build across all scales?

MODULE 5 · SCALE LADDER, NSP, AND STICKY

Repeated failure becomes the all-scale hypothesis that sticky Kakeya needs.

WHY

One good intermediate cover is not yet stickiness, and Katz–Tao all-scale control may describe too few tubes.

INPUT

The factorized alternative of Proposition 9.1, repeated whenever the direct gain fails.

MOVE

Build a scale ladder, amplify a sparse family by rigid copies, and invoke the Frostman all-scale sticky theorem.

OUTPUT

The Katz–Tao-at-every-scale volume bound and the definite ω-improvement in Proposition 1.7.

1 · REPEAT THE ONLY SURVIVING BRANCH

One parent cover becomes a ladder

Whenever Module 4 does not already give the volume gain, its factorized parent cover becomes the next rung. Repetition supplies enough rungs to approximate every intermediate scale.

2 · COMPOSE THE CONTROLS

The ladder becomes Katz–Tao at every scale

Inner-fiber and outer-cover constants multiply under nested factoring. With the loss parameters chosen in order, their total remains a permitted δ^−ε error.

3 · REPAIR THE CONTRACT MISMATCH

KT is absolute; sticky asks for relative Frostman control

Katz–Tao bounds absolute tube density in convex containers, but the imported sticky theorem requires a Frostman family that stays relatively distributed inside its parent and initially protects |U| rather than Nv. The NSP rigid-copy amplifier supplies exactly this missing bridge.

Advanced bookkeeping · why finitely many ladder losses stay inside δ^−ε

Lemma 11.1 says that, for an integer M≥1, either the improved volume bound already holds or a refinement admits scales. Here M is the iteration depth—called N in the paper’s lemma—and is not the tube count N=#𝕋 used above.

δ=ρ_J<⋯<ρ₀=1, J≤2^M, ρ_i+1/ρ_i ≥ δ^{(1−ω/100)^M}.

Why the errors do not explode. Lemma 4.12 composes nested covers multiplicatively: the Katz–Tao constant of the fine family is bounded, up to constants, by the inner-fiber constant times the outer-cover constant. Across J≤2^M rungs, per-level losses of about δ^−2ε₁ total at most δ^{−2^M+1ε₁}. Fix M first and then choose ε₁ so that this is at most δ^−ε. Also choose M with (1−ω/100)^M<ε; every requested scale then lies within a δ^−ε multiplicative gap of a ladder rung. Lemma 11.2 consequently yields a direct gain or Katz–Tao convex Wolff control at every scale.

LAB 6 · THE MULTISCALE TREE

Do nearby fine tubes stay together when you zoom out?

A fine δ-tube has one parent at each coarser scale ρ. Missing branching creates a scale at which refined induction can gain; persistent controlled branching models the ladder produced by Lemmas 11.1–11.2. This visualization is intuition, while “Katz–Tao at every scale” is the formal output.

Toy branching completeness 72% Visible scales 5

branching profile: —
available move: —

Static takeaway. A branching deficit exposes a scale where induction can gain. Persistent controlled branching across the ladder is the rigid all-scale alternative that feeds the sticky theorem.

near-maximal branching exploitable deficit

IMPORTED INPUT · THEOREM 6.2

Frostman convex at every scale

For every ε>0 there are η,κ>0 such that, for every δ>0, a δ^η-dense shading on a family with Frostman convex Wolff control at every scale, error δ^−η, satisfies

|U| ≥ κδ^ε. (6.1)

Imported, with no circularity. Theorem 6.2 is Theorem 5.2 of the earlier Assouad-dimension paper [S8], a mild generalized sticky theorem built on the sticky paper [S6]. It predates and does not depend on Theorem 1.9. Section 6 uses it directly in Lemma 6.4’s all-scale branch; Section 10 uses it again after NSP to derive Theorem 10.2.

NEEDED OUTPUT · THEOREM 10.2

Katz–Tao convex at every scale

Under the analogous Katz–Tao-at-every-scale hypothesis, the paper must prove the cardinality-sensitive estimate

|U| ≥ κδ^εNv. (10.1)

NIKISHIN–STEIN–PISIER AMPLIFIER

Amplify, select, invoke sticky, then divide back.

The rigid copies create enough ambient mass for a Frostman subfamily; they are controlled images of the original family, but they are not asserted to be pairwise disjoint.

1 · ORIGINAL

Start with the sparse family 𝕋.

2 · RIGID COPIES

A₁(𝕋),…,A_R(𝕋), with R controlled.

3 · SELECT

Select the certified subfamily from the union.

4 · STICKY

Apply the Frostman all-scale sticky theorem.

5 · DIVIDE BACK

Use R≲N⁻¹v⁻¹ to recover mass.

Visible text equivalent. Begin with 𝕋; form at most K_εN⁻¹v⁻¹ rigid copies; inside their union choose an essentially distinct Frostman-convex-at-every-scale subfamily. Apply Theorem 6.2 to that selected family. Rigid-motion invariance gives |A_j(U)|=|U|, and subadditivity—not disjointness—gives |⋃_jA_j(U)|≤Σ_j|A_j(U)|=R|U|. Hence κδ^ε≤R|U|; combining this with R≤K_εN⁻¹v⁻¹ yields |U|≳δ^εNv.

Proof locator. Original code-native schematic for Wang–Zahl 2025, Proposition 10.3 and Theorem 10.2 in §10 (beginning on page 104), with the strategy overview in §2.4, page 22. The shapes are explanatory symbols, not paper artwork.

