\documentclass[11pt,letterpaper]{article} \usepackage[T1]{fontenc} \usepackage{lmodern} \usepackage[left=.93in,right=.93in,top=.79in,bottom=.80in]{geometry} \usepackage{amsmath,amssymb,amsthm,mathtools,microtype} \usepackage{xcolor} \usepackage[colorlinks=true,linkcolor=blue!40!black,citecolor=blue!40!black, urlcolor=blue!45!black,bookmarksnumbered=true]{hyperref} \hypersetup{pdftitle={Irrational values of Dirichlet series with growing period}, pdfauthor={twinheterodox},pdfsubject={Irrational values of periodic Dirichlet series}} \newtheorem{theorem}{Theorem}[section] \newtheorem{lemma}[theorem]{Lemma} \newtheorem{proposition}[theorem]{Proposition} \newtheorem{corollary}[theorem]{Corollary} \newcommand{\Q}{\mathbb Q} \newcommand{\Z}{\mathbb Z} \newcommand{\N}{\mathbb N} \newcommand{\grids}{\mathcal Q} \newcommand{\moment}{\mathcal L_n} \newcommand{\lcm}{\operatorname{lcm}} \newcommand{\dd}{\,\mathrm d} \numberwithin{equation}{section} \allowdisplaybreaks[1] \title{\fontsize{18}{22}\selectfont Irrational values of Dirichlet series with growing period} \author{twinheterodox} \date{18 November 2026} \begin{document} \maketitle \begin{abstract} Let \(f\) be a nonzero rational-valued function of period \(D\). We count the exponents \(2\le j\le s\) of prescribed parity for which \(L_f(j)=\sum_{n\ge1}f(n)n^{-j}\) is irrational. For fixed \(0<\eta<1\) our lower bound of order \(\sqrt{s\log\log s/(D\log(sD))}\) has explicit constants and holds simultaneously for all \(D\le s^{1-\eta}\) and all such \(f\). For the real and imaginary parts of the quartic character modulo five we improve the previous lower-bound coefficient at even exponents by a factor of \(2/\sqrt3\). \end{abstract} \setlength{\abovedisplayskip}{9.5pt plus 2.5pt minus 5pt} \setlength{\belowdisplayskip}{9.5pt plus 2.5pt minus 5pt} \section{The count when the period can change}\label{sec:intro} Fix a positive integer \(D\). Assign rational numbers to its residue classes, at least one nonzero, and extend them periodically to a function \(f\). Its Dirichlet series is \(L_f(x)=\sum_{n\ge1} f(n)n^{-x}\) for \(x>1\). For \(\delta\in\{0,1\}\) write \[ N_{f,\delta}(s)= \#\{j\in\Z:2\le j\le s,\ j\equiv\delta\pmod 2,\ L_f(j)\notin\Q\}. \] The fixed-period bound \cite[Theorem~1.1]{Periodic} holds for all coefficients once \(s\) is sufficiently large in terms of \(D\). When \(D\) varies with \(s\), we need one starting cutoff for a whole range of periods. Here \(D\) can be any period of \(f\), and all logarithms are natural. Writing \(\varphi\) for Euler's totient function, set \begin{equation}\label{eq:constants} C_D=\prod_{p\nmid D}\left(1+\frac1{p(p-1)}\right),\qquad K_D=\frac{C_D}{\varphi(D)},\qquad R_D=\prod_{p\mid D}\left(1+\frac1p\right). \end{equation} \begin{theorem}\label{thm:main} For every \(0<\eta,\varepsilon<1\) there is \(s_0(\eta,\varepsilon)\) such that, for every integer \(s\ge s_0\), every \(D\le s^{1-\eta}\), every nonzero rational-valued function \(f\) of period \(D\), and \(\delta=0,1\), \begin{equation}\label{eq:main} N_{f,\delta}(s)\ \ge\ (1-\varepsilon)\,2\sqrt{\frac{K_D}{R_D}}\, \sqrt{\frac{s\log\log