OctoInformBeta

OpenAI preprint, September 2026 · Section 5. Pressure from weighted source layers

Lemma 5.2: An angular estimate for positive solutions and a layer estimate for measurable nonnegative sources

Averaged losses admit explicit energy bounds.

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

Angular loss measures the difference between a cap weight at a source point and its retained weight on a line component of a source layer. Let ℰ\mathcal E be a real Euclidean space of dimension nn, with volume measure μ\mu, unit sphere SS, and induced unnormalized surface measure σ\sigma. Balls are centered at the origin. Write r+=max⁡{r,0}r_+=\max\{r,0\} and ∫+\int^+ for the nonnegative extended integral; integrals without this superscript are real integrals. Real integrals of nonintegrable functions have value zero. Quotients with zero denominator have value zero, and real powers at zero have the values 00=10^0=1 and 0t=00^t=0 for t≠0t\ne0.

For real parameters p,q,A,Bp,q,A,B, a positive solution (u,v)(u,v) on ℝn\mathbb R^n is a pair of continuous, strictly positive functions, both C2C^2 away from the origin, satisfying (1) there: −Δu=|x|Avp,−Δv=|x|Buq(x≠0).-\Delta u=|x|^A v^p,\qquad -\Delta v=|x|^B u^q \quad (x\ne0).(1)✦ The weighted sources f,gf,g are the functions in (2): f(x)=|x|Bu(x)q,g(x)=|x|Av(x)p(x≠0),f(0)=g(0)=0.f(x)=|x|^B u(x)^q,\quad g(x)=|x|^A v(x)^p\quad(x\ne0), \qquad f(0)=g(0)=0.(2)✦

Cutoffs and Cap Weights

The auxiliary integers m,k,Nm,k,N, radial cutoff profile ηn\eta_n, and boundary gap profile d0d_0 are defined in (3); their radial extensions are η\eta and dd, and ww is the cutoff derivative weight: m=n+3,k=n(m+1),N=k+n+2,ηn(r)=(1−(r−1)+3)+N,d0(r)=min⁡{1,(2−r)+},η(x)=ηn(|x|),d(x)=d0(|x|),w(x)=−Dη(x)[x].\begin{gathered} m=n+3,\qquad k=n(m+1),\qquad N=k+n+2,\\ \eta_n(r)=\bigl(1-(r-1)_+^3\bigr)_+^N, \qquad d_0(r)=\min\{1,(2-r)_+\},\\ \eta(x)=\eta_n(|x|),\qquad d(x)=d_0(|x|), \qquad w(x)=-D\eta(x)[x]. \end{gathered}(3) For a real weight χ\chi on ℝn\mathbb R^n, the weighted nonlinear energy EχE_\chi is the real integral in (4); Ef,EgE_f,E_g are the two cutoff energies and EE is their sum: Eχ[u;q,B]=∫ℝnχ(x)f(x)u(x)dx,Ef=Eη[u;q,B],Eg=Eη[v;p,A],E=Ef+Eg.\begin{gathered} E_\chi[u;q,B]=\int_{\mathbb R^n}\chi(x)f(x)u(x)\,dx,\\ E_f=E_\eta[u;q,B],\quad E_g=E_\eta[v;p,A],\quad E=E_f+E_g. \end{gathered}(4)✦ The cutoff first and second derivative constants are C1,C2C_1,C_2 in (5); CwC_w is the cutoff weight derivative constant, and ℓn\ell_n is the low-level cutoff constant: C1=3N3N−1,C2=(6N+9N(N−1))3N−2,Cw=C1+2C2,ℓn=6N3N−1.\begin{gathered} C_1=3N3^{N-1},\qquad C_2=\bigl(6N+9N(N-1)\bigr)3^{N-2},\\ C_w=C_1+2C_2,\qquad \ell_n=6N3^{N-1}. \end{gathered}(5)

For x≠0x\ne0, its radial direction is θ(x)=x/|x|\theta(x)=x/|x|.✦ The tent cap profile ϕθ,l\phi_{\theta,l} with center θ∈S\theta\in S and parameter l∈ℝl\in\mathbb R is the function on SS defined in (6); Zθ,lZ_{\theta,l} is its normalization factor and ϕˆθ,l\widehat\phi_{\theta,l} its normalized weight: ϕθ,l(e)=(1−|e−θ|l)+,Zθ,l=∫Sϕθ,l(e)dσ(e),ϕˆθ,l(e)=ϕθ,l(e)Zθ,l.\begin{gathered} \phi_{\theta,l}(e)=\left(1-\frac{|e-\theta|}{l}\right)_+,\\ Z_{\theta,l}=\int_S\phi_{\theta,l}(e)\,d\sigma(e),\qquad \widehat\phi_{\theta,l}(e)=\frac{\phi_{\theta,l}(e)}{Z_{\theta,l}}. \end{gathered}(6) For a real parameter κ\kappa, write Wκ(e,x)W_\kappa(e,x) for the adaptive cap weight, a real-valued function of e∈Se\in S and x∈ℰx\in\mathcal E.

