betaFeedback

WorksMathematics

An exhaustion-aware product moment and zero-free half-planes for Hecke L-functions

Abstract

We refine the ordinary product moment in the cubic-theta comparison by allowing a returned prime product to be exhausted. The return stops either when the positive-prime length inequality holds or when the established prime-free ordinary estimate applies. Keeping the extracted central coefficients gives a local normalization inequality on a larger domain. On the critical ordinary-length interval this yields a net first-mean gain of 1/200001/20000. We prove the complementary-class, short-sum and algebraic comparison inequalities directly, and obtain the endpoint B∗−1/160000=0.87446205769353…B_*-1/160000=0.87446205769353\ldots, where B∗B_* is Liu's algebraic boundary. The zero-free conclusion is conditional on the cited general-field theta, detector, prior-bound and Mellin input theorems in their stated uniform forms. The proof includes the actual return weights, both approximate-functional-equation arms, the principal-residue normalization and descent to arbitrary number fields.

Introduction

The cubic-theta method relates zero-free half-planes to moments of ordinary character sums and to the number of character parameters in a detected amplitude class. Liu’s general-field formulation [1] supplies the ordinary moment and the zero-detecting comparison. His subsequent amplified comparison [2] reaches the algebraic boundary B∗=0.87446830769353045955…B_* = 0.87446830769353045955\ldots. We enlarge the admissible additional prime product by changing the stopping rule in the ordinary return. The change concerns the complete positive norm, before restriction to a detected class.

Let r∗r_* be the unique root in (9/20,23/50)(9/20, 23/50) of

p(r)=218582r3−376469r2−11376r+61560=0.(1)p(r) = 218582r^3 - 376469r^2 - 11376r + 61560 = 0. \tag*{(1)}

Define

ℓ∗=3410r∗2−7127r∗+25202286r∗2−5011r∗+1800,B∗=1112−ℓ∗4,k∗=56−ℓ∗2.(2)\ell_* = \frac{3410r_*^2 - 7127r_* + 2520}{2286r_*^2 - 5011r_* + 1800}, \qquad B_* = \frac{11}{12} - \frac{\ell_*}{4}, \qquad k_* = \frac{5}{6} - \frac{\ell_*}{2}. \tag*{(2)}

For independent variables rr, kk, put

nb(r,k)=(−6858r2+15033r−5400)k+600r2−1837r+720,ne(r,k)=(−702r2+34995r−40500)k−408r2−3775r+5400,n0(r,k)=(4806r2−10887r+8100)k−696r2+1403r−1080,\begin{aligned} n_b(r,k) &= (-6858r^2 + 15033r - 5400)k + 600r^2 - 1837r + 720,\\ n_e(r,k) &= (-702r^2 + 34995r - 40500)k - 408r^2 - 3775r + 5400,\\ n_0(r,k) &= (4806r^2 - 10887r + 8100)k - 696r^2 + 1403r - 1080, \end{aligned}

and set

b∗=−ℓ∗∂rne(r∗,k∗)+∂rn0(r∗,k∗)∂rnb(r∗,k∗).(3)b_* = -\frac{\ell_*\partial_r n_e(r_*, k_*) + \partial_r n_0(r_*, k_*)}{\partial_r n_b(r_*, k_*)}. \tag*{(3)}

The derivatives in this definition hold kk fixed. Our new length and boundary are

ℓ=ℓ∗+140000,B=1112−ℓ4=B∗−1160000.(4)\ell= \ell_* + \frac{1}{40000}, \qquad B = \frac{11}{12} - \frac{\ell}{4} = B_* - \frac{1}{160000}. \tag*{(4)}

Theorem 1.1. Assume the general-field analytic inputs specified in § 2. For each fixed number field FF containing μ12\mu_{12}, every finite-order Hecke LL-function over FF is zero-free in Re⁡s>B\operatorname{Re} s > B. If these inputs hold for every such FF, the same conclusion holds for every finite-order Hecke LL-function over every number field. In particular, it holds for the Riemann zeta function and for Dirichlet LL-functions under those inputs.

The main new estimate is Theorem 4.1. Its proof retains the signed divisor expansion until the actual ordinary children have been formed. The pointwise prime cap is then applied to whole factors. At exhaustion the remaining product is estimated by the prime-free theorem, even when its formal positive-prime defect is still positive. An exact normalization identity pays this operation. The numerical improvement follows by using a slightly longer selected product on 41/100≤m≤43/10041/100 \le m \le43/100.

The terminal ordinary estimates and theta construction are invoked from the cited papers. The local return, the scalar comparison, and the passage from a positive complete saving to the new endpoint are developed below. In particular, the scalar inequalities are proved by rational differentiation and a double-root identity rather than by an auxiliary certificate table. All constants, derivative orders and thresholds may depend on the fixed field and target character; the boundary BB does not.

Arithmetic families and analytic inputs

Fix F⊃μ12F \supset\mu_{12}. Choose a finite set SS of places, admissible generators of the relevant ideals, and a finite group Θ\Theta of finite-place and unit characters as in [1 Part III]. Write qa=Naq_a = Na. The sextic residue character χa\chi_a is extended by zero at every nonunit of its displayed modulus. The displayed moving radical is retained throughout a divisor extraction, including factors that cancel algebraically. We use the normalized full Gauss sum

G(a,h)=qa−1/2∑v mod aχa(v)e(hv/a),G(1,h)=1,\mathcal{G}(a,h) = q_a^{-1/2} \sum_{v \bmod a} \chi_a(v)e(hv/a), \qquad\mathcal{G}(1,h) = 1,

with the trace additive character and local measures of [1]. The imprimitive formula is used also at nonunit frequencies.

For a primitive periodic character λ\lambda of modulus mm and (D,m)=1(D,m)=1, the nonzero part of orbit Poisson, with physical norm scale YY and Φ(0)=0\Phi(0)=0, is

∑d∣Dμ(d)λ(d)Yτ(λ)qdqm∑[h]≠0λ(h)‾KΦ(Yqhqdqm).(5)\sum_{d\mid D} \frac{\mu(d)\lambda(d)Y\tau(\lambda)}{q_dq_m}\sum_{[h]\ne0}\overline{\lambda(h)}K_\Phi\left(\frac{Yq_h}{q_dq_m}\right). \tag*{(5)}

This follows by expanding coprimality, writing the row as dkdk, applying lattice Poisson to each residue class, and evaluating its finite Fourier transform. If the primitive modulus is one, add the origin Yvol⁡(G/Γ)Φ^(0)∑d∣Dμ(d)/qdY\operatorname{vol}(G/\Gamma)\widehat{\Phi}(0)\sum_{d\mid D}\mu(d)/q_d; otherwise that origin vanishes. For an imprimitive character retain its full periodic Fourier transform instead of replacing it by τ(λ)λ‾\tau(\lambda)\overline{\lambda}. The origin has the compact-quotient measure, whereas the nonzero orbits unfold to GG. These are the two measures in the theta normalization of [1].

An ordinary annular factor has the form

SX,V(u)=X−1/2∑ac(a)χu(a)V(qa/X).(6)S_{X,V}(u)=X^{-1/2}\sum_a c(a)\chi_u(a)V(q_a/X). \tag*{(6)}

where the sum is over the prescribed generator classes, cc belongs to the fixed finite character span, and the common coprimality conditions are written with the same zero extension. Different factors may have different admissible profiles and independent Mellin heights. A prime factor is the corresponding normalized sum on a full prime annulus, with the original finite-Θ\Theta coefficients. Its logarithmic width relative to UU is denoted by ziz_i. Products below always use disjoint prime slots. These definitions do not allow coefficients chosen as a function of the current row.

We recall precisely the external results used. The theta coefficients a0a_0, their twisted multiplicativity, reciprocity, full local Gauss identities and orbit Poisson summation are those of [1 Part III], with the metaplectic normalization originating in [3, 4]. The inverse moment and the ordinary product moment, including arbitrary bounded ordinary lengths when no prime factor remains, are the theorems in the sections “The unweighted product moment” and its finite-width induction in [1]. They include fixed divisor weights, natural zeros, finite twists and polynomial control of all requested finite-order profile seminorms.

