OctoInformBeta

OpenAI preprint, September 2026 · §3. Finite affine potentials and prescribed cell masses

Proposition 3.1: Joint regularity of cell masses and directional formulas for moments

Cell variations and the canonical Laplacian.

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

Prescribing the masses of a finitely supported target measure leads to the response of Laguerre cells when their sites and intercepts vary. Let K⊂ℝdK\subset\mathbb R^d be a compact convex body of positive Lebesgue volume |K||K|, and let II be a finite nonempty index set. The uniform measure ρ\rho on KK is normalized Lebesgue measure, as in (1). ρ=1|K|ℒd↾K.\rho=\frac{1}{|K|}\mathcal L^d\!\restriction_K.(1)✦

For sites yi∈ℝdy_i\in\mathbb R^d and intercepts ci∈ℝc_i\in\mathbb R, the finite affine potential uu is the function on ℝd\mathbb R^d given by the maximum in (2), and its closed Laguerre cell PiP_i is the subset of KK where the affine function ℓi\ell_i attains that maximum; ties belong to every maximizing cell. ℓi(x)=x⋅yi+ci,u(x)=maxi∈Iℓi(x),Pi={x∈K:ℓj(x)≤ℓi(x) for every j∈I}.\ell_i(x)=x\cdot y_i+c_i,\qquad u(x)=\max_{i\in I}\ell_i(x),\qquad P_i=\{x\in K:\ell_j(x)\leq\ell_i(x)\text{ for every }j\in I\}.(2)✦ The normalized cell mass MiM_i and first moment BiB_i are the integrals in (3). Mi(y,c)=ρ(Pi),Bi(y,c)=∫Pixdρ(x),ℳ(y,c)=(Mi(y,c))i∈I.M_i(y,c)=\rho(P_i),\qquad B_i(y,c)=\int_{P_i}x\,d\rho(x),\qquad \mathcal M(y,c)=(M_i(y,c))_{i\in I}.(3) A positive mass configuration is a pair (y,c)(y,c) with distinct sites and strictly positive normalized cell masses, as specified by (4). 𝒰(K)={(y,c):yi≠yj for i≠j,Mi(y,c)>0 for every i∈I}.\mathcal U(K)=\{(y,c):y_i\neq y_j\text{ for }i\neq j, \ M_i(y,c)>0\text{ for every }i\in I\}.(4)✦

Face Measures and Variations

For a linear subspace W⊂ℝdW\subset\mathbb R^d and a basepoint zz, the intrinsic affine plane measure λz+W\lambda_{z+W} is the pushforward of Lebesgue measure on WW under translation, as in (5). λz+W=(a↦z+a)#ℒW.\lambda_{z+W}=(a\mapsto z+a)_{\#}\mathcal L_W.(5)✦ The tangent space TnT_n, canonical hyperplane basepoint z(n,s)z(n,s) and intrinsic hyperplane measure μn,s\mu_{n,s} associated with a nonzero normal nn and offset ss are defined by (6). Tn=n⟂={a:a⋅n=0},z(n,s)=−s∥n∥2n,μn,s=λz(n,s)+Tn.T_n=n^\perp=\{a:a\cdot n=0\},\qquad z(n,s)=-\frac{s}{\|n\|^2}n,\qquad \mu_{n,s}=\lambda_{z(n,s)+T_n}.(6) For distinct sites, the intrinsic interface measure σij\sigma_{ij} is the hyperplane measure with normal yi−yjy_i-y_j and offset ci−cjc_i-c_j.✦ The canonical face weight gijg_{ij} and normalized face first moment bijb_{ij} of the common closed face FijF_{ij} are respectively the scalar and vector integrals in (7). Fij=Pi∩Pj,σij=μyi−yj,ci−cj,gij=σij(Fij)|K|∥yi−yj∥,bij=1|K|∥yi−yj∥∫Fijxdσij(x).\begin{gathered} F_{ij}=P_i\cap P_j,\qquad \sigma_{ij}=\mu_{y_i-y_j,c_i-c_j},\\ g_{ij}=\frac{\sigma_{ij}(F_{ij})}{|K|\|y_i-y_j\|},\qquad b_{ij}=\frac{1}{|K|\|y_i-y_j\|}\int_{F_{ij}}x\,d\sigma_{ij}(x). \end{gathered}(7) The canonical cell Laplacian GG is the operator on ℝI\mathbb R^I defined by (8). (Gq)i=∑j≠igij(qi−qj).(Gq)_i=\sum_{j\neq i}g_{ij}(q_i-q_j).(8)✦

