OctoInformBeta

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

Lemma 4.2: Pointwise derivatives in centers and positive variances together with variance and dissipation identities integrated over constant-sum center planes

Gaussian heat derivatives and plane dissipation for star energies.

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 an index set ι\iota, a family A=(Av)v∈ℤ/3ℤA = (A_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of complex square matrices indexed by ι\iota on the edges of a triangle whose vertices form the triangle vertex set ℤ/3ℤ\mathbb{Z}/3\mathbb{Z} (the three-element set {0,1,2}\{0, 1, 2\} under cyclic addition, so v−1=v+2v - 1 = v + 2), real node coordinates u:ι→ℝu : \iota \to \mathbb{R}, and a step size h>0h > 0. Modulating the matrix entries by Gaussian heat weights with center pvp_v and variance σv>0\sigma_v > 0 at the nodes uiu_i (i∈ιi \in \iota) produces a star Gram matrix at each vertex v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z}, whose matrix energy varies with the center vector p=(p0,p1,p2)p = (p_0, p_1, p_2) and variance vector σ=(σ0,σ1,σ2)\sigma = (\sigma_0, \sigma_1, \sigma_2).

The problem is to express the first and second partial derivatives in the centers and the derivatives in the variances (and along positive rays σv=αvs\sigma_v = \alpha_v s with αv,s>0\alpha_v, s > 0) of this vertex star energy in terms of the matrix energy gradient and Hessian quadratic cost, and to relate the variance derivative to the center second derivatives and the Hessian cost after integrating over a constant-sum center plane.

Matrix Energy and Hessian Costs

The real trace tr⁡ℝ(M)\operatorname{tr}_{\mathbb{R}}(M) of a complex square matrix MM indexed by an index set 𝒥\mathcal{J} is the real number Re⁡tr⁡(M)\operatorname{Re}\operatorname{tr}(M). The real Hilbert-Schmidt inner product ⟨X,Y⟩HS\langle X, Y \rangle_{\mathrm{HS}} of two complex square matrices XX and YY indexed by an index set 𝒥\mathcal{J} is the real number tr⁡ℝ(X*Y)\operatorname{tr}_{\mathbb{R}}(X^* Y) (equal to Re⁡tr⁡(X*Y)\operatorname{Re}\operatorname{tr}(X^* Y)).✦ The matrix energy Φ(P)\Phi(P) of a complex square matrix PP indexed by 𝒥\mathcal{J} with matrix square root P1/2P^{1/2} is the real number Φ(P)=Re⁡tr⁡(PP1/2).\Phi(P) = \operatorname{Re}\operatorname{tr}(P P^{1/2}).(1)✦ At a complex square matrix PP indexed by 𝒥\mathcal{J} with square root P1/2P^{1/2}, the energy gradient ∇Φ(P)\nabla\Phi(P) is the continuous real-linear functional on complex square matrices indexed by 𝒥\mathcal{J} that sends each matrix KK to ∇Φ(P)(K)=32Re⁡tr⁡(P1/2K).\nabla\Phi(P)(K) = \frac{3}{2}\operatorname{Re}\operatorname{tr}(P^{1/2} K).(2)✦

For a complex square matrix RR indexed by an index set 𝒥\mathcal{J}, the Sylvester operator SRS_R is the operator mapping each complex square matrix XX indexed by 𝒥\mathcal{J} to SR(X)=RX+XRS_R(X) = R X + X R✦, and the Sylvester penalty ρR(X)\rho_R(X) of RR and XX is the real number ⟨X,SR(X)⟩HS\langle X, S_R(X) \rangle_{\mathrm{HS}}.✦ Given complex square matrices RR, KK, and XX indexed by 𝒥\mathcal{J}, the Sylvester variational functional ΥR,K(X)\Upsilon_{R,K}(X) of RR, KK, and XX is the real number 2⟨K,X⟩HS−ρR(X)2\langle K, X \rangle_{\mathrm{HS}} - \rho_R(X)✦, and the root variational cost croot(R,K)c_{\mathrm{root}}(R, K) of RR and KK is the real number croot(R,K)=32supXΥR,K(X),c_{\mathrm{root}}(R, K) = \frac{3}{2}\sup_{X} \Upsilon_{R,K}(X),(3)✦ where the supremum ranges over all complex square matrices XX indexed by 𝒥\mathcal{J}. For complex square matrices PP, KK, and LL indexed by 𝒥\mathcal{J}, the Hessian quadratic cost bP(K)b_P(K) of PP and KK is the real number croot(P1/2,K)c_{\mathrm{root}}(P^{1/2}, K)✦, and the polarized Hessian pairing ℬP(K,L)\mathcal{B}_P(K, L) of KK and LL at PP is the real number ℬP(K,L)=14(bP(K+L)−bP(K−L)).\mathcal{B}_P(K, L) = \frac{1}{4}\bigl(b_P(K + L) - b_P(K - L)\bigr).(4)✦