The cap normalization lower and upper constants c−,c+c_-,c_+, and normalized cap oscillation constant Ω0\Omega_0, are given by (7): c−=μ(B1)4n2n−12n,c+=4nμ(B1),Ω0=2c−/2n−1+22nc+(c−/2n−1)2.\begin{gathered} c_-=\frac{\mu(B_1)}{4^n2^{n-1}2^n},\qquad c_+=4^n\mu(B_1),\\ \Omega_0=\frac{2}{c_-/2^{n-1}} +\frac{2\,2^n c_+}{(c_-/2^{n-1})^2}. \end{gathered}(7) For real parameters δ,M,P\delta,M,P, the adaptive cap bound constant BκB_\kappa, cap weight oscillation constant Ωκ\Omega_\kappa, and ray integration constant RnR_n determine the angular loss coefficient Λκ,δ(M,P)\Lambda_{\kappa,\delta}(M,P) in (8): Bκ=ℓnc−κn−1,Rn=2n+1n,Ωκ=Cw2N−2c−κn−1+6ℓn2N−1Ω0κn,Λκ,δ(M,P)=Ωκδσ(S)+BκRnMδnP.\begin{gathered} B_\kappa=\frac{\ell_n}{c_-\kappa^{n-1}},\qquad R_n=2^{n+1}n,\\ \Omega_\kappa= \frac{C_w2^{N-2}}{c_-\kappa^{n-1}} +\frac{6\ell_n2^{N-1}\Omega_0}{\kappa^n},\\ \Lambda_{\kappa,\delta}(M,P) =\Omega_\kappa\delta\,\sigma(S) +\frac{B_\kappa R_nM}{\delta^nP}. \end{gathered}(8)

Source Layers and Line Components

For a real function ff on ℰ\mathcal E, the positive source subgraph 𝒮f\mathcal S_f is the subset of ℰ×ℝ\mathcal E\times\mathbb R in (9); its source layer measure νf\nu_f is the product measure restricted to this set. The line layer UfU_f of a real function ff on ℰ\mathcal E is the subset of (S×ℰ×ℝ)×ℝ(S\times\mathcal E\times\mathbb R)\times\mathbb R defined in (9), with ss the displacement in direction e∈Se\in S at fixed level t∈ℝt\in\mathbb R: 𝒮f={(x,t):0<t<f(x)},dνf(x,t)=𝟏𝒮f(x,t)dμ(x)dt,Uf={((e,x,t),s):e∈S,0<t<f(x+se)}.\begin{gathered} \mathcal S_f=\{(x,t):0<t<f(x)\},\qquad d\nu_f(x,t)=\mathbf1_{\mathcal S_f}(x,t)\,d\mu(x)\,dt,\\ U_f=\{((e,x,t),s):e\in S,\ 0<t<f(x+se)\}. \end{gathered}(9)

