---
title: Two-Ends Furstenberg Estimates
url: https://www.emergentmind.com/topics/two-ends-furstenberg-estimates
type: topic
---

# Two-Ends Furstenberg Estimates

Searching arXiv for recent papers on two-ends Furstenberg estimates and related Furstenberg set results.
Two-ends Furstenberg estimates are lower bounds for the size of unions of line- or curve-adapted shadings under a Furstenberg-type incidence hypothesis together with a non-concentration condition along each individual geometric object. In the planar setting, these estimates appear in several closely related forms: as discretized incidence lemmas, as lower bounds for unions of \(\delta\)-balls on \(\delta\)-tubes, as multiscale tube-point inequalities, and as Katz–Tao-type shading inequalities for line families. They have become central in the analysis of Furstenberg sets, the resolution of the planar Furstenberg set conjecture, restriction theory, projection theorems, discretized sum-product phenomena, and recent extensions from lines to transversal curves [2308.08819], [2509.21869], [2607.08461].

## 1. Definitions and formal variants of the two-ends condition

In the continuous planar Furstenberg problem, one fixes parameters
\[
s\in(0,1],\qquad t\in(0,2],
\]
and calls a set \(E\subset \mathbb{R}^2\) an \((s,t)\)-Furstenberg set if there exists a family of affine lines
\[
\mathcal L\subset \mathcal A(2,1)
\]
with
\[
\dim_H \mathcal L\ge t,
\]
such that for every \(\ell\in\mathcal L\),
\[
\dim_H(E\cap \ell)\ge s.
\]
Here \(\mathcal A(2,1)\) denotes the space of all lines in \(\mathbb{R}^2\), equipped with its natural Hausdorff dimension [2308.08819].

The phrase “two ends” enters through discretized models. In the evenly-spaced tube formulation, one works in \([0,1]^2\) at scale \(0<\delta\ll 1\), with a collection \(T\) of essentially distinct \(\delta\)-tubes and, for each \(T\in T\), a set \(Y(T)\) of essentially distinct \(\delta\)-balls meeting \(T\). The pair \((T,\{Y(T)\})\) satisfies the evenly-spaced \((\alpha;W,X)\)-Furstenberg conditions when the tubes obey a spacing condition at scale \(W^{-1}\), \(|T|\sim XW\), each \(Y(T)\) has \(\#Y(T)\sim \delta^{-\alpha}\), and the centers of balls in \(Y(T)\) are \(\delta^\alpha\)-separated along \(T\). In that setting, condition \(2(c)\) is explicitly called the “two-ends spacing” on each tube [2107.11937].

A later shading-based formulation starts with a \(\delta\)-separated family \(L\) of lines and a shading
\[
Y:L\to \{\text{unions of }\delta\text{-balls}\},\qquad Y(\ell)\subset N_\delta(\ell)\cap B^2(0,1).
\]
If \(0<\delta<r_2<r_1<1\), the shading is \((r_1,r_2,C)\)-two-ends when for every \(\ell\in L\) and every tube \(J\subset N_\delta(\ell)\) of dimensions \(r_1\times 1\),
\[
|Y(\ell)\cap J|\le C\,(r_2/r_1)^2\,|Y(\ell)|.
\]
This version encodes a scale-sensitive prohibition against concentrating most of the shading inside a single long thin subregion of \(N_\delta(\ell)\) [2509.21869].

An analogous curve-based variant replaces lines by graphs \(f\in C^2(I)\) and uses dyadic \(\delta\)-squares. There one says a shading \(\mathcal P\subset \mathcal D_\delta(\mathbb R^2)\) is \((\alpha,\beta)\)-two-ends if for every Euclidean ball \(B(x,r)\), with \(\delta\le r\le 1\),
\[
|\mathcal P\cap B(x,r)|_\delta \le C\Bigl(\frac r\delta\Bigr)^\beta |\mathcal P| \qquad (\beta<1).
\]
The accompanying heuristic is that the shading does not concentrate in the middle of the curve-tube but “occupies both ends” at all scales [2607.08461].

These formulations are not identical. A precise reading of the literature shows that “two ends” can mean \(\delta^\alpha\)-spacing of points along a tube, a quantitative non-concentration inequality on subsegments, or the extraction of two well-separated endpoints carrying many incidences. This suggests that the common structural content is multiscale non-concentration along one-dimensional geometric supports rather than a single rigid definition.

