OctoInformBeta

OpenAI preprint, September 2026 · §6. Exact transmission on the regular harmonic subspaces

Lemma 6.4: Positive propagators bound normalization and restart the regular harmonic solution at each pulse endpoint

Diagonal leg growth converges along prescribed periods.

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

Harmonic radial growth is governed by a second-order equation on each angular block. In logarithmic radius t=log⁡rt=\log r, a diagonal leg runs from the end of a control pulse to the next period, including the transitions surrounding its long dwell interval.

Angular and radial pulse data consist of finite-dimensional real inner product spaces EbE_b, indexed by b∈Ξb\in\Xi, with orthonormal bases (eb,i)i∈Ib(e_{b,i})_{i\in I_b} on finite index sets, an angular label set 𝒦\mathcal K, and real parameters ρ,p0,γ,M,Bb,βb,θb\rho,p_0,\gamma,M,B_b,\beta_b, \theta_b. Here γ,θb>0\gamma,\theta_b>0, M,Bb≥0M,B_b\ge0, and the real linear endomorphisms Db,κ:Eb→EbD_{b,\kappa}\colon E_b\to E_b satisfy (1) for every v∈Ebv\in E_b and κ∈𝒦\kappa\in\mathcal K: −Bb∥v∥2≤⟨v,Db,κv⟩≤0,βbBb=θb(θb+p0),p0+2θb≠0.\begin{gathered} -B_b\|v\|^2\le \langle v,D_{b,\kappa}v\rangle\le0, \qquad \beta_bB_b=\theta_b(\theta_b+p_0), \qquad p_0+2\theta_b\ne0. \end{gathered}(1)

A radial history is a choice of symmetric background operators Db,0D_{b,0}, endpoint anisotropy q*≥1q_*\ge1, positive central degrees ℓb\ell_b, and a central drift pcp_c, with Bb=ℓb(ℓb+pc)B_b=\ell_b(\ell_b+p_c). Its controlled angular operators are symmetric, and its background operators satisfy the quadratic bounds in (1). The history also specifies a smooth nonnegative radial factor α\alpha and a continuous drift pp, with a nonnegative bound MpM_p, satisfying (2) for every block bb: $$\begin{gathered} 0\le\alpha(t)\le M,\qquad |p(t)|\le M_p,\qquad \alpha(t)=1,\quad p(t)=p_c\quad(t\le0),\\ \alpha(t)\longrightarrow\beta_b,\qquad p(t)\longrightarrow p_0 \quad(t\longrightarrow\infty),\qquad p(t)\ge\gamma\quad\hbox{for all sufficiently large }t. \end{gathered}$$(2)

The Control Stream

Period j+1j+1, indexed by integers j≥0j\ge0, has pulse duration τj\tau_j, dwell duration δj\delta_j, and start time σj\sigma_j given by (3); eje_j is its pulse endpoint and hjh_j is the length of its whole diagonal leg: τj=(j+1)6,δj=(j+1)7,σj=10+∑k=0j−1(δk+4τk),ej=σj+τj,hj=σj+1−ej.\begin{gathered} \tau_j=(j+1)^6,\qquad \delta_j=(j+1)^7,\qquad \sigma_j=10+\sum_{k=0}^{j-1}(\delta_k+4\tau_k),\\ e_j=\sigma_j+\tau_j,\qquad h_j=\sigma_{j+1}-e_j. \end{gathered}(3)

Let χ\chi be the fixed smooth transition function. The transition step SaS_a of slot a≥0a\ge0, its smooth control pulse ψa\psi_a, and the smooth transition ramp RR are the functions in (4): Sa(x)=χ(x−(2a+1)),ψa(x)=Sa′(x),R(x)=χ(2x−1/2).S_a(x)=\chi(x-(2a+1)),\qquad \psi_a(x)=S_a'(x),\qquad R(x)=\chi(2x-1/2).(4)

The smooth dwell window wjw_j, the period anisotropy increment ηjz\eta_j^z of a control pulse family zz, and its pulsed anisotropy profile qzq_z, starting at integer threshold j0≥0j_0\ge0, are defined by (5). The diagonal anisotropy profile q0q_0 is qzq_z with zz identically zero; it is round initially and throughout every pulse.✦ wj(t)=R(t−σj−τjτj)−R(t−σj−2τj−δjτj),ηjz(t)=τj−1zj(t−σjτj)+(q*−1)wj(t),qz(t)=1+∑j≥j0ηjz(t).\begin{aligned} w_j(t)&=R\!\left(\frac{t-\sigma_j-\tau_j}{\tau_j}\right) -R\!\left(\frac{t-\sigma_j-2\tau_j-\delta_j}{\tau_j}\right),\\ \eta_j^z(t)&=\tau_j^{-1}z_j\!\left(\frac{t-\sigma_j}{\tau_j}\right) +(q_*-1)w_j(t),\qquad q_z(t)=1+\sum_{j\ge j_0}\eta_j^z(t). \end{aligned}(5)

