OctoInformBeta

OpenAI preprint, September 2026 · §5. Output-dependent truncations and energy jumps

Proposition 5.3: On finite integer sets, jump maximal sums and energy jumps or masked dissipation from the squared mesh step satisfy uniform cubic bounds

Cubic bounds on energy jumps and output-dependent masked dissipation.

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 set D⊂ℤD \subset \mathbb{Z}, a mesh step h>0h > 0, and a triple A=(A0,A1,A2)A = (A_0, A_1, A_2) of complex D×DD \times D matrices. Under Gaussian heat evolution on the zero-sum plane of three center coordinates, a matrix energy functional dissipates along scales through a Hessian quadratic cost, while output-dependent cutoff endpoints aij≤bija_{ij} \le b_{ij} retain each entry A0(i,j)A_0(i, j) only across the half-open scale window [aij,bij)[a_{ij}, b_{ij}).

The problem is to bound both the total energy jumps produced by bounded-variation replacements of A0A_0 and the scale integral of the masked Hessian cost in terms of a cubic norm of AA, uniformly in DD, hh, the cutoff matrices aa and bb, and the number of distinct endpoint values.

Triangle Matrices and Matrix Energy

The triangle vertex set ℤ/3ℤ\mathbb{Z}/3\mathbb{Z} is the additive group {0,1,2}\{0, 1, 2\} under cyclic addition. For an index set ι\iota, the triangle block space (ℤ/3ℤ)×ι(\mathbb{Z}/3\mathbb{Z}) \times \iota is the Cartesian product of the triangle vertex set ℤ/3ℤ\mathbb{Z}/3\mathbb{Z} and ι\iota. For a complex square matrix PP on ι\iota, the real trace Re⁡tr⁡(P)\operatorname{Re}\operatorname{tr}(P) is the real part of the trace of PP. For a complex square matrix PP on ι\iota, the matrix energy Φ(P)\Phi(P) is the real number Re⁡tr⁡(PP1/2)\operatorname{Re}\operatorname{tr}(P P^{1/2}), where P1/2P^{1/2} is the functional-calculus square root of PP. For complex square matrices XX and YY on ι\iota, the real Hilbert-Schmidt inner product ⟨X,Y⟩HS\langle X, Y \rangle_{\mathrm{HS}} is the real number Re⁡tr⁡(X*Y)\operatorname{Re}\operatorname{tr}(X^* Y).

Given a complex square matrix PP on ι\iota, the Sylvester operator SPS_P is the linear operator on complex square matrices on ι\iota defined on each complex square matrix XX on ι\iota by SP(X)=PX+XPS_P(X) = P X + X P. Given complex square matrices PP and XX on ι\iota, the Sylvester penalty is the real number ⟨X,SP(X)⟩HS\langle X, S_P(X) \rangle_{\mathrm{HS}}. For complex square matrices PP and KK on ι\iota, the Sylvester variational functional VP,KV_{P,K} is the real functional defined on each complex square matrix XX on ι\iota by VP,K(X)=2⟨K,X⟩HS−⟨X,SP(X)⟩HSV_{P,K}(X) = 2\langle K, X \rangle_{\mathrm{HS}} - \langle X, S_P(X) \rangle_{\mathrm{HS}}. For complex square matrices PP and KK on ι\iota, the root Hessian cost is the value 32sup⁡X∈ℂι×ιVP,K(X)\frac{3}{2}\sup_{X \in \mathbb{C}^{\iota \times \iota}} V_{P,K}(X). For complex square matrices PP and KK on ι\iota, the Hessian cost bP(K)b_P(K) is the root Hessian cost evaluated at P1/2P^{1/2} and KK.