## 2. Early two-ends lemmas in the \((\alpha,2\alpha)\) regime

A major precursor appears in the Katz–Tao-style analysis of planar \((\alpha,2\alpha)\)-Furstenberg sets. In the \(\delta\)-discretized model, a set \(E_\delta\subset B^2(0,1)\) is called a \(\delta\)-discretized \((\alpha,2\alpha)\)-Furstenberg set if there is a \(\delta\)-separated family of lines \(\mathcal L_\delta\subset A(2,1)\) with \(|\mathcal L_\delta|\asymp \delta^{-2\alpha}\), and for each \(w\in\mathcal L_\delta\) a set \(R_w\subset w(\delta)\) with \(|R_w|\asymp \delta^{2-\alpha}\) satisfying
\[
|R_w\cap B(x,r)|\le C(r/\delta)^\alpha,\qquad r\in[\delta,1],
\]
such that
\[
E_\delta=\bigcup_{w\in\mathcal L_\delta} R_w.
\]
Within this framework, the incidence relation \(x\sim x'\) means that \(x,x'\in R_w\) for some \(w\in\mathcal L_\delta\), and one studies the double-incidence set
\[
A=\{(x_0,x_1,x_2)\in E_\delta^3:x_0\sim x_1,\ x_0\sim x_2\}.
\]
The Two-Ends Lemma then says, roughly, that under a contradictory small-size assumption on \(E_\delta\), there exist two points \(y_1,y_2\in E_\delta\) with \(|y_1-y_2|\gtrsim 1\) such that many points \(x_0\in E_\delta\) are simultaneously incident to both \(y_1\) and \(y_2\) [2001.11304].

In that argument, the two-ends phenomenon is a structural extraction principle rather than a final union estimate. The proof combines pair-counting on lines, non-concentration on each \(R_w\), elimination of near-diagonal contributions, and a Fubini step that forces two macroscopic ends. The significance is that once these two ends are fixed, a projective change of coordinates turns one family of lines into a nearly Cartesian configuration, after which Bourgain’s discretized projection theorem can be applied to obtain a gain beyond the naïve \(2\alpha\) lower bound [2001.11304].

A quantitatively stronger version was later obtained for \((\alpha,2\alpha)\)-Furstenberg sets. Writing
\[
y(\alpha,2\alpha)=\inf\{\dim_H E:E\text{ is an }(\alpha,2\alpha)\text{-Furstenberg set}\},
\]
one has
\[
y(\alpha,2\alpha)\ge 2\alpha + c(\alpha),\qquad
c(\alpha)=\frac{\alpha(1-\alpha)}{6(155+68\alpha)}.
\]
In particular,
\[
y(1/2,1)\ge 1+\frac1{4536}.
\]
The proof passes through a discretized incidence theorem in which, at each point \(p\), the directions of incident lines form a \((\delta,\alpha)\)-set on \(S^1\) and satisfy an additional “robust transversality” condition, followed by a two-ends stopping-time argument that decomposes \(E\) into bushes of lines [2112.08249].

A common misconception is that two-ends arguments in the \((\alpha,2\alpha)\) literature were already direct, global lower bounds on \(|E|\). In these papers, the two-ends mechanism is instead embedded in a longer incidence-and-projection scheme: it supplies nontrivial geometry at two well-separated endpoints, which is then converted into dimension growth by projection or sum-product methods.

## 3. Evenly-spaced discretized estimates and sharp examples

A distinct two-ends Furstenberg estimate was proved for evenly-spaced \(\delta\)-tube configurations. If \((T,\{Y(T)\})\) satisfies the evenly-spaced \((\alpha;W,X)\)-Furstenberg conditions with \(|T|\sim XW\), and
\[
B=\bigcup_{T\in T} Y(T),
\]
then for every \(0<\varepsilon\ll 1\) there is a constant \(C=C(\varepsilon)\) so that
\[
|B|
\;\ge\;
C^{-1}(\log(1/\delta))^{-3.5}
\min\Bigl\{
\delta^{-(\alpha+1)},
\ \delta^{-3\alpha/2}(XW)^{1/2},
\ \delta^{-\alpha}XW
\Bigr\}.
\]
When \(W=1\) and \(X=\delta^{-1}\), this recovers the regime relevant to Furstenberg’s original \(\alpha\)-set question [2107.11937].