SPARSE INPUT

KT at every scale

Katz–Tao control is an upper nonconcentration bound; it does not force N≍v⁻¹.

PROPOSITION 10.3

Take controlled rigid copies

R ≤ K_εN⁻¹v⁻¹

The union of A₁(𝕋),…,A_R(𝕋) contains an essentially distinct Frostman-convex-at-every-scale subfamily.

DIVIDE BACK

Recover the missing mass factor

κδ^ε ≤ |⋃A_j(U)| ≤ R|U| ⟹ |U| ≳ δ^εNv

Rigid motions preserve |U| and subadditivity supplies the upper bound; the copies need not be disjoint. This is the Nikishin–Stein–Pisier bridge from Theorem 6.2 to Theorem 10.2.

Optional deeper explanation · why the imported sticky theorem is rigid

SHARED INPUT · STICKY KAKEYA · WANG–ZAHL 2022/26

Why sticky configurations cannot stay tiny.

Self-similarity

For a direction-separated family, |𝕋[T_ρ]|≈(ρ/δ)² is the sticky branching benchmark. Formally, the proof uses Frostman and Katz–Tao control at each scale; shadings and refinements preserve the density and multiplicity statistics needed to zoom repeatedly.

Planiness

Inside a small ball, tubes cluster near planes. The relevant plane may rotate as position changes, so “planar” is a local statement, not one global plane.

Graininess

At the δ^1/2 scale, the set resembles a union of plate-like grains of dimensions δ × δ^1/2 × δ^1/2.

Projection rigidity

Incidence loops constrain a slope function. Modern projection theorems rule out the sparse additive-and-multiplicative behavior suggested by an approximate-subring heuristic.

Twisted projection

π_f(x,y,z)=(x+yf(z),z)

Slice geometry predicts a small image, while direction separation turns projected tubes into a curved family whose maximal estimate forces a large image.

Contradiction

The sparse model gives an upper bound δ^c for some fixed c>0, while projection theory gives a lower bound δ^ε for every ε>0. Choosing ε<c makes δ^ε much larger than δ^c as δ→0, an impossibility.

Research boundary. These six boxes expose the dependency chain, not the technical proof. Establishing planiness/graininess, regularizing the slope, and proving the projection estimate require the full sticky paper. “Approximate subring” describes the Katz–Tao historical strategy; it is not a substitute for Wang–Zahl’s projection-theoretic argument. [S6]

Lemma 11.2 now has only two exits. Its direct branch gives the loss improvement α₂. In the other branch, the refined family has Katz–Tao control at every scale, so Theorem 10.2—applied with output loss ω/2—gives the second improvement. Taking the smaller of α₂ and ω/2 packages both outcomes into Proposition 1.7:

E(σ,ω) ⟹ D(σ,ω−g(σ,ω)), g(σ,ω)=min(α₂,ω/2)>0.

[S4, §§10–11]

ORIGINAL ROUTE CHECK · MODULE 5 OUTPUT

What leaves the scale ladder and enters endpoint closure?

NEXT QUESTIONHow does one strict ω-gain force both parameters all the way to zero?

MODULE 6 · SELF-IMPROVEMENT AND ENDPOINT

A local gain opens the whole parameter interval.

WHY

Proposition 1.7 improves ω at one positive pair; the theorem needs the endpoint (σ,ω)=(0,0).

INPUT

The independent hairbrush seed D(1/2,0), D⇔E, and the definite ω-gain.

MOVE

Trade ω-gain for σ-gain, then use relative openness and closedness.

OUTPUT

D(0,0), E(0,0), Corollary 1.10, and Theorem 1.2.

Fix ω>0 and define S_ω={σ∈[0,2/3]:D(σ,ω)}. The point is to show this set is all of [0,2/3], not merely to iterate an unspecified numerical sequence.

WHY FOUR?

An essentially distinct unit tube has two angular parameters for its direction and two transverse parameters for its position. A δ-net in this four-parameter space has O(δ⁻⁴) elements, so N≲δ⁻⁴ and Q=Nv^1/2≤N≲δ⁻⁴.

STAY INSIDE σ≥0

Given g=g(σ,ω)>0, set h=min{g,2σ}. The g-gain implies the weaker h-gain, while 0<σ−h/4<σ.

PAY FOR THE LEFT STEP

Q^h/4≲δ^−h ⟹ D(σ,ω−h)⇒D(σ−h/4,ω)

The exponent 1/4 comes directly from the four-parameter tube count.

SEED
Hairbrush
Appendix B proves D(1/2,0), hence the weaker D(1/2,ω) for each fixed ω>0.
TRANSPORT
Proposition 1.6
D(σ,ω)⇒E(σ,ω).
GAIN
Proposition 1.7
E(σ,ω)⇒D(σ,ω−g).
TRADE
Use the four-parameter tube count
Set h=min{g(σ,ω),2σ}. Then N≲δ⁻⁴ and Q^h/4≲δ^−h, so D(σ,ω−h)⇒D(σ−h/4,ω) without leaving the domain σ≥0.
CLOSE σ
Upward closed + a domain-safe left neighbor
S_ω is nonempty and upward closed because D(σ,ω)⇒D(σ′,ω) for σ′≥σ. At every positive σ∈S_ω, transport, gain, and the h-trade above give D(σ−h/4,ω) with 0<σ−h/4<σ, so S_ω is relatively open.
CLOSE ω
Spend ε-slack at the boundary
For closedness, use ε/2 at nearby parameters and absorb their small exponent difference into the remaining ε/2 using polynomial tube-count control. Thus S_ω is closed; connectedness gives S_ω=[0,2/3], hence D(0,ω). The same ε-slack as ω↓0 gives D(0,0), and Proposition 1.6 gives E(0,0).