The triangle block space on an index set ι\iota is the Cartesian product ℤ/3ℤ×ι\mathbb{Z}/3\mathbb{Z} \times \iota.✦ For a family W=(Wv)v∈ℤ/3ℤW = (W_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of complex square matrices indexed by ι\iota, the triangle matrix T(W)T(W) is the complex square matrix on ℤ/3ℤ×ι\mathbb{Z}/3\mathbb{Z} \times \iota whose ι×ι\iota \times \iota block in position (i,j)(i, j) (for i,j∈ℤ/3ℤi, j \in \mathbb{Z}/3\mathbb{Z}) equals 00 when i=ji = j, equals WiW_i when j=i+1j = i + 1, and equals Wj*W_j^* when 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}.(5)✦ For each vertex v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z}, the vertex block projection 𝒫v\mathcal{P}_v is the diagonal complex matrix on ℤ/3ℤ×ι\mathbb{Z}/3\mathbb{Z} \times \iota whose diagonal entry at (a,i)∈ℤ/3ℤ×ι(a, i) \in \mathbb{Z}/3\mathbb{Z} \times \iota is 11 when a=va = v and 00 otherwise.✦ Given a family W=(Wv)v∈ℤ/3ℤW = (W_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of complex square matrices indexed by ι\iota and a vertex v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z}, the vertex star row Rv(W)R_v(W) of WW at vv is the complex matrix 𝒫vT(W)\mathcal{P}_v T(W) on ℤ/3ℤ×ι\mathbb{Z}/3\mathbb{Z} \times \iota✦, and the vertex star Gram matrix Gv(W)G_v(W) of WW at vv is the complex matrix Rv(W)*Rv(W)R_v(W)^* R_v(W).

Given a family A=(Av)v∈ℤ/3ℤA = (A_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of complex square matrices indexed by ι\iota and a family g=(gv)v∈ℤ/3ℤg = (g_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of real node weights on ι\iota, the weighted matrix array A(g)A^{(g)} is the family of complex square matrices indexed by ι\iota whose (i,j)(i, j) entry at v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z} (for i,j∈ιi, j \in \iota) is (A(g))v(i,j)=gv(i)Av(i,j)gv+1(j).\bigl(A^{(g)}\bigr)_v(i, j) = \sqrt{g_v(i)}\, A_v(i, j)\, \sqrt{g_{v+1}(j)}.(6)✦ For a family g=(gv)v∈ℤ/3ℤg = (g_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of real node weights on ι\iota, the vertex root matrix DgD_g is the diagonal complex matrix on ℤ/3ℤ×ι\mathbb{Z}/3\mathbb{Z} \times \iota with diagonal entry ga(i)\sqrt{g_a(i)} at each (a,i)∈ℤ/3ℤ×ι(a, i) \in \mathbb{Z}/3\mathbb{Z} \times \iota.✦ For a family A=(Av)v∈ℤ/3ℤA = (A_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of complex square matrices indexed by ι\iota, a family g=(gv)v∈ℤ/3ℤg = (g_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of real node weights on ι\iota, and a vertex v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z}, the vertex star base row of AA and gg at vv is the complex matrix Rv(A)DgR_v(A) D_g on ℤ/3ℤ×ι\mathbb{Z}/3\mathbb{Z} \times \iota✦, and the vertex star base column of AA and gg at vv is the complex matrix Rv(A)*DgR_v(A)^* D_g on ℤ/3ℤ×ι\mathbb{Z}/3\mathbb{Z} \times \iota.✦ The weighted Gram matrix Gram⁡(B,ω)\operatorname{Gram}(B, \omega) of a complex square matrix BB indexed by an index set 𝒥\mathcal{J} and a real weight function ω:𝒥→ℝ\omega : \mathcal{J} \to \mathbb{R} is the complex matrix B*diag⁡(ω)BB^* \operatorname{diag}(\omega) B.✦

