OctoInformBeta

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

Lemma 6.2: Invertible radial evolution and lower barriers, with uniform incoming limits over admitted angular histories

Regular values have controlled Riccati slopes.

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

At logarithmic radius t=log⁡rt=\log r, the radial harmonic equation on a finite-dimensional real inner product space EE has the form u″(t)=A(t)u(t)−p(t)u′(t).u''(t)=A(t)u(t)-p(t)u'(t).(1) Here A(t)A(t) is the positive angular operator and p(t)p(t) is the radial drift. The regular data for (1) are u(0)=xu(0)=x and u′(0)=ℓxu'(0)=\ell x, where ℓ\ell is the central degree.

For Cartesian dimension nn and link dimension m=n−1m=n-1, put Bℓ=ℓ(ℓ+m−1)B_\ell=\ell(\ell+m-1). Here II denotes the identity operator on the space under consideration. The exact Berger coefficient ℬρ(H,D;q)\mathcal B_\rho(H,D;q), for operators H,DH,D on EE, a scalar qq and an exponent ρ\rho, is the operator defined by ℬρ(H,D;q)=qρH+(qρ−qρ−1)D.\mathcal B_\rho(H,D;q)=q^\rho H+(q^\rho-q^{\rho-1})D.(2)✦ For radial scale a(t)>0a(t)>0, anisotropy q(t)>0q(t)>0 and a Hopf square DJ2D_J^2 with angular label JJ, (2) gives the convention Aℓ(t)=a(t)−2ℬ1/m(BℓI,DJ2;q(t))=c(t)−1[BℓI+(1−q(t)−1)DJ2],c(t)=a(t)2q(t)−1/m.A_\ell(t)=a(t)^{-2}\mathcal B_{1/m}(B_\ell I,D_J^2;q(t)) =c(t)^{-1}\bigl[B_\ell I+(1-q(t)^{-1})D_J^2\bigr], \qquad c(t)=a(t)^2q(t)^{-1/m}.(3) The profiles a(t)a(t) and q(t)q(t) in (3) vary with radius; the endpoint anisotropy q*q_* used below is fixed.

Bounded coefficient data consist of continuous functions A:ℝ→End⁡(E)A:\mathbb R\to\operatorname{End}(E) and p:ℝ→ℝp:\mathbb R\to\mathbb R, with constants MA,Mp≥0M_A,M_p\ge0 satisfying ∥A(t)∥≤MA,|p(t)|≤Mp(t∈ℝ).\|A(t)\|\le M_A,\qquad |p(t)|\le M_p\qquad(t\in\mathbb R).(4) All operator norms in (4) are induced by the inner product.

Regular Evolution

For an operator coefficient KK on EE and an initial vector z∈Ez\in E, the Peano-Baker iterates Fk(K,z;t)F_k(K,z;t) and their evolution ℱK(z;t)\mathcal F_K(z;t) are EE-valued functions defined by F0(K,z;t)=z,Fk+1(K,z;t)=∫0tK(s)Fk(K,z;s)ds,ℱK(z;t)=∑k=0∞Fk(K,z;t).\begin{aligned} F_0(K,z;t)&=z,& F_{k+1}(K,z;t)&=\int_0^t K(s)F_k(K,z;s)\,ds,\\ \mathcal F_K(z;t)&=\sum_{k=0}^{\infty}F_k(K,z;t). \end{aligned}(5) The first-order block operator for (1) is KA,p(t)(u,v)=(v,A(t)u−p(t)v)K_{A,p}(t)(u,v)=(v,A(t)u-p(t)v). The regular phase ΦA,p,ℓ,x\Phi_{A,p,\ell,x} is an E×EE\times E-valued function defined by ΦA,p,ℓ,x(t)=ℱKA,p((x,ℓx);t)\Phi_{A,p,\ell,x}(t)=\mathcal F_{K_{A,p}}((x,\ell x);t), using (5); its first component uA,p,ℓ,xu_{A,p,\ell,x} is the regular value and its second component vA,p,ℓ,xv_{A,p,\ell,x} is the regular velocity.