|U| ≥ δ^ελ^Km⁻¹Nv. Corollary 1.10

Corollary 1.10 is the arbitrary-average-density reformulation of E(0,0). Theorem 1.2’s per-tube shading condition implies average λ-density, while its rectangular-prism hypothesis gives m≤1000. Absorbing that fixed constant yields Theorem 1.2, and the covering argument in the previous chapter yields Hausdorff and Minkowski dimension 3.

Endpoint boundary. Lemma 6.4 and the Section 11 gain mechanism are stated for σ>0. The proof reaches σ=0 through the closedness argument; it does not substitute σ=0 into those technical lemmas. [S4, pp. 7–8, 110–114, 125–127]

OPTIONAL CODA · CONSEQUENCES OF THE ENDPOINT

Two further ℝ³ theorems, not dependencies of the Kakeya proof.

THEOREM 1.12 · TUBE DOUBLING IN ℝ³

If T̃ is the two-fold dilate of a δ-tube T, then for every ε>0 and all sufficiently small δ, every δ-tube family in ℝ³ obeys

|⋃_T∈𝕋T̃| ≤ δ^−ε|⋃_T∈𝕋T|.

This proves the tube-doubling conjecture only in dimension three; the conjecture remains open in dimensions four and higher.

THEOREM 1.14 · LINE-SEGMENT EXTENSION IN ℝ³

For every collection ℒ of positive-length line segments in ℝ³, writing ℓ̃ for the full line containing ℓ,

dim(⋃_ℓ∈ℒℓ̃)=dim(⋃_ℓ∈ℒℓ).

This follows from the three-dimensional Kakeya theorem via Keleti–Máthé; it is a corollary, not a step used to prove Theorem 1.2.

[S4, pp. 14–15 and §12]

ORIGINAL ROUTE CHECK · MODULE 6 OUTPUT

What has been proved before the guide switches to the later route?

NEXT QUESTIONHow does the later proof reorganize this engine without calling its contracts D and E?

ROUTE CHANGE · LATER STREAMLINED PROOF · 2026

Restart from the robust target with different contracts.

The original 127-page dependency chain is now complete. The workflow below follows Guth–Wang–Zahl’s later 47-page organization. Its K_KT(β) and K_F(β) assertions play analogous bookkeeping roles, but they are new definitions—not renamed versions of D and E. Polynomial partitioning supplies the original grains input and is not needed in this later route. [S5]

STATIC PRIMER · READ THIS BEFORE THE INTERACTION

Two contracts feed two lemmas.

The roadmap below unpacks this fixed logical loop; none of its definitions depend on clicking a tab.

Shared quantifier order. For every ε>0, the assertion chooses η,δ₀>0 and must hold for every 0<δ≤δ₀ and every sufficiently full shaded family in its named concentration class.

ABSOLUTE-DENSITY CONTRACT

K_KT(β)

Δ_max(𝕋)≤δ^−η, λ≥δ^η ⟹ μ≤δ^−ε(#𝕋)^β

The trivial pointwise bound μ≤#𝕋 supplies K_KT(1).

RELATIVE-DENSITY CONTRACT

K_F(β)

C_F(𝕋,B₁)≤δ^−η, λ≥δ^η ⟹ μ≤δ^−εδ^−2β((#𝕋)|T|)^1−β/2

Here C_F(𝕋,B₁)=sup_K′⊂B₁Δ(𝕋,K′)/Δ(𝕋,B₁): a relative Frostman density ratio.

MAIN LEMMA 1 · SUPPLY

K_KT(β) ⟹ K_F(β)

Factoring repairs the relative-density input at the same exponent.

MAIN LEMMA 2 · IMPROVE

β>0, K_KT(β) ∧ K_F(β) ⟹ K_KT(β−ν(β)), ν(β)>0

This is the paper's statement. For iteration, one may weaken its output to a smaller positive decrement so the next exponent stays positive.

How one pass works. Main Lemma 1 proves K_F(β) from K_KT(β) at the same exponent. Main Lemma 2 then proves K_KT(β−ν(β)), supplying a definite decrement. For iteration use the weaker d=min{ν(β),β/2}; because #𝕋≥1, the exact output implies K_KT(β−d), and β−d remains positive.

2026 STREAMLINED ROUTE · INTERACTIVE WORKFLOW

Eight streamlined moves, with a fork and reunion.

Every node in this interactive roadmap follows the 2026 Guth–Wang–Zahl organization. Select a node; “Risk” names the logical gap that the next move must repair.

The 2026 streamlined fork: Steps 1–3 prepare and diagnose. Step 4 ends the sticky arm; Step 5 begins the alternative non-sticky arm and continues through Step 7. Both arms meet at Step 8.

STEP 1 / 8

Regularize without losing the example

Repeated dyadic pigeonholing makes tube counts, shading sizes, and pointwise multiplicities roughly constant. Only logarithmic factors are discarded, and those fit inside an arbitrary δ^−ε allowance.