For zero detection we use the original inverse detector, its amplitude-class count, the same-object contour comparison, both actual-conductor approximate functional equations, and the direct estimate of [1, 2]. The prior result β∗(F)≤B∗<0.8745\beta_\ast(F) \le B_\ast< 0.8745 from [2] is assumed before any new prime cap is used. Here

β∗(F)=sup⁡{Re⁡ρ:LF(ρ,η)=0, 0<Re⁡ρ<1, η of finite order}.\beta_\ast(F)=\sup\{\operatorname{Re}\rho:L_F(\rho,\eta)=0,\ 0<\operatorname{Re}\rho<1,\ \eta\text{ of finite order}\}.

This prior bound licenses the local cap with κ=749/1000\kappa=749/1000 and the count with parameter k∗k_\ast. Its proof is an input, rather than an inference from the improved endpoint.

Finally we use the normalized theta/Mellin identity in [1, 2], the principal contour and continuation argument. With c0=(4+b)/6c_0=(4+b)/6 it identifies the Mellin transform of fη(Z)f_\eta(Z) with

exp⁡((s−5/6)2)Hη(s)LF(s,η)(7)\exp((s-5/6)^2)\frac{H_\eta(s)}{L_F(s,\eta)} \tag*{(7)}

where HηH_\eta is holomorphic and nonzero for Re⁡s>437/500\operatorname{Re}s>437/500. The direct estimate for the normalized average JηJ_\eta has exponent B(ℓ)−c0+εB(\ell)-c_0+\varepsilon, and the principal contour produces the same fηf_\eta. These input statements include their local corrections and their allowed order of choices. Theorem 1.1 is an implication from these analytic results; it does not require the reader to adopt unrecorded numerical computations.

Return weights and profile closure

We first describe why the two transforms return the families just defined. This also fixes how their weights are used in the ordinary argument.

The inverse family

The canonical inverse energy of [1] is

E=Z−N−ν∑f squarefreeqf<Zνw(f)∑0<qk≤CZM∣∑n squarefreea0(n)ν(n)ρ(n)χk(n)χn(f)4W(qn/ZN)∣2(8)\mathcal{E}=Z^{-N-\nu} \sum_{\substack{f\ \mathrm{squarefree}\\q_f<Z^\nu}}w(f) \sum_{0<q_k\le CZ^M} \left| \sum_{n\ \mathrm{squarefree}}a_0(n)\nu(n)\rho(n)\chi_k(n)\chi_n(f)^4W(q_n/Z^N) \right|^2 \tag*{(8)}

where ρ(n)=1(n,Rρ)=1\rho(n)=\mathbf{1}_{(n,R_\rho)=1}, w(f)≪τ(f)Lw(f)\ll\tau(f)^L is independent of kk, and ν\nu is a fixed finite twist. Complete cube extraction writes N=r+3aN=r+3a and gives the original squared normalization Z−N−νYbZ^{-N-\nu}Y_b, with YrYb3=ZNY_rY_b^3=Z^N. Möbius coefficients in this extraction remain signed. In the first Poisson transform write ni=AiCuin_i=A_iC u_i and expand 1(u1,u2)=1\mathbf{1}_{(u_1,u_2)=1} by its full divisor identity. The cube-prime valuation states give ideals q0q_0, J=sJ2J=sJ_2, EE, R1R_1, with b1b2=q02J22sb_1b_2=q_0^2J_2^2s. The actual residual coefficient is

μ(u)ν1(u)ρ(u)χu(y)‾χu(CJ)4χu(dk)‾1(u,q0)=1,y=hf2E.\mu(u)\nu_1(u)\rho(u)\overline{\chi_u(y)}\chi_u(CJ)^4\overline{\chi_u(d_k)}\mathbf{1}_{(u,q_0)=1}, \qquad y=hf^2E.

For fixed yy the old labels f,hf,h have divisor-bounded multiplicity, since f2E∣yf^2E\mid y determines hh. This bound is taken on the original support, before positive enlargement. Thus it does not incur a second factor ZvZ^v.

For the second transform write x_j=g^'z_j, zj=v′njz_j=v'n_j, d2∣g′d_2\mid g', and put

k′=dkd2k′′,f′=JCd2v′,ργ(n)=ρ(n)1(n,rad⁡(q0t′g′/d2))=1.k'=d_kd_2k'', \qquad f'=JCd_2v', \qquad\rho_\gamma(n)=\rho(n)\mathbf{1}_{(n,\operatorname{rad}(q_0t'g'/d_2))=1}.

The returned coefficient is exactly

a0(n)ν2(n)ργ(n)χn(dkk′′)χn(d2)‾χn(CJv′)4=a0(n)ν2(n)ργ(n)χn(k′)χn(f′)‾4.a_0(n)\nu_2(n)\rho_\gamma(n)\chi_n(d_kk'')\overline{\chi_n(d_2)}\chi_n(CJv')^4 = a_0(n)\nu_2(n)\rho_\gamma(n)\chi_n(k')\overline{\chi_n(f')}^4.

Indeed χn(d2)5=χn(d2)‾\chi_n(d_2)^5=\overline{\chi_n(d_2)} when the arguments are coprime; otherwise both sides vanish. The fixed unit Gauss factors give ν2=νξ1ξ2\nu_2=\nu\xi_1\xi_2. On the support forced by twisted multiplicativity, J,C,d2,v′J,C,d_2,v' are pairwise coprime and squarefree. Hence f′f' is an admissible label, including at every natural zero.

Lemma 3.1. The return uses one common positive structural projection measure. If each child satisfies Echild≤AZF′+λSchildE_{\mathrm{child}}\le AZ^{F'+\lambda}S_{\mathrm{child}} uniformly over the permitted data, the parent bound contains AZλAZ^\lambda once.

Proof. Fix γ=(q0,t′,g′/d2)\gamma=(q_0,t',g'/d_2). For fixed (γ,f′,k′)(\gamma,f',k'), splitting f′f' into J,C,d2,v′J,C,d_2,v' and then splitting JJ has divisor-bounded multiplicity; the old (f,h)(f,h) fiber was already counted. If j,t,g,ϑ,Rj,t,g,\vartheta,R denote the respective logarithmic scales of J,t′,g′,d2,R1J,t',g',d_2,R_1, the structural mass exponent is

Cc=a+R/2−j+t+g−ϑ.C_c=a+R/2-j+t+g-\vartheta.

Write its weights as AγA_\gamma and use the same measure

dν(γ)=Z−Cc∑γAγδγd\nu(\gamma)=Z^{-C_c}\sum_\gamma A_\gamma\delta_\gamma

in both Cauchy integrals. The column and label centers are

Nc=r−Ai−B−t−g−v,Vc=B+ϑ+v+j,Fc=Nc+Vc.N_c=r-A_i-B-t-g-v,\qquad V_c=B+\vartheta+v+j,\qquad F_c=N_c+V_c.

The exact Poisson prefactors satisfy

κi+λc+Cc+2Fc=r+3a+V=F.(9)\kappa_i+\lambda_c+C_c+2F_c=r+3a+V=F. \tag*{(9)}

One F′F' restores the normalization in (8); the other comes from the child bound. If the two resulting positive integrals are Ii=∫Ei dνiI_i=\int\mathcal{E}_i\,d\nu_i and Ti=∫Si dνiT_i=\int\mathcal{S}_i\,d\nu_i, use

(I1I2)1/2≤AZF′+λ(T1T2)1/2.(I_1I_2)^{1/2}\le AZ^{F'+\lambda}(T_1T_2)^{1/2}.

The measures and twists on the two sides need not coincide. Their common uniform bound and (9) transmit AA and λ\lambda once. In particular, no interchange of the geometric mean with the integrals is used.

Profiles and the ordinary return

All differentiation is performed in actual fixed-ratio coordinates ui=log⁡(qi/Yi)u_i=\log(q_i/Y_i). A typical separated profile is

PZ(u)=χ(u)K(ZζeL(u)eR(u))∏jWj(cjeLj(u))eit⋅Au.P_Z(u)=\chi(u)K\left(Z^\zeta e^{L(u)}e^{R(u)}\right)\prod_j W_j\left(c_j e^{L_j(u)}\right)e^{it\cdot Au}.

