OctoInformBeta

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

Lemma 6.3: Uniform estimates for parameter derivatives and outgoing slopes over admitted histories

Regular pulses approximate their ideal transfers.

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

A regular harmonic function evolves across an angular pulse through its radial value and radial velocity. The time variable is logarithmic radius, t=log⁡rt=\log r. On a finite-dimensional real inner-product space EE, write u,w∈Eu,w\in E for the radial value and velocity; the radial equation with angular coefficient A(t)A(t) and drift p(t)p(t) has the first-order block operator in (1). For a time-dependent linear operator GG and an initial vector zz in its space, the evolution ℱG(z,t)\mathcal F_G(z,t) is the vector given by the Peano-Baker series in (2). ℬA,p(t)(u,w)=(w,A(t)u−p(t)w).\mathcal B_{A,p}(t)(u,w)=(w,A(t)u-p(t)w).(1)✦ T0G(z,t)=z,Tk+1G(z,t)=∫0tG(s)TkG(z,s)ds,ℱG(z,t)=∑k=0∞TkG(z,t).\begin{aligned} T_0^G(z,t)&=z,& T_{k+1}^G(z,t)&=\int_0^t G(s)T_k^G(z,s)\,ds,\\ \mathcal F_G(z,t)&=\sum_{k=0}^{\infty}T_k^G(z,t). \end{aligned}(2) The Peano-Baker iterates TkG(z,t)T_k^G(z,t) in (2) start with the constant initial vector zz; each successive iterate is the oriented integral from zero to tt of GG applied to the preceding iterate.

For a positive central degree dd and x∈Ex\in E, the regular solution phase ΦA,p,d(x,t)\Phi_{A,p,d}(x,t) is the pair in E×EE\times E obtained by evolution from (x,dx)(x,dx); its components are the regular value uu and velocity ww in (3).✦ The regular value operator FA,p,d(t)F_{A,p,d}(t), velocity operator VA,p,d(t)V_{A,p,d}(t) and Riccati logarithmic derivative PA,p,d(t)P_{A,p,d}(t) are endomorphisms of EE defined by (4), using left composition LA(t)U=A(t)UL_A(t)U=A(t)U on the endomorphism algebra of EE. Here U[−1]U^{[-1]} is the inverse of UU when UU is invertible and is zero otherwise. ΦA,p,d(x,t)=ℱℬA,p((x,dx),t)=(uA,p,d(x,t),wA,p,d(x,t)).\Phi_{A,p,d}(x,t)=\mathcal F_{\mathcal B_{A,p}}((x,dx),t) =(u_{A,p,d}(x,t),w_{A,p,d}(x,t)).(3) FA,p,d(t)=uLA,p,d(I,t),VA,p,d(t)=wLA,p,d(I,t),PA,p,d(t)=VA,p,d(t)FA,p,d(t)[−1].\begin{aligned} F_{A,p,d}(t)&=u_{L_A,p,d}(I,t),& V_{A,p,d}(t)&=w_{L_A,p,d}(I,t),\\ P_{A,p,d}(t)&=V_{A,p,d}(t)F_{A,p,d}(t)^{[-1]}. \end{aligned}(4)

Fix an orthonormal basis e=(ei)e=(e_i) of EE. The orthonormal matrix representation [U]e[U]_e of an endomorphism UU is defined by (5). [U]e,ij=⟨ei,Uej⟩.[U]_{e,ij}=\langle e_i,Ue_j\rangle.(5)

The Pulse and Its Normalization

For a real exponent σ\sigma, a real parameter qq and endomorphisms b,Db,D of EE, the exact Berger angular coefficient 𝒜σ(b,D;q)\mathscr A_\sigma(b,D;q) is the endomorphism of EE given by (6). In Cartesian dimension nn, put m=n−1m=n-1. For an orthogonal complex structure JJ on ℝn\mathbb R^n, let DJD_J denote differentiation along the vector field x↦Jxx\mapsto Jx. For the degree-ℓ\ell Euclidean harmonic polynomial space VℓV_\ell with its Fischer inner product, the geometric specialization of (6) uses Bℓ=ℓ(ℓ+m−1)B_\ell=\ell(\ell+m-1), σ=1/m\sigma=1/m and D=DJ2D=D_J^2; a(t)>0a(t)>0 and q(t)>0q(t)>0 are the radial scale and anisotropy. 𝒜σ(b,D;q)=qσb+(qσ−qσ−1)D,Aℓ(t)=c(t)−1[BℓI+(1−q(t)−1)DJ2],c(t)=a(t)2q(t)−1/m.\begin{aligned} \mathscr A_\sigma(b,D;q)&=q^\sigma b+(q^\sigma-q^{\sigma-1})D,\\ A_\ell(t)&=c(t)^{-1}\bigl[B_\ell I+(1-q(t)^{-1})D_J^2\bigr],& c(t)&=a(t)^2q(t)^{-1/m}. \end{aligned}(6)✦