s}{\log(sD)}} . \end{equation} If \(f\) vanishes on the nonunits modulo \(D\), one may replace \(2\sqrt{K_D/R_D}\) by \(2\sqrt{K_D}\). \end{theorem} There is also a bound independent of the factorization of \(D\). Let \(\zeta(x)=\sum_{n\ge1}n^{-x}\) for \(x>1\) be the Riemann zeta function and set \(\kappa=C_1=\zeta(2)\zeta(3)/\zeta(6)\). Since \((p^2-p+1)(p+1)=p^3+1\) and \(\prod_p(1+p^{-3})=\zeta(3)/\zeta(6)\), \[ D K_D/R_D=\kappa\prod_{p\mid D}(1+p^{-3})^{-1} \ \ge\ \zeta(2). \] Consequently every profile in Theorem~\ref{thm:main} satisfies \begin{equation}\label{eq:universal} N_{f,\delta}(s)\ \ge\ (1-\varepsilon)\,\pi\sqrt{\frac23}\, \sqrt{\frac{s\log\log s}{D\log(sD)}} . \end{equation} If \(\log D=o(\log s)\), the square root in \eqref{eq:main} has the fixed-period scale \(\sqrt{s\log\log s/\log s}\). For \(D\) of order \(s^\theta\) the logarithm costs a factor tending to \(1/\sqrt{1+\theta}\). Even at \(D=\lfloor s^{1-\eta}\rfloor\), \eqref{eq:universal} with \(\varepsilon=\tfrac12\) gives \(\gg_\eta s^{\eta/2}\sqrt{\log\log s/\log s}\) irrational exponents of each parity. The coefficients and their denominators may change with \(s\). \paragraph{Why two counts enter.} We use grids of fractions \(j/q\) with \(D\mid q\) and \(\varphi(q)\le B\). Too few irrational exponents would let a linear combination of the grids eliminate every irrational value. Clearing denominators gives integer determinants, at least one nonzero. To show its absolute value is less than one, we count the different fractions sampled. Take \(D=4\). The \(8\)- and \(12\)-grids both have totient \(4\), so each supplies one elimination weight at \(B=4\). Assign each fraction to its smallest supporting grid. The \(8\)-grid receives four fractions, all of reduced denominator \(8\). The \(12\)-grid receives eight, of reduced denominators \(3,6,12\). We carry the fraction count into the determinant estimate. A coarse bound of order \(B^2\) would lose the range of large periods. For sparse coefficients even the numerator matters, as the quartic character illustrates in Section~\ref{sec:five}. \paragraph{Earlier work.} Fischler's theorem for periodic coefficients \cite[Theorem~2]{Fischler} allows one to count values outside a fixed finite-dimensional rational vector space, with a cutoff depending on the period. For real characters of conductor \(T\ge2\), Kumar and Tosi \cite[Theorem~1]{KT} state a lower bound of order \(\sqrt{s/\log s}\) at exponents of opposite parity.\footnote{The proof of \cite[Theorem~1]{KT} in \href{https://arxiv.org/abs/2610.41458v1}{arXiv:2610.41458v1} has a gap in its nonvanishing argument. In Lemma~18 (pp.~12--13), take \(a=b=1\) and \(\ell=h=y=0\). For every sufficiently large admissible \(n\), the factor at \(j=rn-1\) is \(t\), so \(K_0(0)=0\). The empty product gives \(g_{0,1,1}(0)=1\), so the lemma would assign a finite \(p_n\)-adic valuation to zero. Our proof uses \cite[Proposition~5.4]{FLSZ}.