The norm-sheet kernels from the fixed Schwartz–Bruhat lifts obey logarithmic bounds as their argument tends to zero and arbitrary polynomial decay at infinity. Integration by parts on the fixed uu-box therefore gives, for each prescribed AA, some finite J,CJ,C such that

∫(1+∣ξ∣)A∣PZ^(ξ)∣ dξ≪(log⁡Z)dF(1+∣t∣)CSJ.(10)\int(1+|\xi|)^A|\widehat{P_Z}(\xi)|\,d\xi\ll(\log Z)^{d_F}(1+|t|)^C S_J. \tag*{(10)}

The bound is uniform in the actual divisor scales cjc_j. It has no factor ZJηZ^{J\eta} from a coarse logarithmic mesh.

The physical positive row lift need not be compactly supported. Apply the second exact Poisson transform to that lift before identifying a child ball. Its compact dual support gives the actual upper bound on the new row. A negative formal center alone does not delete a nonempty bounded row.

We spell out the ordinary paired weights. In a final artificial-divisor dyad Ut−≤qt<2Ut−U^{t_-}\le q_t<2U^{t_-}, let r~j\widetilde r_j be the logarithmic removed ordinary length, ωj\omega_j the fixed removed prime-slot length, eje_j the support-conversion error, and πj,πj′\pi_j,\pi'_j the nominal endpoint allowances on side jj. The actual central multiplier on that side is

U(−rj++ej+πj′)/2,rj+=r~j−(πj−πj′),r~j+ωj≥t−.U^{(-r_j^++e_j+\pi'_j)/2},\qquad r_j^+=\widetilde r_j-(\pi_j-\pi'_j),\qquad\widetilde r_j+\omega_j\ge t_-.

Thus if ct,αc_{t,\alpha} is the frozen coefficient of the complete last-divisor expansion, the weight in both positive integrals is

bt,α=∣ct,α∣U(−r1++e1+π1′−r2++e2+π2′)/2.b_{t,\alpha}=|c_{t,\alpha}|U^{(-r_1^++e_1+\pi'_1-r_2^++e_2+\pi'_2)/2}.

The original endpoint rules give

t−+log⁡U(bt,α/∣ct,α∣)≤e1+ω1+e2+ω2+π1+π22≤2θN.t_-+\log_U(b_{t,\alpha}/|c_{t,\alpha}|)\le\frac{e_1+\omega_1+e_2+\omega_2+\pi_1+\pi_2}{2}\le2\theta_N.

Here θN=O(1/log⁡U)\theta_N=O(1/\log U) records the total fixed-ratio endpoint width of the finite factor list. With divisor multiplicities included, ∑bt,α≪U2θN+ε\sum b_{t,\alpha}\ll U^{2\theta_N+\varepsilon}. The common structural radical sBs_B contributes its actual count Up2−s0+εU^{p_2-s_0+\varepsilon} and its opposite normalization U−(p2−s0)U^{-(p_2-s_0)}. Hence each Cauchy-side measure is the structural measure times this same bt,αb_{t,\alpha} and the absolute Fourier density. Integrate that density using (10), then insert the same low-rank uniform estimate separately on the two sides. In particular, the factor Ut−U^{t_-} is never charged again as a bare divisor count.

Reflection also respects row dependence. For primitive inducing character ψk∗\psi_k^*, redundant radical R0,kR_{0,k} and actual conductor CkC_k, the Euler expansion of the functional equation is

Sk0(X;W)=ϵk∑d∣R0,krad⁡(h)∣R0,kμ(d)ψk∗(d)ψk∗(h)‾qdqhSψk0/ψk0(CkqdXqh;W♯).S_k^0(X;W)=\epsilon_k\sum_{\substack{d\mid R_{0,k}\\ \operatorname{rad}(h)\mid R_{0,k}}}\frac{\mu(d)\psi_k^*(d)\overline{\psi_k^*(h)}}{\sqrt{q_dq_h}}S_{\psi_k^0/\psi_k}^{0}\left(\frac{C_kq_d}{Xq_h};W^\sharp\right).

Its absolute Euler mass is bounded, separately on each row, by

Lk≤∏p∣R0,k1+qp−1/21−qp−1/2≪εUε.L_k\le\prod_{p\mid R_{0,k}}\frac{1+q_p^{-1/2}}{1-q_p^{-1/2}}\ll_\varepsilon U^\varepsilon.

To see uniformity, first fix a sufficiently large prime cutoff; above it the logarithm of each factor is at most εlog⁡qp\varepsilon\log q_p, and below it only a fixed constant remains. Decompose W♯W^\sharp into actual annuli. Their central coefficients are summable, including the logarithmic small-argument terms, and depend on the incoming profiles rather than the row. For each profile apply rowwise Cauchy with LkL_k before the common two-scale Sobolev estimate. This retains the mixed derivative and does not require a common Euler-divisor set for all rows. Equation (10) now closes any prescribed finite-order child seminorm. The strict width decrease bounds the depth in advance, so these orders can all be fixed before the external height exponent.

An exhaustion-aware ordinary norm

Put

κ=7491000,K=6κ=2247500,cK=K−1K.\kappa= \frac{749}{1000}, \qquad K = 6\kappa= \frac{2247}{500}, \qquad c_K = \frac{K-1}{K}.

The prior half-plane in Section 2 permits the pointwise squared prime cost Uκz+εU^{\kappa z+\varepsilon}. For m∈[m−,ro]m \in[m_-,r_o] form

P=SUm,V1SUm,V2Q0,Pc=P−Pcmp.P = S_{Um,V_1}S_{Um,V_2}Q_0, \qquad P^c = P-P^{\mathrm{cmp}}.

The comparison replaces the ordinary scales (m,m)(m,m) by (1/4,2m−1/4)(1/4,2m-1/4), preserving their product, both profiles, and all prime factors. The positive row measure is the prescribed smooth measure Wrow(qu/U)W_{\mathrm{row}}(q_u/U) of [2].

Theorem 4.1. Assume the local arithmetic and ordinary input theorems of Section 2. Suppose

41100≤m−≤ro≤43100,0≤W≤140,1−2ro−7W>0,52−4m−+KW<1.\frac{41}{100} \le m_- \le r_o \le\frac{43}{100}, \qquad0 \le W \le\frac{1}{40}, \qquad1-2r_o-7W>0, \qquad\frac{5}{2}-4m_-+KW<1.

Let Q0,EQ_0,E be complete disjoint prime products of widths 0≤z≤(1−2m)/K0\le z\le(1-2m)/K and 0≤w≤W0\le w\le W. Then, for every ε>0\varepsilon>0,

∑uWrow(qu/U)∣Pc(u)E(u)∣2≪U1+εSJ(1+H)C,(11)\sum_u W_{\mathrm{row}}(q_u/U)|P^c(u)E(u)|^2 \ll U^{1+\varepsilon}S_J(1+H)^C, \tag*{(11)}
∑u∈UWrow(qu/U)∣P(u)E(u)∣2≪U1+εSJ(1+H)C.\sum_{u\in\mathcal{U}} W_{\mathrm{row}}(q_u/U)|P(u)E(u)|^2 \ll U^{1+\varepsilon}S_J(1+H)^C.

Here U\mathcal{U} is the source sixth-power-free nonunit family with primitive inducing character outside Θ\Theta. The estimates retain the independent profiles, finite twists, natural zeros and common punctures. They are uniform on fixed compact parameter families with positive margins in (4.1).

