OctoInformBeta

OpenAI preprint, September 2026 · §1. Introduction

Theorem 1.1 [The critical dimension]: The De Silva-Jerison cone and comparison with all nonnegative Sobolev competitors of the same trace

A nonflat Bernoulli minimizer in dimension seven.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

The dimension-seven construction uses a spherical profile in the last coordinate and unit cost for the positivity set. Let D⊂ℝdD\subset\mathbb R^d, with Euclidean norm |⋅||\cdot| and Lebesgue measure |E||E| for a measurable set EE.

A test function φ:ℝd→ℝ\varphi:\mathbb R^d\to\mathbb R on DD is a smooth function with compact support contained in DD.✦ A weak gradient G:ℝd→ℝdG:\mathbb R^d\to\mathbb R^d of u:ℝd→ℝu:\mathbb R^d\to\mathbb R on DD is a vector field satisfying (1) for every test function φ\varphi on DD and every h∈ℝdh\in\mathbb R^d, where ∂hφ\partial_h\varphi denotes the directional derivative of φ\varphi in the direction hh. ∫Du∂hφdx=−∫D(G⋅h)φdx.\int_D u\,\partial_h\varphi\,dx =-\int_D (G\cdot h)\varphi\,dx.(1) A Sobolev pair (u,G)∈H1(D)(u,G)\in H^1(D) consists of u∈L2(D)u\in L^2(D) and a weak gradient G∈L2(D;ℝd)G\in L^2(D;\mathbb R^d).✦ A pair (u,G)(u,G) belongs to Hloc1(D)H^1_{\mathrm{loc}}(D) if it belongs to H1(Br(c))H^1(B_r(c)) for every ball Br(c)B_r(c) with center c∈ℝdc\in\mathbb R^d, radius r>0r>0, and Br(c)¯⊂D\overline{B_r(c)}\subset D.✦

A pair (u,G)∈H01(D)(u,G)\in H^1_0(D) is a Sobolev pair in H1(D)H^1(D) for which, for every ε>0\varepsilon>0, there is a test function φ\varphi on DD satisfying the simultaneous approximations (2): ∥u−φ∥L2(D)<ε,∥G−∇φ∥L2(D)<ε.\|u-\varphi\|_{L^2(D)}<\varepsilon, \qquad \|G-\nabla\varphi\|_{L^2(D)}<\varepsilon.(2)✦

For a set B⊂ℝdB\subset\mathbb R^d and a Sobolev pair (v,V)(v,V) on BB, with ∇v=V\nabla v=V, the one-phase Bernoulli energy is (3): J(v;B)=∫B(|∇v|2+𝟏{v>0})dx.J(v;B)=\int_B\bigl(|\nabla v|^2+\mathbf 1_{\{v>0\}}\bigr)\,dx.(3) A local minimizer u:ℝd→ℝu:\mathbb R^d\to\mathbb R in an open set DD is a function having a representative weak gradient GG with (u,G)∈Hloc1(D)(u,G)\in H^1_{\mathrm{loc}}(D), nonnegative almost everywhere in DD, and satisfying (4) on every ball B=Br(c)B=B_r(c) with r>0r>0 and B¯⊂D\overline B\subset D, for every Sobolev pair (v,V)∈H1(B)(v,V)\in H^1(B) such that v≥0v\ge0 almost everywhere in BB and (v−u,V−G)∈H01(B)(v-u,V-G)\in H^1_0(B): J(u;B)≤J(v;B).J(u;B)\le J(v;B).(4)✦ A global minimizer u:ℝd→ℝu:\mathbb R^d\to\mathbb R is a local minimizer in D=ℝdD=\mathbb R^d.

A function u:ℝd→ℝu:\mathbb R^d\to\mathbb R is nonzero if it is not almost everywhere zero.✦ A one-homogeneous function u:ℝd→ℝu:\mathbb R^d\to\mathbb R satisfies (5), with the exceptional null set allowed to depend on rr: for every r>0,u(rx)=ru(x)for almost every x∈ℝd.\text{for every }r>0,\qquad u(rx)=r u(x)\quad\text{for almost every }x\in\mathbb R^d.(5)✦ A flat function u:ℝd→ℝu:\mathbb R^d\to\mathbb R is a half-space solution of the form (6), for a unit vector e∈ℝde\in\mathbb R^d: u(x)=(x⋅e)+=max⁡{x⋅e,0}for almost every x∈ℝd,|e|=1.u(x)=(x\cdot e)_+=\max\{x\cdot e,0\} \quad\text{for almost every }x\in\mathbb R^d, \qquad |e|=1.(6)✦