} Mistiaen \cite[Th\'eor\`eme~1.0.5]{Mistiaen} gives a dimension bound in that parity for a character of modulus \(N\ge3\) with \(4\nmid N\). In the determinant approach, Fischler, Lai, Sprang and Zudilin \cite{FLSZ} combined the elimination method of Fischler, Sprang and Zudilin \cite{FSZ}, the inverse-totient grids of Lai and Yu \cite{LY}, and the positive Hankel determinant used by Fauzan \cite{Fauzan}. Their nonvanishing theorem permits signed measures. The reflected-tail construction for even beta values \cite{Beta} and its periodic extension \cite{Periodic} use this additional freedom. We use the arithmetic normalization and the largest-grid witness of \cite{Periodic}. The new steps are the uniform weighted grid count and a size estimate in terms of the actual number of fractions, including a separate cost for each retained numerator. \section{Counting grids and fractions uniformly}\label{sec:count} Writing \(q=Dr\) isolates the effect of an increasing period. The ratio \(\varphi(pD)/\varphi(D)\) is \(p\) if \(p\mid D\) and \(p-1\) if \(p\nmid D\). Thus each prime divisor of \(r\) new to \(D\) contributes one factor \(1-1/p\) to \(\varphi(Dr)/(r\varphi(D))\). Accordingly define \begin{equation}\label{eq:weights} h_D(r)=\prod_{\substack{p\mid r\\p\nmid D}}(1-1/p)^{-1}, \qquad \lambda_D(r)=\frac{r}{h_D(r)} =\frac{\varphi(Dr)}{\varphi(D)},\qquad v_D(r)=\prod_{\substack{p\mid r\\p\mid D}}(1-1/p). \end{equation} The grid cutoff is \(\lambda_D(r)\le x:=B/\varphi(D)\). A reduced denominator \(b\) first appears on the grid \(\lcm(b,D)\). Summing \(\varphi(b)\) over the denominators first appearing at \(Dr\) gives \(D\lambda_D(r)v_D(r)\) fractions, including the endpoint for \(b=1\) (Section~\ref{sub:fractions}). We therefore need a weighted first moment as well as the grid count. \begin{lemma}\label{lem:means} For \(w_D=1\) or \(w_D=v_D\), put \(V_D(x;w)=\sum_{\lambda_D(r)\le x}w_D(r)\), where \(r\ge1\). As \(x\to\infty\), uniformly over all positive integers \(D\), \begin{align} V_D(x;w)&=x(c_D(w)+o(1)),\label{eq:mean}\\ \sum_{\lambda_D(r)\le x}\lambda_D(r)w_D(r) &=\frac{x^2}{2}(c_D(w)+o(1)),\label{eq:firstmoment}\\ c_D(1)&=C_D,\qquad c_D(v)=C_D\prod_{p\mid D}(1-p^{-2}).\notag \end{align} Both constants lie between \(1/\zeta(2)\) and \(\kappa\), so the errors are also uniform in relative terms. \end{lemma} \begin{proof} For \(D=1\), Erd\H{o}s credits the linear asymptotic to joint work with Tur\'an \cite[p.~540]{Erdos} and controls the tail by a second moment of \(r/\varphi(r)\) \cite[pp.~543--544]{Erdos}. Dressler \cite[Theorem~1.1]{Dressler} computes \(\kappa\). For varying \(D\) and both weights, we truncate at a fixed prime cutoff \(y\). Both products then have one common period, regardless of \(D\). We first bound the truncation error. Write \(h(r)=r/\varphi(r)\). Expanding \(h(r)^2\) over its squarefree divisors and using \(\lfloor T/d\rfloor\le T/d\) gives \[ \sum_{r\le T}h(r)^2\le JT,\qquad J=\prod_p\left(1+\frac{(1-1/p)^{-2}-1}{p}\right)<\infty . \] The coefficients are nonnegative and the factors of \(J\) are \(1+O(p^{-2})\). For \(R\ge2\) and \(x\ge1\) this implies \begin{equation}\label{eq:tail} \#\{r>Rx:r\le xh(r)\}\ll x/R. \end{equation} Indeed on \(2^kRx2^kR\). Summing the