Gaussian Heat Weights and Plane Integrals

For a real variance σ\sigma and a real argument tt, the Gaussian density gσ(t)g_\sigma(t) is the real number gσ(t)=(2πσ)−1/2exp⁡(−t22σ).g_\sigma(t) = (2\pi\sigma)^{-1/2} \exp\!\left(-\frac{t^2}{2\sigma}\right).(7)✦ For real numbers uu, pp, and σ\sigma with σ>0\sigma > 0, the Gaussian score η(u,p,σ)\eta(u, p, \sigma) is the real number (u−p)/σ(u - p)/\sigma✦, and for a real step size hh, the Gaussian weight wt(h,u,p,σ)\mathrm{wt}(h, u, p, \sigma) is the real number hgσ(u−p)h\, g_\sigma(u - p).✦ Given a step size h∈ℝh \in \mathbb{R}, node coordinates u:ι→ℝu : \iota \to \mathbb{R}, centers p∈ℝℤ/3ℤp \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}}, and variances σ∈ℝℤ/3ℤ\sigma \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}}, the vertex heat weights form the family w=(wv)v∈ℤ/3ℤw = (w_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of real functions on ι\iota given at each v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z} and i∈ιi \in \iota by wv(i)=wt(h,ui,pv,σv)=hgσv(ui−pv).w_v(i) = \mathrm{wt}(h, u_i, p_v, \sigma_v) = h\, g_{\sigma_v}(u_i - p_v).(8)✦

Given a step size h∈ℝh \in \mathbb{R}, coordinates ũ:𝒥→ℝ\tilde{u} : \mathcal{J} \to \mathbb{R} on an index set 𝒥\mathcal{J}, a group assignment ℓ:𝒥→ν\ell : \mathcal{J} \to \nu to a group index set ν\nu, centers p̃:ν→ℝ\tilde{p} : \nu \to \mathbb{R}, and variances σ̃:ν→ℝ\tilde{\sigma} : \nu \to \mathbb{R}, the grouped heat weight ω(0)\omega^{(0)} is the real function on 𝒥\mathcal{J} defined at each a∈𝒥a \in \mathcal{J} by ω(0)(a)=wt(h,ũa,p̃ℓ(a),σ̃ℓ(a)).\omega^{(0)}(a) = \mathrm{wt}\bigl(h, \tilde{u}_a, \tilde{p}_{\ell(a)}, \tilde{\sigma}_{\ell(a)}\bigr).(9)✦ Given h∈ℝh \in \mathbb{R}, ũ:𝒥→ℝ\tilde{u} : \mathcal{J} \to \mathbb{R}, ℓ:𝒥→ν\ell : \mathcal{J} \to \nu, p̃,σ̃:ν→ℝ\tilde{p}, \tilde{\sigma} : \nu \to \mathbb{R}, and a group v∈νv \in \nu, the grouped first variation weight ωv(1)\omega^{(1)}_v and grouped second variation weight ωv(2)\omega^{(2)}_v are the real functions on 𝒥\mathcal{J} that vanish at a∈𝒥a \in \mathcal{J} when ℓ(a)≠v\ell(a) \neq v and satisfy ωv(1)(a)=η(ũa,p̃v,σ̃v)wt(h,ũa,p̃v,σ̃v),ωv(2)(a)=(η(ũa,p̃v,σ̃v)2−σ̃v−1)wt(h,ũa,p̃v,σ̃v)\omega^{(1)}_v(a) = \eta(\tilde{u}_a, \tilde{p}_v, \tilde{\sigma}_v)\, \mathrm{wt}(h, \tilde{u}_a, \tilde{p}_v, \tilde{\sigma}_v), \qquad \omega^{(2)}_v(a) = \bigl(\eta(\tilde{u}_a, \tilde{p}_v, \tilde{\sigma}_v)^2 - \tilde{\sigma}_v^{-1}\bigr)\, \mathrm{wt}(h, \tilde{u}_a, \tilde{p}_v, \tilde{\sigma}_v)(10) when ℓ(a)=v\ell(a) = v.