For a 3×33 \times 3 matrix MM indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z} whose entries are complex ι×ι\iota \times \iota matrices, the block matrix assembly of MM is the complex matrix on (ℤ/3ℤ)×ι(\mathbb{Z}/3\mathbb{Z}) \times \iota whose ((v,i),(w,j))((v, i), (w, j)) entry is Mvw(i,j)M_{vw}(i, j). For a family W=(W0,W1,W2)W = (W_0, W_1, W_2) of complex ι×ι\iota \times \iota matrices indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z}, the triangle matrix blocks form the 3×33 \times 3 block matrix T̃(W)\widetilde{T}(W) whose (v,w)(v, w) block is 00 when v=wv = w, WvW_v when w=v+1w = v + 1, and Ww*W_w^* when v=w+1v = w + 1, and the triangle matrix T(W)T(W) is the block matrix assembly of T̃(W)\widetilde{T}(W). For v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z}, the vertex block projection QvQ_v on (ℤ/3ℤ)×ι(\mathbb{Z}/3\mathbb{Z}) \times \iota is the diagonal matrix with entry 11 at (w,i)(w, i) when w=vw = v and 00 otherwise. For a family W=(W0,W1,W2)W = (W_0, W_1, W_2) of complex ι×ι\iota \times \iota matrices indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z} and v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z}, the vertex star row Rv(W)R_v(W) is the matrix QvT(W)Q_v T(W). The vertex star Gram matrix of a family W=(W0,W1,W2)W = (W_0, W_1, W_2) of complex ι×ι\iota \times \iota matrices indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z} and v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z} is the matrix Rv(W)*Rv(W)R_v(W)^* R_v(W).

For a family A=(A0,A1,A2)A = (A_0, A_1, A_2) of complex ι×ι\iota \times \iota matrices indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z} and a complex ι×ι\iota \times \iota matrix BB, the zero-component replacement A[B]A^{[B]} is the matrix family indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z} equal to BB at 00 and to AvA_v at v≠0v \ne 0. Given a matrix family AA on ι\iota and real node weights g=(g0,g1,g2)g = (g_0, g_1, g_2) on ι\iota indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z}, the weighted matrix array A(g)A^{(g)} is the matrix family on ι\iota whose (i,j)(i, j) entry at vertex v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z} is gv(i)Av(i,j)gv+1(j)\sqrt{g_v(i)}\, A_v(i, j)\, \sqrt{g_{v+1}(j)}. For real node weights gg on ι\iota indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z}, the vertex root matrix DgD_g on (ℤ/3ℤ)×ι(\mathbb{Z}/3\mathbb{Z}) \times \iota is the diagonal matrix with entry gv(i)\sqrt{g_v(i)} at (v,i)(v, i). Given a matrix family AA on ι\iota, real node weights gg on ι\iota, and v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z}, the vertex star base row is the matrix Rv(A)DgR_v(A) D_g. The vertex star base column of a matrix family AA on ι\iota, real node weights gg on ι\iota, and v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z} is the matrix Rv(A)*DgR_v(A)^* D_g. For a complex square matrix MM on an index set κ\kappa and a real weight function ω:κ→ℝ\omega : \kappa \to \mathbb{R}, the weighted Gram matrix of MM with weight ω\omega is the matrix M*diag⁡(ω)MM^* \operatorname{diag}(\omega) M.