second-moment bounds \(2Jx/(2^kR)\) proves \eqref{eq:tail}. A subscript \(y\) will mean that only primes \(p\le y\) are kept. Writing \(\mu\) for the M\"obius function, the expansion \(h(r)=\sum_{d\mid r}\mu(d)^2/\varphi(d)\) gives \[ \frac1T\sum_{r\le T}(h_D(r)-h_{D,y}(r))\le e_y,\qquad e_y=\sum_{\substack{d\ge1\\\exists p\mid d,\ p>y}} \frac{\mu(d)^2}{d\varphi(d)}\longrightarrow0 . \] The full series converges to \(\kappa\). Also \(0\le v_{D,y}(r)-v_D(r)\le\sum_{p\mid r,\,p>y}1/p\). Its mean over \(r\le T\) is at most \(u_y=\sum_{p>y}p^{-2}\). These bounds hold for all \(D\). Let \(P_y=\prod_{p\le y}p\). Both truncated products are periodic modulo \(P_y\). In a residue class \(a\) the cutoff is now the ordinary interval \(r\le xh_{D,y}(a)\). Summing over classes gives \[ \sum_{r\le xh_{D,y}(r)} w_{D,y}(r) =x M_{D,y}+O(P_y),\qquad M_{D,y}=\frac1{P_y}\sum_{a\bmod P_y}h_{D,y}(a)w_{D,y}(a). \] By the Chinese remainder theorem the mean factors over primes. For \(p\nmid D\) its factor is \(1+1/[p(p-1)]\). For \(p\mid D\) it is \(1\) at weight \(1\), and \(1-p^{-2}\) at weight \(v_D\). Thus \(\lvert M_{D,y}-c_D(w)\rvert\ll u_y\) uniformly. To restore the full cutoff, discard \(r>Rx\) from both counts by \eqref{eq:tail}. Restoring the weights below \(Rx\) costs at most \(Rxu_y\). Given \(\tau>0\), at most \(Rxe_y/\tau\) integers there have \(h_D-h_{D,y}>\tau\). Every other change of membership lies in \[ xh_{D,y}(r)t\) is at most \(\log b/(t-1)\). Summing over the \(h\) fractions gives \eqref{eq:Aprice}. \section{An integer determinant on the selected fractions}\label{sec:det} Fix a cutoff \(s\equiv\delta\bmod 2\) and put \(e=(-1)^s\). If \(f(j)=ef(-j)\) on every residue class, Section~\ref{sec:choose} uses Fourier series to give the count directly. Here a class with \(f(j)\ne ef(-j)\) will give nonvanishing. Multiplication by a nonzero integer preserves irrationality, so assume \(f\) is integral. Fix \(B\) with \(\grids\ne\varnothing\). Here we allow any finite reflection-closed \(\Theta\subset\Q\cap(0,1)\) containing every proper fraction \(j/q\) with \(f(j)\ne0\) on the grids. This permits the smaller choice of numerators in Section~\ref{sec:five}. If \(N_{f,\delta}(s)<|\grids|\), there are integers \(w_q\), not all zero, satisfying \begin{equation}\label{eq:eliminate} \sum_{q\in\grids}w_q q^j=0 \quad\text{for every irrational \(L_f(j)\) counted by \(N_{f,\delta}(s)\).} \end{equation} These are fewer homogeneous rational equations than unknowns. Let \(A_0\) be a positive integer clearing all rational \(L_f(j)\) in the same range and parity. The quantities \(s,D,f,B,w_q,A_0\) stay fixed as the auxiliary integer \(n\) grows. Take even \(n\) divisible by each \(p-1\) for the primes dividing the denominators of \(\Theta\). Then \(A^n\) is integral. For the rising factorial \((t)_r=t(t+1)\cdots(t+r-1)\) put \[ R_n(t)= A^n(n!)^{s-h} \frac{\prod_{\theta\in\Theta}(t+\theta)_n}{(t)_{n+1}^{s}} . \] Reflection gives \(R_n(-n-t)=eR_n(t)\). For \(\deg U\le(n+1)s-nh-2\) define the convergent linear functional \begin{equation}\label{eq:moment} \moment(U)= A_0\sum_{q\in\grids}w_q\sum_{j=1}^q f(j) \sum_{v\ge0}\left. R_n(t)\bigl(U(t+n/2)+U(-t-n/2)\bigr)\right|_{t=v+j/q}. \end{equation} Let \(d_n=\lcm(1,\ldots,n)\). For each \(i\) put \(E_{n,i}=(n!)