OctoInformBeta

OpenAI preprint, September 2026 · §4. Gaussian heat dissipation

Lemma 4.5: Mesh sums with variance at least the squared step size, energy estimates from weight bounds, and entrywise monotonicity of the cubic array norm

Bounds on Gaussian weight sums and plane heat energy.

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

Take a finite index set ι\iota, real node coordinates on ι\iota, and three complex ι×ι\iota \times \iota matrices weighted by discrete Gaussian masses of step size h>0h > 0 and variances σ0,σ1,σ2>0\sigma_0, \sigma_1, \sigma_2 > 0 centered at real triples summing to a level dd. Assembling those weighted matrices into a 3×33 \times 3 block matrix yields a matrix energy at each of the three vertex rows, and the problem is to bound the plane integral of their sum by the scaled cubic sum of entry moduli. Alongside this energy estimate lie uniform bounds on discrete Gaussian weight sums over grids of spacing hh when variances are at least h2h^2, as well as monotonicity of the cubic entry sum under entrywise shrinkage.

Matrix Energy and Cubic Array Norms

The triangle vertex set Fin3\mathrm{Fin}\, 3 is the set {0,1,2}\{0, 1, 2\}, equipped with cyclic addition v+1v + 1 and v−1=v+2v - 1 = v + 2 (written ℤ/3ℤ\mathbb{Z}/3\mathbb{Z} in the paper).✦ The triangle block space over an index set ι\iota is the Cartesian product Fin3×ι\mathrm{Fin}\, 3 \times \iota. Given an index set ι\iota and a family W=(Wv)v∈Fin3W = (W_v)_{v \in \mathrm{Fin}\, 3} of complex ι×ι\iota \times \iota matrices, the triangle matrix T(W)T(W) on Fin3×ι\mathrm{Fin}\, 3 \times \iota is the block matrix (1) whose (i,j)(i, j) triangle matrix block is 00 for i=ji = j, WiW_i for j=i+1j = i + 1, and Wj*W_j^* for i=j+1i = j + 1: T(W)=(0W0W2*W0*0W1W2W1*0).T(W) = \begin{pmatrix} 0 & W_0 & W_2^* \\ W_0^* & 0 & W_1 \\ W_2 & W_1^* & 0 \end{pmatrix}.(1)

For an index set ι\iota and a vertex v∈Fin3v \in \mathrm{Fin}\, 3, the vertex block projection PvP_v on Fin3×ι\mathrm{Fin}\, 3 \times \iota is the diagonal complex matrix whose diagonal entry at (k,i)∈Fin3×ι(k, i) \in \mathrm{Fin}\, 3 \times \iota is 11 if k=vk = v and 00 otherwise.✦ For a finite index set ι\iota, a family W=(Wv)v∈Fin3W = (W_v)_{v \in \mathrm{Fin}\, 3} of complex ι×ι\iota \times \iota matrices, and v∈Fin3v \in \mathrm{Fin}\, 3, the vertex star row Rv(W)R_v(W) on Fin3×ι\mathrm{Fin}\, 3 \times \iota is the complex matrix PvT(W)P_v T(W).✦ The vertex star Gram matrix Qv(W)Q_v(W) of a family W=(Wv)v∈Fin3W = (W_v)_{v \in \mathrm{Fin}\, 3} of complex ι×ι\iota \times \iota matrices at v∈Fin3v \in \mathrm{Fin}\, 3, for a finite index set ι\iota, is the complex matrix Rv(W)*Rv(W)R_v(W)^* R_v(W) on Fin3×ι\mathrm{Fin}\, 3 \times \iota. The real trace Re⁡tr⁡(M)\operatorname{Re}\operatorname{tr}(M) of a complex κ×κ\kappa \times \kappa matrix MM on a finite index set κ\kappa is the real part of the trace of MM.✦ Given a finite index set κ\kappa and a complex κ×κ\kappa \times \kappa matrix PP, the matrix energy Φ(P)\Phi(P) is the real number given by (2), where P1/2P^{1/2} is the square root of PP: Φ(P)=Re⁡tr⁡(PP1/2).\Phi(P) = \operatorname{Re}\operatorname{tr}(P P^{1/2}).(2)✦