The Explicit Cone

The spherical profile ff and its derivative f′f' use the inverse hyperbolic tangent AA, with the formulas (7) on |t|<1|t|<1: A(t)=log⁡(1+t)−log⁡(1−t)2,f(t)=8−25t2+15t48(1−t2)2−158tA(t),f′(t)=−t(1−t2)3−5t4(1−t2)2−15t8(1−t2)−158A(t).\begin{aligned} A(t)&=\frac{\log(1+t)-\log(1-t)}2,\\ f(t)&=\frac{8-25t^2+15t^4}{8(1-t^2)^2}-\frac{15}{8}tA(t),\\ f'(t)&=-\frac{t}{(1-t^2)^3}-\frac{5t}{4(1-t^2)^2} -\frac{15t}{8(1-t^2)}-\frac{15}{8}A(t). \end{aligned}(7) There is a profile root in the interval (8); the first positive profile root t7t_7 is selected in that interval: t7∈[5173301000000,5173311000000],f(t7)=0.t_7\in\left[\frac{517330}{1000000},\frac{517331}{1000000}\right], \qquad f(t_7)=0.(8) The truncated profile p:ℝ→ℝp:\mathbb R\to\mathbb R, the squared boundary slope c7c_7, and the Bernoulli normalization κ\kappa are defined by (9): p(t)=f(min⁡{|t|,t7}),c7=(1−t72)(f′(t7))2,κ=(c7)−1.p(t)=f\bigl(\min\{|t|,t_7\}\bigr), \qquad c_7=(1-t_7^2)\bigl(f'(t_7)\bigr)^2, \qquad \kappa=(\sqrt{c_7})^{-1}.(9) Writing x7x_7 for the last coordinate of x∈ℝ7x\in\mathbb R^7, the unscaled conical profile Z:ℝ7→ℝZ:\mathbb R^7\to\mathbb R and the normalized De Silva-Jerison cone U:ℝ7→ℝU:\mathbb R^7\to\mathbb R are given by (10): Z(x)={|x|p(x7/|x|),x≠0,0,x=0,U(x)=κZ(x).Z(x)= \begin{cases} |x|\,p(x_7/|x|),&x\ne0,\\ 0,&x=0, \end{cases} \qquad U(x)=\kappa Z(x).(10)

The Claim

Theorem 1. The normalized cone UU in (10) satisfies the scaling and nontriviality conditions (11) and (12): for every r>0,U(rx)=rU(x)for almost every x∈ℝ7.\text{for every }r>0,\qquad U(rx)=rU(x)\quad\text{for almost every }x\in\mathbb R^7.(11)✦ |{x∈ℝ7:U(x)≠0}|>0.\bigl|\{x\in\mathbb R^7:U(x)\ne0\}\bigr|>0.(12)✦ For every unit vector e∈ℝ7e\in\mathbb R^7, its disagreement with the corresponding half-space solution has positive measure as in (13): |{x∈ℝ7:U(x)≠(x⋅e)+}|>0.\bigl|\{x\in\mathbb R^7:U(x)\ne(x\cdot e)_+\}\bigr|>0.(13)✦ It also has the full global minimality property (14): U is a global minimizer of J on ℝ7.U\text{ is a global minimizer of }J\text{ on }\mathbb R^7.(14)✦ The dimension-seven existence assertions are (15) and (16): ∃u:ℝ7→ℝ,u is a global minimizer of J,u is nonzero, one-homogeneous and nonflat.\exists u:\mathbb R^7\to\mathbb R,\qquad \begin{gathered} u\text{ is a global minimizer of }J,\\ u\text{ is nonzero, one-homogeneous and nonflat}. \end{gathered}(15)✦ {u:ℝ7→ℝ:u is a nonzero one-homogeneousnonflat global minimizer of J}≠⌀.\left\{u:\mathbb R^7\to\mathbb R: \begin{array}{l} u\text{ is a nonzero one-homogeneous}\\ \text{nonflat global minimizer of }J \end{array}\right\}\ne\varnothing.(16)✦

In the paper, the result reads: every nonzero one-homogeneous global minimizer of JJ in ℝd\mathbb R^d, for 1≤d≤61\le d\le6, is flat; there is a nonflat one-homogeneous global minimizer in ℝ7\mathbb R^7; consequently the critical dimension, the least dimension admitting a nonzero one-homogeneous nonflat global minimizer, is d*=7d_*=7.

What Is Checked

The claim is stated over real-valued functions on ℝ7\mathbb R^7, with Sobolev competitors and zero-trace differences of both the functions and their weak gradient representatives. The normalized profile has the local Sobolev regularity (17) with the specified gradient ∇U\nabla U: (U,∇U)∈Hloc1(ℝ7).(U,\nabla U)\in H^1_{\mathrm{loc}}(\mathbb R^7).(17)✦

Fix c∈ℝ7c\in\mathbb R^7 and r>0r>0, and put B=Br(c)B=B_r(c). Let (v,V)∈H1(B)(v,V)\in H^1(B) satisfy (v−U,V−∇U)∈H01(B)(v-U,V-\nabla U)\in H^1_0(B). For 0≤v≤U0\le v\le U almost everywhere in BB, the comparison from below is (18): J(U;B)≤J(v;B).J(U;B)\le J(v;B).(18)✦ For U≤vU\le v almost everywhere in BB, the comparison from above is (19): J(U;B)≤J(v;B).J(U;B)\le J(v;B).(19)✦