INPUT

An arbitrary shaded δ-tube family satisfying convex nonconcentration.

MOVE

Refine to uniform branching and essentially constant multiplicity/fullness.

OUTPUT

A structured family whose scale statistics can be multiplied.

RISK

Throwing away too much mass or choosing incompatible good subsets at different scales.

(𝕋′,Y′) is a δ^ε-refinement of (𝕋,Y)

Source route: Streamlined proof §§2 and 5; uniform sets and refinement.

2026 STREAMLINED REDUCTION · GUTH–WANG–ZAHL

How the later proof forces an arbitrary family toward sticky.

Where the roadmap fits. The static primer above defines both assertions and both implications. Steps 1–7 now explain the geometric work required to prove those arrows.

What the last step must justify. The paper supplies a positive monotone choice of ν(β). For fixed 0<b<1, d_b=min{ν(b),ν(1),b/2}>0 is a uniform usable decrement while b≤β≤1.

VISUAL BRIDGE · IMPROVE, REPEAT, THEN ABSORB

Aim at a positive target—never jump to β=0.

The two main lemmas form a reusable loop. Start from the trivial estimate at β=1; each pass supplies the missing Frostman input and then lowers the Katz–Tao exponent.

KNOWNK_KT(β)β=1 on the first pass
Main Lemma 1
SUPPLIEDK_F(β)at that same β
Main Lemma 2 uses both
IMPROVEDK_KT(β−d)0<d≤min{ν(β),β/2}

Repeat with the improved positive exponent. Fix 0<b<1 and set d_b=min{ν(b),ν(1),b/2}. Since ν is positive and monotone, ν(β)≥min{ν(b),ν(1)} throughout b≤β≤1. Thus each pass may use the same d_b, stays positive, and after finitely many passes lands below b. No β=0 endpoint or attained infimum is used.

THE LAST HANDOFF · WHY POSITIVE b IS ENOUGH

Tube count

Convex nonconcentration gives #𝕋≲δ^−(2+η).

Choose b>0 small

K_KT(b) gives μ≲δ^−ε(#𝕋)^b≲δ^{−[ε+(2+η)b]}.

Put the shaded mass back

|U| ≥ μ⁻¹Σ_T∈𝕋|Y(T)| ≳ δ^{η+ε+(2+η)b}#𝕋|T|

Choose η, ε, and b>0 so η+ε+(2+η)b<α. Since 0<δ<1, this is stronger than |U|≥δ^α#𝕋|T| for all sufficiently small δ.

Logical finish: arbitrarily small b>0 + cardinality control. It does not assert K_KT(0), take a limit of inequalities, or assume an infimum is attained.

β-improvement loop. Main Lemma 1 turns K_KT(β) into K_F(β). Main Lemma 2 uses both to supply a definite decrement. Truncate that decrement when necessary to keep the next exponent positive; monotonicity then gives a uniform decrement above any fixed positive target, so finitely many passes reach it. The convex-density tube-count bound absorbs the remaining (#𝕋)^β factor into the requested δ-loss.

A. Zoom

Choose an intermediate radius ρ and group fine δ-tubes by parent ρ-tubes.

→

B. Factor

μ(fine) ≲ μ(inside parent) · μ(active parents)

Shadings make “active” precise.

→

C. Diagnose

The localized family may violate convex nonconcentration. Find a convex container W where its density is maximal.

→

D. Normalize

John’s theorem replaces the maximizing convex container W by a comparable ellipsoid; the ellipsoid’s semiaxis ratios determine whether the relevant normalized geometry is ball-, slab-, plank-, or tube-like.

ConditionMeaningUseful inheritanceProof role

Katz–TaoNo convex set has abnormally high absolute tube density.Downward: a subfamily remains Katz–Tao.Apply the definition of β after refinement.

FrostmanNo subcontainer is much denser than the ambient container.Upward: well-distributed inner objects yield well-distributed containers.Control high-density families created by localization.

FROSTMAN HIGH-DENSITY BOUND · EXACT 2026 FORM

Excess density must buy a multiplicity saving.

K_F(β): μ(𝕋,Y) ≤ δ^−εδ^−2β((#𝕋)|T|)^1−β/2.

This is the streamlined paper’s normalized Frostman inequality, under its Frostman and fullness hypotheses. It complements K_KT(β): localization can create a dense inner family for which relative Frostman control survives even when absolute Katz–Tao control does not. Crucially, Main Lemma 1 supplies the handoff K_KT(β)⇒K_F(β); Main Lemma 2 combines both statements to lower β by a definite ν(β). [S5] [S7]

THICK

If the maximizing convex container is ball-like or a high-density plank family, its occupied fraction is already too large for an extremizer.

THIN

If it is slab-like, rescale the planks to tubes. Low density invokes β; high density invokes the high-density lemma.

MANY SCALES

Choose scale intervals strategically. Gaps yield inductive savings; blocks of near-maximal branching invoke sticky Kakeya.

Iteration gives K_KT(β) for every β>0. The cardinality bound implicit in Δ_max(𝕋) then absorbs (#𝕋)^β into any requested δ-loss; no unsupported endpoint inference is needed.

PROOF VERSION LEDGER · KEEP THE ROUTES DISTINCT

The 2025 paper is the original proof. The 2026 paper is a later streamlined proof of that result; it is not the paper in which Wang–Zahl first resolved the conjecture.