The proof has two principal steps. First, a dyadic pigeonhole reduction finds a single gap length \(\delta^\beta\) so that on a large subcollection \(T'\subset T\), each tube has many consecutively \(\delta^\beta\)-separated pairs of balls. Second, one forms a graph whose vertices are the centers of all \(\delta\)-balls in \(B\), and whose edges come from these nearby pairs. Since any two tubes in \(T'\) cross at most once, the plane drawing of this graph has crossing number \(O(|T'|^2)\). The crossing-number bound then yields
\[
|B|\gtrsim \min\{|T|\delta^{-\beta},\, |T|^{1/2}\delta^{-3\beta/2}\},
\]
while a trivial density estimate contributes the third term \(\delta^{-(\alpha+1)}\) [2107.11937].

The sharpness theory is unusually explicit. Three constructions attain the three terms in the minimum: a “trivial height set,” a “full bush,” and a “rational two-ends” configuration. In the rational example, choosing \(r\sim \delta^{-\alpha}(XW)^{1/2}\) gives
\[
|B|\approx \frac{W^2X^2}{r^3}
=
\delta^{-3\alpha/2}(XW)^{1/2},
\]
showing that the middle term is optimal. This establishes that the two-ends exponent is not an artifact of proof technology but is already forced by extremal arithmetic-combinatorial examples [2107.11937].

Conceptually, this result isolates a regime in which no Fourier-analytic machinery is needed in the core argument once a uniform gap scale has been fixed. The “heart of the lower bound” is the combination of two-ends spacing and the crossing-number method.

## 4. Multiscale incidence theory and the resolution of the planar conjecture

The planar Furstenberg set conjecture was fully resolved by Ren and Wang. If \(E\subset\mathbb R^2\) is an \((s,t)\)-Furstenberg set, then
\[
\dim_H(E)\ge \min\Bigl\{s+t,\ \frac{3s+t}{2},\ s+1\Bigr\}.
\]
This matches the conjectured lower bound throughout the full range \(s\in(0,1]\), \(t\in(0,2]\) [2308.08819].

A central ingredient is a multiscale point-tube incidence inequality, described as Wolff’s two-ends lemma. At a single scale \(\delta\), for a set \(P\) of \(\delta\)-cubes and a family \(\mathcal T\) of \(\delta\)-tubes, one considers
\[
I(P,\mathcal T)=\#\{(p,T)\in P\times \mathcal T:p\subset T\}.
\]
For any intermediate parameter \(S\) with \(\delta^{-\alpha}\le S\le \delta^{-1}\),
\[
I(P,\mathcal T)
\lesssim
\Bigl(S\delta^{-1}|\mathcal T|\sum_{p\in P}1\Bigr)^{1/2}
+
S^{-1+}I(P,\mathcal T^{\,S\delta}).
\]
Here \(\mathcal T^{\,S\delta}\) denotes the natural covering of each \(\delta\)-tube by \(S\delta\)-tubes; the first term is the “high-frequency” term and the second the “low-frequency” term. Choosing \(S\) appropriately and iterating across scales yields an extra power saving whenever a tube is rich at both ends of its length [2308.08819].

The proof strategy proceeds through several discretized models. First, one reformulates the problem using a \((\delta,t)\)-set \(P\subset[0,1]^2\) of \(\delta\)-cubes and a family \(\mathcal T\) of \(\delta\)-tubes, seeking lower bounds on \(|\mathcal T|\) in terms of \(|P|\) and the multiplicity \(M\) of tubes through each \(p\in P\). In a semi-well-spaced case, where \(P\) has dimension \(2-s\) at large scales and dimension \(s\) at small scales, the two-ends incidence estimate gives
\[
|\mathcal T|
\gtrsim
\min\{M|P|,\ M^{3/2}|P|^{1/2},\ \delta^{-1}M\}
\approx_\delta
\delta^{-\min\{t,\frac{s+t}{2},1\}+}M.
\]
In an almost Ahlfors–David regular case, an exceptional-set estimate for orthogonal projections yields
\[
|\mathcal T|\gtrsim \delta^{-(s+t)/2+}M.
\]
The final induction on scales uses an Orponen–Shmerkin decomposition of the branching function and a multiscale incidence lemma to reassemble the contributions [2308.08819].