Let NN be the number of chronological slots, indexed by j=0,…,N−1j=0,\ldots,N-1. The smooth transition step sjs_j and smooth control pulse bjb_j are defined from the fixed smooth transition χ\chi in (7); the packet duration ΔN\Delta_N leaves an open round gap at each endpoint. sj(τ)=χ(τ−(2j+1)),bj(τ)=sj′(τ),ΔN=2N+1.s_j(\tau)=\chi(\tau-(2j+1)),\qquad b_j(\tau)=s_j'(\tau),\qquad \Delta_N=2N+1.(7)

Let HH be a finite-dimensional real inner-product parameter space with a fixed orthonormal basis, and let K⊂HK\subset H be compact. The amplitude functions aj:H→ℝa_j:H\to\mathbb R are differentiable at every point of KK; both aja_j and their Fréchet derivatives are continuous on KK. For a scalar BB, endomorphisms DjD_j of EE and duration TT, the individual packet stretches qjq_j are scalar functions of T,h,sT,h,s, the nonlinear angular packet 𝒫σ\mathscr P_\sigma is an endomorphism of EE depending on 𝒒\mathbf q, and the nonlinear Berger control word 𝒲T,h\mathscr W_{T,h} is an endomorphism-valued function of ss, all given by (8). qj(T,h,s)=1+ΔNTaj(h)bj(ΔNs),𝒫σ(𝒒)=BI+∑j(𝒜σ(BI,Dj;qj)−BI),𝒲T,h(s)=𝒫σ((qj(T,h,s))j).\begin{aligned} q_j(T,h,s)&=1+\frac{\Delta_N}{T}a_j(h)b_j(\Delta_Ns),\\ \mathscr P_\sigma(\mathbf q) &=BI+\sum_j\bigl(\mathscr A_\sigma(BI,D_j;q_j)-BI\bigr),\\ \mathscr W_{T,h}(s)&=\mathscr P_\sigma((q_j(T,h,s))_j). \end{aligned}(8)

For a preceding angular history A0A^0, a scalar profile α\alpha and a starting time t0t_0, the inserted angular pulse AhA_h is the coefficient in (9). Ah(t)=A0(t)+α(t)(𝒲T,h((t−t0)/T)−BI).A_h(t)=A^0(t)+\alpha(t) \bigl(\mathscr W_{T,h}((t-t_0)/T)-BI\bigr).(9)✦

For endomorphism directions XjX_j on EE, the control packet coefficient ChXC_h^X and its unit-interval version CˆhX\widehat C_h^X are endomorphism-valued functions of τ\tau and ss, respectively, given by (10). The Berger slow generators Gj(z)G_j(z) and ideal transfer W(h)W(h) are endomorphisms of EE given by (10) for real constants β,p0\beta,p_0, a positive constant θ\theta with p0+2θ≠0p_0+2\theta\ne0 and a real variable zz; LGL_G denotes left multiplication by GG on the endomorphism algebra of EE. ChX(τ)=∑jaj(h)bj(τ)Xj,CˆhX(s)=ΔNChX(ΔNs),Gj(z)=βzp0+2θ(σBI+Dj),W(h)=ℱLCˆhG(1)(I,1).\begin{aligned} C_h^X(\tau)&=\sum_j a_j(h)b_j(\tau)X_j,& \widehat C_h^X(s)&=\Delta_N C_h^X(\Delta_Ns),\\ G_j(z)&=\frac{\beta z}{p_0+2\theta}(\sigma BI+D_j),& W(h)&=\mathcal F_{L_{\widehat C_h^{G(1)}}}(I,1). \end{aligned}(10)

