The Falconer distance conjecture and distance measures

Source and verification

This article explains family 073 of the October 6, 2026 OpenAI collection. The research conclusions are attributed to its manuscripts. The source review covers statements, introductions, and the proof mechanisms described below; it is not an independent verification of the complete proofs.

Overview

The distance-set problem asks whether a fractal that is sufficiently large in dimension must generate a positive-length set of distances. Family 073 claims the strict Falconer threshold in every dimension. The central passage is from geometric nonconcentration to an absolutely continuous distance distribution.

🏷️ Preliminaries: dimension and Frostman measures

For a compact set $E\subset {\mathbb{R}}^d$, its distance set is

$$\Delta (E)=\{|x-y|:x,y\in E\}.$$

Hausdorff dimension measures the cost of covers by sets with varying diameters. A useful analytic replacement for a dimension lower bound is a Frostman measure: if $s<{\dim }_{H}E$, there is a probability measure $\mu$ supported on $E$ such that

$$\mu (B(x,r))\le C{r}^s.$$

The upper bound prevents too much mass from fitting into a small ball. It does not give a matching lower bound and does not imply that $\mu$ has a density. This matters for highly irregular fractals.

The $a$-energy is $I_a(\mu )=\iint |x-y{|}^{-a}d\mu (x)d\mu (y)$. A Frostman bound of exponent $s$ implies $I_a(\mu )<\infty$ for $0<a<s$: split distances into dyadic intervals and sum a constant times $2^{j(a-s)}$. This elementary summation is one reason to work strictly below the set’s dimension.

🏷️ The claimed threshold

Family 073, as stated in the manuscript

If $d\ge 2$, $E\subset {\mathbb{R}}^d$ is compact, and ${\dim }_{H}E>d/2$, then $\Delta (E)$ has positive one-dimensional Lebesgue measure. (OpenAI, 2026)

The conclusion is positive measure, which is stronger than Hausdorff dimension one. It is unpinned: the source does not claim at this threshold that one fixed $y\in E$ has a positive-measure pinned distance set $\{|x-y|:x\in E\}$. Neither the equality case ${\dim }_{H}E=d/2$ nor additional regularity of $E$ is part of the assertion.

🏷️ Distance measures and the decisive analytic criterion

Push the product measure $\mu \otimes \mu$ forward under $(x,y)\mapsto |x-y|$ to obtain a probability measure $\nu$ on $\Delta (E)$. If $d\nu =h(t)dt$ with $h\in L^2$, then

$$1={\int }_{\Delta (E)}h(t)dt\le |\Delta (E){|}^{1/2}\Vert h{\Vert }_2,$$

so $|\Delta (E)|\ge \Vert h{\Vert }_2^{-2}>0$. More generally, any nonzero submeasure of the distance distribution with an $L^2$ density suffices: replace $1$ by its positive mass.

This criterion explains why it is permissible to discard some pairs. The retained pairs must still carry a fixed positive mass, and the limiting density estimate must hold uniformly through the smoothing process. Obtaining an attractive estimate at each scale on unrelated retained sets would not establish either condition.

🏷️ The source’s multiscale mechanism

Choose an exponent between $d/2$ and ${\dim }_{H}E$, then restrict to separated pieces of a Frostman measure. Separation keeps the distance phase away from its singularity at coincident points. The manuscript constructs decreasing admissible pair sets; their intersection is required to have positive product mass.

At each scale, further small exceptional sets are removed to control directional fibers. A directional fiber fixes one endpoint and pushes the other endpoint to the sphere by radial projection. Its mass on a small spherical cap measures concentration along a narrow cone. In dimensions where the original measure has dimension less than $d-1$, such a fiber cannot simply be assumed to have a density on the sphere. The source instead claims the cap estimates needed for its Fourier argument.

Angular cap control and control of projections onto a line are different requirements: the latter involves belts on the sphere. The paper therefore adds a separate Fourier estimate to pass between the relevant tests. Its subsequent oscillatory analysis keeps the original Euclidean distance in ${\mathbb{R}}^d$; auxiliary projections used to obtain angular information do not replace that distance.

The recursion tracks losses at two separated depths. The role of this accounting is to keep geometric selections from exhausting the retained pair mass while allowing enough decay in the distance estimates. The final assembly reconstructs a positive distance submeasure and supplies the analytic regularity needed for positive length.

The source’s angular preparation, recursion, and final reconstruction are substantial claims. The elementary $L^2$ implication above is a complete deduction once its hypothesis is available; this article does not supply an independent proof of that hypothesis at $s>d/2$.