The significance of the two-ends mechanism here is explicit: it is the pivot that raises the exponent from the classical
\[
\max(t/2+s,\ s+1)
\]
to the conjectured
\[
\min\{s+t,\ (3s+t)/2,\ s+1\}.
\]
The same framework yields the exceptional-set bound
\[
\dim_H\{\theta\in S^1:\dim_H\pi_\theta(K)<u\}\le \max\{2u-\dim_H K,0\},
\]
resolving an orthogonal projection question of Oberlin, and also produces a discretized sum-product estimate
\[
\max\{|A+A|_\delta,\ |A\cdot A|_\delta\}\ge \delta^{-5s/4+\eta}\qquad (\eta>0,\ s<2/3)
\]
for \((\delta,s,\delta^{-})\)-sets \(A\subset[1,2]\) [2308.08819].

## 5. Modern planar two-ends inequalities for Katz–Tao line sets

Recent work reformulates two-ends Furstenberg estimates directly as lower bounds for the union of shaded neighborhoods of lines. In one such setup, a family \(L\) of \(\delta\)-separated lines in \(\mathbb R^n\) is directional if any two distinct \(\ell,\ell'\in L\) make an angle \(\gtrsim \delta\), and a shading \(Y\) is \(\lambda\)-dense if \(|Y(\ell)|\ge \lambda |N_\delta(\ell)|\) for every \(\ell\). If \(Y\) is \((\alpha_1,\alpha_2)\)-two-ends, then in the planar case one has
\[
\Bigl|\bigcup_{\ell\in L}Y(\ell)\Bigr|
\ge
c_{\alpha_1}\,\lambda^{1/2}\,\delta^{-O(\varepsilon)}\sum_{\ell\in L}|Y(\ell)|
\]
for every \(\varepsilon>0\). The key feature is the gain of \(\lambda^{1/2}\), which is stated to match the classical Furstenberg estimate in the regime of “upper dimension-range” \(s+t>2\) [2411.08871].

A sharper full-range planar statement was later obtained for Katz–Tao \((\delta,t)\)-sets of lines. If \(L\) is a \(\delta\)-separated Katz–Tao \((\delta,t)\)-set, \(Y\) is an \((r_1,r_2)\)-two-ends shading with \(r_2<r_1\), \(Y\) is \(\lambda\)-dense, and \(|Y(\ell)|\ge \delta^2\) for every \(\ell\), then for every \(\varepsilon>0\) there exists \(c_\varepsilon>0\) such that
\[
\Bigl|\bigcup_{\ell\in L}Y(\ell)\Bigr|
\ge
c_\varepsilon\,\delta^\varepsilon\,\lambda^{1/2}\,
(r_2/r_1)^{(t-1)/2}\,
\gamma_{Y,t^*}^{-1/2}
\sum_{\ell\in L}|Y(\ell)|,
\]
where
\[
t^*=\min\{t,2-t\}
\]
and
\[
\gamma_{Y,t}=\sup_{\ell\in L}\ \sup_{\substack{\delta\le r\le1\\x\in \ell}}
\bigl(r/\delta\bigr)^t\,
\bigl(\delta^{-2}|B(x,r)\cap Y(\ell)|\bigr).
\]
When \(t=1\), the factor \(\gamma_{Y,t^*}^{-1/2}\) disappears, recovering the earlier Wang–Wu inequality [2509.21869].

The proof is multiscale and broad–narrow. Its stated ingredients are a two-ends reduction, uniformization of branching functions, multiscale decomposition, intermediate-scale incidence via a theorem of Demeter–Wang together with an “upper-range” Furstenberg incidence bound, and a broad–narrow argument that handles narrow clusters by induction and broad points by incidence geometry. The paper also emphasizes that the \(\gamma_{Y,t^*}^{-1/2}\) factor is genuinely necessary when \(t\neq 1\): anisotropic rescalings of Wolff’s “hairbrush” example show that it cannot in general be removed [2509.21869].

This line of work changes the role of two-ends estimates. Instead of serving only as an internal lemma in a dimension argument, they become standalone inequalities with explicit density gains and full \(t\)-dependence.