Valuation allocation and completed lengths

Expand the complete signed coefficient product first. The valuation classes in the full Gauss correlation formulas [2 the Gauss-norm construction] partition the added width as

w=wf+f0+f1+u,u=ggood+v+c,t=f0+u,L1=f1+4T1+e1,L=v+4T+e.w = w_f+f_0+f_1+u, \qquad u=g_{\mathrm{good}}+v+c, \qquad t=f_0+u, \qquad L_1=f_1+4T_1+e_1, \qquad L=v+4T+e.

where all variables are nonnegative. Write At=2m+z+wA_t=2m+z+w. The completed lengths and coefficient exponent are

A∗=At−f1−t−L1−L−5c,(12)A^* = A_t-f_1-t-L_1-L-5c, \tag*{(12)}
M∗=1−6t−f1−L1−L−5c+T1+5v+T,M^* = 1-6t-f_1-L_1-L-5c+T_1+5v+T,
q∗=f1+v+c,m∗=M∗−q∗,q^* = f_1+v+c, \qquad m^* = M^*-q^*,
Rloc=−f0−L1+T1−S1−u−L+v−4c+T−S.R_{\mathrm{loc}}=-f_0-L_1+T_1-S_1-u-L+v-4c+T-S.

For the original common-support ideals C,DC,D, put μcom=min⁡(log⁡UqC,log⁡UqD)\mu_{\mathrm{com}}=\min(\log_U q_C,\log_U q_D), so that wf≤μcomw_f\le\mu_{\mathrm{com}}. Let zCz_C be the extracted part of the initial product and ηK=1−2m−Kz≥0\eta_K=1-2m-Kz\ge0. The ordinary positive-prime defect is exactly

Dfree=A∗+(K−1)(z−zC)−M∗=6w−ηK−(K−1)zC−5wf−5f1−T1−5v−T.(13)D_{\mathrm{free}}=A^*+(K-1)(z-z_C)-M^*=6w-\eta_K-(K-1)z_C-5w_f-5f_1-T_1-5v-T. \tag*{(13)}

No length inequality has been used in this identity. Ordinary removal gives L1+L+5c≤2m+o(1)L_1+L+5c\le2m+o(1), whence

M∗≥1−2ro−6W−o(1),m∗≥1−2ro−7W−o(1)>0.M^*\ge1-2r_o-6W-o(1),\qquad m^*\ge1-2r_o-7W-o(1)>0.

Support edges are chosen inside this fixed margin.

Keep the actual outer ideal DD and its Fourier measure. The numerical benchmark d^=min⁡(log⁡UqD,A∗)\widehat d=\min(\log_U q_D,A^*) changes its complementary exponent by the opposite amount; it does not introduce a second family of DD's. With the source complementary divisor exponents R,E0R,E_0, split the actual formal ordinary length

mf=m∗−log⁡UqC+d^−R−E0m_f=m^*-\log_U q_C+\widehat d-R-E_0

at 1/1001/100. The large-formal branch has an actual ordinary return. The other branch will be bounded directly.

Removing whole factors and stopping at exhaustion

For the large-formal branch the two-transform return [1, 2] has

M′=M+J−g−g2+t2≤M−σ,Δ′=b2−p2+a+BC+ℓa≥a+ℓa.M'=M+J-g-g_2+t_2\le M-\sigma,\qquad\Delta'=b_2-p_2+a+B_C+\ell_a\ge a+\ell_a.

The main choice is g=J++2σg=J_++2\sigma, ℓa=σ/3\ell_a=\sigma/3, a=ao=0a=a_o=0; fixed amplifier errors use g=J++σg=J_++\sigma, ℓa=0\ell_a=0. For a prime-free input no auxiliary-prime average is introduced. First expand the complete last-divisor identity and all exact sixth-power punctures. Only then split the final inducing character according to membership in Θ\Theta.

Lemma 4.2. For an actual ordinary child outside Θ\Theta, let DchD_{\mathrm{ch}} be its positive-prime affine defect, including the assigned support allowance. If the largest prime-slot width is ω\omega, removal of whole slots until admissibility or exhaustion has squared cost at most

U(Dch)+/6+κω+ε.(14)U^{(D_{\mathrm{ch}})_+/6+\kappa\omega+\varepsilon}. \tag*{(14)}

The original central extraction coefficient is preserved.

Proof. Let zremz_{\mathrm{rem}} be the actual remaining width. Each removed slot of width bib_i lowers the positive-prime defect by KbiK b_i. Before the last necessary removal their total is less than (Dch)+/K(D_{\mathrm{ch}})_+/K; the last slot costs at most ω\omega. If all slots disappear earlier, their total is zremz_{\mathrm{rem}}. Thus

b≤min⁡{zrem,(Dch)+/K+ω},κb≤(Dch)+/6+κω.b\le\min\{z_{\mathrm{rem}},(D_{\mathrm{ch}})_+/K+\omega\},\qquad\kappa b\le(D_{\mathrm{ch}})_+/6+\kappa\omega.

If a prime remains, the stopping condition is the ordinary moment’s length condition. At exhaustion use the prime-free ordinary theorem of [1], which permits arbitrary bounded ordinary lengths. It is not necessary that Dch≤0D_{\mathrm{ch}}\le0 in that case. The multiplier already extracted in the central ordinary representation remains outside this estimate; removing a prime does not renormalize either plain factor. The prime-free theorem is a preceding input, so this call is not an induction on the estimate being proved.

The actual common-support maximum in the return is

−d−(K0−kf)−ao+J+=max⁡{−d−(K0−kf)−ao,−cC−2a}≤−μcom+δ1.-d - (K_{0} - k_{f}) - a_{o} + J_{+} = \max\left\{-d - (K_{0} - k_{f}) - a_{o}, -c_{C} - 2a\right\} \leq-\mu_{\mathrm{com}} + \delta_{1}.

Support fixing and last-divisor fixing only decrease the defect. Combining this fact with Δ′≥a+ℓa\Delta^{\prime} \geq a + \ell_{a} pays the amplifier’s own portion once and leaves

(Dfree−μcom)+/6+εloc.(15)(D_{\mathrm{free}} - \mu_{\mathrm{com}})_{+}/6 + \varepsilon_{\mathrm{loc}}. \tag*{(15)}

Both Cauchy integrals receive the same paired divisor weight and are bounded separately as in Lemma 3.1. No bare divisor count is inserted afterwards.

Normalization and the remaining branches

Set D0=Dfree−μcomD_{0} = D_{\mathrm{free}} - \mu_{\mathrm{com}}. Direct substitution in (12) gives

Rloc=−f0−f1−u−3(T1+T)−4c−S1−S−e1−e≤0,R_{\mathrm{loc}} = -f_{0} - f_{1} - u - 3(T_{1} + T) - 4c - S_{1} - S - e_{1} - e \leq0,
Rloc+D0/6=−5f16−19(T1+T)6−5v6−4c−S1−S−e1−e−ηK6−(K−1)zC6+wf−μcom6≤0.\begin{aligned} R_{\mathrm{loc}} + D_{0}/6 &= -\frac{5f_{1}}{6} - \frac{19(T_{1} + T)}{6} - \frac{5v}{6} - 4c - S_{1} - S - e_{1} - e \\ &\quad- \frac{\eta K}{6} - \frac{(K - 1)z_{C}}{6} + \frac{w_{f} - \mu_{\mathrm{com}}}{6} \leq0. \end{aligned}

Consequently Rloc+(D0)+/6≤0R_{\mathrm{loc}} + (D_{0})_{+}/6 \leq0, which pays (15) for either sign of D0D_{0}. This includes the cases z=0z = 0, w=0w = 0, and complete common support.

For mf≤1/100m_{f} \leq1/100, use the original Gauss polynomial and ∣G(n,h)∣≤qn\lvert G(n,h)\rvert\leq\sqrt{q_{n}}, retaining unit frequencies. The direct calculation in [2 the small-formal-length estimate], after the original coefficient energy and benchmark have been subtracted, gives

3At−72+18w+7200+o(1).3A_{t} - \frac{7}{2} + 18w + \frac{7}{200} + o(1).

Put d0=1−2ro∈[7/50,9/50]d_{0} = 1 - 2r_{o} \in[7/50, 9/50]. Since At≤1−cKd0+WA_{t} \leq1 - c_{K}d_{0} + W and W<d0/7W < d_{0}/7, this is less than

−12+3d0K+7200≤−93200+90749<−13.-\frac{1}{2} + \frac{3d_{0}}{K} + \frac{7}{200} \leq-\frac{93}{200} + \frac{90}{749} < -\frac{1}{3}.

This argument includes negative formal centers and does not apply an ordinary theorem to an unformed Gauss polynomial.