The scalar solution phase Ψf,p(y0,z0;s)\Psi_{f,p}(y_0,z_0;s) is the evolution from (y0,z0)(y_0,z_0) for the scalar potential ff and drift pp.✦ For a scalar potential ff, drift pp and initial slope θ\theta, the scalar Riccati baseline at time ss is the scalar function obtained by dividing the velocity component by the value component of the scalar solution phase Ψf,p(1,θ;s)\Psi_{f,p}(1,\theta;s). The shifted baseline vv with initial slope θ\theta is a scalar function of tt, and its integrated phase ϕ\phi is a real number, both specified in (11). Ψf,p(y0,z0;s)=ℱℬfI,p((y0,z0),s),(y(s),z(s))=Ψα(t0+⋅)B,p(t0+⋅)(1,θ;s),v(t)=z(max⁡{0,t−t0})y(max⁡{0,t−t0}),ϕ=∫t0t0+Tv(t)dt.\begin{aligned} \Psi_{f,p}(y_0,z_0;s)&=\mathcal F_{\mathcal B_{fI,p}}((y_0,z_0),s),\\ (y(s),z(s))&=\Psi_{\alpha(t_0+\cdot)B,\,p(t_0+\cdot)}(1,\theta;s),\\ v(t)&=\frac{z(\max\{0,t-t_0\})}{y(\max\{0,t-t_0\})},& \phi&=\int_{t_0}^{t_0+T}v(t)\,dt. \end{aligned}(11)

For h∈Hh\in H, the regular value transfer 𝒯(h)\mathcal T(h) and normalized transfer Y(h)Y(h) are endomorphisms of EE, and the outgoing slope error Q(h)Q(h) is a matrix in the orthonormal basis ee, all given by (12). Derivatives DhD_h below are Fréchet derivatives with respect to the parameter hh. 𝒯(h)=FAh,p,d(t0+T)FAh,p,d(t0)[−1],Y(h)=e−ϕ𝒯(h),Q(h)=[PAh,p,d(t0+T)]e−θI.\begin{aligned} \mathcal T(h)&=F_{A_h,p,d}(t_0+T)F_{A_h,p,d}(t_0)^{[-1]},\\ Y(h)&=e^{-\phi}\mathcal T(h),& Q(h)&=[P_{A_h,p,d}(t_0+T)]_e-\theta I. \end{aligned}(12)

The Claim

The fixed single-block data satisfy (13) for every slot jj and x∈Ex\in E, with σ,β,p0\sigma,\beta,p_0 real, B,M≥0B,M\geq0, and θ,γ>0\theta,\gamma>0. −B∥x∥2≤⟨x,Djx⟩≤0,βB=p0θ+θ2,p0+2θ≠0.-B\|x\|^2\leq\langle x,D_jx\rangle\leq0,\qquad \beta B=p_0\theta+\theta^2,\qquad p_0+2\theta\ne0.(13)

An admitted history (A0,α,p)(A^0,\alpha,p) is a triple consisting of a smooth endomorphism-valued coefficient A0A^0 on EE, a smooth scalar profile α\alpha and a continuous scalar drift pp, with global bounds ∥A0(t)∥≤MH\|A^0(t)\|\leq M_H, |p(t)|≤Mp|p(t)|\leq M_p and 0≤α(t)≤M0\leq\alpha(t)\leq M, where MH,Mp≥0M_H,M_p\geq0, and with ⟨x,A0(t)x⟩≥0\langle x,A^0(t)x\rangle\geq0 for every t,xt,x. For some d>0d>0 and real B0B_0, the central conditions are A0(t)=d(d+B0)IA^0(t)=d(d+B_0)I and p(t)=B0p(t)=B_0 for t≤0t\leq0. On a pulse interval IT=[t0,t0+T]I_T=[t_0,t_0+T] with t0≥0t_0\geq0, admission with errors δ,ε≥0\delta,\varepsilon\geq0 requires (14). A0(t)=α(t)BI,|α(t)−β|≤δ,|p(t)−p0|≤δ,p(t)≥γ(t∈IT),∥[PA0,p,d(t0)]e−θI∥≤ε.\begin{gathered} A^0(t)=\alpha(t)BI,\qquad |\alpha(t)-\beta|\leq\delta,\qquad |p(t)-p_0|\leq\delta,\qquad p(t)\geq\gamma\quad(t\in I_T),\\ \|[P_{A^0,p,d}(t_0)]_e-\theta I\|\leq\varepsilon. \end{gathered}(14)