For an index set ι\iota, a family A=(Av)v∈Fin3A = (A_v)_{v \in \mathrm{Fin}\, 3} of complex ι×ι\iota \times \iota matrices, and a family g=(gv)v∈Fin3g = (g_v)_{v \in \mathrm{Fin}\, 3} of real functions gv:ι→ℝg_v : \iota \to \mathbb{R}, the weighted matrix array A(g)=(Av(g))v∈Fin3A^{(g)} = (A^{(g)}_v)_{v \in \mathrm{Fin}\, 3} is the family of complex ι×ι\iota \times \iota matrices defined entrywise by (3): Av(g)(i,j)=gv(i)Av(i,j)gv+1(j)(v∈Fin3,i,j∈ι).A^{(g)}_v(i, j) = \sqrt{g_v(i)}\, A_v(i, j)\, \sqrt{g_{v+1}(j)} \qquad (v \in \mathrm{Fin}\, 3,\; i, j \in \iota).(3)✦

At scale h∈ℝh \in \mathbb{R}, the cubic array norm 𝒮(A)\mathcal{S}(A) of a family A=(Av)v∈Fin3A = (A_v)_{v \in \mathrm{Fin}\, 3} of complex ι×ι\iota \times \iota matrices on a finite index set ι\iota is the real number given by (4): 𝒮(A)=∑v∈Fin3∑i,j∈ιh2|Av(i,j)|3.\mathcal{S}(A) = \sum_{v \in \mathrm{Fin}\, 3} \sum_{i, j \in \iota} h^2 |A_v(i, j)|^3.(4)✦ For an index set ι\iota, a family A=(Av)v∈Fin3A = (A_v)_{v \in \mathrm{Fin}\, 3} of complex ι×ι\iota \times \iota matrices, and a complex ι×ι\iota \times \iota matrix BB, the zero-component replacement A(0←B)A^{(0 \leftarrow B)} is the family of complex ι×ι\iota \times \iota matrices indexed by Fin3\mathrm{Fin}\, 3 equal to BB at 00 and to AvA_v at v≠0v \neq 0.✦

Gaussian Weights and Plane Integrals

The variance-σ\sigma Gaussian density gσ(t)g_\sigma(t) for real σ\sigma and tt is the real value given by (5): gσ(t)=(2πσ)−1/2exp⁡(−t2/(2σ)).g_\sigma(t) = (2\pi\sigma)^{-1/2} \exp\bigl(-t^2 / (2\sigma)\bigr).(5)✦ The Gaussian weight w(h,u,p,σ)w(h, u, p, \sigma) for real hh, uu, pp, and σ\sigma is the real number hgσ(u−p)h\, g_\sigma(u - p).✦ Given an index set ι\iota, h∈ℝh \in \mathbb{R}, node coordinates u:ι→ℝu : \iota \to \mathbb{R}, centers p:Fin3→ℝp : \mathrm{Fin}\, 3 \to \mathbb{R}, and variances σ:Fin3→ℝ\sigma : \mathrm{Fin}\, 3 \to \mathbb{R}, the vertex heat weights w=(wv)v∈Fin3w = (w_v)_{v \in \mathrm{Fin}\, 3} form the family of real functions on ι\iota defined by (6): wv(i)=w(h,ui,pv,σv)=hgσv(ui−pv).w_v(i) = w(h, u_i, p_v, \sigma_v) = h\, g_{\sigma_v}(u_i - p_v).(6) The mesh mass constant M0M_0 is the real number defined by (7): M0=2e1/2.M_0 = 2 e^{1/2}.(7)✦

For an index set ι\iota, a family A=(Av)v∈Fin3A = (A_v)_{v \in \mathrm{Fin}\, 3} of complex ι×ι\iota \times \iota matrices, h∈ℝh \in \mathbb{R}, u:ι→ℝu : \iota \to \mathbb{R}, and p,σ:Fin3→ℝp, \sigma : \mathrm{Fin}\, 3 \to \mathbb{R}, the heat matrix array H(A,h,u,p,σ)H(A, h, u, p, \sigma) is the weighted matrix array A(w)A^{(w)} formed from AA and the vertex heat weights w=(wv)v∈Fin3w = (w_v)_{v \in \mathrm{Fin}\, 3} of (6).✦ For a finite index set ι\iota, a family A=(Av)v∈Fin3A = (A_v)_{v \in \mathrm{Fin}\, 3} of complex ι×ι\iota \times \iota matrices, h∈ℝh \in \mathbb{R}, u:ι→ℝu : \iota \to \mathbb{R}, p,σ:Fin3→ℝp, \sigma : \mathrm{Fin}\, 3 \to \mathbb{R}, and v∈Fin3v \in \mathrm{Fin}\, 3, the heat matrix energy ev(A,h,u,p,σ)e_v(A, h, u, p, \sigma) is the matrix energy of the vertex star Gram matrix of H(A,h,u,p,σ)H(A, h, u, p, \sigma) at vv, given by (8): ev(A,h,u,p,σ)=Φ(Qv(H(A,h,u,p,σ))).e_v(A, h, u, p, \sigma) = \Phi\bigl(Q_v(H(A, h, u, p, \sigma))\bigr).(8)✦