At the first Poisson origin, the character mean vanishes unless the product modulus is powerful. Its divisor-bounded allocations, central square normalization and row mean give O(U1+ε)O(U^{1+\varepsilon}), as in [1]. The second origin has ceiling At−1+4t+o(1)A_{t} - 1 + 4t + o(1). The sparse exceptional-character branch has ceiling At−1+3t+o(1)A_{t} - 1 + 3t + o(1) after the two matching fixed-character leading means cancel at the exact common scale product. Their respective bounds are

−cKd0+5W<−(cK−5/7)d0<0,−cKd0+4W<−(cK−4/7)d0<0.-c_{K}d_{0} + 5W < -(c_{K} - 5/7)d_{0} < 0,\qquad-c_{K}d_{0} + 4W < -(c_{K} - 4/7)d_{0} < 0.

Here K>7/2K > 7/2. The exceptional means are handled in this branch, never by Lemma 4.2. Subunit rectangles retain the original volume alternative.

Integrate in the original weighted Fourier measures and use (10) for tails. This proves the centered bound. To restore PP, reflect the longer ordinary factor of PcmpP^{\mathrm{cmp}}, retaining its conductor and Euler deletion. Its ordinary total length is at most 3/2−2m3/2 - 2m, and

3/2−2m+K(z+w)≤5/2−4m+Kw<1.3/2 - 2m + K(z + w) \leq5/2 - 4m + Kw < 1.

The ordinary input therefore bounds PcmpEP^{\mathrm{cmp}}E on U\mathcal{U}. The squared triangle inequality proves the second estimate of (11), including its cross term. This completes the proof of Theorem 4.1.

Prime selection and the first mean

On the main amplitude rectangle

R=[7/20,3/5]×[2/5,1/2],g=δx,\mathcal{R}=[7/20,3/5]\times[2/5,1/2],\qquad g=\delta x,

the detector count has exponent Rk∗(δ,x)R_{k_*}(\delta,x), and the screened prime inventory has length at least L0=21/100L_0=21/100 in UU units. The whole-slot selection argument of [2 the amplitude-selection lemma] gives residual amplitude

a(z)=δxL0−z/2L0−z,a′(z)=δL0(x−1/2)(L0−z)2≤0.(16)a(z)=\delta\frac{xL_0-z/2}{L_0-z},\qquad a'(z)=\frac{\delta L_0(x-1/2)}{(L_0-z)^2}\leq0. \tag*{(16)}

Choose, at first without strict shortening,

zK(m)=1−2mK,WK(m)=cK(1−2m)6,Wnew(m)=WK(m)+11000.z_K(m)=\frac{1-2m}{K},\qquad W_K(m)=\frac{c_K(1-2m)}{6},\qquad W_{\mathrm{new}}(m)=W_K(m)+\frac{1}{1000}.

For 41/100≤m≤43/10041/100\leq m\leq43/100,

1−2m−7Wnew≥405236741000>3500,1−(5/2−4m+KWnew)≥15343500000>3100.(17)\begin{aligned} 1-2m-7W_{\mathrm{new}}&\geq\frac{40523}{6741000}>\frac{3}{500},\\ 1-(5/2-4m+KW_{\mathrm{new}})&\geq\frac{15343}{500000}>\frac{3}{100}. \tag*{(17)} \end{aligned}

Cover this interval by finitely many closed intervals [m−,r0][m_-,r_0] of width less than 1/10001/1000 and use W=Wnew(m−)W=W_{\mathrm{new}}(m_-). The first margin remains larger than 1/2501/250. This supplies legal uniform instances of Theorem 4.1, rather than combining the maximum width at the left endpoint with the rightmost ordinary length.

Proposition 5.1. On this band, the full first mean gains a net ν=1/20000\nu=1/20000 in UU units over the source amplified-contour envelope, retaining its original physical numerator and both incoming profiles.

Proof. We have zK<41/1000z_K<41/1000, Wnew<1/40W_{\mathrm{new}}<1/40, and a(zK)>13/100a(z_K)>13/100. For the last inequality, after multiplication by L0−zK>0L_0-z_K>0, the difference from 13/10013/100 is at least

2110000−9zK200>51200000>0.\frac{21}{10000}-\frac{9z_K}{200}>\frac{51}{200000}>0.

The new first-mean prime gain is Gnew=(gzK+a(zK)Wnew)/2G_{\mathrm{new}}=(gz_K+a(z_K)W_{\mathrm{new}})/2. Relative to width WKW_K it improves by

Gnew−GK=a(zK)2000>13200000.G_{\mathrm{new}}-G_K=\frac{a(z_K)}{2000}>\frac{13}{200000}.

The previous widths are z0=2(1−2m)/9z_0=2(1-2m)/9, W0=7(1−2m)/54W_0=7(1-2m)/54. Direct substitution in (16) shows GK≥G0G_K\geq G_0: writing v=1−2mv=1-2m, zK=svz_K=sv, z0=s0vz_0=s_0v, s=1/Ks=1/K, s0=2/9s_0=2/9, one has

2(GK−G0)=g(s−s0)v+δv6{(1−s)(xL0−sv/2)L0−sv−(1−s0)(xL0−s0v/2)L0−s0v}≥0.2(G_K-G_0)=g(s-s_0)v+\frac{\delta v}{6}\left\{\frac{(1-s)(xL_0-sv/2)}{L_0-sv}-\frac{(1-s_0)(xL_0-s_0v/2)}{L_0-s_0v}\right\}\geq0.

Indeed, writing the total saving as F(s)=GKF(s)=G_K with z=svz=sv, its derivative is

F′(s)=δv2{x−112+(1/2−x)L0(L0−v)6(L0−sv)2}>0,F'(s)=\frac{\delta v}{2}\left\{x-\frac{1}{12}+\frac{(1/2-x)L_0(L_0-v)}{6(L_0-sv)^2}\right\}>0,

since v≤9/50<L0v\leq9/50<L_0 and x≥2/5x\geq2/5. This compares against the actual-mm source selection, which is at least its uniform contact envelope. Thus only the extra 1/10001/1000 is counted as new gain.

Shorten both widths by 1−10−41-10^{-4}. Since Gnew<1/100G_{\mathrm{new}}<1/100, this costs less than 10−610^{-6}. The inventory remains sufficient because 2xL0−zK>127/1000>W2xL_0-z_K>127/1000>W. Choose whole-slot rounding, support edges, count and norm errors with aggregate first-mean cost below 6×10−66\times10^{-6}, and local height cost below 2×10−62\times10^{-6}. The remaining gain exceeds 5×10−55\times10^{-5}. For completeness, if #B≪UεR+εR\#\mathcal{B}\ll U^{\varepsilon_R+\varepsilon_R} and the selected product has amplitude at least Ugz+a(z)w−εAU^{g z+a(z)w-\varepsilon_A}, the complete norm gives

∑u∈B∣Sm(u)∣4≪U1−2gz−2a(z)w+εN+2εA.\sum_{u\in\mathcal{B}}|S_m(u)|^4\ll U^{1-2gz-2a(z)w+\varepsilon_N+2\varepsilon_A}.

Hölder gives first-mean loss (3εR+εN+2εA)/4(3\varepsilon_R+\varepsilon_N+2\varepsilon_A)/4. The physical factor U−θmU^{-\theta m} stays outside, unchanged. Restriction to B\mathcal{B} occurs after the positive estimate, so the signed coefficient expansion and all cross terms are retained.

The complete scalar comparison

For k∈[2/3,3/4]k\in[2/3,3/4], define

c=13k,A=56−δ,Dc=3−(1+2c)x,Pc=2−2(1+c)x+2cx2,c=\frac{1}{3k},\qquad A=\frac{5}{6}-\delta,\qquad D_c=3-(1+2c)x,\qquad P_c=2-2(1+c)x+2cx^2,
Jc=ADc+δPc,Rk=1−δ+AδPc2Jc.J_c=AD_c+\delta P_c,\qquad R_k=1-\delta+\frac{A\delta P_c}{2J_c}.