Let vi∈ℝdv_i\in\mathbb R^d and ri∈ℝr_i\in\mathbb R be site and intercept directions. The nonnegative cut indicator χ+\chi_+ equals one on [0,∞)[0,\infty) and zero on (−∞,0)(-\infty,0).✦ For a scalar weight ww and a point z∈ℝdz\in\mathbb R^d, the signed face variation 𝒱ijw(z)\mathcal V_{ij}^w(z) is the tangent integral in (9), where 𝟏K\mathbf 1_K is the indicator function of KK. 𝒱ijw(z)=∫Tyi−yjri−rj+(vi−vj)⋅(z+a)∥yi−yj∥𝟏K(z+a)w(z+a)×∏k∈I\{i,j}χ+(ℓi(z+a)−ℓk(z+a))dℒTyi−yj(a).\begin{aligned} \mathcal V_{ij}^w(z) =\int_{T_{y_i-y_j}}& \frac{r_i-r_j+(v_i-v_j)\cdot(z+a)}{\|y_i-y_j\|} \mathbf 1_K(z+a)w(z+a)\\ &{}\times\prod_{k\in I\setminus\{i,j\}} \chi_+\bigl(\ell_i(z+a)-\ell_k(z+a)\bigr) \,d\mathcal L_{T_{y_i-y_j}}(a). \end{aligned}(9)✦

The Claim

Theorem 1. Suppose that the sites are distinct and that ν\nu is a measure absolutely continuous with respect to Lebesgue measure, with ν(Pi)>0\nu(P_i)>0 for every ii. For every site direction vv, intercept direction rr and index ii, ddtMi(y+tv,c+tr)|t=0=∑j≠i[gij(ri−rj)+bij⋅(vi−vj)].\left.\frac{d}{dt}M_i(y+tv,c+tr)\right|_{t=0} =\sum_{j\neq i}\bigl[g_{ij}(r_i-r_j) +b_{ij}\cdot(v_i-v_j)\bigr].(10)✦ Fix ii and choose, for every j≠ij\neq i, a point zj∈int⁡Kz_j\in\operatorname{int}K such that ℓi(zj)=ℓj(zj)\ell_i(z_j)=\ell_j(z_j). Then ddtBi(y+tv,c+tr)|t=0=1|K|(∑j≠i𝒱ijx↦xk(zj))k=1d.\left.\frac{d}{dt}B_i(y+tv,c+tr)\right|_{t=0} =\frac{1}{|K|} \left(\sum_{j\neq i}\mathcal V_{ij}^{\,x\mapsto x_k}(z_j) \right)_{k=1}^d.(11)✦ The joint mass map satisfies ℳ∈C1(𝒰(K);ℝI).\mathcal M\in C^1\bigl(\mathcal U(K);\mathbb R^I\bigr).(12)✦ There is a connected simple graph HH on II whose edges have positive canonical face weights, as in (13); the canonical cell Laplacian has the kernel in (14), where 𝟏\mathbf 1 is the vector with all coordinates equal to one. ∃ a connected simple graph H on Isuch that{i,j}∈E(H)⇒gij>0,\exists\text{ a connected simple graph }H\text{ on }I \quad\text{such that}\quad \{i,j\}\in E(H)\ \Longrightarrow\ g_{ij}>0,(13)✦ ker⁡G={a𝟏:a∈ℝ}.\ker G=\{a\mathbf 1:a\in\mathbb R\}.(14)✦ For real vectors m,h∈ℝIm,h\in\mathbb R^I with ∑imi=1\sum_i m_i=1 and ∑ihi=0\sum_i h_i=0, there is a unique vector qq satisfying the equation and normalization in (15). ∃!q∈ℝI:Gq=h,∑i∈Imiqi=0.\exists!\,q\in\mathbb R^I:\qquad Gq=h,\qquad \sum_{i\in I}m_iq_i=0.(15)✦

In the paper, the result reads: for a general compact convex source KK, on the open locus of distinct sites with Mi>0M_i>0, all MiM_i and BiB_i are jointly C1C^1 in sites and intercepts, with the directional formulas (16) for every intercept direction rr and site direction vv, DMi[r,v]=∑j≠i[gij(ri−rj)+bij⋅(vi−vj)],DBi[r,v]=∑j≠ibij(ri−rj)+1|K|∑j≠i1∥yi−yj∥∫Fijx(x⋅(vi−vj))dσij(x),\begin{aligned} DM_i[r,v] &=\sum_{j\neq i}\bigl[g_{ij}(r_i-r_j) +b_{ij}\cdot(v_i-v_j)\bigr],\\ DB_i[r,v] &=\sum_{j\neq i}b_{ij}(r_i-r_j) +\frac{1}{|K|}\sum_{j\neq i}\frac{1}{\|y_i-y_j\|} \int_{F_{ij}}x\bigl(x\cdot(v_i-v_j)\bigr)\,d\sigma_{ij}(x), \end{aligned}(16) and the graph with edges gij>0g_{ij}>0 is connected, while the symmetric operator GG has kernel ℝ𝟏\mathbb R\mathbf 1 and range {h∈ℝI:∑ihi=0}\{h\in\mathbb R^I:\sum_i h_i=0\}.

What Is Checked

The directional assertions (10) and (11) concern the normalized Lebesgue integrals (3), with positivity supplied by an arbitrary absolutely continuous measure ν\nu. The vector moment assertion uses the signed tangent integrals (9) with interior pairwise tie points. The joint regularity statement (12) concerns the mass map on the positive mass configurations (4).

The graph and operator assertions (13), (14) and (15) apply to the canonical coefficients (7) and their Laplacian (8). The normalized solution statement allows arbitrary real weights mm whose coordinate sum is one.