For a family A=(Av)v∈ℤ/3ℤA = (A_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of complex square matrices indexed by ι\iota, step size h∈ℝh \in \mathbb{R}, node coordinates u:ι→ℝu : \iota \to \mathbb{R}, centers p∈ℝℤ/3ℤp \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}}, and variances σ∈ℝℤ/3ℤ\sigma \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}}, the heat matrix array Ap,σA_{p,\sigma} is the weighted matrix array A(w)A^{(w)} formed from AA and the vertex heat weights ww✦, and for each v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z} the heat matrix energy ev(p,σ)e_v(p, \sigma) is the real number ev(p,σ)=Φ(Gv(Ap,σ)).e_v(p, \sigma) = \Phi\bigl(G_v(A_{p,\sigma})\bigr).(11)✦ For a family A=(Av)v∈ℤ/3ℤA = (A_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of complex square matrices indexed by ι\iota, a family g=(gv)v∈ℤ/3ℤg = (g_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of real node weights on ι\iota, and vertices v,w∈ℤ/3ℤv, w \in \mathbb{Z}/3\mathbb{Z}, the fixed base matrix Bv,w(A,g)B_{v,w}(A, g) is the complex matrix on ℤ/3ℤ×ι\mathbb{Z}/3\mathbb{Z} \times \iota that equals Rv(A)DgR_v(A) D_g when w=vw = v and Rv(A)*DgR_v(A)^* D_g when w≠vw \neq v.✦ For AA, hh, uu, pp, σ\sigma, and v,w∈ℤ/3ℤv, w \in \mathbb{Z}/3\mathbb{Z}, write B=Bv,w(A,w)B = B_{v,w}(A, w) with the vertex heat weights ww, and let ω(0),ωw(1),ωw(2)\omega^{(0)}, \omega^{(1)}_w, \omega^{(2)}_w be the grouped weights on ℤ/3ℤ×ι\mathbb{Z}/3\mathbb{Z} \times \iota for group set ℤ/3ℤ\mathbb{Z}/3\mathbb{Z}, coordinates (a,i)↦ui(a, i) \mapsto u_i, group assignment (a,i)↦a(a, i) \mapsto a, centers pp, and variances σ\sigma; the local Gram matrix Pv,w(p,σ)P_{v,w}(p, \sigma), local Gram first variation Kv,w(p,σ)K_{v,w}(p, \sigma), and local Gram second variation Mv,w(p,σ)M_{v,w}(p, \sigma) are the complex square matrices on ℤ/3ℤ×ι\mathbb{Z}/3\mathbb{Z} \times \iota given by Pv,w(p,σ)=Gram⁡(B,ω(0)),Kv,w(p,σ)=Gram⁡(B,ωw(1)),Mv,w(p,σ)=Gram⁡(B,ωw(2)).P_{v,w}(p, \sigma) = \operatorname{Gram}(B, \omega^{(0)}), \qquad K_{v,w}(p, \sigma) = \operatorname{Gram}(B, \omega^{(1)}_w), \qquad M_{v,w}(p, \sigma) = \operatorname{Gram}(B, \omega^{(2)}_w).(12)