At sum level d∈ℝd \in \mathbb{R}, the constant-sum center plane Πd\Pi_d is the subset {p∈ℝFin3:p0+p1+p2=d}\{p \in \mathbb{R}^{\mathrm{Fin}\, 3} : p_0 + p_1 + p_2 = d\} of ℝFin3\mathbb{R}^{\mathrm{Fin}\, 3}. Given i∈Fin3i \in \mathrm{Fin}\, 3, a sum level d∈ℝd \in \mathbb{R}, and z=(z1,z2)∈ℝ2z = (z_1, z_2) \in \mathbb{R}^2, the plane coordinate chart value ϕi,d(z)\phi_{i, d}(z) is the triple in ℝFin3\mathbb{R}^{\mathrm{Fin}\, 3} whose entries sum to dd, equal to d−z1−z2d - z_1 - z_2 at coordinate ii and to z1z_1 and z2z_2 at the remaining two vertices in increasing order.✦ The plane integral ℐd(f)\mathcal{I}_d(f) of a function f:ℝFin3→ℝf : \mathbb{R}^{\mathrm{Fin}\, 3} \to \mathbb{R} at sum level d∈ℝd \in \mathbb{R} is the Lebesgue integral over ℝ2\mathbb{R}^2 given by (9): ℐd(f)=∫ℝ2f(ϕ2,d(z))dz.\mathcal{I}_d(f) = \int_{\mathbb{R}^2} f\bigl(\phi_{2, d}(z)\bigr)\, dz.(9)✦ The total plane heat energy ℰd(A,h,u,σ)\mathcal{E}_d(A, h, u, \sigma) of a family A=(Av)v∈Fin3A = (A_v)_{v \in \mathrm{Fin}\, 3} of complex ι×ι\iota \times \iota matrices on a finite index set ι\iota, with h∈ℝh \in \mathbb{R}, u:ι→ℝu : \iota \to \mathbb{R}, d∈ℝd \in \mathbb{R}, and σ:Fin3→ℝ\sigma : \mathrm{Fin}\, 3 \to \mathbb{R}, is the real number defined by (10): ℰd(A,h,u,σ)=∑v∈Fin3ℐd(p↦ev(A,h,u,p,σ)).\mathcal{E}_d(A, h, u, \sigma) = \sum_{v \in \mathrm{Fin}\, 3} \mathcal{I}_d\bigl(p \mapsto e_v(A, h, u, p, \sigma)\bigr).(10)✦