These are the class-count functions of [2]. Set b=b∗b=b_* and

h=1+b+3ℓ2,ly=1+b−ℓ2,lx=1−b−ℓ2,h=\frac{1+b+3\ell}{2},\qquad l_y=\frac{1+b-\ell}{2},\qquad l_x=\frac{1-b-\ell}{2},
C0=−14+b6+5ℓ4,L=23ℓ−1−b18,θ0=2δx9,C_0=-\frac{1}{4}+\frac{b}{6}+\frac{5\ell}{4},\qquad L=\frac{23\ell-1-b}{18},\qquad\theta_0=\frac{2\delta x}{9},
E=C0+δ2+δxL−3h4(1−Rk∗),E=C_0+\frac{\delta}{2}+\delta xL-\frac{3h}{4}(1-R_{k_*}),
S=3Rk∗+14+θ02−1750.(18)S=\frac{3R_{k_*}+1}{4}+\frac{\theta_0}{2}-\frac{17}{50}. \tag*{(18)}

The full unamplified exponent at row scale U=ZdU=Z^d is E+(d−h)SE+(d-h)S. In particular EE already includes the functional-equation and outer normalization factors.

Elementary bounds and the algebraic contact

Differentiation gives

∂kRk=δxA2(1−x)23k2Jc2≥0,1−δ≤Rk≤R3/4≤1−2δ3−22δx81.\partial_kR_k=\frac{\delta xA^2(1-x)^2}{3k^2J_c^2}\ge0,\qquad1-\delta\le R_k\le R_{3/4}\le1-\frac{2\delta}{3}-\frac{22\delta x}{81}.

On 1/8≤δ≤5/61/8\le\delta\le5/6, 0≤x≤1/20\le x\le1/2,

∣Eb∣<1/5,5/16≤Eℓ≤5/4.(19)|E_b|<1/5,\qquad5/16\le E_\ell\le5/4. \tag*{(19)}

The derivative formulas are Eb=1/6−δx/18−3(1−R)/8E_b=1/6-\delta x/18-3(1-R)/8 and Eℓ=5/4+(23/18)δx−9(1−R)/8E_\ell=5/4+(23/18)\delta x-9(1-R)/8. The polynomial pp is strictly decreasing on [.45,.46][.45,.46]. With t=r−.45057t=r-.45057, Taylor expansion gives

p=.162414095686326−217502.0989101246t−81009.52478t2+218582t3.p=.162414095686326-217502.0989101246t-81009.52478t^2+218582t^3.

At t=74/108t=74/10^8 and 75/10875/10^8 its signs are positive and negative. Rational substitution in (1.2)–(1.3) then yields

.45057074<r∗<.45057075,.16879<ℓ∗<.16880,.09175<b∗<.09185.(20).45057074<r_*<.45057075,\qquad.16879<\ell_*<.16880,\qquad.09175<b_*<.09185. \tag*{(20)}

For example, writing t=r−.45t=r-.45 gives denominator 7.965−2953.6t+2286t27.965-2953.6t+2286t^2 and numerator 3.375−4058t+3410t23.375-4058t+3410t^2 for ℓ∗\ell_*. The fixed-kk derivatives defining b∗b_* lie respectively in (21592.47,21592.66)(21592.47,21592.66), (−4134.33,−4134.28)(-4134.33,-4134.28) and (5333.96,5334.03)(5333.96,5334.03). These rational enclosures suffice for every bound below.

Put a□=δ(189x−25)/139a_{\square}=\delta(189x-25)/139 and

Q=E−7a□108(2ℓy−h).Q=E-\frac{7a_{\square}}{108}(2\ell y-h).

Lemma 6.1. At ℓ=ℓ∗\ell=\ell_*, Q≤0Q\leq0 on R\mathcal{R}, with equality only at (r∗,1/2)(r_*,1/2). On this rectangle S≥71/300S\geq71/300.

Proof. Let T=PcDc′−Pc′DcT=P_cD'_c-P'_cD_c. Then

T=4+2c−12cx+2c(1+2c)x2,(T/Dc2)′=8(1−c)2Dc3>0.T=4+2c-12cx+2c(1+2c)x^2,\qquad(T/D_c^2)'=\frac{8(1-c)^2}{D_c^3}>0.

Since Pc/Dc≥3/8P_c/D_c\geq3/8, T/Dc2<2/3T/D_c^2<2/3 and δ/A≥21/29\delta/A\geq21/29,

T[Dc+(δ/A)Pc]2≤107648261075<512.\frac{T}{[D_c+(\delta/A)P_c]^2}\leq\frac{107648}{261075}<\frac{5}{12}.

Thus

Qx/δ=L−49556(2ℓy−h)−3h8T[Dc+(δ/A)Pc]2>0,Q_x/\delta=L-\frac{49}{556}(2\ell y-h)-\frac{3h}{8}\frac{T}{[D_c+(\delta/A)P_c]^2}>0,

using L>3/20L>3/20, 0<2ℓy−h<1/80<2\ell y-h<1/8, h<4/5h<4/5. At x=1/2x=1/2 exact expansion gives

Q(δ,1/2)=b∗nb(δ,k∗)+ℓ∗ne(δ,k∗)+n0(δ,k∗)432(54k∗δ−75k∗−6δ+10).Q(\delta,1/2)=\frac{b_*n_b(\delta,k_*)+\ell_*n_e(\delta,k_*)+n_0(\delta,k_*)}{432(54k_*\delta-75k_*-6\delta+10)}.

Write D(r)=2286r2−5011r+1800D(r)=2286r^2-5011r+1800 and K0(r)=600r2−1837r+720K_0(r)=600r^2-1837r+720. For ℓ(r)\ell(r) as in (2), k(r)=K0(r)/(3D(r))k(r)=K_0(r)/(3D(r)) one has

nb(r,k(r))=0,3D(r)2{ℓ(r)ne(r,k(r))+n0(r,k(r))}=−1944r(6r−5)2p(r).n_b(r,k(r))=0,\qquad3D(r)^2\{\ell(r)n_e(r,k(r))+n_0(r,k(r))\}=-1944r(6r-5)^2p(r).

The numerator therefore has a double zero at r∗r_*, by (3), and equals AQ(δ−r∗)2A_Q(\delta-r_*)^2, where

AQ=(−6858k∗+600)b∗+(−702k∗−408)ℓ∗+4806k∗−696>2200.A_Q=(-6858k_*+600)b_*+(-702k_*-408)\ell_*+4806k_*-696>2200.

Its denominator is negative on R\mathcal{R}. This proves the assertion about QQ. Finally S≥33/50−3δ/4+δx/9≥71/300S\geq33/50-3\delta/4+\delta x/9\geq71/300.

Complementary rectangles and the full box

Use the rational reference b0=.0918b_0=.0918, ℓc=.16880\ell_c=.16880 and count k=3/4k=3/4. Then h0=.7991h_0=.7991, C00=−.0237C_{00}=-.0237, L0′=4651/30000L'_0=4651/30000. Direct differentiation gives

Eδδ=−(3h0/4)(5/6)2DcPc2Jc3<0.E_{\delta\delta}=-\frac{(3h_0/4)(5/6)^2D_cP_c^2}{J_c^3}<0.

For δ≤23/100\delta\leq23/100 the elementary count bound gives E≤−1193/2000000E\leq-1193/2000000. For δ≥23/100\delta\geq23/100, EE increases with xx: indeed

T[Dc+(δ/A)Pc]2≤856⋅1812766322<12,\frac{T}{[D_c+(\delta/A)P_c]^2}\leq\frac{856\cdot181^2}{76632^2}<\frac{1}{2},

whereas L0′>3/20L'_0 > 3/20, h0<4/5h_0 < 4/5. At the two relevant xx-boundaries the exact expressions are

E(δ,1/2)=−2659452δ2+2425492δ−526140120000(185−138δ).E(\delta,1/2)=\frac{-2659452\delta^2+2425492\delta-526140}{120000(185-138\delta)}.
E(δ,2/5)=−201477144δ2+167193610δ−35905500600000(2525−1698δ).E(\delta,2/5)=\frac{-201477144\delta^2+167193610\delta-35905500}{600000(2525-1698\delta)}.