For A,h,u,p,σA, h, u, p, \sigma and vertices v,w,z∈ℤ/3ℤv, w, z \in \mathbb{Z}/3\mathbb{Z}, writing Pv,w=Pv,w(p,σ)P_{v,w} = P_{v,w}(p, \sigma), Kv,w=Kv,w(p,σ)K_{v,w} = K_{v,w}(p, \sigma), and Mv,w=Mv,w(p,σ)M_{v,w} = M_{v,w}(p, \sigma), the heat first variation Fv,w(p,σ)F_{v,w}(p, \sigma), heat variance term Λv,w(p,σ)\Lambda_{v,w}(p, \sigma), heat Hessian cost 𝒞v,w(p,σ)\mathcal{C}_{v,w}(p, \sigma), pure heat second variation Qv,w(p,σ)Q_{v,w}(p, \sigma), and mixed heat Hessian pairing ℋv,w,z(p,σ)\mathcal{H}_{v,w,z}(p, \sigma) are the real numbers defined by Fv,w(p,σ)=∇Φ(Pv,w)(Kv,w),Λv,w(p,σ)=12∇Φ(Pv,w)(Mv,w),𝒞v,w(p,σ)=bPv,w(Kv,w),Qv,w(p,σ)=bPv,w(Kv,w)+∇Φ(Pv,w)(Mv,w),ℋv,w,z(p,σ)=ℬPv,w(Kv,w,Kv,z).\begin{aligned} F_{v,w}(p, \sigma) &= \nabla\Phi(P_{v,w})(K_{v,w}), & \Lambda_{v,w}(p, \sigma) &= \tfrac{1}{2}\nabla\Phi(P_{v,w})(M_{v,w}), \\ \mathcal{C}_{v,w}(p, \sigma) &= b_{P_{v,w}}(K_{v,w}), & Q_{v,w}(p, \sigma) &= b_{P_{v,w}}(K_{v,w}) + \nabla\Phi(P_{v,w})(M_{v,w}), \\ \mathcal{H}_{v,w,z}(p, \sigma) &= \mathcal{B}_{P_{v,w}}(K_{v,w}, K_{v,z}). && \end{aligned}(13)

The constant-sum center plane Πd\Pi_d at sum level d∈ℝd \in \mathbb{R} is the subset {p∈ℝℤ/3ℤ:p0+p1+p2=d}\{p \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}} : p_0 + p_1 + p_2 = d\} of ℝℤ/3ℤ\mathbb{R}^{\mathbb{Z}/3\mathbb{Z}}. Given i∈ℤ/3ℤi \in \mathbb{Z}/3\mathbb{Z} and d∈ℝd \in \mathbb{R}, the plane coordinate chart χi,d:ℝ×ℝ→Πd\chi_{i,d} : \mathbb{R} \times \mathbb{R} \to \Pi_d is the map sending each z=(z1,z2)∈ℝ×ℝz = (z_1, z_2) \in \mathbb{R} \times \mathbb{R} to the triple equal to d−z1−z2d - z_1 - z_2 at ii and to z1z_1 and z2z_2 at the remaining two vertices in increasing order.✦ For a function f:ℝℤ/3ℤ→ℝf : \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}} \to \mathbb{R} and d∈ℝd \in \mathbb{R}, the plane integral ∫Πdf(p)dp\int_{\Pi_d} f(p)\, dp is the Lebesgue integral ∫Πdf(p)dp=∫ℝ×ℝf(χ2,d(z))dz.\int_{\Pi_d} f(p)\, dp = \int_{\mathbb{R} \times \mathbb{R}} f\bigl(\chi_{2,d}(z)\bigr)\, dz.(14)✦ For a matrix family A=(Av)v∈ℤ/3ℤA = (A_v)_{v \in \mathbb{Z}/3\mathbb{Z}} on ι\iota, step size h∈ℝh \in \mathbb{R}, node coordinates u:ι→ℝu : \iota \to \mathbb{R}, a sum level d∈ℝd \in \mathbb{R}, a vector α∈ℝℤ/3ℤ\alpha \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}}, and s∈ℝs \in \mathbb{R}, writing tαt\alpha (for any t∈ℝt \in \mathbb{R}) for the rescaled variance vector with aa-th entry αat\alpha_a t, the total plane time derivative Θd(α,s)\Theta_d(\alpha, s) is the real number Θd(α,s)=∑v∈ℤ/3ℤ∑w∈ℤ/3ℤαw∫ΠdΛv,w(p,sα)dp.\Theta_d(\alpha, s) = \sum_{v \in \mathbb{Z}/3\mathbb{Z}} \sum_{w \in \mathbb{Z}/3\mathbb{Z}} \alpha_w \int_{\Pi_d} \Lambda_{v,w}(p, s\alpha)\, dp.(15)✦

The Claim

Theorem 1. Fix an index set ι\iota, a family A=(Av)v∈ℤ/3ℤA = (A_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of complex square matrices indexed by ι\iota, a step size h>0h > 0, and real node coordinates u:ι→ℝu : \iota \to \mathbb{R}; for a vector y∈ℝℤ/3ℤy \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}}, a vertex w∈ℤ/3ℤw \in \mathbb{Z}/3\mathbb{Z}, and q∈ℝq \in \mathbb{R}, write y[w↦q]y[w \mapsto q] for the vector equal to qq at ww and yay_a for a≠wa \neq w.