Gaussian Heat Costs and Masked Truncations

For real variance σ\sigma and t∈ℝt \in \mathbb{R}, the Gaussian density gσ(t)g_\sigma(t) is gσ(t)=(2πσ)−1/2exp⁡(−t22σ).g_\sigma(t) = (2\pi\sigma)^{-1/2} \exp\!\left(-\frac{t^2}{2\sigma}\right).(1) Given step size h∈ℝh \in \mathbb{R}, point x∈ℝx \in \mathbb{R}, center p∈ℝp \in \mathbb{R}, and variance σ∈ℝ\sigma \in \mathbb{R}, the heat weight is the real number hgσ(x−p)h g_\sigma(x - p). Given point x∈ℝx \in \mathbb{R}, center p∈ℝp \in \mathbb{R}, and variance σ∈ℝ\sigma \in \mathbb{R}, the Gaussian score is the real number (x−p)/σ(x - p)/\sigma. For a step size h∈ℝh \in \mathbb{R}, node coordinates u:ι→ℝu : \iota \to \mathbb{R}, centers p:ℤ/3ℤ→ℝp : \mathbb{Z}/3\mathbb{Z} \to \mathbb{R}, and variances σ:ℤ/3ℤ→ℝ\sigma : \mathbb{Z}/3\mathbb{Z} \to \mathbb{R}, the vertex heat weights w=(w0,w1,w2)w = (w_0, w_1, w_2) form the family of node weights on ι\iota defined at v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z} and i∈ιi \in \iota by wv(i)=hgσv(ui−pv)w_v(i) = h g_{\sigma_v}(u_i - p_v). Given a matrix family AA on ι\iota indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z} and parameters (h,u,p,σ)(h, u, p, \sigma), the heat matrix array is the weighted matrix array A(w)A^{(w)} formed with the vertex heat weights ww for (h,u,p,σ)(h, u, p, \sigma). For a vertex v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z}, the heat matrix energy Eheat(A,h,u,p,σ,v)E_{\mathrm{heat}}(A, h, u, p, \sigma, v) is the real number Φ(Rv(A(w))*Rv(A(w)))\Phi\bigl(R_v(A^{(w)})^* R_v(A^{(w)})\bigr).

The fixed base matrix Fv,v′(A,g)F_{v,v'}(A, g) of a matrix family AA on ι\iota indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z}, real node weights gg on ι\iota, and vertices v,v′∈ℤ/3ℤv, v' \in \mathbb{Z}/3\mathbb{Z} is the vertex star base row Rv(A)DgR_v(A) D_g when v′=vv' = v, and the vertex star base column Rv(A)*DgR_v(A)^* D_g when v′≠vv' \ne v. Given h∈ℝh \in \mathbb{R}, coordinates ũ:κ→ℝ\tilde{u} : \kappa \to \mathbb{R}, a group assignment ℓ:κ→ν\ell : \kappa \to \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 at x∈κx \in \kappa is the real number hgσ̃ℓ(x)(ũx−p̃ℓ(x))h g_{\tilde{\sigma}_{\ell(x)}}(\tilde{u}_x - \tilde{p}_{\ell(x)}). For group q∈νq \in \nu, the grouped first variation weight at x∈κx \in \kappa is the real number equal to ũx−p̃qσ̃qhgσ̃q(ũx−p̃q)\frac{\tilde{u}_x - \tilde{p}_q}{\tilde{\sigma}_q} h g_{\tilde{\sigma}_q}(\tilde{u}_x - \tilde{p}_q) if ℓ(x)=q\ell(x) = q and 00 otherwise. For a matrix family AA on ι\iota indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z}, h∈ℝh \in \mathbb{R}, u:ι→ℝu : \iota \to \mathbb{R}, p,σ:ℤ/3ℤ→ℝp, \sigma : \mathbb{Z}/3\mathbb{Z} \to \mathbb{R}, and v,v′∈ℤ/3ℤv, v' \in \mathbb{Z}/3\mathbb{Z}, let ww denote the vertex heat weights for (h,u,p,σ)(h, u, p, \sigma), and set κ=(ℤ/3ℤ)×ι\kappa = (\mathbb{Z}/3\mathbb{Z}) \times \iota, ν=ℤ/3ℤ\nu = \mathbb{Z}/3\mathbb{Z}, ũ(z,i)=ui\tilde{u}_{(z,i)} = u_i, and ℓ(z,i)=z\ell(z, i) = z: then the local Gram matrix Pv,v′(A,h,u,p,σ)P_{v,v'}(A, h, u, p, \sigma) is the matrix given by the weighted Gram matrix of Fv,v′(A,w)F_{v,v'}(A, w) with the grouped heat weight, and the local first variation matrix Kv,v′(A,h,u,p,σ)K_{v,v'}(A, h, u, p, \sigma) is the matrix given by the weighted Gram matrix of Fv,v′(A,w)F_{v,v'}(A, w) with the grouped first variation weight for group v′v'. The heat Hessian cost cheat(A,h,u,p,σ,v,v′)c_{\mathrm{heat}}(A, h, u, p, \sigma, v, v') of (A,h,u,p,σ)(A, h, u, p, \sigma) at (v,v′)(v, v') is the real number bPv,v′(Kv,v′)b_{P_{v,v'}}(K_{v,v'}).