For a set U⊂X×ℝU\subset X\times\mathbb R and y∈Xy\in X, segment containment at a real displacement rr means that (y,s)∈U(y,s)\in U for every ss in the closed interval joining 00 and rr.✦ Given a set U⊂X×ℝU\subset X\times\mathbb R and y∈Xy\in X, let QU(y)Q_U(y) be the set of rational displacements rr for which segment containment holds. The forward and backward exit distances RU+,RU−R_U^+,R_U^- and the line component length ℒU\mathcal L_U of a set U⊂X×ℝU\subset X\times\mathbb R are the extended nonnegative functions on XX defined in (10), with empty supremum zero: RU+(y)=supr∈QU(y)max⁡{r,0},RU−(y)=sup−r∈QU(y)max⁡{r,0},ℒU(y)=RU+(y)+RU−(y).\begin{gathered} R_U^+(y)=\sup_{r\in Q_U(y)}\max\{r,0\},\qquad R_U^-(y)=\sup_{-r\in Q_U(y)}\max\{r,0\},\\ \mathcal L_U(y)=R_U^+(y)+R_U^-(y). \end{gathered}(10) The component infimum IUHI_UH associated with a set U⊂X×ℝU\subset X\times\mathbb R and a real function HH on X×ℝX\times\mathbb R is the extended nonnegative function on XX defined by (11), with empty infimum +∞+\infty: IUH(y)=infr∈QU(y)(H(y,r))+.I_UH(y)=\inf_{r\in Q_U(y)}\bigl(H(y,r)\bigr)_+.(11) The component weight JU,H,D,δJ_{U,H,D,\delta} for a set U⊂X×ℝU\subset X\times\mathbb R, real functions H,DH,D on X×ℝX\times\mathbb R, and a real parameter δ\delta is the extended nonnegative function on XX defined in (12): JU,H,D,δ(y)={IUH(y),(y,0)∈U,ℒU(y)<∞,ℒU(y)≤(2δ)+IUD(y),0,otherwise.J_{U,H,D,\delta}(y)= \begin{cases} I_UH(y),& \begin{gathered} (y,0)\in U,\quad \mathcal L_U(y)<\infty,\\ \mathcal L_U(y)\le(2\delta)_+I_UD(y), \end{gathered}\\ 0,&\text{otherwise}. \end{cases}(12)

The source line component weight VfV_f of a real function ff on ℰ\mathcal E is the real-valued function on S×ℰ×ℝS\times\mathcal E\times\mathbb R defined in (13). Here ℛ(a)=a\mathcal R(a)=a for finite a≥0a\ge0 and ℛ(+∞)=0\mathcal R(+\infty)=0: H((e,x,t),s)=Wκ(e,x+se),D((e,x,t),s)=d(x+se),Vf(e,x,t)=ℛ(JUf,H,D,δ(e,x,t)).\begin{gathered} H((e,x,t),s)=W_\kappa(e,x+se),\qquad D((e,x,t),s)=d(x+se),\\ V_f(e,x,t)=\mathcal R\bigl(J_{U_f,H,D,\delta}(e,x,t)\bigr). \end{gathered}(13) The directionally averaged weight loss λ¯f\overline\lambda_f is the real-valued function on ℰ×ℝ\mathcal E\times\mathbb R defined by the real integral in (14); the directional weighted loss Lu,fL_{u,f} is the extended nonnegative function on SS defined by the nonnegative extended integral there: λ¯f(x,t)=∫S(Wκ(e,x)−Vf(e,x,t))dσ(e),Lu,f(e)=∫𝒮f+((Wκ(e,x)−Vf(e,x,t))u(x))+dνf(x,t).\begin{gathered} \overline\lambda_f(x,t)= \int_S\bigl(W_\kappa(e,x)-V_f(e,x,t)\bigr)\,d\sigma(e),\\ L_{u,f}(e)=\int^+_{\mathcal S_f} \bigl((W_\kappa(e,x)-V_f(e,x,t))u(x)\bigr)_+\,d\nu_f(x,t). \end{gathered}(14) For the two sources in (2), set Lf=Lu,fL_f=L_{u,f} and Lg=Lv,gL_g=L_{v,g}; dependence on κ,δ\kappa,\delta is suppressed.