Given p∈ℝℤ/3ℤp \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}}, σ∈ℝℤ/3ℤ\sigma \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}} with σa>0\sigma_a > 0 for every a∈ℤ/3ℤa \in \mathbb{Z}/3\mathbb{Z}, and v,w∈ℤ/3ℤv, w \in \mathbb{Z}/3\mathbb{Z}, the function q↦ev(p[w↦q],σ)q \mapsto e_v\bigl(p[w \mapsto q], \sigma\bigr) is differentiable at q=pwq = p_w with derivative given by (16): ddq|q=pwev(p[w↦q],σ)=Fv,w(p,σ).\left.\frac{d}{dq}\right|_{q = p_w} e_v\bigl(p[w \mapsto q], \sigma\bigr) = F_{v,w}(p, \sigma).(16)✦ For those same p,σ,v,wp, \sigma, v, w, the function q↦Fv,w(p[w↦q],σ)q \mapsto F_{v,w}\bigl(p[w \mapsto q], \sigma\bigr) is differentiable at q=pwq = p_w and satisfies (17): ddq|q=pwFv,w(p[w↦q],σ)=Qv,w(p,σ).\left.\frac{d}{dq}\right|_{q = p_w} F_{v,w}\bigl(p[w \mapsto q], \sigma\bigr) = Q_{v,w}(p, \sigma).(17)✦ In the variance variable σw\sigma_w, for the same p,σ,v,wp, \sigma, v, w, the map q↦ev(p,σ[w↦q])q \mapsto e_v\bigl(p, \sigma[w \mapsto q]\bigr) is differentiable at q=σwq = \sigma_w, yielding (18): ddq|q=σwev(p,σ[w↦q])=Λv,w(p,σ).\left.\frac{d}{dq}\right|_{q = \sigma_w} e_v\bigl(p, \sigma[w \mapsto q]\bigr) = \Lambda_{v,w}(p, \sigma).(18)✦ When p∈ℝℤ/3ℤp \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}}, α∈ℝℤ/3ℤ\alpha \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}} satisfies αa>0\alpha_a > 0 for every a∈ℤ/3ℤa \in \mathbb{Z}/3\mathbb{Z}, s>0s > 0, and v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z}, the scaled-variance function t↦ev(p,tα)t \mapsto e_v(p, t\alpha) is differentiable at t=st = s with (19): ddt|t=sev(p,tα)=∑w∈ℤ/3ℤαwΛv,w(p,sα).\left.\frac{d}{dt}\right|_{t = s} e_v(p, t\alpha) = \sum_{w \in \mathbb{Z}/3\mathbb{Z}} \alpha_w \Lambda_{v,w}(p, s\alpha).(19)✦ If p∈ℝℤ/3ℤp \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}}, σ∈ℝℤ/3ℤ\sigma \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}} satisfies σa>0\sigma_a > 0 for every a∈ℤ/3ℤa \in \mathbb{Z}/3\mathbb{Z}, and v,w,z∈ℤ/3ℤv, w, z \in \mathbb{Z}/3\mathbb{Z} satisfy w≠vw \neq v, z≠vz \neq v, and w≠zw \neq z, then q↦Fv,w(p[z↦q],σ)q \mapsto F_{v,w}\bigl(p[z \mapsto q], \sigma\bigr) is differentiable at q=pzq = p_z as in (20): ddq|q=pzFv,w(p[z↦q],σ)=ℋv,w,z(p,σ).\left.\frac{d}{dq}\right|_{q = p_z} F_{v,w}\bigl(p[z \mapsto q], \sigma\bigr) = \mathcal{H}_{v,w,z}(p, \sigma).(20)✦ For any sum level d∈ℝd \in \mathbb{R}, any σ∈ℝℤ/3ℤ\sigma \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}} with σa>0\sigma_a > 0 for all a∈ℤ/3ℤa \in \mathbb{Z}/3\mathbb{Z}, and any v,w∈ℤ/3ℤv, w \in \mathbb{Z}/3\mathbb{Z}, the plane integrals over Πd\Pi_d satisfy (21): 2∫ΠdΛv,w(p,σ)dp=∫ΠdQv,w(p,σ)dp−∫Πd𝒞v,w(p,σ)dp.2 \int_{\Pi_d} \Lambda_{v,w}(p, \sigma)\, dp = \int_{\Pi_d} Q_{v,w}(p, \sigma)\, dp - \int_{\Pi_d} \mathcal{C}_{v,w}(p, \sigma)\, dp.(21)✦ Given a sum level d∈ℝd \in \mathbb{R}, a vector α∈ℝℤ/3ℤ\alpha \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}} with αa>0\alpha_a > 0 for all a∈ℤ/3ℤa \in \mathbb{Z}/3\mathbb{Z}, and s>0s > 0, the total plane time derivative obeys (22): 2Θd(α,s)=∑v∈ℤ/3ℤ((∑w∈ℤ/3ℤαw)∫Πdℋv,v+1,v+2(p,sα)dp−∑w∈ℤ/3ℤαw∫Πd𝒞v,w(p,sα)dp).2\,\Theta_d(\alpha, s) = \sum_{v \in \mathbb{Z}/3\mathbb{Z}} \left( \left(\sum_{w \in \mathbb{Z}/3\mathbb{Z}} \alpha_w\right) \int_{\Pi_d} \mathcal{H}_{v,v+1,v+2}(p, s\alpha)\, dp - \sum_{w \in \mathbb{Z}/3\mathbb{Z}} \alpha_w \int_{\Pi_d} \mathcal{C}_{v,w}(p, s\alpha)\, dp \right).(22)✦