^{\lfloor i/n\rfloor}(i\bmod n)!\) and \(U_i(z)=(z)_i/E_{n,i}\). Write \(\Delta_{n,m}=\det_{0\le i,j1\), reindexing each partial fraction gives \[ \sum_{v\ge0}G_U(v+\theta) = E_U(\theta) +\sum_{\substack{2\le\ell\le s\\ \ell\equiv s\bmod 2}} \rho_\ell\,\zeta(\ell,\theta). \] The coefficients \(\rho_\ell\) are the same at every shift. Summing on the \(q\)-grid gives \(\sum_{j=1}^q f(j)\zeta(\ell,j/q)=q^\ell L_f(\ell)\) by periodicity. Equation~\eqref{eq:eliminate} therefore removes the irrational terms, and the factors \(A_0,d_n^s\) clear the rational ones. This proves the entrywise assertion in \eqref{eq:J}. \subsection{A nonzero determinant}\label{sub:nonzero} The weights \(w_q\) in \eqref{eq:eliminate} may have either sign. Let \(j_0\le D\) be the first positive integer with \(f(j_0)\ne ef(-j_0)\), and \(q_*\) the largest denominator with \(w_{q_*}\ne0\). Its grid detects this mismatch at \(\theta_*=j_0/q_*\). Every smaller grid that samples \(\theta_*\) sees a positive integer \(q\theta_*n/2,\\ eA_0W(-x)S(x),&x<-n/2,\\ 0,& |x|\le n/2 . \end{cases} \] Extend \(a_x\) by zero off \(X\). Since \(S(x)=O(|x|^{-T-1})\), the absolute moments through degree \(T-1\) exist. The quotient of successive values of \(S\) gives the recurrence \(P(x)a_{x+1}=Q(x)a_x\) with \begin{align*} P(z)&=(z+1+n/2)^{s+1}(z-n/2)\prod_{\theta\in\Theta}(z+\theta-n/2),\\ Q(z)&=(z-n/2)^{s+1}(z+1+n/2)\prod_{\theta\in\Theta}(z+\theta+n/2). \end{align*} Indeed \(S(z+1)/S(z)=Q(z)/P(z)\). At the fractional boundaries \(x=n/2-1+\theta\) and \(x=-n/2-\theta\), the required zero occurs in \(P\) and \(Q\), respectively, by reflection of \(\Theta\). The extra linear factors handle \(x=n/2\) and \(x=-n/2-1\). Both polynomials have degree \(d\). By \cite[Lemma~5.1]{FLSZ}, the Cauchy transform \(F(z)=\sum_{x\in X}a_x/(z-x)\) satisfies \(P(z)F(z+1)-Q(z)F(z)=C(z)\) for a polynomial \(C\). A rational solution \(r\) of the same equation would make \((F-r)/S\) one-periodic. At the fraction \(\theta_*\) chosen above, every smaller grid contributes zero to the difference. Hence \[ W(\theta_*)-eW(-\theta_*)=w_{q_*}\bigl(f(j_0)-e f(-j_0)\bigr)\ne0. \] The class of \(\theta_*\) is in \(X\), even if only its reflection has nonzero weight or if \(\theta_*=1\). At \(k+\theta_*\) and \(-k+\theta_*\) for large integers \(k\), the residues of \((F-r)/S\) are \(A_0W(\theta_*)\) and \(eA_0W(-\theta_*)\). Here a missing mass contributes residue zero. The two points differ by \(2k\), so these unequal residues contradict one-periodicity. This excludes a rational solution. Proposition~5.4 of \cite{FLSZ} now gives \eqref{eq:window}. \section{Bounding the determinant by the number of shifts}\label{sec:size} The grids supply \(|\grids|\) elimination variables, but their active fractions at a fixed level lie at just \(H\) possible positions on each tail. A polynomial of degree \(2H\) can vanish at all these positions. Repeating this for