2022 → 2026

Sticky theorem

Wang–Zahl solve the multiscale self-similar special case, using projection and sum-product technology.

Primary source [S6]

2024 / 2025

Assouad theorem

Every 3D Kakeya set has Assouad dimension 3. Ahlfors–David regular Kakeya sets, and Kakeya sets with stably equal Hausdorff and packing dimensions, have Hausdorff dimension 3. Important, but not yet the full conjecture.

Primary source [S8]

2025 · 127 pages · 14 figures

Original full proof

Wang–Zahl prove the conjecture. Their Theorem 1.2 uses rectangular-prism nonclustering and a per-tube λ condition to obtain the δ^ελ^K union bound.

Source entry [S4] Original PDF ↗

2026 · 47 pages · no figures

Later streamlined proof

Guth–Wang–Zahl reorganize the same main result around Δ_max, average shading density, and two exponent lemmas. Polynomial partitioning from the original route is explicitly no longer needed.

Source entry [S5] Streamlined PDF ↗

STAGE CHECK 5 · PROOF ENGINE

Why search for a maximizing convex container W after zooming into a ball?

06 · WHY FOURIER ANALYSTS SEE TUBES

A small frequency cap becomes a long spatial packet.

Fourier uncertainty turns curved surfaces in frequency space into needle-like regions in physical space. Kakeya geometry controls where packets overlap; orthogonality and decoupling control how their oscillations add.

FREQUENCY + PHASE

A pure wave sin(k·x+φ) has frequency vector k and phase φ. Superposition means adding many such waves; aligned phases reinforce, while mismatched phases can cancel.

CAP + NORMAL

A cap is a small patch on a curved frequency surface. Its normal points in the long direction of the corresponding physical packet; a narrower cap means less directional uncertainty and a wider packet.

THE EXTENSION OPERATOR

In E_Sf(x)=∫_Se^ix·ξf(ξ)dσ(ξ), S is the surface, ξ a frequency on it, f its amplitude, dσ surface area, and x the physical-space location where the waves are summed.

LAB 7 · FREQUENCY ↔ SPACE

Frequency cap width R^−1/2 0.22 Relative phases mixed

narrower cap: —
physical packet: R^1/2 × R tube
phase schematic: —

Schematic uncertainty visualization. Packet localization is approximate, with rapidly decaying tails; the coherence meter is not a full oscillatory integral.

Static takeaway. A narrow frequency cap produces a long spatial packet normal to the surface. Kakeya geometry controls where packets overlap; phase-sensitive estimates control how their oscillations add.

FOURIER RESTRICTION

Can data on a curved surface spread without concentrating too much?

E_Sf(x)=∫_Se^ix·ξf(ξ)dσ(ξ)

Split S into caps. Each cap produces packets in tubes normal to S. A restriction estimate bounds the global L^p size of their oscillatory sum. In a fixed formulation, admitting smaller p is stronger because it controls a more concentration-sensitive norm.

L^∞→L^p · Wang 2018p > 42/13diagonal conjecturep > 3diagonal · Wang–Wu 2022p > 45/14diagonal · Wang–Wu 2024p > 22/7

Wang’s 2018 theorem is an L^∞→L^p estimate for the truncated paraboloid, so its exponent is not part of the diagonal record. Wang–Wu’s 2022 theorem is a diagonal L^p→L^p estimate for the paraboloid. Their 2024 theorem, which supersedes it, covers compact C² hypersurfaces, possibly with boundary, having strictly positive second fundamental form. The conjectural range remains p>3. [S9] [S10] [S11]

WANG’S BROOMS

Concentrated at one end, dispersed at the other.

Polynomial partitioning cuts space into cells separated by the zero set of a low-degree polynomial. Tubes either cross cells, meet the wall transversely, or run tangentially near it. In the difficult tangential case, Wang found planar fans of packets rooted near a common region. Their mass cannot remain concentrated far from the root, giving a global L² saving. [S9]

Polynomial partitioning remains central to this restriction work. It appeared in the original 2025 Kakeya route but is absent from the 2026 streamlined Kakeya proof.

CLICKABLE DEPENDENCY MAP

Shared tools are not automatic implications.

Select a node to trace its safest connection.

→ direct analytic reduction↝ shared tool / method≠ not equivalent

Curved surface → caps → wave packets: a direct analytic decomposition.
Wave packets ↝ Kakeya: tube geometry constrains where packets overlap, but the set theorem does not imply restriction.
Decoupling → restriction/local smoothing: it controls how separated frequency pieces add.
Falconer, projections, and Furstenberg ↝ Kakeya/restriction: they share projection and incidence tools, not automatic theorem implications.

STAGE CHECK 6 · FOURIER BRIDGE

Once a curved frequency cap becomes a spatial tube, what still remains before a restriction estimate follows?

07 · THE OTHER FOUR STRANDS

Same instinct, different obstruction.

Each result isolates the configuration that defeats a naive average, then uses a scale-aware decomposition tailored to that obstruction.

GUTH · WANG · ZHANG, 2020

Local smoothing for the 2+1 dimensional wave equation

CONJECTURED RANGE ACHIEVED

A wave can be rough at each instant yet smoother when averaged through space-time. The sharp cone square-function estimate closed the difficult 4≤p<6 range.