In the paper, the result reads: fix a positive vector α∈ℝℤ/3ℤ\alpha \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}} and set σi=αis\sigma_i = \alpha_i s; then for each v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z}, all centers p∈ℝℤ/3ℤp \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}}, and all s>0s > 0, one has 2∂sev=∑i∈ℤ/3ℤαi∂pi2ev−αvVv−∑w≠vαwYv,w2\partial_s e_v = \sum_{i \in \mathbb{Z}/3\mathbb{Z}} \alpha_i \partial_{p_i}^2 e_v - \alpha_v V_v - \sum_{w \neq v} \alpha_w Y_{v,w} and ∂pv+1∂pv−1ev=Zv\partial_{p_{v+1}}\partial_{p_{v-1}} e_v = Z_v, where eve_v stands for the heat matrix energy, VvV_v stands for the heat Hessian cost 𝒞v,v(p,sα)\mathcal{C}_{v,v}(p, s\alpha), Yv,wY_{v,w} (w≠vw \neq v) stands for 𝒞v,w(p,sα)\mathcal{C}_{v,w}(p, s\alpha), and ZvZ_v stands for the mixed heat Hessian pairing ℋv,v+1,v+2(p,sα).\mathcal{H}_{v,v+1,v+2}(p, s\alpha). The library separates the pointwise derivative identities (16) through (20) from the dissipation identities (21) and (22), stating (21) after integrating over the constant-sum plane Πd\Pi_d and (22) after also summing over the vertices.

What Is Checked

The seven claims are stated over an index set ι\iota, a family A=(Av)v∈ℤ/3ℤA = (A_v)_{v \in \mathbb{Z}/3\mathbb{Z}} of complex square matrices indexed by ι\iota, a step size h>0h > 0, real node coordinates u:ι→ℝu : \iota \to \mathbb{R}, center vectors p∈ℝℤ/3ℤp \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}}, positive variance vectors σ\sigma (including rescaled vectors sαs\alpha with αa>0\alpha_a > 0 and s>0s > 0), and sum levels d∈ℝd \in \mathbb{R}, together with the Gaussian heat weights, vertex star Gram matrices, matrix energy and energy gradient, variational Hessian costs and pairings, and plane integrals over Πd\Pi_d. For these constructions, the library proves that the real trace tr⁡ℝ\operatorname{tr}_{\mathbb{R}} on complex square matrices indexed by an index set 𝒥\mathcal{J} is additive and homogeneous under multiplication by real scalars.