successive levels forces high-degree moments to start farther out where \(R_n\) is small. We keep this degree cost \(2H\) in the analytic basis of \cite[Section~4]{FLSZ}, as adapted in \cite[Section~5]{Periodic}. \begin{proposition}\label{prop:size} Fix the data of Section~\ref{sec:det}. There is a constant \(C_0>0\) independent of \(n,m\) such that, for sufficiently large admissible \(n\) and \(2m\le(s-H-10)n\), \begin{equation}\label{eq:size} |J_{n,m}|\le (C_0m)^m A^{mn} \exp\!\bigl(m^2+(2s+10H)mn\bigr) \left(\frac{10Hn}{m}\right)^{(s-H)mn-2m^2}. \end{equation} \end{proposition} \begin{proof} Let \(\nu_i=\lfloor i/(2H)\rfloor\) and choose the monic polynomial \[ V_i(z)=z^{i-2H\nu_i} \prod_{\theta\in\Theta\cup\{1\}}\prod_{0\le v<\nu_i} \bigl(z^2-(n/2+v+\theta)^2\bigr). \] On each later level \(t=v+j/q\), every positive root in the product is at most \(t+n/2\). Each quadratic factor at either sign of \(t+n/2\) thus has absolute value at most \((t+n/2)^2\), giving \(\lvert V_i(\pm(t+n/2))\rvert\le(t+n)^i\). Let \(S_i\) be the sum of these squared polynomial values with the absolute masses of the individual summands in \eqref{eq:moment}. Cauchy--Schwarz gives \(\lvert\moment(V_iV_j)\rvert\le(S_iS_j)^{1/2}\). The monic change of basis preserves the moment determinant. Its expansion therefore gives \(\lvert\Delta_{n,m}\rvert\le m!\prod_{i0\) with \(2C1\) as the Mellin transform of \[ t\longmapsto\frac{\sum_{r=1}^D f(r)e^{-rt}}{1-e^{-Dt}}. \] The numerator divided by \(e^{-t}\) is a nonzero polynomial in \(e^{-t}\) of degree at most \(D-1\). Its number of changes of sign bounds the number of distinct real zeros of the transform. Indeed, if \(t_0\) is a sign-change point, multiplication of the integrand by \(\log(t/t_0)\) removes that sign change. With \(F(x)\) denoting the original transform, the new transform is \(t_0^x(t_0^{-x}F(x))'\). Rolle's theorem and induction give the bound, starting with an integrand of constant sign. Differentiation is valid locally for \(x>1\) after any finite number of logarithmic factors. It follows that at most \(D-1\) eligible exponents give zero. For \(D\le s^{1-\eta}\) this leaves at least \(s/3\) for all sufficiently large \(s\). This exceeds the required bound, since \(K_D/\sqrt G\le\sqrt\kappa\). Finally let \(\sigma\) be the largest integer at most \(s\) of parity \(\delta\). Since \(s-1\le\sigma\le s\), we have \(s^{1-\eta}\le\sigma^{1-\eta/2}\) for large \(s\). Apply the parity-aligned argument with \(\eta/2\) at \(\sigma\). The ratio of the square-root factors at \(\sigma\) and \(s\) tends to one uniformly because \(\log(sD)\ge\log s\). Take \(\rho=\varepsilon/4\) and enlarge \(s_0\) to obtain \eqref{eq:main} for either parity. \end{proof} \section{Two sparse profiles modulo five}\label{sec:five} The real and imaginary parts of the quartic character \(\chi\) with \(\chi(2)=i\) are supported on \(\{1,4\}\) and \(\{2,3\}\) modulo five, respectively. Their Dirichlet values at real arguments are the real and imaginary parts of \(L_\chi\), and \(\chi(-1)=i^2=-1\) gives both profiles odd parity. Both series are positive