OctoInformBeta

OpenAI preprint, September 2026 · §2. Reduction to a smooth link and its suspension

§2 (setting): allowing signed competitors: Positive-part truncation on open sets of finite measure with nonnegative prescribed trace

Signed competitors give the same comparison.

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

Work in real Euclidean space ℝd\mathbb{R}^{d} with Lebesgue measure dxdx, and let D⊂ℝdD\subset\mathbb{R}^{d} have finite measure. A nonnegative prescribed Sobolev function fixes the trace of the competitors; their values in the interior may have either sign.

A test function φ\varphi on DD is a smooth real-valued function on ℝd\mathbb{R}^{d} with compact closed support contained in DD.✦ Write Cc∞(D)C_c^\infty(D) for these functions, and denote the directional derivative of φ\varphi at xx in a direction h∈ℝdh\in\mathbb{R}^{d} by Dφ(x)(h)\mathrm{D}\varphi(x)(h). A vector field GG is a weak gradient of a real-valued function uu on DD if, for every φ∈Cc∞(D)\varphi\in C_c^\infty(D) and every constant direction h∈ℝdh\in\mathbb{R}^{d}, ∫Du(x)Dφ(x)(h)dx=−∫D(G(x)⋅h)φ(x)dx.\int_D u(x)\,\mathrm{D}\varphi(x)(h)\,dx =-\int_D \bigl(G(x)\cdot h\bigr)\varphi(x)\,dx.(1)✦ A Sobolev pair (u,G)∈H1(D)(u,G)\in H^1(D) consists of a function u∈L2(D)u\in L^2(D) and a vector field G∈L2(D;ℝd)G\in L^2(D;\mathbb{R}^{d}) satisfying (1).✦

A zero-trace Sobolev pair (w,W)∈H01(D)(w,W)\in H^1_0(D) is a pair in H1(D)H^1(D) whose function and weak gradient admit simultaneous approximation by test functions and their gradients: for every ε>0 there is φ∈Cc∞(D) such that∥w−φ∥L2(D)<ε,∥W−∇φ∥L2(D)<ε.\begin{gathered} \text{for every }\varepsilon>0\text{ there is } \varphi\in C_c^\infty(D)\text{ such that}\\ \|w-\varphi\|_{L^2(D)}<\varepsilon, \qquad \|W-\nabla\varphi\|_{L^2(D)}<\varepsilon. \end{gathered}(2)✦

For a function and vector-field pair (v,V)(v,V), the Bernoulli energy is J(v;D)=∫D(|V(x)|2+mathbf1{v(x)>0})dx,J(v;D)=\int_D\left(|V(x)|^2+mathbf{1}_{\{v(x)>0\}}\right)\,dx,(3)✦ where VV is the explicit representative of ∇v\nabla v when (v,V)(v,V) is a Sobolev pair. The positive-part truncation of (v,V)(v,V) is the pair (v+,V+)(v_+,V_+) defined by v+(x)=max⁡{v(x),0},V+(x)=𝟏{v(x)>0}V(x).v_+(x)=\max\{v(x),0\}, \qquad V_+(x)=\mathbf{1}_{\{v(x)>0\}}V(x).(4)

The Claim

Theorem 1. Let v∈L2(D)v\in L^2(D) and V∈L2(D;ℝd)V\in L^2(D;\mathbb{R}^{d}). For the energy (3) and the truncation (4), J(v+;D)≤J(v;D).J(v_+;D)\leq J(v;D).(5)✦

Theorem 2. Suppose DD is open, (u,G)∈H1(D)(u,G)\in H^1(D), and u≥0u\geq0 almost everywhere in DD. Assume that J(u;D)≤J(w;D)J(u;D)\leq J(w;D) for every pair (w,W)∈H1(D)(w,W)\in H^1(D) such that w≥0w\geq0 almost everywhere in DD and (w−u,W−G)∈H01(D)(w-u,W-G)\in H^1_0(D). Then every pair (v,V)∈H1(D)(v,V)\in H^1(D) with (v−u,V−G)∈H01(D)(v-u,V-G)\in H^1_0(D) satisfies J(u;D)≤J(v;D).J(u;D)\leq J(v;D).(6)✦

What Is Checked

The inequality (5) is stated over square-integrable functions and vector fields on a set of finite measure; the comparison (6) is stated over Sobolev pairs on an open such set, with a nonnegative prescribed function and trace differences defined by (2).