At vertex v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z} and sum level d∈ℝd \in \mathbb{R}, the plane coordinate chart γv,d:ℝ2→ℝℤ/3ℤ\gamma_{v,d} : \mathbb{R}^2 \to \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}} is the map sending each z=(z1,z2)∈ℝ2z = (z_1, z_2) \in \mathbb{R}^2 to the triple summing to dd equal to d−z1−z2d - z_1 - z_2 at vertex vv and to z1,z2z_1, z_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 over the constant-sum center plane Πd={p∈ℝℤ/3ℤ:p0+p1+p2=d}\Pi_d = \{p \in \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}} : p_0 + p_1 + p_2 = d\} is ∫Πdf(p)dπ(p)=∫ℝ2f(γ2,d(z))dz.\int_{\Pi_d} f(p)\,d\pi(p) = \int_{\mathbb{R}^2} f(\gamma_{2,d}(z))\,dz.(2) For a matrix family AA on ι\iota indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z}, step size h∈ℝh \in \mathbb{R}, node coordinates u:ι→ℝu : \iota \to \mathbb{R}, sum level d∈ℝd \in \mathbb{R}, and variances σ:ℤ/3ℤ→ℝ\sigma : \mathbb{Z}/3\mathbb{Z} \to \mathbb{R}, the total plane heat energy is ℰtot(A,h,u,d,σ)=∑v∈ℤ/3ℤ∫ΠdEheat(A,h,u,p,σ,v)dπ(p).\mathcal{E}_{\mathrm{tot}}(A, h, u, d, \sigma) = \sum_{v \in \mathbb{Z}/3\mathbb{Z}} \int_{\Pi_d} E_{\mathrm{heat}}(A, h, u, p, \sigma, v)\,d\pi(p).(3)

Given scale endpoints a,b∈ℝa, b \in \mathbb{R}, scale s∈ℝs \in \mathbb{R}, and z∈ℂz \in \mathbb{C}, the window mask Ma,b(s,z)M_{a,b}(s, z) is the complex number equal to zz when a≤s<ba \le s < b and 00 otherwise.✦ Given a matrix family AA on ι\iota indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z}, real cutoff matrices a,b∈ℝι×ιa, b \in \mathbb{R}^{\iota \times \iota}, and scale s∈ℝs \in \mathbb{R}, the masked array A[a,b](s)A^{[a,b]}(s) is the zero-component replacement A[B(s)]A^{[B(s)]} with B(s)∈ℂι×ιB(s) \in \mathbb{C}^{\iota \times \iota} given by B(s)ij=Maij,bij(s,A0(i,j))B(s)_{ij} = M_{a_{ij}, b_{ij}}(s, A_0(i, j)). For a vertex r∈ℤ/3ℤr \in \mathbb{Z}/3\mathbb{Z}, the strict diffusion rate α(r):ℤ/3ℤ→ℝ\alpha^{(r)} : \mathbb{Z}/3\mathbb{Z} \to \mathbb{R} is the function defined by αz(r)=1\alpha^{(r)}_z = 1 for z=rz = r and αz(r)=11/10\alpha^{(r)}_z = 11/10 for z≠rz \ne r. For a matrix family AA on ι\iota indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z}, real cutoff matrices a,b∈ℝι×ιa, b \in \mathbb{R}^{\iota \times \iota}, step size h∈ℝh \in \mathbb{R}, node coordinates u:ι→ℝu : \iota \to \mathbb{R}, vertex r∈ℤ/3ℤr \in \mathbb{Z}/3\mathbb{Z}, and scale s∈ℝs \in \mathbb{R}, the masked heat cost is 𝒞mask(A,a,b,h,u,r,s)=∑v∈ℤ/3ℤ∫Π0cheat(A[a,b](s),h,u,p,sα(r),v,v)dπ(p).\mathcal{C}_{\mathrm{mask}}(A, a, b, h, u, r, s) = \sum_{v \in \mathbb{Z}/3\mathbb{Z}} \int_{\Pi_0} c_{\mathrm{heat}}\bigl(A^{[a,b]}(s), h, u, p, s\alpha^{(r)}, v, v\bigr)\,d\pi(p).(4) For a matrix family AA on ι\iota indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z} and step size h∈ℝh \in \mathbb{R}, the cubic array norm is 𝒮(A)=∑v∈ℤ/3ℤ∑i,j∈ιh2|Av(i,j)|3.\mathcal{S}(A) = \sum_{v \in \mathbb{Z}/3\mathbb{Z}} \sum_{i,j \in \iota} h^2 |A_v(i, j)|^3.(5)