for real \(x>1\), since their successive blocks are \((5k+1)^{-x}-(5k+4)^{-x}\) and \((5k+2)^{-x}-(5k+3)^{-x}\), respectively, for \(k\ge0\). At every odd integer \(j\ge3\), the two values are therefore nonzero algebraic multiples of \(\pi^j\) by Section~\ref{sec:choose}, hence transcendental. For the even-exponent count, consider the first support pair and reduced denominator \(10\). On the \(10\)-grid, \(1/10\) is available but \(3/10\) samples the other pair. The \(20\)-grid supplies \(3/10=6/20\), since \(6\equiv1\bmod5\), at a totient cost of \(8\) instead of \(4\). On an odd denominator the same doubling leaves its totient unchanged. Keeping track of this difference reduces the number of shifts while leaving the number of eliminating weights unchanged. \begin{corollary}\label{cor:five} Let \(f\) be a nonzero rational-valued function of period five whose support is contained in either \(\{1,4\}\) or \(\{2,3\}\) modulo five. For every \(\varepsilon>0\), for all sufficiently large \(s\) independent of \(f\), and for either parity, \[ N_{f,\delta}(s)\ge (1-\varepsilon)\,4\sqrt{K_5/3}\, \sqrt{\frac{s\log\log s}{\log s}}, \qquad K_5=\frac{5\kappa}{21}. \] For \(\operatorname{Re}\chi\) and \(\operatorname{Im}\chi\) this improves the previous fixed-period lower-bound coefficient \(2\sqrt{K_5}\) at even exponents by the factor \(2/\sqrt3\). \end{corollary} \begin{proof} Use the same grids \(5\mid q,\ \varphi(q)\le B\). For \(\Theta\) retain the fractions sampled by the chosen residue pair. This set is reflection-closed even if \(f\) is nonzero at just one member of the pair. A proper reduced \(a/b\) occurs if \(q=kb\) is admissible and \(ka\) belongs to the pair modulo five. This forces \(5\nmid k\) and \(5\mid b\). For a numerator in the chosen pair, \(k=1\) works as soon as \(\varphi(b)\le B\). For the complementary pair, \(k=2\) works at a cost ratio \(\varphi(2b)/\varphi(b)\) of one on odd \(b\) and two on even \(b\). Every \(k\ge3\) has \[ \frac{\varphi(kb)}{\varphi(b)} = k\prod_{\substack{p\mid k\\p\nmid b}}(1-1/p)\ge\varphi(k)\ge2 . \] Thus odd denominators use all numerators up to the full cutoff. For even denominators exactly the complementary half of numerators has the stricter cutoff \(\varphi(b)\le B/2\). It is half because reduction of units modulo \(b\) onto units modulo five is surjective with equal fibers. Write \(U_D(y)=\sum_{D\mid b,\,\varphi(b)\le y}\varphi(b)\). Equation~\eqref{eq:firstmoment} at fixed \(D\) gives \(U_D(y)=(K_D/2+o(1))y^2\). Since the even denominators are precisely the multiples of ten among the multiples of five, the number of retained proper fractions is \[ h=U_5(B)-\tfrac12\bigl(U_{10}(B)-U_{10}(B/2)\bigr) =\frac{B^2}{2}\bigl(K_5-\tfrac38K_{10}+o(1)\bigr). \] By \eqref{eq:constants}, \(K_{10}/K_5=2/3\), so this gives \(G_*=3K_5/4\) in \eqref{eq:H}. Restricting to one residue pair keeps every grid and uses asymptotically three quarters as many shifts as retaining all units. The cost bound \eqref{eq:Aprice} and the integrality argument of Section~\ref{sub:integer} both apply to individual selected numerators. Use \(G_*\) in \eqref{eq:BC}. The proof of \eqref{eq:aligned} now yields \(2K_5/\sqrt{G_*} =4\sqrt{K_5/3}\), and \(\log(5s)/\log s\to1\). At matching parity the same zero bound applies. \end{proof} \paragraph{Verification.