Left composition by A(t)A(t) is the operator LA(t)T=A(t)∘TL_A(t)T=A(t)\circ T on End⁡(E)\operatorname{End}(E).✦ For coefficient data A,pA,p on EE and central degree ℓ\ell, the regular value operator UU, velocity operator WW and Riccati logarithmic derivative PP are functions with values in End⁡(E)\operatorname{End}(E) defined by UA,p,ℓ(t)=uLA,p,ℓ,I(t),WA,p,ℓ(t)=vLA,p,ℓ,I(t),PA,p,ℓ(t)=WA,p,ℓ(t)UA,p,ℓ(t)−1.\begin{aligned} U_{A,p,\ell}(t)&=u_{L_A,p,\ell,I}(t),& W_{A,p,\ell}(t)&=v_{L_A,p,\ell,I}(t),\\ P_{A,p,\ell}(t)&=W_{A,p,\ell}(t)U_{A,p,\ell}(t)^{-1}. \end{aligned}(6) The inverse in (6) is defined to be zero for singular UU.

For a chosen orthonormal basis ebe_b of a space EbE_b, write 𝖬b(T)\mathsf M_b(T) for the matrix of TT in that basis; this is a continuous linear map of operator spaces.✦

Angular Histories

A pulse system on a block set Ξ\Xi consists of finite-dimensional real inner product spaces EbE_b, indexed by b∈Ξb\in\Xi and equipped with chosen orthonormal bases, a set κ\kappa of angular labels, and operators Db,hD_{b,h} on EbE_b for h∈κh\in\kappa.✦ Its scalar data are an exponent ρ\rho, limiting drift p0p_0, constants M≥0M\ge0 and γ>0\gamma>0, and block parameters Bb≥0B_b\ge0, βb\beta_b and θb>0\theta_b>0 satisfying −Bb∥x∥2≤⟨x,Db,hx⟩≤0,βbBb=θb(θb+p0),p0+2θb≠0.\begin{gathered} -B_b\|x\|^2\le\langle x,D_{b,h}x\rangle\le0,\\ \beta_bB_b=\theta_b(\theta_b+p_0),\qquad p_0+2\theta_b\ne0. \end{gathered}(7)✦

A radial history for (7) specifies symmetric background operators Db,0D_{b,0} satisfying the same quadratic bounds, and requires every Db,hD_{b,h} to be symmetric. It specifies q*≥1q_*\ge1, degrees ℓb>0\ell_b>0, a central drift B0B_0, a smooth nonnegative radial coefficient α\alpha, and a continuous drift pp, with $$\begin{gathered} B_b=\ell_b(\ell_b+B_0),\qquad |\alpha(t)|\le M,\qquad |p(t)|\le M_p,\qquad M_p\ge0,\\ \alpha(t)=1,\quad p(t)=B_0\quad(t\le0),\\ \lim_{t\to\infty}\alpha(t)=\beta_b,\qquad \lim_{t\to\infty}p(t)=p_0,\qquad p(t)\ge\gamma \quad\hbox{for all sufficiently large }t. \end{gathered}$$(8) In the geometric specialization of (8), α=a−2\alpha=a^{-2}.

Let σ\sigma be the fixed smooth transition function. For NN chronological slots, an amplitude vector a=(ai)i=0N−1a=(a_i)_{i=0}^{N-1}, a stretch parameter TT and a packet coordinate uu, the packet duration is dN=2N+1d_N=2N+1; the smooth transition step hih_i, its control pulse ηi\eta_i, and the individual packet stretch QiQ_i are hi(u)=σ(u−(2i+1)),ηi(u)=hi′(u),Qi(a,T,u)=1+T−1dNaiηi(dNu),0≤i<N.h_i(u)=\sigma(u-(2i+1)),\qquad \eta_i(u)=h_i'(u),\qquad Q_i(a,T,u)=1+T^{-1}d_Na_i\eta_i(d_Nu),\quad 0\le i<N.(9) The duration leaves an open round gap at each endpoint. For an operator HH, a family of operators D=(Di)i=0N−1D=(D_i)_{i=0}^{N-1} and scalar parameters q=(qi)i=0N−1q=(q_i)_{i=0}^{N-1}, the nonlinear angular packet 𝒩ρ\mathcal N_\rho and nonlinear Berger control word 𝒲ρ\mathcal W_\rho are operator-valued functions defined using (2) and (9): 𝒩ρ(H,D;q)=H+∑i=0N−1(ℬρ(H,Di;qi)−H),𝒲ρ(H,D,a;T,u)=𝒩ρ(H,D;(Qi(a,T,u))i=0N−1).\begin{aligned} \mathcal N_\rho(H,D;q)&=H+\sum_{i=0}^{N-1} \bigl(\mathcal B_\rho(H,D_i;q_i)-H\bigr),\\ \mathcal W_\rho(H,D,a;T,u)&= \mathcal N_\rho(H,D;(Q_i(a,T,u))_{i=0}^{N-1}). \end{aligned}(10)