Concavity and the derivative signs at δ=7/20,3/5\delta=7/20,3/5 imply

E(δ,1/2)<−9/50000(δ≤7/20),E(δ,1/2)<−1/500(δ≥3/5)E(\delta,1/2)<-9/50000\quad(\delta\le7/20),\qquad E(\delta,1/2)<-1/500\quad(\delta\ge3/5)

in the range under consideration. To bound the other edge, the numerator of E(δ,2/5)+1/1000E(\delta,2/5)+1/1000, with y=δ−2/5y=\delta-2/5, is

−201477144y2+(24965474/5)y−3922976/25≤−156000+5000000∣y∣−200000000y2<0.-201477144y^2+(24965474/5)y-3922976/25\le-156000+5000000|y|-200000000y^2<0.

This proves the reference bound E<−9/50000E<-9/50000 on the three complete rectangles

[1/8,5/6]×[0,2/5],[1/8,7/20]×[2/5,1/2],[3/5,5/6]×[2/5,1/2].[1/8,5/6]\times[0,2/5],\qquad[1/8,7/20]\times[2/5,1/2],\qquad[3/5,5/6]\times[2/5,1/2].

Restoring b∗b_* costs less than 1/1000001/100000 by (20); increasing ℓ\ell up to .16883.16883 costs at most 3/800003/80000. Returning to k∗k_* lowers EE. Hence throughout these rectangles

E<−53400000.(21)E<-\frac{53}{400000}. \tag*{(21)}

We also need a full-box upper bound. Subtract 19/1000019/10000 from E(δ,1/2)E(\delta,1/2) and clear the positive denominator. At y=δ−23/50y=\delta-23/50 its numerator is

−2659452y2+(256504/25)y−537677/625≤−860+11000∣y∣−2600000y2<0.-2659452y^2+(256504/25)y-537677/625\le-860+11000|y|-2600000y^2<0.

The preceding monotonicity and small-δ\delta estimate show Eref<19/10000E_{\mathrm{ref}}<19/10000 on the entire box. Restoring the parameters gives

E<779400000<1500(ℓ∗≤ℓ≤.16883).(22)E<\frac{779}{400000}<\frac{1}{500}\qquad(\ell_*\le\ell\le.16883). \tag*{(22)}

These arguments cover the boundary edges as well as the interiors.

The actual short sums

For m≤3/40m\le3/40, ideal counting in (6) gives the original numerator bound Um/2+εU^{m/2+\varepsilon} on each AFE profile. Relative to N=(3R+1)/4−θ0/2N=(3R+1)/4-\theta_0/2, its first-mean difference is bounded by

Δs=(R−1)/4+3/80+17θ0/40\Delta_s=(R-1)/4+3/80+17\theta_0/40
≤3/80−δ/6+43δx/1620≤−104964800((δ,x)∈R).\le3/80-\delta/6+43\delta x/1620\le-\frac{1049}{64800}\qquad((\delta,x)\in\mathcal{R}).

At d=hd=h, (22) and h>.79h>.79 give

E+hΔs<−699116480000<−1/100.(23)E+h\Delta_s<-\frac{69911}{6480000}<-1/100. \tag*{(23)}

The actual slope for any m≤3/40m\le3/40 is R−17/50+θ0+(1/2−θ0)mR-17/50+\theta_0+(1/2-\theta_0)m, which is positive on R\mathcal{R} and less than one. Thus the endpoint d=hd=h bounds every licensed smaller dd. Small rows retain their separate direct estimate. This bound applies to the complete short numerator, with its original outer factors.

Insertion in the physical comparison

For a fixed amplitude class and a single physical dyad, displacement of the contour and a first-mean moment saving γ\gamma change the exponent by

(θ−θ0){d(1−m)−ly}−dγ,(24)(\theta-\theta_0)\{d(1-m)-l_y\}-d\gamma, \tag*{(24)}

as in [2 the simultaneous comparison]. Both operations apply to the same sum. The comparison does not assume that the detector and numerator have the same twist or Mellin height.

The AFE is applied at the actual conductor. Writing its index as abab, with aa supported in SS and qb≍Dq_b \asymp D, both arms retain the scalar qa−1/2−θD−θq_a^{-1/2-\theta}D^{-\theta}. Their conductor-dependent profiles are handled by the common log-scale Sobolev estimate. On reflection the two AFE arms exchange; the root numbers, conductor powers and Gamma ratios cancel. The entire holomorphic dyad is moved before taking absolute values or conjugating the whole dual arm. Long ordinary factors with m>1/2m>1/2 use the prime-free arbitrary-length theorem, rather than the selected range of Theorem 4.1.

For comparison, first absorb the independent incoming heights into V1,V2V_1,V_2, then construct the equal-product comparison with those same profiles. An additional Fourier mode acts on the whole product on a given centered side. The fixed-character leading means cancel at the exact common product of scales. Reflection may increase the long factor after a divisor is extracted; its central extraction coefficient must therefore be retained. For the actual top comparison the displayed modulus is its natural radical, and the capacity is the last inequality of (4.1).

Proposition 7.1. With b=b∗b=b_* and ℓ=ℓ∗+1/40000\ell=\ell_*+1/40000, the complete nonprincipal comparison has a scalar reserve greater than 7/1067/10^6, after a single aggregate global expense smaller than 10−710^{-7}.

Proof. From the count bound and (19),

Qℓ=5/4+(23/18)δx−9(1−R)/8+35a□/216≤5/4−79δ/432<593/500.Q_\ell=5/4+(23/18)\delta x-9(1-R)/8+35a_{\square}/216\leq5/4-79\delta/432<593/500.

Lemma 6.1 therefore gives Q(ℓ)<593/20000000Q(\ell)<593/20000000. On [.41,.43][.41,.43], Proposition 5.1 subtracts hνh\nu in ZZ units. As h>.79h>.79, the exponent is less than

59320000000−.79120000=−9.85×10−6.\frac{593}{20000000}-.79\frac{1}{20000}=-9.85\times10^{-6}.

For all other ordinary lengths use the original selected and unamplified alternatives of [2 the left, central and right envelope comparison]. To make their coverage explicit, put λ=ly/h∈(.577,.578)\lambda=l_y/h\in(.577,.578) and m0=1−λ∈(.422,.423)m_0=1-\lambda\in(.422,.423). The selected envelope has descending slope at least 7/2257/225, ascending slope at least 49/43249/432, and its excess over the contact envelope outside [.41,.43][.41,.43] is at least 7/1500007/150000. The central unamplified range [3/40,3/8][3/40,3/8] has the same reserve; the outer range m>9/20m>9/20 has the larger excess 7δ/21607\delta/2160. These are consequences of the same piecewise-affine envelope, before any class restriction. Its outside-band reserve is thus greater than

.797150000−59320000000−10−7=42760000000>7106..79\frac{7}{150000}-\frac{593}{20000000}-10^{-7}=\frac{427}{60000000}>\frac{7}{10^6}.

The complementary classes use (21); their slope SS is positive. The short sums use (23). For d≤h−1/80d\leq h-1/80 on R\mathcal{R}, Lemma 6.1 and (22) give

E+(d−h)S<779400000−7124000=−12131200000.E+(d-h)S<\frac{779}{400000}-\frac{71}{24000}=-\frac{1213}{1200000}.

All lie farther from zero than the outside-band reserve. The low-amplitude, principal elementary and very-small-row branches in [2] are preserved. Their sufficient scalar conditions follow already from the coarse box .0917<b<.0919.0917 < b < .0919, .1687<ℓ<.1689.1687 < \ell< .1689:

lx>ℓ,ly−ℓ>11b/6,h<2ly,ly<1/2,l_x > \ell,\quad l_y-\ell> 11b/6,\quad h < 2l_y,\quad l_y < 1/2,
ℓ(1/2+2ℓ/h)−11h/54<−1/200,C0+(1−h)/16<−1/250,\ell(1/2+2\ell/h)-11h/54 < -1/200,\qquad C_0+(1-h)/16 < -1/250,
C0+(1/2+3ℓ/2)/50<−1/125,13h/75−ly/2+1/50<−1/20.C_0+(1/2+3\ell/2)/50 < -1/125,\qquad13h/75-l_y/2+1/50 < -1/20.