} Independent AI agents reviewed the 18 November 2026 research proof and first manuscript draft for correctness and exposition. A final independent audit of the exposition revision verified Theorem~\ref{thm:main} and Corollary~\ref{cor:five}. The accompanying program \texttt{checks/check\_grids.py} in the paper project reconstructs the fractions directly and verifies the finite grid identities in Sections~\ref{sec:count} and~\ref{sec:five} for the documented test cases. {\small \begin{thebibliography}{99} \setlength{\itemsep}{.18\baselineskip} \bibitem{Dressler} R.~E.~Dressler, \href{https://doi.org/10.2140/pjm.1970.34.371}{A density which counts multiplicity}, \emph{Pacific J. Math.} \textbf{34} (1970), 371--378. \bibitem{Erdos} P.~Erd\H{o}s, \href{https://doi.org/10.1090/S0002-9904-1945-08390-6}{Some remarks on Euler's \(\varphi\) function and some related problems}, \emph{Bull. Amer. Math. Soc.} \textbf{51} (1945), 540--544. \bibitem{Fauzan} A.~Fauzan, \href{https://doi.org/10.5281/zenodo.22826419}{\(\zeta(5)\) is irrational}, Zenodo, 17 September 2026. \bibitem{Fischler} S.~Fischler, \href{https://doi.org/10.1112/jlms.12290}{Irrationality of values of \(L\)-functions of Dirichlet characters}, \emph{J. London Math. Soc.} \textbf{101} (2020), 857--876. \bibitem{FLSZ} S.~Fischler, L.~Lai, J.~Sprang and W.~Zudilin, \href{https://arxiv.org/abs/2611.16156}{Many more odd zeta values are irrational}, arXiv:2611.16156 (2026). \bibitem{FSZ} S.~Fischler, J.~Sprang and W.~Zudilin, \href{https://doi.org/10.1112/S0010437X1900722X}{Many odd zeta values are irrational}, \emph{Compos. Math.} \textbf{155} (2019), 938--952. \bibitem{KT} V.~Kumar and R.~Tosi, \href{https://arxiv.org/abs/2610.41458}{Irrationality of values of \(L\)-functions for real Dirichlet characters}, arXiv:2610.41458 (2026). \bibitem{LY} L.~Lai and P.~Yu, \href{https://arxiv.org/abs/1911.08458}{A note on the number of irrational odd zeta values}, \emph{Compos. Math.} \textbf{156} (2020), 1699--1717. \bibitem{Mistiaen} L.~Mistiaen, \href{https://arxiv.org/abs/2512.02612}{Ind\'ependance lin\'eaire de valeurs de fonctions \(L\) de Dirichlet}, arXiv:2512.02612 (2025). \bibitem{DLMF} F.~W.~J.~Olver et al., eds., \href{https://dlmf.nist.gov/24.8}{NIST Digital Library of Mathematical Functions}, \S24.8(i), Fourier series for Bernoulli polynomials. \bibitem{Sprang} J.~Sprang, \href{https://arxiv.org/abs/1802.09410}{Infinitely many odd zeta values are irrational. By elementary means}, arXiv:1802.09410 (2018). \bibitem{Beta} twinheterodox, \href{https://hexagonmath.org/2611.00600v1}{An asymptotic lower bound for the number of irrational even beta values}, Hexagon 2611.00600v1 (2026). \bibitem{Periodic} twinheterodox, \href{https://hexagonmath.org/2611.00615v1}{Irrational values of Dirichlet series with periodic coefficients}, Hexagon 2611.00615v1 (2026). \end{thebibliography}} \end{document}