Constants for Positive Solutions

With Lebesgue measure on ℝn\mathbb R^n and a real weight exponent ξ\xi, let the unit sphere area sns_n, Newton normalization DnD_n, ball kernel constant cnc_n, local Newton kernel mass KnK_n, and unit ball weight mass 𝒲n\mathcal W_n be as defined in (15): sn=σ(S),Dn=(n−2)sn,cn=22−n/Dn,Kn=∫B12|x|2−ndx,𝒲n(ξ)=∫B1|x|ξdx.\begin{gathered} s_n=\sigma(S),\qquad D_n=(n-2)s_n,\qquad c_n=2^{2-n}/D_n,\\ K_n=\int_{B_{12}}|x|^{2-n}\,dx,\qquad \mathcal W_n(\xi)=\int_{B_1}|x|^\xi\,dx. \end{gathered}(15) The scaling exponents α,β\alpha,\beta and source mass growth constants Cg,CfC_g,C_f are given by (16); CfC_f uses the exchanged parameters (q,p,B,A)(q,p,B,A): α=2+A+p(2+B)pq−1,β=2+B+q(2+A)pq−1,Cg=[(cn𝒲n(A))(cn𝒲n(B))p]1/(1−pq)cn,Cf=Cg|(p,q,A,B)↦(q,p,B,A).\begin{gathered} \alpha=\frac{2+A+p(2+B)}{pq-1},\qquad \beta=\frac{2+B+q(2+A)}{pq-1},\\ C_g=\frac{\bigl[(c_n\mathcal W_n(A)) (c_n\mathcal W_n(B))^p\bigr]^{1/(1-pq)}}{c_n}, \qquad C_f=C_g\big|_{(p,q,A,B)\mapsto(q,p,B,A)}. \end{gathered}(16) For real parameters C,ρ,sC,\rho,s, the dyadic tail number 𝒯\mathcal T, uniform tail constant TgT_g, source mass constant M*M_* in B4B_4, and local mass constant U*U_* are defined by (17): 𝒯(C,ρ,s)=2ρC2s8s−ρ1−2s−ρ,Tg=𝒯(Cg,n−2,n−2−α),M*=(Cg4n−2−α)++(Cf4n−2−β)+,U*=KnCg8n−2−α+Tg|B4|Dn.\begin{gathered} \mathcal T(C,\rho,s)=2^\rho\frac{C2^s8^{s-\rho}}{1-2^{s-\rho}},\qquad T_g=\mathcal T(C_g,n-2,n-2-\alpha),\\ M_*=(C_g4^{n-2-\alpha})_++(C_f4^{n-2-\beta})_+,\\ U_*=\frac{K_n C_g8^{n-2-\alpha}+T_g|B_4|}{D_n}. \end{gathered}(17)

The Claim

Theorem 1. Let f,u:ℰ→ℝf,u:\mathcal E\to\mathbb R be measurable and nonnegative, with 𝒮f\mathcal S_f open. Suppose 0<κ≤10<\kappa\le1, 0<δ≤120<\delta\le\tfrac12, M≥0M\ge0, P>0P>0, and ∫B4+fdμ≤M\int^+_{B_4}f\,d\mu\le M. Then ∫𝒮f+(u(x)λ¯f(x,t))+dνf(x,t)≤(Λκ,δ(M,P))+∫ℰ+(η(x)f(x)u(x))+dμ(x)+(ℓnP)+∫B2+udμ.\begin{aligned} &\int^+_{\mathcal S_f} \bigl(u(x)\overline\lambda_f(x,t)\bigr)_+\,d\nu_f(x,t)\\ &\quad\le \bigl(\Lambda_{\kappa,\delta}(M,P)\bigr)_+ \int_{\mathcal E}^+\bigl(\eta(x)f(x)u(x)\bigr)_+\,d\mu(x)\\ &\qquad{}+(\ell_nP)_+\int_{B_2}^+u\,d\mu. \end{aligned}(18)✦ For a positive solution (u,v)(u,v) on ℝn\mathbb R^n with n≥2n\ge2 and p,q>0p,q>0, under the same restrictions on κ,δ,P\kappa,\delta,P, one has ∫S+Lf(e)dσ(e)≤(Λκ,δ(M*,P)Ef+ℓnP(U*)+)+.\int_S^+L_f(e)\,d\sigma(e) \le\bigl(\Lambda_{\kappa,\delta}(M_*,P)E_f +\ell_nP(U_*)_+\bigr)_+.(19)✦

In the paper, the result reads: for a positive solution with n≥3n\ge3, p,q>0p,q>0, A,B>−2A,B>-2, pq>1pq>1, 0<κ<10<\kappa<1 and 0<δ<140<\delta<\tfrac14, every P≥1P\ge1 satisfies (20), 𝔼e(Lf(e)+Lg(e))≤(Cκδ+Cκ,δP)E+CP,\mathbb E_e\bigl(L_f(e)+L_g(e)\bigr) \le\left(C_\kappa\delta+\frac{C_{\kappa,\delta}}{P}\right)E+CP,(20) where 𝔼e\mathbb E_e denotes the directional average, all constants are independent of the solution, the constants in the relative coefficient are independent of PP, and CC is independent of κ,δ,P\kappa,\delta,P.

What Is Checked

The layer estimate (18) is stated over measurable nonnegative functions with an open positive source subgraph and the cap and line component weights above; the angular estimate (19) is stated over positive solutions and the explicit equation-dependent constants in (17).