∥u∥_{Lᵖ(ℝ²×[1,2])} ≤ C_α(∥u₀∥_W^{α,p} + ∥u₁∥_W^{α−1,p}), p≥4, α>½−2/p.

Geometry

R^−1/2 angular sectors near the cone dualize to 1×R^1/2×R planks. Their packing must be tracked across every intermediate aperture.

New mechanism

A multiscale square-function constant remembers actual plank concentration, combines a Kakeya-type L² overlap estimate with Lorentz rescaling, and approximates short cone pieces by a parabola.

Boundary

Two spatial dimensions plus time; p≥4; strict Sobolev inequality. This does not claim all-dimensional local smoothing or the endpoint.

Read the primary paper ↗

GUTH · IOSEVICH · OU · WANG, 2019

Falconer’s planar distance problem

MAJOR PROGRESS · OPEN

If a planar fractal is large, must it determine a positive-measure set of distances? The conjectural threshold is dimension greater than 1.

E⊂ℝ² compact, dim_HE > 5/4 ⟹ ∃x∈E : |{ |x−y| : y∈E }| > 0.

Obstruction

Train-track configurations create narrow, tall spikes in the pinned-distance distribution and make a global L² approach fail.

New mechanism

Split the Frostman measure into good and bad wave packets. Radial projection theory makes the discarded part small in L¹ for most pins; refined decoupling controls the good part in L².

Finish

The good distribution is L¹-close to the original pinned measure and has finite L² norm. Substantial good mass therefore remains on the actual pinned distance set; Cauchy–Schwarz there forces positive measure.

Read the primary paper ↗

REN · WANG, 2023

The planar Furstenberg set conjecture

SHARP PLANAR SOLUTION

Let 0<s≤1 and 0<t≤2. If a line family ℒ has dim_Hℒ≥t and dim_H(E∩ℓ)≥s for every ℓ∈ℒ, how large must E be?

dim_HE ≥ min { s+t, (3s+t)/2, s+1 }.

slice dimension s 0.60line-family dimension t 1.00

active lower bound—

—

Two regimes

Almost Ahlfors-regular configurations yield to geometric/additive-combinatorial tools. Semi-well-spaced configurations yield to a Fourier high–low incidence method.

Glue

A branching function records effective dimension across dyadic scales. The scale interval is cut into regular or semi-well-spaced pieces; sharp local estimates multiply through induction.

Feedback

This regular-versus-spaced scale decomposition became a methodological template around the later sticky/non-sticky Kakeya reduction. [S12] [S7]

Read the primary paper ↗

WANG · WU, 2024

Restriction through decoupling + two-ends Furstenberg

p > 22/7 · OPEN TO p > 3

Refined decoupling turns an analytic norm into a question about how many shaded tubes pass through each ball. A two-ends condition prevents each tube’s activity from hiding in one short segment.

refined decouplingoscillatory sum

two-ends incidencetube multiplicity M

⇒

restrictionp>22/7 in ℝ³

The paper proposes a higher-dimensional two-ends Furstenberg conjecture that would imply the restriction conjecture. This is a proposed route, not a solved equivalence. [S11]

Read the primary paper ↗

STAGE CHECK 7 · AWARD PORTFOLIO

Why can these five strands share a toolkit without one theorem automatically proving the others?

08 · A CENTURY OF IDEAS

The breakthrough is a relay, not an isolated leap.

Click a milestone for the mathematical object it contributed. Exact historical claims link to primary papers wherever practical.

1917

The rotating-needle question

Sōichi Kakeya asks for the least-area region allowing a unit needle to reverse direction. The later set problem keeps every direction but drops the continuous-motion requirement.

Besicovitch retrospective ↗

WHAT IS SETTLED?

A precise frontier map

SOLVED

Kakeya set dimension

n=1 is trivial; n=2 is classical; n=3 is Wang–Zahl.

OPEN

Higher-dimensional Kakeya

The dimension conjecture remains open for every n≥4.

OPEN

3D Kakeya maximal function

Full set dimension is weaker than the conjectured maximal-operator inequality.

OPEN

3D restriction

Wang–Wu reach p>22/7 in the diagonal formulation; p>3 is conjectured.

SOLVED RANGE

Planar local smoothing

The 2+1 wave equation has the conjectured range p≥4, α>½−2/p.

OPEN

Planar Falconer

5/4 is the highlighted positive-measure threshold; the conjectural threshold is 1.

SOLVED

Planar (s,t)-Furstenberg

The Ren–Wang piecewise lower bound is sharp.

STAGE CHECK 8 · FRONTIER

Which boundary does the 3D full-dimension theorem leave intact?

09 · MODEL CHECK

Can you reconstruct the argument?

Begin with a 12-question reconstruction of the main logic. Switch to the full 29-question research mode when you want the original-paper details as well.

NOTATION DESK

Search the vocabulary.

Filter terms

Besicovitch set

A compact set containing a unit line segment in every direction; often called a Kakeya set.

δ-tube

The δ-neighborhood of a unit segment. In ℝ³ its volume is comparable to δ².

Minkowski dimension

The exponent governing equal-scale covering numbers N_δ(E).

Hausdorff dimension

The critical exponent for variable-size covers weighted by diameter^s.

Shading Y(T)

The active measurable subset of a tube retained after localization or refinement.

λ(𝕋,Y)

Σ_T|Y(T)| / Σ_T|T|, the family-average shaded fraction. It does not require the same fraction on every tube.