For NN chronological slots, the control packet duration is H=2N+1H=2N+1, with an open round gap at each endpoint.✦ The individual packet stretches Qau,TQ_a^{u,T} for amplitudes u(a)u(a), time scale TT, and normalized time xx are defined by (6): Qau,T(x)=1+T−1Hu(a)ψa(Hx),0≤a<N.Q_a^{u,T}(x)=1+T^{-1}H u(a)\psi_a(Hx),\qquad 0\le a<N.(6)✦

For a background BB and finite family D=(Da)aD=(D_a)_a in a real vector space, the exact Berger angular coefficient 𝒜ρ\mathcal A_\rho, nonlinear angular packet 𝒲B,D\mathcal W_{B,D}, and nonlinear Berger control word WB,Du,TW_{B,D}^{u,T} are defined by (7): 𝒜ρ(B,D;q)=qρB+(qρ−qρ−1)D,𝒲B,D((qa)a)=B+∑a=0N−1(𝒜ρ(B,Da;qa)−B),WB,Du,T(x)=𝒲B,D((Qau,T(x))a).\begin{aligned} \mathcal A_\rho(B,D;q)&=q^\rho B+(q^\rho-q^{\rho-1})D,\\ \mathcal W_{B,D}((q_a)_a)&=B+\sum_{a=0}^{N-1} \bigl(\mathcal A_\rho(B,D_a;q_a)-B\bigr),\\ W_{B,D}^{u,T}(x)&=\mathcal W_{B,D}((Q_a^{u,T}(x))_a). \end{aligned}(7)

An amplitude stream sj(a)s_j(a) and slot assignment a↦κa∈𝒦a\mapsto\kappa_a\in\mathcal K specify Qj,a=Qasj,τjQ_{j,a}=Q_a^{s_j,\tau_j} and Wb,j=WRb,(Db,κa)asj,τjW_{b,j}=W_{R_b,(D_{b,\kappa_a})_a}^{s_j,\tau_j}, where Rb=BbIR_b=B_bI. The period angular increment Δb,j\Delta_{b,j}, active angular increment Δb,jact\Delta^{\mathrm{act}}_{b,j}, infinite angular coefficient 𝒞b\mathcal C_b, and radial equation coefficient AbA_b are the operator families in (8): Δb,j(t)=Wb,j(t−σjτj)−Rb,Δb,jact(t)={Δb,j(t),j≥j0,0,j<j0,𝒞b(t)=𝒜ρ(Rb,Db,0;q0(t))+∑j≥0Δb,jact(t),Ab(t)=α(t)𝒞b(t).\begin{aligned} \Delta_{b,j}(t)&=W_{b,j}\!\left(\frac{t-\sigma_j}{\tau_j}\right)-R_b,\\ \Delta^{\mathrm{act}}_{b,j}(t)&= \begin{cases}\Delta_{b,j}(t),&j\ge j_0,\\0,&j<j_0,\end{cases}\\ \mathcal C_b(t)&=\mathcal A_\rho(R_b,D_{b,0};q_0(t)) +\sum_{j\ge0}\Delta^{\mathrm{act}}_{b,j}(t),\qquad A_b(t)=\alpha(t)\mathcal C_b(t). \end{aligned}(8)

Regular Solutions and Leg Propagators

For a time-dependent endomorphism C(t)C(t) of a real normed space EE, the Peano-Baker iterates FkF_k and linear evolution ΦC\Phi_C are defined by (9). For a potential AA and drift pp, the first-order block operator is ℬA,p\mathcal B_{A,p} in that equation. F0(C,x;t)=x,Fk+1(C,x;t)=∫0tC(s)Fk(C,x;s)ds,ΦC(t)x=∑k=0∞Fk(C,x;t),ℬA,p(t)(u,v)=(v,A(t)u−p(t)v).\begin{gathered} F_0(C,x;t)=x,\qquad F_{k+1}(C,x;t)=\int_0^t C(s)F_k(C,x;s)\,ds,\qquad \Phi_C(t)x=\sum_{k=0}^\infty F_k(C,x;t),\\ \mathcal B_{A,p}(t)(u,v)=(v,A(t)u-p(t)v). \end{gathered}(9)