Discrete Maximal Functions

The discrete interval Ir(n)I_r(n) of radius r∈ℕr \in \mathbb{N} centered at an integer n∈ℤn \in \mathbb{Z} is the set {m∈ℤ:n−r≤m≤n+r}\{m \in \mathbb{Z} : n - r \le m \le n + r\}. The support radius RD(n)R_D(n) of a finite set D⊂ℤD \subset \mathbb{Z} around n∈ℤn \in \mathbb{Z} is the supremum sup⁡m∈D|m−n|\sup_{m \in D} |m - n|, taken to be 00 when DD is empty.✦ Given a finite set D⊂ℤD \subset \mathbb{Z}, a function f:ℤ→ℝf : \mathbb{Z} \to \mathbb{R}, n∈ℤn \in \mathbb{Z}, and r∈ℕr \in \mathbb{N}, the discrete interval average is Avg⁡D,rf(n)=12r+1∑m∈D∩Ir(n)f(m),\operatorname{Avg}_{D,r} f(n) = \frac{1}{2r + 1} \sum_{m \in D \cap I_r(n)} f(m),(6) and the discrete maximal function of ff over DD at nn is MDℤf(n)=max⁡0≤r≤RD(n)Avg⁡D,rf(n)M_D^{\mathbb{Z}} f(n) = \max_{0 \le r \le R_D(n)} \operatorname{Avg}_{D,r} f(n).

Given a finite set D⊂ℤD \subset \mathbb{Z} and a function f:D→ℝf : D \to \mathbb{R}, the extension by zero f̃:ℤ→ℝ\tilde{f} : \mathbb{Z} \to \mathbb{R} is the function equal to f(n)f(n) for n∈Dn \in D and 00 for n∉Dn \notin D.✦ For a finite set D⊂ℤD \subset \mathbb{Z}, a function f:D→ℝf : D \to \mathbb{R}, and n∈ℤn \in \mathbb{Z}, the line maximal function MDf(n)M_D f(n) is the real number MDℤf̃(n)M_D^{\mathbb{Z}} \tilde{f}(n). Given a finite set D⊂ℤD \subset \mathbb{Z}, a family A=(A0,A1,A2)A = (A_0, A_1, A_2) of complex D×DD \times D matrices indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z}, and indices i,j∈Di, j \in D, the jump maximal function is MjumpA(i,j)=2MD(|A0(i,⋅)|2)(j)+2MD(|A0(⋅,j)|2)(i)+MD(|A2(⋅,i)|2)(−i−j)+MD(|A1(j,⋅)|2)(−i−j).M_{\mathrm{jump}} A(i, j) = 2 M_D\bigl(|A_0(i, \cdot)|^2\bigr)(j) + 2 M_D\bigl(|A_0(\cdot, j)|^2\bigr)(i) + M_D\bigl(|A_2(\cdot, i)|^2\bigr)(-i - j) + M_D\bigl(|A_1(j, \cdot)|^2\bigr)(-i - j).(7)