For a finite family of blocks EbE_b, impose these conditions on each block, allowing Bb,βb,θb,Db,j,db,MH,bB_b,\beta_b,\theta_b,D_{b,j},d_b,M_{H,b} to vary with bb and retaining common σ,p0,γ,M,α,p,B0,t0,T\sigma,p_0,\gamma,M,\alpha,p,B_0,t_0,T. The amplitudes are common to all blocks. Choose an orthonormal basis ebe_b in each block. The finite matrix algebra product 𝔐\mathfrak M and the simultaneous transfers are defined in (15), using the block versions of (12) and (10). 𝔐=∏bMat⁡dim⁡Eb(ℝ),𝒀(h)=([Yb(h)]eb)b,𝑾(h)=([Wb(h)]eb)b,𝑸(h)=(Qb(h))b.\mathfrak M=\prod_b\operatorname{Mat}_{\dim E_b}(\mathbb R),\qquad \mathbf Y(h)=([Y_b(h)]_{e_b})_b,\qquad \mathbf W(h)=([W_b(h)]_{e_b})_b,\qquad \mathbf Q(h)=(Q_b(h))_b.(15)

Theorem 1. Fix the compact parameter set and amplitude functions above, and fixed single-block data satisfying (13). There are constants T0≥1T_0\geq1, ρ>0\rho>0 and C≥1C\geq1, chosen before the history, its bounds, central data, starting time and duration. For every admitted history, every T≥T0T\geq T_0 with δ+T−1+ε≤ρ\delta+T^{-1}+\varepsilon\leq\rho, and every open connected set S⊂KS\subset K, the maps Y,QY,Q are differentiable at each h∈Sh\in S and satisfy (16). max⁡{∥Y(h)−W(h)∥,∥DhY−DhW∥,∥Q(h)∥,∥DhQ∥}≤C(δ+T−1+ε),h∈S.\max\!\left\{ \|Y(h)-W(h)\|,\, \|D_hY-D_hW\|,\, \|Q(h)\|,\, \|D_hQ\| \right\} \leq C(\delta+T^{-1}+\varepsilon),\qquad h\in S.(16)✦ For fixed data on a finite family of blocks satisfying the corresponding conditions, there are T0≥1T_0\geq1, ρ>0\rho>0 and C≥1C\geq1, chosen before the admitted family and its variable data. For every such family, every T≥T0T\geq T_0 with δ+T−1+ε≤ρ\delta+T^{-1}+\varepsilon\leq\rho, and every open connected S⊂KS\subset K, the maps 𝒀,𝑸\mathbf Y,\mathbf Q are differentiable at each h∈Sh\in S and satisfy (17). max⁡{∥𝒀(h)−𝑾(h)∥,∥Dh𝒀−Dh𝑾∥,∥𝑸(h)∥,∥Dh𝑸∥}≤C(δ+T−1+ε),h∈S.\max\!\left\{ \|\mathbf Y(h)-\mathbf W(h)\|,\, \|D_h\mathbf Y-D_h\mathbf W\|,\, \|\mathbf Q(h)\|,\, \|D_h\mathbf Q\| \right\} \leq C(\delta+T^{-1}+\varepsilon),\qquad h\in S.(17)✦

In the paper, the result reads: for the fixed compact ball KK and pulse family, with pulse durations TjT_j, ηj=j−2+Tj−1\eta_j=j^{-2}+T_j^{-1}, regular value transfer Bℓ,j(h)B_{\ell,j}(h), ideal transfer Hℓ(1,h)H_\ell(1,h), outgoing slope Pℓ,e(h)P_{\ell,e}(h), baseline slope θℓ\theta_\ell and integrated parameter-independent scalar baseline ϕℓ,j\phi_{\ell,j}, one has e−ϕℓ,jBℓ,j(h)=Hℓ(1,h)+OC1(ηj)e^{-\phi_{\ell,j}}B_{\ell,j}(h)=H_\ell(1,h)+O_{C^1}(\eta_j) and Pℓ,e(h)=θℓI+OC1(ηj)P_{\ell,e}(h)=\theta_\ell I+O_{C^1}(\eta_j) uniformly on KK, with constants and threshold independent of every admitted preceding history.

What Is Checked

The claims are stated over the inserted angular coefficients (9), the regular solution operators (4), and their parameter derivatives on open connected subsets of a compact parameter set. The orthonormal matrix representation (5) is continuous and respects addition and real scalar multiplication.