The regular solution phase (ux(t),vx(t))(u_x(t),v_x(t)) for central degree ℓ∈ℝ\ell\in\mathbb R and initial value x∈Ex\in E is ΦℬA,p(t)(x,ℓx)\Phi_{\mathcal B_{A,p}}(t)(x,\ell x); its components uxu_x and vxv_x are the regular value and velocity. For each time tt, the left composition coefficient LA(t)L_A(t) is the linear operator on endomorphisms of EE defined by LA(t)T=A(t)∘TL_A(t)T=A(t)\circ T.✦ The operator-valued regular solution YY, regular velocity operator ZZ, and Riccati logarithmic derivative PP are defined by (10); the inverse in that formula is assigned zero when Y(t)Y(t) is not invertible. (Y(t),Z(t))=ΦℬLA,p(t)(I,ℓI),P(t)=Z(t)Y(t)−1.(Y(t),Z(t))=\Phi_{\mathcal B_{L_A,p}}(t)(I,\ell I),\qquad P(t)=Z(t)Y(t)^{-1}.(10)

For a scalar potential λ\lambda, drift pp, initial displacement a∈ℝa\in\mathbb R, and initial velocity v∈ℝv\in\mathbb R, the scalar solution phase ϕλ,pa,v\phi_{\lambda,p}^{a,v}, displacement propagator Uλ,pU_{\lambda,p}, and velocity propagator Vλ,pV_{\lambda,p} are given by (11). For a real parameter θ\theta, write Dθ=Uλ,p+θVλ,pD_\theta=U_{\lambda,p}+\theta V_{\lambda,p} and kθ=Vλ,p/Dθk_\theta=V_{\lambda,p}/D_\theta for their scalar normalizer and normalized velocity propagator. ϕλ,pa,v(t)=Φℬλ,p(t)(a,v),Uλ,p(t)=(ϕλ,p1,0(t))1,Vλ,p(t)=(ϕλ,p0,1(t))1.\begin{gathered} \phi_{\lambda,p}^{a,v}(t)=\Phi_{\mathcal B_{\lambda,p}}(t)(a,v),\qquad U_{\lambda,p}(t)=(\phi_{\lambda,p}^{1,0}(t))_1,\qquad V_{\lambda,p}(t)=(\phi_{\lambda,p}^{0,1}(t))_1. \end{gathered}(11)

The orthonormal matrix representation [T][T] of an endomorphism TT of a real inner product space EE is its matrix in a chosen orthonormal basis (ei)(e_i) of EE, with entry [T]ik=⟨ei,Tek⟩[T]_{ik}=\langle e_i,Te_k\rangle. Write 𝒀=[Y]\mathbf Y=[Y], 𝒁=[Z]\mathbf Z=[Z], and 𝑷=[P]\mathbf P=[P]. For scalar coefficients λi\lambda_i, a restart time cc defines diagonal propagators Uc,VcU_c,V_c by (12): Uc(s)=diag⁡iUλi(c+⋅),p(c+⋅)(s),Vc(s)=diag⁡iVλi(c+⋅),p(c+⋅)(s).\begin{aligned} U_c(s)&=\operatorname{diag}_i U_{\lambda_i(c+\cdot),p(c+\cdot)}(s),\\ V_c(s)&=\operatorname{diag}_i V_{\lambda_i(c+\cdot),p(c+\cdot)}(s). \end{aligned}(12)

For a radial history with [Db,0]=diag⁡id0,b,i[D_{b,0}]=\operatorname{diag}_i d_{0,b,i}, the diagonal leg coefficient λb,i\lambda_{b,i}, whole-leg propagators Ub,j,Vb,jU_{b,j},V_{b,j}, and normalizer 𝒟b,j\mathcal D_{b,j} are defined by (13). Write 𝒀b,𝑷b\mathbf Y_b,\mathbf P_b for the regular matrices belonging to Ab,p,ℓbA_b,p,\ell_b. λb,i(t)=α(t)[q0(t)ρBb+(q0(t)ρ−q0(t)ρ−1)d0,b,i],Ub,j=diag⁡iUλb,i(ej+⋅),p(ej+⋅)(hj),Vb,j=diag⁡iVλb,i(ej+⋅),p(ej+⋅)(hj),𝒟b,j=Ub,j+θbVb,j.\begin{aligned} \lambda_{b,i}(t)&=\alpha(t)\bigl[q_0(t)^\rho B_b+ (q_0(t)^\rho-q_0(t)^{\rho-1})d_{0,b,i}\bigr],\\ U_{b,j}&=\operatorname{diag}_i U_{\lambda_{b,i}(e_j+\cdot),p(e_j+\cdot)}(h_j),\\ V_{b,j}&=\operatorname{diag}_i V_{\lambda_{b,i}(e_j+\cdot),p(e_j+\cdot)}(h_j),\qquad \mathcal D_{b,j}=U_{b,j}+\theta_bV_{b,j}. \end{aligned}(13)