Dissipation and Jump Constants

The strict positivity constant is the real number cstrict=12min⁡(1−352,1110−122).c_{\mathrm{strict}} = \frac{1}{2} \min\!\left(1 - \frac{3}{5}\sqrt{2},\, \frac{11}{10} - \frac{1}{2}\sqrt{2}\right).(8)✦ The mesh mass constant cmassc_{\mathrm{mass}} is the real number 2e1/22 e^{1/2}.✦ The jump scalar constant cjumpc_{\mathrm{jump}} is the real number 2+4822 + 48\sqrt{2}.✦ The masked constant cmaskc_{\mathrm{mask}} is the real number cmass(22+3cjump)c_{\mathrm{mass}}(2\sqrt{2} + 3 c_{\mathrm{jump}}).✦

The Claim

Theorem 1. Let D⊂ℤD \subset \mathbb{Z} be a finite set and A=(A0,A1,A2)A = (A_0, A_1, A_2) be a family of complex D×DD \times D matrices indexed by ℤ/3ℤ\mathbb{Z}/3\mathbb{Z}. Then the jump maximal sum satisfies ∑i,j∈D|A0(i,j)|MjumpA(i,j)≤cjump∑v∈ℤ/3ℤ∑i,j∈D|Av(i,j)|3.\sum_{i,j \in D} |A_0(i, j)|\, M_{\mathrm{jump}} A(i, j) \le c_{\mathrm{jump}} \sum_{v \in \mathbb{Z}/3\mathbb{Z}} \sum_{i,j \in D} |A_v(i, j)|^3.(9)✦ Suppose further that h>0h > 0, and define the uniform grid u:D→ℝu : D \to \mathbb{R} by u(m)=hmu(m) = h m for m∈Dm \in D. For any real cutoff matrices a,b∈ℝD×Da, b \in \mathbb{R}^{D \times D} with aij≤bija_{ij} \le b_{ij} for all i,j∈Di, j \in D, any nonempty finite set S⊂ℝS \subset \mathbb{R} containing every aija_{ij} and bijb_{ij} with h2≤sh^2 \le s for all s∈Ss \in S, and any r∈ℤ/3ℤr \in \mathbb{Z}/3\mathbb{Z}, the masked heat cost satisfies cstrict∫min⁡Smax⁡S𝒞mask(A,a,b,h,u,r,s)ds≤cmask𝒮(A).c_{\mathrm{strict}} \int_{\min S}^{\max S} \mathcal{C}_{\mathrm{mask}}(A, a, b, h, u, r, s)\,ds \le c_{\mathrm{mask}} \mathcal{S}(A).(10)✦ Let n∈ℕn \in \mathbb{N}, and let B:ℕ→ℂD×DB : \mathbb{N} \to \mathbb{C}^{D \times D} be a matrix sequence satisfying |Bk(i,j)|≤|A0(i,j)||B_k(i, j)| \le |A_0(i, j)| for all k∈ℕk \in \mathbb{N} and i,j∈Di, j \in D and ∑k=0n−1|Bk+1(i,j)−Bk(i,j)|≤2|A0(i,j)|\sum_{k=0}^{n-1} |B_{k+1}(i, j) - B_k(i, j)| \le 2 |A_0(i, j)| for all i,j∈Di, j \in D. For any variance profiles σ:ℕ→ℝℤ/3ℤ\sigma : \mathbb{N} \to \mathbb{R}^{\mathbb{Z}/3\mathbb{Z}} with h2≤σk(v)h^2 \le \sigma_k(v) for all k∈ℕk \in \mathbb{N} and v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z}, the total plane energy jumps satisfy ∑k=0n−1|ℰtot(A[Bk+1],h,u,0,σk)−ℰtot(A[Bk],h,u,0,σk)|≤3cmasscjump𝒮(A).\sum_{k=0}^{n-1} \bigl|\mathcal{E}_{\mathrm{tot}}(A^{[B_{k+1}]}, h, u, 0, \sigma_k) - \mathcal{E}_{\mathrm{tot}}(A^{[B_k]}, h, u, 0, \sigma_k)\bigr| \le 3 c_{\mathrm{mass}} c_{\mathrm{jump}} \mathcal{S}(A).(11)✦ For any nondecreasing sequence t:ℕ→ℝt : \mathbb{N} \to \mathbb{R} with h2≤t0h^2 \le t_0 and any r∈ℤ/3ℤr \in \mathbb{Z}/3\mathbb{Z}, the integrated masked dissipation satisfies cstrict∑k=0n−1∫tktk+1∑v∈ℤ/3ℤ∫Π0cheat(A[Bk],h,u,p,sα(r),v,v)dπ(p)ds≤cmask𝒮(A).c_{\mathrm{strict}} \sum_{k=0}^{n-1} \int_{t_k}^{t_{k+1}} \sum_{v \in \mathbb{Z}/3\mathbb{Z}} \int_{\Pi_0} c_{\mathrm{heat}}\bigl(A^{[B_k]}, h, u, p, s\alpha^{(r)}, v, v\bigr)\,d\pi(p)\,ds \le c_{\mathrm{mask}} \mathcal{S}(A).(12)✦