For indices j≥0j\ge0, corresponding to paper periods j+1j+1, the pulse duration τj\tau_j, dwell duration δj\delta_j and period start TjT_j are τj=(j+1)6,δj=(j+1)7,Tj=10+∑i=0j−1(δi+4τi).\tau_j=(j+1)^6,\qquad \delta_j=(j+1)^7,\qquad T_j=10+\sum_{i=0}^{j-1}(\delta_i+4\tau_i).(11) The smooth ramp RR and dwell window wjw_j for (11) are R(u)=σ(2u−1/2),wj(t)=R(t−Tj−τjτj)−R(t−Tj−2τj−δjτj).R(u)=\sigma(2u-1/2),\qquad w_j(t)=R\left(\frac{t-T_j-\tau_j}{\tau_j}\right) -R\left(\frac{t-T_j-2\tau_j-\delta_j}{\tau_j}\right).(12) For control pulses zjz_j, the period anisotropy increment Δj[z]\Delta_j[z] and pulsed profile q[z]q[z] with starting threshold j0j_0 are Δj[z](t)=τj−1zj(t−Tjτj)+(q*−1)wj(t),q[z](t)=1+∑j≥j0Δj[z](t).\Delta_j[z](t)=\tau_j^{-1}z_j\left(\frac{t-T_j}{\tau_j}\right) +(q_*-1)w_j(t),\qquad q[z](t)=1+\sum_{j\ge j_0}\Delta_j[z](t).(13) The diagonal profile q[0]q[0] in (13) has zero control pulses and is round initially and throughout every pulse.✦

Fix a slot assignment ν:{0,…,N−1}→κ\nu:\{0,\ldots,N-1\}\to\kappa and amplitude stream s=(sj,i)s=(s_{j,i}). The period angular increment ℐb,j\mathcal I_{b,j}, infinite angular coefficient 𝒞b;s,j0\mathcal C_{b;s,j_0} and radial equation coefficient Ab;s,j0A_{b;s,j_0} are ℐb,j(t)=𝒲ρ(BbI,(Db,ν(i))i,sj;τj,t−Tjτj)−BbI,𝒞b;s,j0(t)=ℬρ(BbI,Db,0;q[0](t))+∑j≥j0ℐb,j(t),Ab;s,j0(t)=α(t)𝒞b;s,j0(t).\begin{aligned} \mathcal I_{b,j}(t)&= \mathcal W_\rho\left(B_bI,(D_{b,\nu(i)})_i,s_j; \tau_j,\frac{t-T_j}{\tau_j}\right)-B_bI,\\ \mathcal C_{b;s,j_0}(t)&= \mathcal B_\rho(B_bI,D_{b,0};q[0](t))+ \sum_{j\ge j_0}\mathcal I_{b,j}(t),\\ A_{b;s,j_0}(t)&=\alpha(t)\mathcal C_{b;s,j_0}(t). \end{aligned}(14) The active angular increment is ℐb,j\mathcal I_{b,j} for j≥j0j\ge j_0 and zero otherwise, so that (14) sums the active increments.✦ An amplitude stream s=(sj,i)s=(s_{j,i}) is admitted for a starting threshold j0j_0 when it satisfies 12≤Qi(sj,τj,u)≤32(j≥j0,u∈ℝ,0≤i<N).\frac12\le Q_i(s_j,\tau_j,u)\le\frac32 \qquad(j\ge j_0,\ u\in\mathbb R,\ 0\le i<N).(15) For (14), abbreviate the slope in (6) by Pb;s,j0=PAb;s,j0,p,ℓbP_{b;s,j_0}=P_{A_{b;s,j_0},p,\ell_b}.