The Claim

Theorem 1. Let λ,p\lambda,p be continuous real functions on ℝ\mathbb R, globally bounded in absolute value by nonnegative constants, and suppose λ(t)≥0\lambda(t)\ge0 for t≥0t\ge0. Their scalar propagators satisfy (14) for t>0t>0: Uλ,p(t)>0,Vλ,p(t)>0.U_{\lambda,p}(t)>0,\qquad V_{\lambda,p}(t)>0.(14)✦ For θ>0\theta>0 and t≥0t\ge0, the normalizer bounds are (15): Dθ(t)>0,0≤kθ(t)≤θ−1.D_\theta(t)>0,\qquad 0\le k_\theta(t)\le\theta^{-1}.(15)✦

On a finite-dimensional real inner product space with a chosen orthonormal basis, let A,p,λiA,p,\lambda_i be continuous on ℝ\mathbb R, with global bounds ∥A(t)∥≤MA\|A(t)\|\le M_A, |p(t)|≤Mp|p(t)|\le M_p, and |λi(t)|≤Ni|\lambda_i(t)|\le N_i, where MA,Mp,Ni≥0M_A,M_p,N_i\ge0. Suppose [A(t)]=diag⁡iλi(t)[A(t)]=\operatorname{diag}_i\lambda_i(t) on [c,d)[c,d). For every central degree ℓ\ell and t∈[c,d]t\in[c,d], diagonal propagation is given by (16): 𝒀(t)=Uc(t−c)𝒀(c)+Vc(t−c)𝒁(c).\mathbf Y(t)=U_c(t-c)\mathbf Y(c)+V_c(t-c)\mathbf Z(c).(16)✦

Let a radial history have diagonal background [Db,0]=diag⁡id0,b,i[D_{b,0}]=\operatorname{diag}_i d_{0,b,i}. Suppose every active packet stretch Qk,a(x)Q_{k,a}(x), for k≥j0k\ge j_0, lies in [1/2,3/2][1/2,3/2] for every slot aa and real xx. For every block bb and j≥j0j\ge j_0, the complete leg restart is (17): 𝒀b(σj+1)=(Ub,j+Vb,j𝑷b(ej))𝒀b(ej).\mathbf Y_b(\sigma_{j+1})= \bigl(U_{b,j}+V_{b,j}\mathbf P_b(e_j)\bigr)\mathbf Y_b(e_j).(17)✦

For a radial history with diagonal background, fix any threshold j0j_0, block bb, axis ii, and nonnegative number dd satisfying d2+p0d=βb[q*ρBb+(q*ρ−q*ρ−1)d0,b,i]d^2+p_0d=\beta_b[q_*^\rho B_b+(q_*^\rho-q_*^{\rho-1})d_{0,b,i}]. The normalized diagonal growth has the limit (18): limj→∞log⁡((𝒟b,j)ii)δj=d.\lim_{j\to\infty}\frac{\log\bigl((\mathcal D_{b,j})_{ii}\bigr)}{\delta_j}=d.(18)✦

In the paper, the result reads: for the prescribed whole diagonal leg on degree ℓ\ell with positive normalizing parameter θℓ\theta_\ell, let U,VU,V propagate the Cauchy data (I,0),(0,I)(I,0),(0,I), set 𝒟=U+θℓV\mathcal D=U+\theta_\ell V and K=𝒟−1VK=\mathcal D^{-1}V; these matrices have positive diagonal entries, satisfy 0≤K≤θℓ−1I0\le K\le\theta_\ell^{-1}I, and are independent of current and earlier pulse parameters hh; writing 𝒀ℓ,h\mathbf Y_{\ell,h} for the regular value matrix and 𝑷ℓ,h\mathbf P_{\ell,h} for its logarithmic derivative matrix, the exact identity is (19), and the asymptotic formula (20), with nonnegative indicial roots dℓ,id_{\ell,i}, holds uniformly over the fixed finite degrees and axes: 𝒟−1𝒀ℓ,h(σj+1)=[I+K(𝑷ℓ,h(ej)−θℓI)]𝒀ℓ,h(ej),\mathcal D^{-1}\mathbf Y_{\ell,h}(\sigma_{j+1})= \bigl[I+K(\mathbf P_{\ell,h}(e_j)-\theta_\ell I)\bigr]\mathbf Y_{\ell,h}(e_j),(19) log⁡𝒟ii=δjdℓ,i+O(1+δj(j+1)−2+τj)=δj(dℓ,i+o(1)).\log\mathcal D_{ii}= \delta_jd_{\ell,i}+O\!\left(1+\delta_j(j+1)^{-2}+\tau_j\right) =\delta_j(d_{\ell,i}+o(1)).(20)

What Is Checked

The scalar and diagonal statements are over globally bounded continuous coefficients; the whole-leg statements are over radial histories and their orthonormal bases, with the packet-stretch restriction for the restart. The orthonormal matrix representation respects sums and real scalar multiples and is continuous.