🏷️ Key steps and proof outlines

Regularization without assuming uniform fractal geometry

The first difficulty is that a Frostman measure may branch very unevenly from one dyadic scale to the next. The source regularizes it by recording, at a finite list of depths, dyadic bins for both leaf masses and child counts. A label therefore specifies not just how much mass a small cube has, but how that mass is distributed through its ancestors. Selecting a common label produces a controlled branching profile. The number of labels is a power of the depth parameter, hence a subpower of the reciprocal spatial scale. Losing one such factor is compatible with the fixed positive margin $s-d/2$.

The proof proceeds upward through the tree: classify fine leaf masses, count children carrying each label, append the dyadic bin of that count, and repeat. It never asserts that the original measure is Ahlfors regular. Each selected piece instead has approximate mass and branching identities at the designated scales. Those identities are what later conditional estimates use. See “Regularization and multiplication of fibre bounds.” (OpenAI, 2026)

Angular gains must survive conditioning

For separated endpoints $x,y$, put $e(x,y)=(x-y)/|x-y|$. An angular test restricts this direction to a set depending on one endpoint and an anchor scale. The source’s angular-gain argument samples directions, uses a planar incidence input after slicing, and checks that the sample retains cap nonconcentration. The sampling estimate is uniform over a finite grid of caps: concentration bounds for each cap are combined with a union bound. The thresholds are chosen large enough to absorb the number of relevant caps, yet remain subpower losses.

These tests are retained at both endpoints. Their role is to prevent a large distance collision count from being supported entirely on nearly parallel or otherwise concentrated configurations. Conditioning on a small endpoint cube can increase all probabilities, so unconditional cap estimates are insufficient. The regularized profile records the normalization cost of each such conditioning. This is also why the proof distinguishes spherical caps from the broader belts arising in a scalar projection test.

Positive collision counts and Fourier energy

If $\eta$ is a finite measure on the distance line and $ϵ>0$, define

$$B_{ϵ}(\eta )=\sum _{j\in \mathbb{Z}}\eta ([jϵ,(j+1)ϵ){)}^2.$$

This counts pairs of distances falling in the same bin. It is bounded above by the mass of $|s-s^{\prime }|\le ϵ$. Conversely, a pair with $|s-s^{\prime }-u|\le Cϵ$ belongs to only boundedly many relative bin offsets. Cauchy–Schwarz for each offset bounds its mass by a constant times $B_{ϵ}(\eta )$, uniformly in the translation $u$.

A smooth Fourier majorant then turns collisions into a truncated integral of $|\hat{\eta }{|}^2$. The source applies this to distance pushforwards of retained endpoint pairs. Oscillatory integral estimates exploit the Euclidean phase $|x-y|$ and the angular tests. The point of keeping collision counts positive is that restricting pairs can only decrease them; this monotonicity is unavailable for a general signed oscillatory expression.

The recursive state includes its earlier tests

The manuscript alternates between a positive distance-collision state and an angular energy state. An angular state remembers a base interval of scales in addition to its current interval. This record specifies which older angular tests must still appear in its endpoint symbols. A recursive reduction that improved the numerical energy but silently dropped those tests would not be an admissible continuation.

The normalization explains the half-dimensional threshold. If $M(a)$ is the approximate mass of an occupied cube at depth $a$, the relevant factor on passing from depth $a$ to $b$ is

$$2^{d(b-a)}{\left(\frac{M(b)}{M(a)}\right)}^2.$$

The first term counts spatial possibilities, while the squared ratio comes from two endpoint laws. At a uniform effective dimension $s$, it behaves like $2^{(d-2s)(b-a)}$. Thus $s>d/2$ supplies a decaying margin. The actual profile need not have constant slope; the two-state recursion distributes its gains and losses over suitable intervals and preserves the inherited tests.

Finally, decreasing pair restrictions give a common positive-mass limit. Uniform smoothed collision or Fourier-energy bounds supply its $L^2$ distance density. Weak compactness of bounded sets in $L^2$ identifies that density with the limiting distance measure by testing against continuous compactly supported functions. The positive-length conclusion then follows from the Cauchy–Schwarz argument above. This last compactness step explains why scale-dependent unrelated pair selections would not suffice.

See Also

• Family 074: Kakeya studies overlapping tubes and the distinction between estimates at one scale and Hausdorff-dimension conclusions.
• Family 077: Fourier extension explains the oscillatory estimates that make directional concentration relevant to harmonic analysis.

🏷️ Manuscripts and verification scope

• The Falconer distance conjecture in all dimensions — the pinned manuscript; its directory supplies the source and citation metadata.

The repository provides a formalization-scope page. Its scope description lists both planar and all-dimensional distance-set statements. These are statements about the repository’s documentation: no Lean build, axiom audit, or comparator-equivalence check was performed for this article.

OpenAI collection gives the reading sequence and source policy; coverage register records which families have articles and which remain to be written.

🐻  OpenAI 2026. The Falconer distance conjecture in all dimensions.