In the paper, the result reads: for every h>0h > 0, integer N≥0N \ge 0, complex arrays A=(A0,A1,A2)A = (A_0, A_1, A_2) on {−N,…,N}2\{-N, \dots, N\}^2, and endpoints h≤εij<ρijh \le \varepsilon_{ij} < \rho_{ij}, taking any permutation α\alpha of (1,11/10,11/10)(1, 11/10, 11/10) and σv=αvs\sigma_v = \alpha_v s gives ∫s−s+∫Π0∑v=02Vv(p,s)dπ(p)ds≤C𝒮(A)\int_{s_-}^{s_+} \int_{\Pi_0} \sum_{v=0}^2 V_v(p, s)\,d\pi(p)\,ds \le C\mathcal{S}(A) with a constant CC independent of the arrays, mesh, endpoints, and number of distinct endpoint values, where the diagonal heat Hessian cost cheatc_{\mathrm{heat}} at (v,v)(v, v) stands for the paper’s Vv(p,s)V_v(p, s) and s−,s+s_-, s_+ are the extremal scale endpoints. The library states this bound on an arbitrary finite set D⊂ℤD \subset \mathbb{Z} rather than {−N,…,N}2\{-N, \dots, N\}^2, uses general half-open cutoffs aij≤bija_{ij} \le b_{ij} in a finite set S⊂[h2,∞)S \subset [h^2, \infty) with integration limits min⁡S\min S and max⁡S\max S in (10), parametrizes the permutations α\alpha by α(r)\alpha^{(r)} for r∈ℤ/3ℤr \in \mathbb{Z}/3\mathbb{Z} with C=cmask/cstrictC = c_{\mathrm{mask}}/c_{\mathrm{strict}}, and states separately the jump maximal bound (9), the total energy jump bound (11), and the stepwise masked dissipation bound (12).

What Is Checked

The jump maximal bound (9) is stated over any finite set D⊂ℤD \subset \mathbb{Z} and complex D×DD \times D matrix family A=(A0,A1,A2)A = (A_0, A_1, A_2); over the same DD and AA with mesh step h>0h > 0 and uniform grid u(m)=hmu(m) = hm, the masked cost integral bound (10) is stated over real cutoff matrices aij≤bija_{ij} \le b_{ij} in a nonempty finite set S⊂[h2,∞)S \subset [h^2, \infty) and r∈ℤ/3ℤr \in \mathbb{Z}/3\mathbb{Z}, while the total energy jump bound (11) and finite masked dissipation bound (12) are stated over any n∈ℕn \in \mathbb{N} and matrix sequence BkB_k satisfying |Bk(i,j)|≤|A0(i,j)||B_k(i, j)| \le |A_0(i, j)| and ∑k=0n−1|Bk+1(i,j)−Bk(i,j)|≤2|A0(i,j)|\sum_{k=0}^{n-1} |B_{k+1}(i, j) - B_k(i, j)| \le 2|A_0(i, j)|, together with variance profiles σk(v)≥h2\sigma_k(v) \ge h^2 in (11) or a nondecreasing scale sequence tkt_k with t0≥h2t_0 \ge h^2 and r∈ℤ/3ℤr \in \mathbb{Z}/3\mathbb{Z} in (12).