Multiplicity μ

Summed shaded volume divided by union volume: average active overlap.

Δ_max

The largest tube density inside any convex container.

Katz–Tao condition

An absolute convex nonconcentration bound; useful because subfamilies inherit it.

Frostman condition

A relative density bound comparing every subcontainer with its ambient container.

Sticky

Near-maximal, coherent branching of fine tubes inside coarse tubes over many scales.

Planiness

Near-extremal tubes locally cluster near planes that may vary with position.

Graininess

Local plate-like organization. The δ×δ^1/2×δ^1/2 shape is the classical benchmark only; formal Proposition 7.5 uses a×b×c grains with b/c=ρ.

Plank

An anisotropic convex body a×b×1, intermediate between a tube and a slab.

Wave packet

A frequency-localized oscillatory piece concentrated near a long spatial tube.

Decoupling

An inequality comparing a Fourier sum with separated frequency pieces, exploiting limited constructive interference.

Two-ends condition

A tube’s active mass is not allowed to concentrate in one short portion.

δ^ε loss

A factor weaker than a constant but stronger, for dimension purposes, than any fixed positive-power deficit.

10 · SOURCE LIBRARY

Follow every major claim.

Official citation first; primary research and supporting exposition follow. Dates and publication status were checked on 21 August 2026.

OFFICIALPRIMARYEXPOSITION

OFFICIAL
International Mathematical Union, “Fields Medals 2026 — Hong Wang.”
The authoritative award scope: a body of work across harmonic analysis and geometric measure theory, including the three-dimensional Kakeya breakthrough—not an official designation of any single “Fields paper.”
IMU page ↗ official citation PDF ↗
PRIMARY
L. Guth, H. Wang, R. Zhang, “A sharp square function estimate for the cone in ℝ³” (2020).
Sharp cone square function and local smoothing in 2+1 dimensions.
arXiv:1909.10693 ↗ Annals DOI ↗
PRIMARY
L. Guth, A. Iosevich, Y. Ou, H. Wang, “On Falconer’s distance set problem in the plane” (published online 2019; Invent. Math. 219, 2020, 779–830).
Pinned positive-measure distance set above dimension 5/4.
arXiv:1808.09346 ↗ Inventiones DOI ↗
PRIMARY · ORIGINAL
H. Wang, J. Zahl, “Volume estimates for unions of convex sets, and the Kakeya set conjecture in three dimensions” (2025 preprint).
The original 127-page full proof, with 14 numbered figures. Theorem 1.2 gives the rectangular-prism / per-tube-λ estimate displayed above; Theorem 1.1 is the Hausdorff and Minkowski dimension conclusion.
arXiv record ↗ 127-page PDF ↗
PRIMARY · LATER PROOF
L. Guth, H. Wang, J. Zahl, “A streamlined proof of the Kakeya set conjecture in ℝ³” (2026 preprint).
A later 47-page proof, with no numbered figures, that reorganizes the Wang–Zahl result around convex density and explicitly removes polynomial partitioning from the reduction.
arXiv record ↗ 47-page PDF ↗ HTML paper ↗
PRIMARY
H. Wang, J. Zahl, “Sticky Kakeya sets and the sticky Kakeya conjecture” (JAMS, 2026).
The sticky case in ℝ³ and its projection-theoretic proof.
arXiv:2210.09581 ↗ JAMS DOI ↗
EXPOSITION
L. Guth, “Introduction to the proof of the Kakeya conjecture” (2025).
Historical obstacles, high-density lemma, and the general-to-sticky architecture.
arXiv:2505.07695 ↗
PRIMARY
H. Wang, J. Zahl, “The Assouad dimension of Kakeya sets in ℝ³” (Inventiones, 2025).
Assouad dimension 3 and structural precursors to the full theorem.
arXiv:2401.12337 ↗ journal DOI ↗
PRIMARY
H. Wang, “A restriction estimate in ℝ³ using brooms” (Duke Math. J. 171, 2022, 1749–1822).
For the truncated paraboloid: polynomial partitioning, two-ends, broom geometry, and the L^∞→L^p estimate for p>42/13.
arXiv:1802.04312 ↗ Duke DOI ↗
PRIMARY
H. Wang, S. Wu, “An improved restriction estimate in ℝ³” (2022 preprint; superseded by the 2024 paper).
Diagonal paraboloid restriction for p>45/14.
arXiv:2210.03878 ↗
PRIMARY
H. Wang, S. Wu, “Restriction estimates using decoupling theorems and two-ends Furstenberg inequalities” (2024 preprint).
Diagonal restriction p>22/7 for compact C² hypersurfaces, possibly with boundary, with strictly positive second fundamental form; and a geometric route toward p>3.
arXiv:2411.08871 ↗
PRIMARY
K. Ren, H. Wang, “Furstenberg sets estimate in the plane” (2023 preprint).
The sharp min{s+t,(3s+t)/2,s+1} theorem.
arXiv:2308.08819 ↗
PRIMARY
T. Wolff, “An improved bound for Kakeya type maximal functions” (1995).
Hairbrush and the 5/2 bound in ℝ³.
DOI ↗
PRIMARY
N. Katz, I. Łaba, T. Tao, “An improved bound on the Minkowski dimension of Besicovitch sets in ℝ³” (2000).
Improvement beyond 5/2 and the structural sticky/plainy/grainy program.
arXiv:math/9903166 ↗
PRIMARY
N. Katz, J. Zahl, “An improved bound on the Hausdorff dimension of Besicovitch sets in ℝ³” (JAMS, 2019).
Hausdorff dimension strictly greater than 5/2.
arXiv:1704.07210 ↗
PRIMARY
R. O. Davies, “Some remarks on the Kakeya problem” (1971).
Full dimension for planar Kakeya sets.
DOI ↗
PRIMARY
C. Fefferman, “The multiplier problem for the ball” (1971).
The landmark Fourier-analysis application of Besicovitch configurations.
DOI ↗
PRIMARY
A. Córdoba, “The Kakeya maximal function and the spherical summation multipliers” (1977).
The L² overlap method behind the planar derivation.
DOI ↗
AUTHOR
Hong Wang, publications and preprints.
Author-maintained list and professional information.
homepage ↗
PRIMARY / RETROSPECTIVE
A. S. Besicovitch, “On Kakeya’s problem and a similar one” (Math. Z. 27, 1928, 312–320) and “The Kakeya Problem” (Amer. Math. Monthly 70, 1963, 697–706).
The original published construction and Besicovitch’s own account of the 1917–1919 history.
1928 DOI ↗ 1963 retrospective DOI ↗
PRIMARY
J. Bourgain, “Besicovitch type maximal operators and applications to Fourier analysis” (GAFA 1, 1991, 147–187).
The 7/3 lower bound and maximal-operator advances in three dimensions.
DOI ↗
EXPOSITION
T. Tao, “Stickiness, graininess, planiness, and a sum-product approach to the Kakeya problem” (2014).
A public outline of the Katz–Tao arithmetic/projection strategy, explicitly presented as a possible approach rather than a proof.
Tao’s outline ↗