They retain the original local corrections and actual support. The affine slopes in the near-hh extension are bounded by two, so this extension can be charged inside the aggregate error.

Choose the numerical meshes, strict contour buffers and support edges first. Fix the finite prime list, tests and final finite character group, then every internal Fourier, Sobolev and tail order. If CL,CGC_L,C_G are the local and still-unpaid global height orders, choose the common external height power below

min⁡{(1/100) 2×10−61+CL,2×10−81+CG}.\min\left\{\frac{(1/100)\,2\times10^{-6}}{1+C_L},\frac{2\times10^{-8}}{1+C_G}\right\}.

The local term is already paid in ν\nu. All remaining global nonheight errors are chosen together below 8×10−88\times10^{-8}. The total global expense is therefore below 10−710^{-7}, as used above.

Principal recovery and Mellin continuation

The preceding scalar reserve concerns the nonprincipal amplitude comparison. The full difference Jη−fηJ_\eta-f_\eta also contains the recovery error of the principal signal. We now choose a positive saving common to all its components.

After the numerical mesh has been fixed, choose a finite even prime list with positive lengths ℓi\ell_i summing to ℓ\ell and satisfying that mesh. Define

rnp=7×10−68,rP=min⁡{ly/20,h/600,(437/1000)min⁡iℓi},s=min⁡(rnp,rP)32>0.(25)r_{\rm np}=\frac{7\times10^{-6}}{8},\qquad r_{\rm P}=\min\{l_y/20,h/600,(437/1000)\min_i\ell_i\},\qquad s=\frac{\min(r_{\rm np},r_{\rm P})}{32}>0. \tag*{(25)}

The principal contour in [1] gives two bulk savings and a selected-prime recovery saving, respectively

ly/20−(20+h)e,h/600−20e,ωP−20e,l_y/20-(20+h)e,\qquad h/600-20e,\qquad\omega_{\rm P}-20e,

with ωP≥(437/1000)min⁡iℓi\omega_{\rm P}\ge(437/1000)\min_i\ell_i. Take ee small after the positive ℓi\ell_i have been chosen. All three savings are then at least rP/2r_{\rm P}/2. In the double residue the factors Z−ℓ/6Z^{-\ell/6} and Zℓ/6Z^{\ell/6} cancel before division by the original normalizer cSATc_{SAT}. Thus the residue gives exactly the same fηf_\eta as in (7).

Choose the normalizing loss and remaining late logarithmic losses below \min(r_{\rm np},r_{\rm P})/64. This is done by increasing the threshold for the already fixed data. It requires no new prime mesh. After all internal height orders have been fixed, shrink the external height exponent once more so that its complete high-side cost is below \min(r_{\rm np},r_{\rm P})/64. Finally choose the independent external tail order large enough for the rapidly decreasing Mellin kernels. These operations give

∣Jη(Z)−fη(Z)∣≪ηZβ−c0−s,β=β∗(F).(26)|J_\eta(Z)-f_\eta(Z)|\ll_\eta Z^{\beta-c_0-s},\qquad\beta=\beta_*(F). \tag*{(26)}

The finite prime list and ss are common to the target family. Its internal orders, constants and thresholds may depend on η\eta. In particular, (25) makes no lower bound of 7×10−67\times10^{-6} for the complete analytic saving. Mesh selection precedes the principal minimum, so the order of choices is noncircular.

Proof of Theorem 1.1 over FF. Suppose β>B\beta>B. Choose 0<εlow<(β−B)/40<\varepsilon_{\rm low}<(\beta-B)/4. The direct input estimate and (26) imply

∣Jη(Z)∣≪ηZB−c0+εlow,∣fη(Z)∣≪ηZσ−c0,σ=max⁡{B+εlow,β−s}<β.|J_\eta(Z)|\ll_\eta Z^{B-c_0+\varepsilon_{\rm low}},\qquad|f_\eta(Z)|\ll_\eta Z^{\sigma-c_0},\qquad\sigma=\max\{B+\varepsilon_{\rm low},\beta-s\}<\beta.

The common Mellin identity (7) therefore continues Hη/LFSH_{\eta}/L_{F}^{S} holomorphically to Re⁡s>σ\operatorname{Re} s>\sigma. Since B>437/500B>437/500 and HηH_{\eta} is nonzero there, each LFS(s,η)L_{F}^{S}(s,\eta) is zero-free in that common half-plane. The deleted finite Euler factors have no zeros or poles in it. This contradicts the definition of β\beta as the supremum over the entire finite-order family. Hence β≤B\beta\le B.

Only positivity of ss is needed: the continuation improves the supposed supremum β\beta, and does not require s≥1/160000s\ge1/160000. The larger length ℓ\ell determines the direct boundary BB, while (25) supplies the strict continuation gap.

Descent to arbitrary number fields

We finish the field argument, using full local Euler factors. Let KK be any number field, η\eta a finite-order Hecke character, and M/KM/K its cyclic class field. Put F=K(ζ12)F=K(\zeta_{12}) and E=MFE=MF. The extensions E/FE/F and E/ME/M are abelian. If the inputs of § 2 hold over FF, the fixed-field theorem implies

ζE(s)=∏ψ∈Gal⁡(E/F)LF(s,ψ)\zeta_{E}(s)=\prod_{\psi\in\operatorname{Gal}(E/F)}L_{F}(s,\psi)

is nonzero in B<Re⁡s<1B<\operatorname{Re}s<1. On the other hand the abelian factorization over MM is

ζE(s)=ζM(s)∏ξ∈Gal⁡(E/M)ξ≠1LM(s,ξ).\zeta_{E}(s)=\zeta_{M}(s)\prod_{\substack{\xi\in\operatorname{Gal}(E/M)\\ \xi\ne1}}L_{M}(s,\xi).

The nontrivial factors are holomorphic there, by the standard Hecke theorem and class-field correspondence [5]. Thus a zero of ζM\zeta_{M} would force a zero of ζE\zeta_{E}. Finally

ζM(s)=∏χ∈Gal⁡(M/K)LK(s,χ)\zeta_{M}(s)=\prod_{\chi\in\operatorname{Gal}(M/K)}L_{K}(s,\chi)

shows that a zero of LK(s,η)L_{K}(s,\eta) in that strip would force a zero of ζM\zeta_{M}. This is impossible. The line Re⁡s=1\operatorname{Re}s=1 and the region to its right are covered by the standard nonvanishing theorem and Euler product. The principal pole at s=1s=1 creates no zero. This proves the arbitrary-field assertion of Theorem 1.1.

The descent uses the fixed-field result over every K(ζ12)K(\zeta_{12}); a theorem restricted to the Eisenstein field would not suffice. Its constants need not be uniform in KK. The numerical boundary is the same because the scalar geometry and the finite-length gain are field independent.

Acknowledgement. AI tools assisted in organizing the local derivations, checking algebraic comparisons and preparing the manuscript. The analytic input results are attributed at their points of use.

References

  1. [1]B. Liu, Zero-free half-planes for Hecke L-functions over number fields and Quasi-Riemann Hypothesis, arXiv:2610.26835v2, 29 October 2026. https://arxiv.org/abs/2610.26835v2. Citations refer to this version, whose source file is number_fields_v4.tex.
  2. [2]B. Liu, An improved zero-free half-plane for Hecke L-functions over number fields and Quasi-Riemann Hypothesis, Hexagon 2610.00655v1, 29 October 2026 (manuscript dated 27 October 2026). https://hexagonmath.org/2610.00655v1.
  3. [3]D. A. Kazhdan and S. J. Patterson, Metaplectic forms, Publ. Math. IHES 59 (1984), 35–142. https://www.numdam.org/article/PMIHES_1984__59__35_0.pdf.DOI
  4. [4]D. A. Kazhdan and S. J. Patterson, Corrections to Metaplectic forms, Publ. Math. IHES 62 (1985), 203. https://doi.org/10.1007/BF02698809.
  5. [5]J. S. Milne, Class Field Theory, version 4.03, 2020. https://www.jmilne.org/math/CourseNotes/CFT.pdf.

Paper details

vlnc:W_rb5ybtva
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

AI tools

View statement
Looking for related papers…
Contents