The Claim

Theorem 1. Let h>0h > 0, σ≥h2\sigma \ge h^2, c∈ℝc \in \mathbb{R}, and let n≥0n \ge 0 be an integer. Then the mesh sum satisfies (11): ∑i=0n−1hgσ(c+hi)≤M0.\sum_{i=0}^{n-1} h\, g_\sigma(c + h i) \le M_0.(11)✦ Given a finite subset D⊂ℤD \subset \mathbb{Z}, a step size h>0h > 0, a variance σ≥h2\sigma \ge h^2, and a center p∈ℝp \in \mathbb{R}, the weight sum over DD obeys (12): ∑i∈Dw(h,hi,p,σ)≤M0.\sum_{i \in D} w(h, h i, p, \sigma) \le M_0.(12)✦ When ι\iota is a finite index set, A=(Av)v∈Fin3A = (A_v)_{v \in \mathrm{Fin}\, 3} is a family of complex ι×ι\iota \times \iota matrices, h>0h > 0, u:ι→ℝu : \iota \to \mathbb{R}, d∈ℝd \in \mathbb{R}, σv>0\sigma_v > 0 for every v∈Fin3v \in \mathrm{Fin}\, 3, and M≥0M \ge 0 satisfies ∑i∈ιwv(i)≤M\sum_{i \in \iota} w_v(i) \le M for every center triple p∈ℝFin3p \in \mathbb{R}^{\mathrm{Fin}\, 3} and every v∈Fin3v \in \mathrm{Fin}\, 3 with vertex heat weights wv(i)w_v(i) determined by h,u,p,σh, u, p, \sigma, then (13) holds: ℰd(A,h,u,σ)≤22M𝒮(A).\mathcal{E}_d(A, h, u, \sigma) \le 2\sqrt{2}\, M\, \mathcal{S}(A).(13)✦ When D⊂ℤD \subset \mathbb{Z} is finite, A=(Av)v∈Fin3A = (A_v)_{v \in \mathrm{Fin}\, 3} is a family of complex D×DD \times D matrices, h>0h > 0, d∈ℝd \in \mathbb{R}, σv≥h2\sigma_v \ge h^2 for every v∈Fin3v \in \mathrm{Fin}\, 3, and ui=hiu_i = h i for i∈Di \in D, the total plane heat energy obeys (14): ℰd(A,h,u,σ)≤22M0𝒮(A).\mathcal{E}_d(A, h, u, \sigma) \le 2\sqrt{2}\, M_0\, \mathcal{S}(A).(14)✦ Take a finite index set ι\iota, any h∈ℝh \in \mathbb{R}, and two families A=(Av)v∈Fin3A = (A_v)_{v \in \mathrm{Fin}\, 3} and B=(Bv)v∈Fin3B = (B_v)_{v \in \mathrm{Fin}\, 3} of complex ι×ι\iota \times \iota matrices satisfying |Av(i,j)|≤|Bv(i,j)||A_v(i, j)| \le |B_v(i, j)| for all v∈Fin3v \in \mathrm{Fin}\, 3 and i,j∈ιi, j \in \iota; then at scale hh the cubic array norms satisfy (15): 𝒮(A)≤𝒮(B).\mathcal{S}(A) \le \mathcal{S}(B).(15)✦ If D⊂ℤD \subset \mathbb{Z} is finite, A=(Av)v∈Fin3A = (A_v)_{v \in \mathrm{Fin}\, 3} is a family of complex D×DD \times D matrices, BB is a complex D×DD \times D matrix with |B(i,j)|≤|A0(i,j)||B(i, j)| \le |A_0(i, j)| for all i,j∈Di, j \in D, and h∈ℝh \in \mathbb{R}, then at scale hh the zero-component replacement obeys (16): 𝒮(A(0←B))≤𝒮(A).\mathcal{S}\bigl(A^{(0 \leftarrow B)}\bigr) \le \mathcal{S}(A).(16)✦

In the paper, the result reads: for h>0h > 0, an integer N≥0N \ge 0, finite complex arrays with entries Av(i,j)A_v(i, j), any d∈ℝd \in \mathbb{R}, and variances σv≥h2\sigma_v \ge h^2, one has ∫Πd∑v∈Fin3evdπd≤C0𝒮(A)\int_{\Pi_d} \sum_{v \in \mathrm{Fin}\, 3} e_v\, d\pi_d \le C_0 \mathcal{S}(A) with C0=22(1+2/π)C_0 = 2\sqrt{2}\bigl(1 + \sqrt{2/\pi}\bigr) (where eve_v stands for the heat matrix energy and πd\pi_d for the measure on Πd\Pi_d), and the same bound holds if any entries are replaced by smaller entries in absolute value before forming the weighted matrices. In the stated theorem, (14) uses ℐd\mathcal{I}_d in place of integration against πd\pi_d and M0M_0 in place of 1+2/π1 + \sqrt{2/\pi} on arbitrary finite subsets D⊂ℤD \subset \mathbb{Z}, while (15) and (16) state entrywise shrinkage directly for 𝒮(A)\mathcal{S}(A).

What Is Checked

The claims (11) through (16) are stated over finite index sets ι\iota and finite subsets D⊂ℤD \subset \mathbb{Z}, complex matrix families indexed by Fin3\mathrm{Fin}\, 3, shifts c∈ℝc \in \mathbb{R}, integers n≥0n \ge 0, centers pp, node coordinates uu, sum levels dd, the Gaussian density gσg_\sigma and weights ww and wvw_v, the mesh mass constant M0M_0, the total plane heat energy ℰd\mathcal{E}_d, the cubic array norm 𝒮\mathcal{S}, and zero-component replacement, under the hypotheses h>0h > 0 with σ≥h2\sigma \ge h^2 or σv>0\sigma_v > 0 and a uniform bound M≥0M \ge 0 on the weight sums, arbitrary h∈ℝh \in \mathbb{R} in the cubic comparisons, and entrywise modulus dominance.