Collaboration is part of the theorem. Full Kakeya: Hong Wang and Joshua Zahl; streamlined proof: Larry Guth, Wang, Zahl; local smoothing: Guth, Wang, Ruixiang Zhang; Falconer: Guth, Alex Iosevich, Yumeng Ou, Wang; Furstenberg: Kevin Ren, Wang; strongest restriction result discussed here: Wang and Shukun Wu.

How small can every direction hide?

One medal. Five connected frontiers.

3D Kakeya

Local smoothing

Restriction

Falconer

Furstenberg

Needles started it. Sets transformed it.

Kakeya’s needle problem

Besicovitch / Kakeya set

Pack the same directional information into less visible area.

Dimension is a scaling exponent.

Cover at scale δ

Read a log–log slope

Solve for the exponent

Translate to blurred volume

Hausdorff dimension allows every covering set to choose its own size.

Turn infinitely thin segments into countable tubes.

Why would |Kδ|≳δε for every ε>0 force full Minkowski dimension in ℝ³?

The planar argument, line by line.

Build the multiplicity function

Use Cauchy–Schwarz on the union

Sum overlaps by angular separation

Convert the logarithm into dimension

Pairwise angles forget collective geometry.

Advance one angular shell at a time.

What creates the logarithm in the 2D proof?

One breakthrough, two precise formulations.

Shading

What is occupied?

How much survives?

Where can tubes cluster?

The anti-clustering hypothesis

The dual quantity

Compare pointwise overlap with convex concentration.

δ-separated directions satisfy the convex condition.

The robust tube bound returns to the set we started with.

Keep the award, original proof, and later streamlined proof separate.

No single “Fields paper”

Wang–Zahl

Guth–Wang–Zahl

Which attribution keeps the chronology, authors, and award scope straight?

What connects the robust convex theorem back to ordinary direction-separated Kakeya tubes?

One shared obstruction, then two complete logical routes.

A shared trunk, two bookkeeping systems, one dimension theorem.

Follow two loss parameters from the hairbrush seed to zero.

What changes—and what must survive—at every stage?

Where each paper block enters this guide.

D is normalized; E carries the price of clustering.

m=CKT-CW(𝕋)

ℓ=CF-SW(𝕋)

Why does E contain the factors m−1, m−3/2, and ℓ?

Every successful move points left or down.

Why can the proof not blindly reapply D after an arbitrary localization?

Turn uncontrolled clustering into controlled containers and fibers.

Assume D(σ,ω)

Know E(σ,ω′), with ω′≥ω+t

Lower ω′ by a uniform α

Prune, package, then audit both directions.

Keep a subpolynomial fraction

Choose balanced convex containers

Do not swap inheritance and construction

What exact statement can now enter the near-extremizer diagnostic?

A hard family must defeat two different ways of winning.

Translate volume into overlap

Count descendants in one ρ-parent

Assume a fixed-power shortage

Convert the shortage into a saving

Partition and refine until active overlap admits a grains cover.

Two zooms, one multiplicity product.

Apply D to the rescaled grains

Turn concentration into density

Enlarge and repeat

What may the formal grains module legitimately take from this two-zoom diagnostic?

Make every hypothesis needed by the next zoom hold at the same time.

Restore grain length

Lengthen, then retest

Repair local KT control

One admissibility condition, one gain exit, one grouping condition.

The proof vignette has two zooms; the induction has three multiplicity levels.

Why would |K_δ|≳δ^ε for every ε>0 force full Minkowski dimension in ℝ³?

m=C_KT-CW(𝕋)

ℓ=C_F-SW(𝕋)

Why does E contain the factors m⁻¹, m^−3/2, and ℓ?

K_KT(β)

K_F(β)