\section{Denominators after differentiation}\label{sec:arithmetic} \subsection{A progression estimate} Put $d_n=\lcm(1,\ldots,n)$ and $\delta_n=4d_n$. We first control the Taylor coefficients of the numerator blocks. Taking a coefficient of degree $j$ removes at most $j$ divisible constants at each prime-power level; $d_n^j$ pays for that loss. \begin{lemma}\label{lem:progression} For $\theta=v/d\in\Theta$ in lowest terms, define \[ B_\theta(t)=\frac{A_d^n}{(n!)^L} \prod_{j=0}^{Ln-1}(t-rn+j+v/d). \] For every integer $v_0$ and every $j\ge0$, \begin{equation}\label{eq:block-integrality} d_n^j[t^j]B_\theta(t+v_0)\in\Z. \end{equation} \end{lemma} \begin{proof} Clear $d$ from each factor using the factor $d^{Ln}$ in $A_d^n$. Fix a prime $q\nmid d$. At each level $q^e\le n$, split the $Ln$ progression terms into $L$ blocks of length $n$. Each block has at least $\lfloor n/q^e\rfloor$ divisible constants. A term in the coefficient of $t^j$ replaces $j$ constants by variable terms and loses at most $j$ contributions at that level. Its valuation is therefore at least \[ L v_q(n!)-j v_q(d_n). \] A term retaining a zero constant is zero and needs no estimate. If $q\mid d$, the cleared constants are units and the extra factor in $A_d^n$ has valuation \[ n\left\lceil\frac{L}{q-1}\right\rceil \ge\frac{Ln}{q-1}\ge L v_q(n!). \] These bounds prove \eqref{eq:block-integrality}. Products and integer powers preserve the same coefficient rule by convolution. \end{proof} \subsection{Partial-fraction coefficients} \begin{proposition}\label{prop:denominators} For every positive odd $n$, \begin{align} \delta_n^{S-i}a_{i,h}(n)&\in\Z,\label{eq:a-integral}\\ \delta_n^{S+k-\nu}A_{\nu,h}(n)&\in\Z, &\delta_n^{S+k-\nu}\rho_\nu(n)&\in\Z, &\delta_n^{S+k}q_b(n)&\in\Z.\label{eq:integral} \end{align} \end{proposition} \begin{proof} Write $B_n(t)=K^{mn}P_n(t)^m/(n!)^{mDL}$ and suppress the index $n$ on the coefficients. Expanding at $-4h$ gives \begin{align} a_{i,h}={}&(n+1)^S(-1)^{Sh}\binom nh^S\notag\\[-2pt] &\quad\times[t^{S-i}] \left\{W_n(t-4h)B_n(t-4h) \prod_{\ell\ne h}\left(1+\frac{t}{4(\ell-h)}\right)^{-S} \right\}.\label{eq:local-coefficient} \end{align} The constants from the other poles have product $4^{Sn}(-1)^{Sh}(h!(n-h)!)^S$. Their cancellation with $C_n$ leaves the displayed integer prefactor. Each $4(\ell-h)$ divides $\delta_n$, and $W_n$ has integer coefficients. Lemma~\ref{lem:progression} and the binomial series prove \eqref{eq:a-integral}. Formula~\eqref{eq:derivative-coefficients} then proves the first two bounds in \eqref{eq:integral}. The finite sums in \eqref{eq:q} can introduce prime powers not covered by $\delta_n$, so the coefficient bounds alone do not prove integrality term by term. We use the zeros that survive differentiation to cancel these denominators after grouping, as in~\cite[Section~4]{Tcount}. Fix an odd $j\le4bh$ and put $t_0=j/b-4h$. This is a grid point in $(-4n,0)$ and is not a pole, so $V_n(t_0)=0$. With $x_\ell=j/b+4(\ell-h)$, this says \begin{equation}\label{eq:grouping} \sum_\nu\frac{A_{\nu,h}}{x_h^\nu} =-\sum_{\ell\ne h}\sum_\nu\frac{A_{\nu,\ell}}{x_\ell^\nu}. \end{equation} At a prime $q$ put $v=v_q(\delta_n)$. At most one $x_\ell$ has valuation greater than $v$: otherwise the difference of two such terms, a nonzero $4(\ell-\ell')$ dividing $\delta_n$, would too. The coefficient bound gives \begin{equation}\label{eq:group-bound} v_q\!\left(\frac{\delta_n^{S+k}A_{\nu,\ell}}{x_\ell^\nu}\right) \ge\nu\bigl(v-v_q(x_\ell)\bigr). \end{equation} If $v_q(x_h)\le v$, the left group in \eqref{eq:grouping} is integral after scaling. Otherwise every group on the right is integral after scaling, so the same conclusion follows from \eqref{eq:grouping}. This holds at every prime. Summing the integral groups with the signs in \eqref{eq:q} proves the last bound in \eqref{eq:integral}. \end{proof} The exponent in the common multiplier is $S+k$. Multiplicity $m$ in the numerator preserves the zeros needed for the grouped argument, while differentiation shifts a coefficient of order $i$ to order $i+k$. Its denominator bound remains $\delta_n^{S-i}$, which is $\delta_n^{S+k-(i+k)}$. This explains the additional arithmetic cost of differentiation.