The Claim

Theorem 1. Let A,pA,p be bounded coefficient data as in (4). Let BB be a real constant, and suppose ℓ>0\ell>0, ⟨x,A(t)x⟩≥0\langle x,A(t)x\rangle\ge0 for all t,xt,x, and A(t)=ℓ(ℓ+B)IA(t)=\ell(\ell+B)I, p(t)=Bp(t)=B for t≤0t\le0. Then UA,p,ℓ(t)∈GL⁡(E)(t∈ℝ).U_{A,p,\ell}(t)\in\operatorname{GL}(E)\qquad(t\in\mathbb R).(16)✦ For arbitrary real ℓ\ell, suppose instead that A(t)A(t) is symmetric at every time and UA,p,ℓ(t)U_{A,p,\ell}(t) is invertible at the time under consideration. Then ⟨PA,p,ℓ(t)x,y⟩=⟨x,PA,p,ℓ(t)y⟩(x,y∈E).\langle P_{A,p,\ell}(t)x,y\rangle =\langle x,P_{A,p,\ell}(t)y\rangle\qquad(x,y\in E).(17)✦ For arbitrary real ℓ\ell, at every time where UA,p,ℓU_{A,p,\ell} is invertible, PA,p,ℓ′(t)=A(t)−p(t)PA,p,ℓ(t)−PA,p,ℓ(t)2.P_{A,p,\ell}'(t)=A(t)-p(t)P_{A,p,\ell}(t)-P_{A,p,\ell}(t)^2.(18)✦ Let c≤ℓc\le\ell and t≥0t\ge0, and suppose UA,p,ℓ(t)U_{A,p,\ell}(t) is invertible. If ⟨z,A(s)z⟩≥(cp(s)+c2)∥z∥2\langle z,A(s)z\rangle\ge(cp(s)+c^2)\|z\|^2 for every s≥0s\ge0 and z∈Ez\in E, then c∥x∥2≤⟨x,PA,p,ℓ(t)x⟩(x∈E).c\|x\|^2\le\langle x,P_{A,p,\ell}(t)x\rangle\qquad(x\in E).(19)✦ Under the hypotheses of (16), with AA also symmetric, ∥PA,p,ℓ(t)∥≤ℓ+MA+Mp+1(t≥0).\|P_{A,p,\ell}(t)\|\le\ell+M_A+M_p+1\qquad(t\ge0).(20)✦ For a pulse system and radial history as above, each ε>0\varepsilon>0 admits an index JJ such that, for every starting threshold j0j_0, every stream satisfying (15), every block bb and every j≥Jj\ge J, ∥𝖬b(Pb;s,j0(Tj))−θbI∥≤ε.\bigl\|\mathsf M_b(P_{b;s,j_0}(T_j))-\theta_bI\bigr\|\le\varepsilon.(21)✦

In the paper, the result reads: for the prescribed finite geometry and controls, with jgeomj_{\mathrm{geom}} the geometric starting threshold, LL the largest retained degree and θℓ\theta_\ell the limiting slope for degree ℓ\ell, there are jreg≥jgeomj_{\mathrm{reg}}\ge j_{\mathrm{geom}} and 0<cP<CP0<c_P<C_P, uniform over histories, such that for every j0≥jregj_0\ge j_{\mathrm{reg}} and retained degree 2≤ℓ≤L2\le\ell\le L there is a unique smooth symmetric PℓP_\ell satisfying Pℓ′=Aℓ−pPℓ−Pℓ2P_\ell'=A_\ell-pP_\ell-P_\ell^2 and Pℓ=ℓIP_\ell=\ell I for t≤2t\le2, with cPI≤Pℓ(t)≤CPIc_PI\le P_\ell(t)\le C_PI for every tt, the solutions of Y′=PℓYY'=P_\ell Y are exactly the center-regular solutions of the second-order harmonic equation, arbitrary value data at one time determine a unique such solution, and for some constant CC, at every pulse start TjT_j, including the first, ∥Pℓ(Tj)−θℓI∥≤C(j+1)−2\|P_\ell(T_j)-\theta_\ell I\|\le C(j+1)^{-2} uniformly over all preceding admitted controls.

What Is Checked

The invertibility statement (16), symmetry (17), Riccati equation (18), lower bound (19) and norm bound (20) concern the constructed regular evolution on finite-dimensional real inner product spaces with bounded continuous coefficients; (21) is stated over the pulse systems and radial histories of (7) and (8), using the actual coefficient (14) and admission condition (15).