## 6. Extensions to transversal curves and analytic applications

The two-ends Furstenberg framework has recently been extended from lines to transversal families of \(C^2\) curves. Let \(I\subset \mathbb R\) be a compact interval and \(f\in C^2(I)\). The vertical \(\delta\)-tube around the graph of \(f\) is
\[
\Gamma_f(\delta)=\{(x,y)\in I\times \mathbb R:|y-f(x)|<\delta\}.
\]
A family \(\mathcal F\subset C^2(I)\) is \(\mathfrak T\)-transversal if for all distinct \(f,g\in\mathcal F\),
\[
\inf_{x\in I}\Bigl(|f(x)-g(x)|+|f'(x)-g'(x)|\Bigr)
\ge
\mathfrak T^{-1}\|f-g\|_{C^2(I)}.
\]
For a \(\delta\)-configuration \((\mathcal F,\mathcal P)_\delta\) with \(\mathcal F\) a \((\delta,t)\)-Katz–Tao set, each shading \(\mathcal P(f)\) \(\lambda\)-dense and \((\alpha,\beta)\)-two-ends, and with
\[
\mathcal E=\bigcup_{f\in\mathcal F}\mathcal P(f),
\]
the two-ends Furstenberg inequality takes the form
\[
|\mathcal E|
\gtrsim_{t,\varepsilon,\mathfrak T}
\delta^\varepsilon\,
\lambda^{1/2}\,
\delta^{(t-1)/2}\,
\gamma_{\mathcal P,t^*}^{-1/2}
\sum_{f\in\mathcal F}|\mathcal P(f)|,
\]
where \(t^*=\min\{t,2-t\}\) and
\[
\gamma_{\mathcal P,t}
=
\sup_{f\in\mathcal F}\ 
\sup_{\substack{x\in\Gamma_f\\ \delta\le r\le1}}
\Bigl(\frac r\delta\Bigr)^t
|\mathcal P(f)\cap B(x,r)|_\delta.
\]
The proof is described as a single induction on scales combined with a sharp “intermediate-scale selection” lemma, replacing the four-page “bush” argument in the earlier line case [2607.08461].

The same paper applies the inequality to Fourier decay on convex curves. If \(f\in C^3([-6,6])\) satisfies \(\|f''\|\ge d>0\) and \(\|f\|_{C^3}\le c\), if
\[
\Gamma=\{(x,f(x)):x\in[-1,1]\},
\]
and if \(\mu\) is an \(s\)-Frostman measure on \(\Gamma\) with \(s\le 2/3\), then for every \(\varepsilon>0\) and large \(R\),
\[
\|\widehat{\mu}\|_{L^6(B_R)}^6
\lesssim_{c,d,s,\varepsilon}
R^{2-\frac{5s}{2}+\varepsilon}.
\]
In particular,
\[
\widehat{\mu}(\xi)=O(|\xi|^{-5s/12+o(1)}).
\]
The key incidence input is the dual form of the two-ends Furstenberg inequality [2607.08461].

Two-ends Furstenberg inequalities also feed into restriction estimates. Using the planar result, one obtains in \(\mathbb R^3\) the incidence bound
\[
M\lesssim m^{-1/2}R^{1/4},
\]
identified there with Wolff’s “\(5/2\) hairbrush” estimate in disguise. Combined with refined decoupling and \(L^2\) orthogonality, this leads to
\[
\|Ef\|_{L^p(B_R)}\le C_\varepsilon R^\varepsilon \|f\|_p,\qquad p>22/7,
\]
and to the conclusion that Besicovitch-type sets in \(\mathbb R^3\) have Hausdorff dimension at least \(5/2\). Higher-dimensional interpolation with geometric bounds yields the range
\[
p>p(n)=\frac{154n+6}{77n-95}=2+\frac{28}{11n}+O(n^{-2})
\]
for \(n\ge 4\) [2411.08871].

| Setting | Output | Source |
|---|---|---|
| Planar Furstenberg sets | \(\dim_H(E)\ge \min\{s+t,\frac{3s+t}{2},s+1\}\) | [2308.08819] |
| Restriction in \(\mathbb R^3\) | \(p>22/7\) and Wolff’s \(5/2\)-hairbrush bound | [2411.08871] |
| Transversal curves | Fourier decay \(O(|\xi|^{-5s/12+o(1)})\) | [2607.08461] |

Taken together, these developments show that two-ends Furstenberg estimates now constitute a general multiscale principle rather than a single lemma. In the line case they control how incidences persist across scales and directions; in the curve case they interact with transversality and Katz–Tao structure; and in both settings they convert one-dimensional non-concentration along geometric objects into global lower bounds for unions, dimension estimates, and analytic inequalities.

Source: https://www.emergentmind.com/topics/two-ends-furstenberg-estimates